Middle Grades Report

RESEARCH REPORT NOVEMBER 2014
Looking Forward to High School and College
Middle Grade Indicators of Readiness

in Chicago Public Schools
Elaine M. Allensworth, Julia A. Gwynne, Paul Moore, and Marisa de la Torre
TABLE OF CONTENTS

1 Executive Summary
7 Introduction
Chapter 1
17 Issues in Developing and

Evaluating Indicators
Chapter 2
23 Changes in Academic Performance

from Eighth to Ninth Grade
Chapter 3
29 Middle Grade Indicators of

High School Course Performance
Chapter 4
47 Who Is at Risk of Being Off-Track
at the End of Ninth Grade?

Chapter 5
55 Who Is at Risk of Earning Less

Than As or Bs in High School?
Chapter 6
63 Indicators of Whether Students

Will Meet Test Benchmarks
Chapter 7
75 Who Is at Risk of Not Reaching

the PLAN and ACT Benchmarks?
Chapter 8
81 How Grades, Attendance, and

Test Scores Change
Chapter 9
93 Interpretive Summary
99 References
104 Appendices A-E
ACKNOWLEDGEMENTS
The authors would like to acknowledge the many people who contributed to this work. We thank Robert Balfanz
and Julian Betts for providing us with very thoughtful review and feedback which were used to revise this report.
We also thank Mary Ann Pitcher and Sarah Duncan, at the Network for College Success, and members of our
Steering Committee, especially Karen Lewis, for their valuable feedback. Our colleagues at UChicago CCSR and
UChicago UEI, including Shayne Evans, David Johnson, Thomas Kelley-Kemple, and Jenny Nagaoka, were
instrumental in helping us think about the ways in which this research would be most useful to practitioners and
policy makers. We were fortunate to receive substantial feedback and assistance from the UChicago CCSR
communications staff, Bronwyn McDaniel, Jessica Puller, and Emily Krone. We thank the Chicago Public Schools
for providing us the data that allowed us to do this work. All work at UChicago CCSR is also supported by operating
grants from the Spencer Foundation and the Lewis-Sebring Family Foundation. This study was made possible by
a grant from the Bill and Melinda Gates Foundation, to which we are very grateful.
This report was produced by UChicago CCSRs publications and

communications staff: Emily Krone, Director for Outreach and
Communication; Bronwyn McDaniel, Senior Manager for Outreach
and Communications; and Jessica Puller, Communications Specialist.
11.2014/pdf/[email protected]
Graphic Design: Jeff Hall Design

Photography: David Schalliol
Editing: Ann Lindner
MIDDLE GRADE INDICATORS OF READINESS
Executive Summary
Across the country, policymakers are raising the expectations for
educational attainment. With changes to the economy resulting in dire
economic prospects for high school dropouts, high school graduation
has become a necessity. In fact, high school graduation is no longer
considered sufficient; policymakers are calling on the nations schools
to graduate all students ready for college and careers.
Much of the pressure to improve educational attain-
messages that policymakers and practitioners hear
ment is on high schools, but focus has also turned to
about what matters for academic success, these
earlier grades. There is a very large population of
findings are intended to clarify which factors are
students who struggle with the transition from the
most strongly related to students educational
middle grades to high school, raising concerns that
attainment.
high school failures are partially a function of poor

middle grade preparation. As a result, middle grade
practitioners are grappling with questions about what
skills students need to succeed in high school, which
markers they should use to gauge whether students are
ready to succeed in high school and beyond, and whether it is possible to identify in middle grades students
who are likely to struggle in high school and college.
This report is designed to provide a detailed picture
of the relationship between students performance in
the middle grades (grades five through eight) and their
subsequent performance in high school and college
among students in the Chicago Public Schools (CPS).
Specifically, the report shows:
Which of many potential middle grade indicators
Which students are likely to be successful in high

schoolpassing their classes, earning high grades
that put them on-track for college enrollment and
graduation, and achieving test scores at ACTs
benchmarksand which students are at risk of
failing to meet those standards. These findings are
intended to help schools develop intervention plans
and practices around specific students to help them
reach particular outcomes. They can also provide a
guide for parents about what level of performance
their child needs to be likely to graduate from high
school and be ready for college.
The degree to which student performance changes
during the middle gradesshowing why students
leave the middle grades with different levels of
including attendance rates, grades in specific classes
readiness. For example, do students need to start
and GPAs, test scores, study habits, grit, discipline
the middle grades with high performance to leave
records, and background characteristicsare most
middle school with high levels of readiness? Or does
predictive of high school success. It also shows the
students performance change considerably while
degree to which high school readiness is a function
they are in the middle grades? These findings are
of the high school students choose or the middle
intended to help practitioners and policymakers set
school they attended. With all of the conflicting
realistic goals for improvement.
Executive Summary
KEY FINDINGS:
GPAs from different middle schools are not equivalent, but the differences are generally less than a half
Middle Grade Information Can Be

Used to Create Simple Indicator
Systems of High School Graduation
and College Readiness
of a GPA point. We often hear that grades are subjec-
Many characteristics of students are associated
to succeed if they came from one school than another. In
with their academic outcomes, from background
fact, students with the same grades, attendance, and test
characteristics to test scores, grades, attendance,
score records upon leaving eighth grade are more likely
and discipline records, to noncognitive factors. With
to succeed in high school if they came from some middle
so many factors associated with later outcomes, it may
schools than from others, net of the effects of which high
seem like practitioners would need complicated models
schools students attend (see Chapter 3). At the same
to gauge students likelihood of success in later years.
time, the differences in success among students with the
However, after taking into account just two or three
same grades from different middle schools are small,
key middle school indicators, other information about
compared to the differences in success among students
students only marginally improves the prediction of
with different grades at the same school. As discussed
later outcomes.
in Chapter 3, an A average from one school represents a
tivethat an A average from one school is not equal to

an A average from another school. This suggests that
students with the same academic records are more likely
better likelihood of success than a B average from any
2
Grades and failures are best predicted by earlier
other school. Differences by schools are as much as half
grades and attendance. High school test scores
a GPA point, at the most.
are strongly predicted by earlier test scores.
While it is common for people to believe that grades
Other information about students provides only
have different meanings across schools, it also appears
negligible improvement in the prediction of their
that test scores also have different meanings across
outcomes, beyond the top predictors. Background
schools. Students with the same middle grade test scores
characteristics, study habits, and grit are not predictive
are more likely to score well on the high school tests if
of high school performance, once students middle
they came from particular middle schools than others
grade GPAs, attendance, and test scores are taken into
(see Chapter 6). People tend to see test scores as objec-
account. Background characteristics (e.g., race, gender,
tive, since everyone takes the same test, but students
neighborhood poverty, free lunch eligibility, being
can be prepared to do well on the middle grade tests in
old-for-grade, and special education status) are all
ways that do not necessarily translate into higher per-
related to high school grades and test scores, but they
formance later on.
do not tell us any more about who will pass, get good
grades, or score well on tests in high school, once
Test scores are much weaker indicators of high school
we take into account students eighth-grade GPAs,
grades than middle school grades and attendance.
attendance, and test scores. Students misconduct
Many high school interventions are based on test score
and suspension records in middle school are also not
proficiencymeeting standards on tests, or reading
predictive of high school performance, once we take
at grade level. This is the reasoning behind programs
into account their attendance, grades, and test scores.
that offer support based on test scores, such as double-
Likewise, students reports of their study habits
dose coursing or grade promotion standards in middle
in eighth grade, and their responses on a grit scale
school that delay students entry into ninth grade based
measuring perseverance in the middle grades, are
on test scores. However, while middle grade test scores
not predictive of their performance in high school
are moderately related to passing classes and getting
beyond their current grades and attendance.
high grades in high school, most of the relationship
between test scores and later performance seems to

work through students grades. That is, students with
UCHICAGO CCSR Research Report | Middle Grade Indicators of Readiness in Chicago Public Schools
strong test scores are more likely to get good grades
middle grades; early intervention might help them get
than students with weak test scores, but it is the grades
on a path to high school and college readiness. Many
that matter for later outcomes. Grades are based on a
other students, however, do not show signs of poor
number of factors in addition to tested skills, including
performance until they get to high school. The change
attendance, assignment completion, and quality of work
in context across the transition to high school brings
over the course of an entire semester. Once we account
new challenges to many students who appear to be
for students GPAs and attendance in the middle grades,
doing fine, academically, during the middle grades.
their test scores do not provide much additional information about their likelihood of passing their classes
Middle grade attendance and grades can be used to
in high school, and they only improve the prediction of
identify a set of students who are at very high risk
getting high grades (As and Bs) in high school among
of failing classes and being off-track in high school,
students who also have high grades in middle school.
High test scores also do not inoculate students
and many of these students can be identified by at

least as early as sixth grade. Students with a very high
against poor course performance in high school. Only
risk of failure in high school are chronically absent
about a third of students with high test scores in eighth
in the middle grades or are already receiving Fs in
grade (meeting 2013 ISAT standards in reading and
their classes in the middle grades (see Chapter 4).
math) receive at least a B average in high school, and
The middle grade indicators of very high likelihood
one-fifth have D averages or lower (see Chapter 5).
of failure in Chicago are almost identical to middle

grade indicators that were found to be very predictive
Whether students are ready for high school depends
of high school dropout in Philadelphia. 3 Many of the
not only on their academic performance in the middle
students who are at high risk of ninth-grade failure can
grades but also on the context that they enter into in
be identified by at least as early as sixth grade, although
ninth grade. Students with the same academic records
some fall into this group as their attendance declines
in middle school also have different high school outcomes
through the middle grade years. Without a dramatic
depending on which high school they attend. Students are
change in their educational experience, these students
more likely to pass their ninth-grade classes, and to make
have very little chance of graduating from high school;
larger test score gains, if they attend some high schools
they and their future teachers are set up for failure.
rather than others (see Chapter 3). Especially for students with moderate GPAs (between a 1.0 and 3.0), their
While some students can be identified as at high
probability of being on-track at the end of ninth grade is
risk of failing in high school, many other students
strongly influenced by which high school they attend.
who fail their ninth-grade classes in Chicago do
Middle Grade Indicators Can

Identify Some Students at High
Risk, but High School Warning
Systems Are Still Critical
not show signs of being in academic trouble in the

middle grade years. The majority of students who
fail their ninth-grade courses, and are off-track for

graduation in Chicago, cannot be identified precisely
in middle school (see Chapter 4). While there are
When thinking about how to address problems with
calls for early identification of dropouts in middle
high school dropout and college readiness, there are
school, the change in context over the transition to
often calls for early intervention (before students get to
high school makes it difficult to predict exactly who
high school). Some students can be identified as at high
will fail in ninth grade, beyond the students with
risk of poor high school outcomes when they are in the
very high risk. Once students start to show signs of
1 See Chapter 3, as well as Appendix E, for models that predict

high school grades, pass rates, and on-track status with middle
grade variables. See Chapter 6, as well as Appendix E, for models
Executive Summary
that predict high school test scores with middle grade variables.
2 See Chapter 3 and Appendix E for these comparisons.

3 Neild and Balfanz (2006).
struggle in ninth grade, with absence from classes or
siderably in the middle grade years, especially among
low grades, they become at high risk of not graduating.
students who started out with low attendance in fifth

grade. GPAs improve or fall for some students by as
Students often leave the middle grades looking like
much as a half of a GPA point as they move from fifth to
they are prepared to do very well in high school,
eighth grade.
but their grades and attendance drop dramatically
All students show growth in test scores over time
in ninth grade. Students without exceptionally high
on the vertically scaled state tests (the ISAT), but
performance in middle school are unlikely to get high
their relative performance (e.g., percentile rank) stays
grades in high school, and even having a very strong
about the same, with very little variation in the amount
record in middle school does not ensure a student of
that they grow over timeespecially in reading (see
high grades in high school (see Chapter 2). In fact, 40
Chapter 8). There is a push for students to end eighth
percent of students who left eighth grade with As or Bs in
grade in the exceeds range on state standards; this
their classes and ISAT test scores of 310 (exceeding stan-
point corresponds closely with the college benchmark
dards) in math earned a C or lower in their ninth-grade
scores on the high school tests. However, students with
math class. Students who enter high school with strong
average scores in fifth grade would need to improve
records but no longer perform well in high school are
their test growth by a rate that is well above the highest
another group of students ripe for interventioninter-
growth currently observed for these students to meet
vention to make sure they reach their college potential.
the exceeds range at the end of eighth grade, except by

random chance.
Students need at least a 3.0 GPA in the middle grades
About half of the changes over time in students
to be college-bound; a 3.5 GPA gives them at least
grades, attendance, and test scores can be attributed
a 50 percent chance. Prior research on high school
to which school they attend during the middle grades.
predictors of college graduation shows that, by far,
At some schools, it is typical for students grades to im-
the most important predictor of college graduation is
prove by about a tenth of a GPA point each year; at other
students high school GPA. Only students who graduate
schools, grades tend to decline by a tenth of a GPA point.
from high school with at least a B average have a moder-
Thus, students GPA might end up as much as 0.6 points
ate chance of earning a college degree. Parallel to this
different by the eighth grade, based on which school they
finding about college, only those students who leave
attend for the middle grade years. At some schools, atten-
eighth grade with GPAs of at least 3.0 have a moderate
dance tends to decrease by about one percentage point
chance of earning a 3.0 GPA in high school. Students
each year; at others attendance tends to improve by a
who plan to go to college need to get the message that
percentage point a year. At some schools, students math
college requires very strong levels of effort and engage-
ISAT scores grow by 15 points a year, and at others they
ment in both the middle grades and in high school.
grow at 11 points a year; reading scores grow by 11 points
Attendance Shows More Variation

in Growth Over the Middle Grades
Than Grades; Test Performance Is
Fairly Constant Over Time
a year at some schools, and nine points a year at others.

Strategies aimed at attendance improvement could
likely have as much or more of a pay-off for high
school and college graduation as efforts aimed at
Some students show declining attendance and grades
improving test scores. While there is considerable
over the middle grades; others show improving atten-
local and national focus on improving test scores as
dance and grades. On average, GPAs and attendance
a mechanism for improving educational attainment,
rates are similar in fifth grade through eighth grade in
attendance is often seen as a low-level goal. Yet, middle
Chicago. Attendance shows the more variability across
school attendance is much more predictive of passing
students over time than either grades or test scores. In
high school classes than test scores and is as predictive
particular, some students attendance declines con-
of high grades in high school as test scores. High school
outcomes are also higher for students who improve
off-track for college readiness. Declining attendance
their attendance during the middle grades than for
occurs among students with high test scores, as well
students who improve their test scores. Students who
as among students with low test scores. It is the most
end their middle grade years with strong attendance
critical factor undermining students educational at-
are much more likely to do well in their high school
tainment. This does not mean that schools should focus
courses than students with weak attendance, regardless
on attendance instead of on challenging instruction;
of what their attendance or test scores were in
attendance is the building block that needs to be strong
fifth grade (see Chapter 8).
to enable higher-level engagement in school and the
Attendance is also the critical factor underlying
development of strong academic skills.
large declines in student performance in the transition
Monitoring high-leverage indicators is not a sub-
to high school. Students unexcused absences quadruple
stitute for improving instruction or creating engaging
in one year, on average, when they go from eighth to
environments. But by monitoring those indicators
ninth grade (see Chapter 2). Their study habits also
that are most strongly associated with later outcomes,
decline. As a result of missing more class time and put-
school practitioners can make sure that students get the
ting in less effort, students grades drop by a half of a
right levels and kinds of support to keep them on-track
GPA point, on average. For some students, the drop in
for high school graduation and college readiness. They
attendance leads to high rates of course failure, putting
can focus attention on indicators that matter the most
them off-track for graduation. For others, it takes them
for later outcomes and they can establish realistic goals
from being B students to C students, putting them
for student and school improvement.
Executive Summary
INTRODUCTION
Research on Indicators of High

School and College Readiness
The expectations for schools have changed over the past 20 years. It is
no longer acceptable to graduate only the most motivated students and
to prepare only the most academically talented ones for college. Schools
are expected to graduate all of their students and to put them on a track
for college readiness.
The labor market is grim for students without high

school diplomas, and the vast majority of students
4
now aim to attain a four-year college degree. The

5
pressure is on high schools to improve their graduation

and college-readiness rates. Educators at the secondary
level, however, often contend that students enter high
school without the skills they need to be on-track for
graduation and college readiness.
While it may seem obvious that students need to
enter high school ready to succeed, it is not clear what
This chapter describes the

organization of the study and
discusses the prior research
on middle school indicators of
high school performance and
high school indicators of college
performance.
it means to be ready. There is little research that ties

middle grade records to high school outcomes, or that
later success, set goals for improvement, and provide
shows how they are related. There is even less work
support for students who are likely to fail. Middle grade
that ties middle school records to college readiness.
practitioners can use indicators of readiness to assess
Such information is needed by middle grade practitio-
their own success at preparing students for high school
ners so that they can gauge students readiness for
and to focus their efforts on those skills and behaviors
School Grade Structures in Chicago: Mostly K-8 Schools

The majority of schools serving middle grades in
Chicago are elementary schools that serve grades
K-8. Unlike districts that have a middle school structure, most students do not go through a large school
transition until the transition to high school. During
the 2008-09 school year, 91 percent of the schools
that served eighth-grade students were K-8 schools
4 Sum et al. (2009); Alliance for Excellent Education (2011);

Day and Newburger (2002); Heckman and LaFontaine (2007).
(419 schools). About 4 percent of schools with

eighth-graders served only the middle grades
some combination of grades four through eight
(20 schools), about 3 percent served middle and
high school gradesstarting at grade six or seven
through 12 (14 schools), and the remainder were
schools with other grade configurations.
5 National Center for Education Statistics (2007); Roderick et al.

(2008).
Introduction | Research on Indicators of High School and College Readiness
most crucial for later success. High schools could use
picture of how Chicago students academic performance
better indicators of students readiness to program
changes from eighth to ninth grade and provides con-
students into appropriate classes and target supports
text for understanding the indicator predictions shown
as they begin high school.
in later chapters.
This report provides the answers to four basic

questions about risk and readiness in the middle grade
years of fifth through eighth grade. It shows which
metrics matter and how schools might use them to
identify students for support.
Course Performance. This chapter evaluates potential
indicators of high school grades, showing how well

students personal characteristics and their academic
performance in the middle grades predict their grades
1 Which middle grade metrics predict ninth-grade failure?
in ninth and eleventh grade. It also evaluates which
combinations of indicators provide the best predictions.
Based on those metrics, which middle grade

students are at risk of failing their ninth-grade
End of Ninth Grade? This chapter shows students risk
when they get to high school?
of failing, based on their middle grade records. This

information could be used by school practitioners to
grades (3.0 or higher) in high school?
identify students who are at high risk of failure and can
Based on those metrics, which middle grade students
be used to discuss goals for student performance with
have little chance of earning high grades in high
students and their families.
school and which students are likely to do so?

3 Which middle school metrics predict reaching ACTs
Chapter 4: Who Is at Risk of Being Off-Track at the
classes, and which are likely to pass their classes

2 Which middle grade metrics predict earning high
Chapter 3: Middle Grade Indicators of High School
Chapter 5: Who Is at Risk of Earning Less Than As

or Bs in High School? This chapter shows students
college-readiness benchmarks?
likelihood of earning high grades in high schoolgrades
Based on those metrics, which middle grade students are at
that will make them eligible for college and likely to
risk of missing the benchmarks when they get to high school
succeed, once there. The information in this chapter
and which students are likely to meet them?
could be used to discuss goals for student performance
4 How does students performance on key indicators
change during the middle grades, and what are the

implications for high school success?
In this chapter, we describe the prior research that
set the stage for the study and how this report provides
new information. Each of the subsequent chapters
provides specific pieces of information on middle grade
indicators. Some readers may be more interested in
particular chapters than others. We offer the following
as a guide to the content of the report, which can point
readers to the information of most interest to them:
Chapter 1: Issues in Developing and Evaluating
Indicators. This chapter addresses some of the issues to
consider when deciding which measures of performance
with students and their families.

Chapter 6: Indicators of Whether Students
Will Meet Test Benchmarks. This chapter evaluates
potential middle grade indicators of performance on

high school tests. It compares predictions of performance on subject-specific tests and composite test
scores. It also examines the differences in predictions,
based on which middle or high school students attend.
Chapter 7: Who is at Risk of Not Reaching the
PLAN and ACT Benchmarks? This chapter shows the
levels of performance in the middle grades that give students a chance of meeting PLAN and ACT benchmarks
in ninth and eleventh grades.
Chapter 8: How Grades, Attendance, and Test
to focus on in early warning indicator systems, how
Scores Change. This chapter addresses the degree to
many to use, and the implications of using different
which students show different amounts of growth and
cut-offs for identifying risk.
decline in attendance, grades, and test scores over the
Chapter 2: Changes in Academic Performance

from Eighth to Ninth Grade. This chapter provides a
middle grade years.

Chapter 9: Interpretive Summary.
The Ninth-Grade On-Track Indicator

The ninth-grade on-track indicator simply shows
whether students are making sufficient progress
in ninth grade to be likely to graduate. A student is
on-track if she ends ninth grade with at least five
full-year credits and no more than one semester F in a
core course (English, math, science, or social science).
Ninth-grade on-track is highly predictive of eventual
graduation (80 percent correct prediction).A It is more

predictive of graduation than any other middle or
high school predictor, other than the combinations
of predictors that include twelfth-grade performance
(i.e., those that are measured during the year of
graduation, rather than several years prior), based on
comparisons of studies from across the country.B
A Allensworth and Easton (2005, 2007).
B Bowers, Sprott, and Taff (2013).
Prior Research Shows Passing

Classes and Earning High Grades
in High School Are Essential
for High School and College
Graduation, While Test Scores
Matter for College Access
while background factors (e.g., race, gender, and
There is often a perception that students performance
Passing classes is essential to graduate high school.
on tests is what matters for high school and college
To obtain a diploma, students need to accumulate cred-
graduation. While there are innumerable studies show-
its. That means they need to pass their classes. For most
ing significant relationships between test scores and
dropouts, the pattern of course failures begins in the
educational attainment, grades are more strongly and
ninth grade; failures then accumulate in later grades,
consistently found to be related to educational attain-
until they are so far behind that they cannot catch up.
ment than test scores. Furthermore, the relationship
As a result, a simple indicator of whether students are
of test scores to high school and college graduation
on-track or not in ninth grade, based on failures in the
becomes small, once we take into account students
ninth-grade year (see box above; The Ninth-Grade
GPAs. It is students grades that ultimately matter
On-Track Indicator), is very predictive of eventual grad-
more for high school and college graduation than their
uation, and accounts for almost all of the differences
test performance because grades capture more of the
in graduation rates by students race, gender, economic
factors relevant for student achievement than test
status, and other background characteristics, including
scores. This does not mean that test scores are irrel-
test scores. 9 Each semester course that a student fails
evantit is easier for students with higher tested skills
in ninth grade lowers the probability of graduating by
to get good gradesbut a focus on test scores as the
15 percentage points.10 For this reason, passing ninth-
dominant factor affecting college readiness would be
grade courses is one of the key outcomes examined in
misplaced. For this reason, this report focuses on more
this report.
economic status) are related to high school graduation, once we take into account students course grades,
background factors and test scores do little to further
predict students future educational attainment. 8
than just test scores as indicators of preparation for

high school and college.
There is also a fairly widely held perception that
High grades in high school are essential for college

graduation. While passing courses is critical for gradu-
students background factors are more deterministic
ating from high school, it is not enough to be ready for
of whether they graduate than their grades. However,
college. Students who are likely to succeed in college
6 Rumburger and Lim (2008); Allensworth and Easton (2007).

7 Geiser and Santelices (2007); Roderick et al. (2006);
8 Allensworth and Easton (2007); Allensworth (2013).

9 Allensworth and Easton (2007); Allensworth (2013).
10 Allensworth and Easton (2007).
Bowen, Chingos, and McPherson (2009); Allensworth (2013).
are not merely passing courses; they are working hard
significant relationships between middle and high
and earning high grades. Research in Chicago, and
school factors, without considering whether the
across the country, has found that students high school
middle school indicators are actually good predictors
grades are, by far, the most important predictor of
of later performance, for example, showing how many
getting into college and eventually graduatingmore
students are correctly identified as likely to succeed/
important than ACT or SAT scores or high school
fail in their classes or on their tests in high school.
coursework. A student who earns at least a B average
A few studies have specifically tried to discern the
in high school has a 50/50 chance of getting a four-year
predictiveness of middle school indicators for high
college degree (see Appendix A for more information).12
school performance. These studies provide a good
Therefore, we focus on students probability of earning
starting point for considering the types of middle
high grades as the leading indicator of college readi-
grade indicators that should be examined as indicators
ness. Test scores also play a role in college graduation,
of high school readiness. They also allow us to compare
particularly in giving students access to more selective
patterns in Chicago to other places, which gives us
colleges. College selection matters for graduation
a sense of whether the same indicators matter in
because students with the same high school records are
different places.
11
13
more likely to graduate at some colleges than others.

Chronic absence or failure in middle school indicates
10
High test scores help students get access to college.
high risk for eventual dropout. There is a growing body
When colleges make entrance decisions, they usu-
of work, across multiple cities, that consistently shows
ally rely on both students grades and their scores on
middle school attendance and course failures are strong
college entrance exams (the ACT and SAT). For this
predictors of whether students eventually obtain a
reason, higher scores on the college entrance exams
high school diploma. Research on middle grade indica-
help students gain access to more selective colleges and
tors in Philadelphia, Baltimore, New York, and three
programs (see Appendix A for more information). High
school districts in California (San Francisco, Fresno,
test scores also can help students obtain scholarships. In
and Long Beach) all found that, among the measures
Chicago, all students take the ACT college entrance exam
that were studied, course failures and low attendance in
in the spring of their junior year, as part of the state test-
the middle grades were the strongest predictors of high
ing system. They can use those scores for applications
school course failure or dropout.15 As shown in Chapter
to college. They also take ACT-aligned EPAS exams in
3, this is similar to what we find in Chicago. Studies in
earlier yearsthe EXPLORE in the fall of their eighth-
Philadelphia and Baltimore, which examined students
and ninth-grade years and the PLAN in the fall of their
sixth-grade records, also found that an out-of-school
tenth-grade year. These tests include ACT college-readi-
suspension in sixth grade was highly predictive of not
ness benchmarksscores that ACT has determined give
graduating from high school. Most of the studies also
students a 50/50 chance of earning grades of B or better
examined test scores and demographic characteristics
when they get to college, if they continue to make normal
showing that, while they were related to high school
progress. Therefore, we also include scores on ACTs
graduation or on-track rates, they were less predictive
EPAS tests as outcomes in this study.
than attendance or grades.

There are also a number of studies that link stu-
Prior Research Shows Links

Between Students Middle Grade
Performance and High School
Outcomes
dents test scores to later grades and high school
There are many studies that have shown relationships
test scores, however, and rarely include students
between students middle grade performance and
grades and attendance in the models for comparison.
later outcomes.14 However, researchers often identify
One example of this is a study by ACT, Inc., which
graduation and suggest that test scores can be used

as early indicators of high school performance. These
studies often fail to note the low predictive power of
Research Consistently Links Absence and Failures in

Middle School to High School Dropout
Philadelphia: Neild and Balfanz (2006) examined
New York: Kieffer and Marinell (2012) examined
eighth-grade indicators, while Balfanz, Herzog, and

MacIver (2007) examined sixth-grade indicators to
determine whether they could identify students with
a very high probability of eventual dropout. At both
grade levels, they found that students with Fs in
math or English, less than 80 percent attendance,
or an out-of-school suspension (in the sixth-grade
study) were at high risk of not graduating. Students
demographic characteristics and their test scores
were not as predictive as grades and attendance.
students fourth-grade test scores and attendance

as predictors of being on-track for graduation in
ninth grade, as well as changes in attendance and
test scores from fourth through eighth grade.
Course grades were not included in this study, but
researchers found that both attendance and test
scores in the early grades were predictive of being
off-track for graduation in the ninth-grade year,
along with declining attendance or test scores.
Declining attendance through the middle grade
years was particularly indicative that students
were at elevated risk of not graduating.
California: Kurlaender, Reardon, and Jackson (2008)

examined the relationships between seventh-grade
achievement indicators and high school graduation.
Attendance was not one of their indicators; but they
found that, among indicators studied, course failures
in middle school were the strongest predictors of
eventually not graduating among those they studied.
Test scores, retention in the elementary and middle
grades, and the timing of when students took
algebra were similarly related to graduationbut
not as strongly as course failures.
Baltimore: The Baltimore Education Research

Consortium (BERC, 2011) showed that chronic
absenteeism, course failures, and suspensions in
sixth grade were strongly associated with not
graduating within one year of the expected date.
Chronic absenteeism had the strongest relationship
of all the indicators.
showed that students eighth-grade EXPLORE exam
students, to determine whether self-reports of conduct,
could predict low grades in students ninth-grade year,
discipline, and relationships with school personnel
using data from 24 school districts that participated
improved the prediction. They found that those three
in their EPAS testing system. The study found that 22
self-reported student characteristics combined doubled
percent of the variation in low ninth-grade GPAs could
the precision of their prediction of low GPAs, beyond
be explained by eighth-grade EXPLORE scores (ACT,
test scores alone. Thus, while the authors conclude that
Inc., 2008, p. 27). Unfortunately, the study did not
test scores predict future grades, their analysis actu-
include indicators of middle grade attendance or course
ally shows that factors other than test performance are
grades to compare with exam scores as indicators of
critical for passing classes; while the focus of the study
later performance. Yet, the study did use an interesting
was on the predictiveness of test scores, the relation-
set of measures of academic behaviors, from surveys of
ship was modest and dwarfed by other factors.
11 Bowen, Chingos, and McPherson (2009); Geiser and Santelices
and California (Kurlaender, Reardon, and Jackson, 2008)

predicted high school graduation/dropout, while the
New York study (Kieffer and Marinell, 2012) predicted
ninth-grade on-track rates.
16 The 22 percent rate is calculated using the percent variance
explained for ninth-grade low GPAs provided in the footnote
to Figure 8. Figure 8 states that the EXPLORE composite
explains 53 percent of the explained variance, while the footnote states that 52 percent of the total variance is explained.
16
(2007); Roderick et al. (2006).

12 Roderick et al. (2006); Allensworth (2006); Bowen, Chingos,
and McPherson (2011).
13 Roderick et al. (2006).
14 For example, Alexander, Entwisle, and Kabbani (2001);
studies cited in Rumburger and Lim (2008).
15 The studies in Philadelphia (Neild and Balfanz, 2006;
Balfanz, Herzog, and MacIver, 2007), Baltimore (BERC, 2011),
11
Low test scores in the middle grades foretell low test
how much do students experiences in school actually
scores at the end of high school. While test scores
matter for test scores at all?
are not strongly predictive of students later grades,
12
As we show in this report, we find only small differ-
students test scores in the middle grades are highly
ences in students test score growth across the middle
predictive of their test scores at the end of high school.
grade years. No students with low performance in fifth
In California, for example, Zau and Betts (2008) exam-
grade make up enough ground to have average perfor-
ined whether indicators from grades four to nine could
mance by eighth grade, especially in reading. The same
predict success on the California exit exam (CAHSEE);
pattern can be seen in Kieffer and Marinells (2012)
Kurlaender, Reardon, and Jackson (2008), likewise,
study in New York, described earlier. They observe very
examined the relationship between seventh-grade in-
little variation in test score growth from fourth through
dicators and risk of failure on the CAHSEE. Both found
eighth grade, relative to the size of the gaps that exist
that students having very weak test scores in the early
initially. We find that, while there is substantial year-
grades were at high risk of failing the exit exam. Zau and
to-year variation in gains, differences tend to even out
Betts (2008) showed that they could predict students
when multiple years are considered and are modest
California exit exam passing almost as well in fourth
relative to the differences in test scores that students
grade as they could in ninth. These two studies also
start out with in the early grades. This brings up ques-
found that students course grades, behavior, and atten-
tions about how effective interventions around test
dance in the primary grades could predict passing the
score growth can be for reducing inequities in perfor-
CAHSEE. It is not clear from these studies the extent
mance, even if students are identified as at risk early on,
to which students grades and attendance improve the
as long as all schools have similar resources, incentives,
prediction of CAHSEE failure, beyond prior test scores,
and approaches.
as the variables were entered together in their statistical

models. However, each was predictive.
ACT, Inc., has shown a strong relationship between
This Study Extends Prior Research

on Middle Grade Indicators
their eighth-grade EXPLORE exam and student perfor-
This study builds off prior research and extends the
mance on the corresponding eleventh-grade ACT exam.
analysis of middle grade performance by asking
Furthermore, they have shown that prior test scores
questions that have not been addressed before:
eclipse other factors in predicting ACT scores. Entering

eighth-grade test scores are far more predictive of
Should schools consider a broader range of
eleventh- and twelfth-grade ACT scores than students
indicators, beyond test scores, grades, attendance,
coursework, grades, or background characteristics dur-
and discipline, such as grit and test score growth?
ing high school.17 In their words, the level of academic
Prior studies have found relationships of middle school
achievement that students attain by eighth grade [their
test scores, grades, attendance, and discipline records with
EXPLORE score] has a larger impact on their [ACT score]
high school outcomes, so we include these as potential
by the time they graduate from high school than anything
indicators of high school success. We also include change
that happens academically in high school. Their inter-
during the middle grade years in these indicators, since
pretation of this relationship is that early intervention is
Kieffer and Marinell (2012) found them to be predictive in
important. They do not study early intervention, howev-
New York. Prior research in Chicago, and elsewhere, has
er, to see if this makes a difference in later performance.
shown that it is not just students absolute skill levels that
All of these findings, however, could also suggest
are related to passing classes and getting good grades, but
that exam performance may not be strongly malleable.
their skills relative to classroom peers; even students with
If students test scores at the end of high school can be
above-average achievement are at elevated risk of failure
predicted as well in fourth grade as in ninth grade, and
if their classroom peers have much higher skill levels.18
if students high school experiences matter little for
Therefore, we include indicators of students skills relative
their ACT scores compared to their prior test scores,
to their school cohort peers.
We also expand the indicators under consideration
climate of the high school they attend will moderate the
to include two noncognitive factorsstudents self-
relationships between readiness indicators and ninth-
reported grit (or tenacity) and their study behaviors in
grade outcomes. It could be, for example, that students of
the middle grades. The concept of academic tenacity
all skill levels are likely to fail in some high schools. Other
has received considerable attention in recent years as
high schools might have very strong student support sys-
an important factor behind students academic achieve-
tems in place, leading students to succeed who would have
ment. Duckworth and her colleagues have found that
been at risk if they attended other schools. Thus, which in-
grit, or perseverance to pursue long-term goals over the
dicators are most important, and the levels of those indica-
long-term, shows relationships with the grades of high-
tors that indicate risk, might depend on which high school
achieving students above and beyond their test scores
students attend. The Kurlaender et al. (2008) study that
(Duckworth et al., 2007). We include a subset of the grit
compared three California districts found, for example,
scale used by Duckworth and colleagues as one of the
that while course failures and test scores were predictive
indicators of middle grade performance. We also include
of later outcomes in all districts, the timing of algebra was
a measure of students study habits in the middle grades.
not. One can also see from their analysis that the threshold
Prior research on ninth-grade indicators found stu-
that puts students at risk of failing the CAHSEE varies by
dents study habits predict ninth-grade pass rates and
which district students attendwith students in Fresno
grades, above and beyond their attendance (Allensworth
at risk if they are simply below basic levels on the eighth-
and Easton, 2007). We might expect that students study
grade assessment, but students in the other districts at
habits in middle school predict their success when they
risk only if they are far below basic levels.
19
get to high school.

Table 1 provides a list of the middle school indicators
It is also possible that the predictiveness of indicators

depends on which middle school a student attends. There
included in this study, and Appendix B lists the defini-
is a prevailing belief that GPAs are not comparable across
tion of all the indicators.
schools, meaning that two students with the same grades

attending different middle schools may not be equally
Which indicators are the most critical indicators to
ready for high school. Often, concerns are raised that
track? Many variables are related to later outcomes, but
GPAs are easily manipulated. Yet the same could be said
not all indicators are useful or necessary for an effec-
of many other potential indicators. For example, some
tive indicator system. With the wide array of data that
schools seem to manipulate test scores, over-preparing
are now available, practitioners often feel that they are
students for the types of questions asked on the stan-
drowning in data. It is important to critically analyze
dardized assessment in a way that would not translate to
the ways which indicators are related to later outcomes,
different examinations taken in later years.20 Thus, there
if practitioners are to focus on the best indicators to
may be a different threshold of risk for an indicator, based
enact change. Chapter 1 discusses some of the issues to
on which middle school a student attends.
consider when evaluating indicators.

Who is at risk for not being college-ready, as well as
Do the indicators depend on where students attend
for not graduating? Prior work on middle school indica-
middle and high school? Students success in ninth grade
tors has primarily focused on minimum expected levels
is determined not only by their preparation but also by
of performancegraduating, passing classes, and pass-
their experiences in the ninth grade. The structure and
ing exit exams. Middle grade educators also want to be
17 ACT, Inc. (2008).

18 Nomi and Allensworth (2013); Farkas, Sheehan, and
and Melinda Gates Foundation (Dweck, Walton, and Cohen,

2011). We recently included it as one of a number of important
noncognitive factors affecting grades in a UChicago CCSR
report (Farrington et al., 2012).
20 Luppescu et al. (2011); Koretz (2005, 2008).
Grobe (1990); Kelly (2008).
19 The importance of perseverance for academic success gained

considerable attention in a New York Times Magazine article
(Tough, 2011), through a manuscript commissioned by the Bill
13
TABLE 1
Middle Grade Indicators Included in the Study

Middle Grade Indicators (Grades Five through Eight)
Course
Performance
Average grades in all core classes (both semesters combined in each year)
Average grades in English/language arts classes (both semesters combined in each year)
Average grades in math (both semesters combined in each year)
Number of course failures in core courses (both semesters combined)
Growth in GPA from grades five through eight
* Grades are measured on a 4-point scale, where F=0 and A=4.
Attendance
Attendance rate (number of days attended out of number of days enrolled in a given year)
Growth in attendance rates from grades five through eight
Test Scores
Reading score on the ISAT in the spring of each year, grades six through eight
Subscales in the reading test in eighth grade: Vocabulary development, reading
strategies, reading comprehension, literature, and extended-response results
Math score on the ISAT in the spring of each year, grades six through eight
Subscales in math test in eighth grade: Number sense, measurement, algebra, geometry, data
analysis, statistics and probability, and extended-response results for mathematical knowledge,
strategic knowledge, and explanation results
Average gain in test scores (reading and math) from grades six through eight
Relative skills: difference in reading/math score from average of eighth-grade peers in the same school
14
* Middle grade test score indicators are based on the Illinois Standards Achievement Test. Information on these tests is available
in ISBE (2012).
Suspensions and
Misconduct
Number of days suspended during the eighth-grade year

Number of serious (levels 4-6) infractions committed during the eighth-grade year
* Serious infractions include battery/fighting/aggravated assault; theft/burglary/robbery; disorderly conduct; trespassing on CPS
property; gang activity; use of firearm; arson; use of alcohol, narcotics; murder.
Noncognitive
Factors
Measures obtained from surveys when students were in the seventh or eighth grade:
Grit (Individual-level reliability in grades six through eight is 0.86.)
To what extent do the following describe you:
A. I finish whatever I begin.
B. I am a hard worker.
C. I continue steadily toward my goals.
D. I dont give up easily.
Not Like Me At All, Not Much Like Me, Somewhat Like Me, Mostly Like Me, Very Much Like Me
Study Habits (Individual-level reliability in grades six through eight is 0.83.)
How much do you agree with the following:
A. I set aside time to do my homework and study.
B. I try to do well on my schoolwork even when it isnt interesting to me.
C. If I need to study I dont go out with my friends.
D. I always study for tests.
Strongly Disagree, Disagree, Agree, Strongly Agree
Background
Characteristics
Race, gender, special education status, neighborhood poverty level, neighborhood

socioeconomic status, free/reduced price lunch status, and whether student entered high
school older than age 14
Neighborhood characteristics come from census data at the block group level on male
unemployment, percentages of families below the poverty line, mean education levels, and
median family income. Other characteristics are derived from school administrative data.
Note: See Appendix B for more information on data used for this study.
able to assess whether their students performance puts
systems that match the level of support to the level of
them on a path to potentially succeed in college. To be
risk. It is also helpful to know students general level
on-track to college readiness, students should be mak-
of risk, given their middle grade performance, to help
ing appropriate gains in learning the skills required for
students and their families understand the connections
acceptance to college. These are generally measured
between students current progress and their likeli-
by tests such as ACTs EPAS system. In Chicago, all
hood of success in later years and to set goals that will
students participate in the EPAS tests and take the
encourage high levels of performance. These issues are
ACT in eleventh grade. Even more importantly, they
discussed in more detail in Chapter 1.
need to be performing at sufficiently high levels in their

examine readiness for high school in three ways: Ready
How We Answered Questions

About Middle Grade Indicators
to pass their ninth-grade classes; ready to meet college-
To examine the ways in which middle grade indicators
readiness benchmarks at the end of ninth grade; and
are predictive of high school success, we followed a
ready to earn high grades (As and Bs).
cohort of students in the Chicago Public Schools from
high school classes to earn high grades. Therefore, we
fifth to eleventh grade, using their academic records,

What is students level of risk, who is at high risk,
survey data, and census information on their residen-
and who might go either way? It is useful to be able
tial neighborhood. The report focuses on the cohort
to identify students who have an extremely high risk
of students who were first-time ninth-graders in
of failing to meet an outcome (e.g., high risk of being
2009-10, and who should have been in the eleventh
off-track, not graduating, missing test benchmarks)
grade in the 2011-12 school year. See Appendix B on
because school personnel know with a high degree of
p.106 for a description of the cohort samples used in
certainty that the students they target are very likely to
each analysis. The analyses focus on performance in the
drop outthey are not wasting resources on students
ninth-grade year, showing whether middle grade factors
who will graduate without any intervention. These are
were related to performance when students arrived in
high-yield indicators that very accurately identify
high school. We include some eleventh-grade outcomes
students who will likely fail without falsely identifying
to show whether it is likely students will eventually
students who will not fail.
graduate and to show the achievement levels that give
21
At the same time, there are limitations to just considering intervention plans for students at very high risk of
students access to college.

We examine the predictiveness of each middle grade
failure. Students at very high risk of failure may not be
indicator for each ninth-grade outcome. We examine
very amenable to change. If a student has a 90 percent
them alone and in combinationwhether each adds
chance of dropping out, it will take a very substantial
new information for predicting the outcome beyond
intervention to get that student to graduate. Yet, a student
the others, or depends on values of others. We try to
with a 50 percent chance of dropping out could go either
determine what minimal combination of predictors is
way. A more modest intervention might be enough to get
the most accurate. We also examine the precision in
that student to succeed. Even though interventions would
the predictionwho can be identified for intervention
need to be targeted to a larger group, those interventions
or success with high certainty, and how accurately we
might be more effective and potentially less costly.
can predict students later performance from their
This suggests that indictor systems might want to
middle grade records. Finally, we examine to what
identify students at moderate risk, as well as those at
degree the predictions depend on which middle school
high risk, so that schools can design tiered intervention
or high school students attend.
21 For more information on high-yield indicators, see a summary

in Heppen and Therriault (2008). Developing Early Warning
Systems to Identify Potential High School Dropouts. Retrieved
from: http://www.betterhighschools.org/pubs/documents/
IssueBrief_EarlyWarningSystemsGuide.pdf.
15
After determining the best predictors of each
for middle school and high school practitioners. The
outcome, we examine the degree to which the best
findings in this report suggest that simplistic calls
indicators change across the middle grade years. We
for early intervention may not be the best solution.
examined student performance from fifth through
Students experiences in high school moderate the
eighth grade, whether performance tended to change
degree to which students succeed, above and beyond
on average, and the degree to which some students
their qualifications upon leaving middle school. Many
showed more growth or decline than others. We also
students who perform well in the middle grades do
examined the extent to which these changes could be
not graduate from high school ready for college. High
attributed to the school that students attended.
schools could make use of middle grade indicators to
Finally, we discuss the implications of this work
make sure students are reaching their potential.
16
CHAPTER 1
Issues in Developing and

Evaluating Indicators
There are myriad studies that show relationships
between students academic performance, family
characteristics, neighborhood factors, and personal
factors in the elementary and middle grades with their
academic performance in high school. Many characteristics of students in the middle grades are related
to their performance in high school, but it is difficult
for practitioners to track a large number of indicators
and develop intervention strategies for all of them. It
is also inefficient to focus efforts on factors that have
only weak, indirect, or spurious relationships with
the high school outcomes, when similar amounts of
This chapter discusses issues to

consider when designing indicator
systemswhich indicators to use
and how many, as well as implications of different cut-offs for
identifying risk. This chapter also
explains how to read the graphs
that are used in subsequent
chapters to compare indicators.
effort could have a bigger pay-off if focused on factors

with stronger direct relationships with high school
outcomes. An effective early warning indicator system
are strongly related to high school graduation because
focuses attention on factors that have strong and direct
they directly affect itwhen students miss class and
relationships with later performance, and that also are
fail classes they do not accumulate the credits they
malleable with school practice.
need to graduate. Therefore, Chicago designed student
Which Indicators Produce the Most

Accurate and Simple System?
monitoring systems around the ninth-grade indicators

of attendance and course failure, and schools used those
systems to improve attendance and pass rates in the
The simpler an indicator system, the easier it is to track
ninth grade. This provided a mechanism for driving
the indicators and develop intervention plans. Many
improvements through use of the indicator system itself.
potential indicators are correlated with high school
In contrast, other indicators are correlated with
outcomes, but it is not necessary to track all of them.
high school outcomes simply because they are related
If a small number of indicators is just as effective at
to other factorstheir relationship with the outcome
indicating students risk of later outcomes as a larger
is spurious and would not be a good lever for moving
number of indicators, school practitioners do not need
student outcomes. Other factors have an indirect
to monitor the myriad elements that may be related to
relationship with the high school outcomeaffecting
the outcome. Thus, one issue to consider is how to get
it by affecting some other factor that has a direct rela-
the best prediction of later outcomes with the smallest
tionship. For example, poverty status is related to high
number of indicators.
school graduation, but the relationship is indirect
Another issue to consider is whether the indicator

is useful only for prediction, or whether it also could
poverty affects students attendance rates and grades,

and their grades affect whether students graduate.
be used as an intermediary outcome to drive change in
Efforts to track and improve indicators that directly
the later outcome. For example, in Chicago, schools use
affect the outcome will generally be most effective for im-
ninth-grade attendance and failure rates as indicators
proving the outcome, because no other intervening vari-
of eventual high school graduation; they provide fairly
able has to change as a result of the intervention. Efforts
accurate predictions of students risk. However, they
to change indicators that have indirect relationships
Chapter 1 | Issues in Developing and Evaluating Indicators
17
could also improve outcomes, but only to the extent that
superfluous. Appendix C describes the methods used
the mediating factor also changes. Often practitioners
for comparing indicators in this study.
are frustrated when the results of their efforts do not
18
To some extent, this approach also makes it likely that
translate into improvements in the outcomes that
indicators that are most directly related to the later out-
were targeted for improvement; this occurs when their
come will be chosen over indicators that have indirect or
efforts are aimed at indicators that are only modestly
spurious relationships with the outcome. Indicators that
or indirectly related to the targeted outcome.
have direct relationships with the outcome will often
It is standard practice for researchers to enter
but not alwayshave the strongest relationships with
potential predictors of an outcome into a statistical
the outcome, because no other intervening factor must
model and then determine which of them show signifi-
change. Indicators with direct relationships will also
cant relationships through the regression coefficients,
continue to add to the prediction when other variables
controlling for other predictors in the model. However,
are included, because their relationship is not dependent
because the predictors are correlated with each other,
on a third factor. Factors that have indirect or spurious
it is not possible to use regression coefficients to decide
relationships with the outcome will no longer contribute
which predictors are actually the best to track.
to the prediction, once the intervening variables (in the
22
This
approach results in unnecessarily complicated indica-
case of indirect effects) or prior variables (in the case
tor systems that include predictors that do not neces-
of spurious relationships) are included in the model. 23
sarily improve the prediction of the outcome beyond
At the same time, just because an indicator has a strong
all of the others. Instead, the model statistics (R )
relationship with an outcome does not necessarily mean
are much more useful than significance levels on
that it has a direct effect; such an interpretation would
coefficients for determining which potential
depend on theory and on research designed to get at
indicators produce the simplest indicator system.
questions of causality. Furthermore, a predictor that
The first step in selecting the best indicators is to
has an indirect or spurious relationship with an outcome
identify which has the strongest relationship to the out-
can still be useful in an indicator system if it adds to the
come. The easiest way to do this is to compare correlation
prediction of the later outcome.
coefficients; although, as discussed in the next section,
Chapters 3 and 6 show the strength of the relation-
researchers should also consider the predictive power
ships of each potential middle grade indicator with each
of the indicators and issues of sensitivity and specificity,
of the high school outcomes, and the degree to which
as well as cut-off levels that are meaningful for predict-
combinations of potential indicators improve the pre-
ing later outcomes. Once the strongest predictor is
diction above and beyond predictions that use smaller
identified, the next issue is whether the prediction could
numbers of indicators. Tables of potential indicators and
be improved by considering other indicators. Model
combinations of indicators are included so that readers
statistics on the amount of variation in the outcome that

2
can compare the strength of the relationship of each
is explained by the predictors (e.g., the R or pseudo-R ),
potential indicator with the high school outcome, and
along with changes in the accuracy of predictions (e.g.,
see how it changes the prediction. In all cases, predic-
the percentage of students correctly classified), can be
tion models that use just two or three indicators are
used to determine whether the prediction is significantly
almost as predictive as models that use all possible
better with any given additional predictor. If the model
indicators. Even though many factors are related to
does not explain more variation, and the percentage of
high school achievement, most are indirectly related
students correctly predicted to succeed or fail does not
to high school achievement through some combination
significantly improve, then the additional indicator is
of middle grade GPAs, attendance, or test scores.
22 The regression coefficients will potentially be strongly affected
in the model. Some variables may be significant predictors

of the outcome, but not add significantly more explanatory
power beyond the other variables.
23 For more information on these issues, see Davis (1985).
by multicollinearity. The distribution of shared variance could

be arbitrarily attributed to two factors, when one factor is the
most directly related, depending on which other variables are
How Accurate Are the Predictions

That Result from the Indicators?
students flagged with high-yield indicators had at
Often, researchers identify factors as related to later
data on Philadelphia schools, the researchers looked
outcomes, without showing the accuracy of predictions
to see whether they could find indicators of eventual
based on the indicator systemthe degree to which
dropout using data on students performance in the
students at risk are correctly identified or misidenti-
middle grades. Using the 75 percent criterion, they
fied using the indicators. Although an indicator may
found that an F in English or math and attendance
be correlated with an outcome, the indicator might not
lower than 80 percent were signals available in middle
substantially improve the prediction of that outcome
school that could be used for precise targeting of
beyond what would occur if we just assumed that all
students at very high risk of not graduating. The
students were at risk or not at risk. This often occurs
Philadelphia indicators are highly useful because
if an outcome is either very rare or very common. If
practitioners do not need to worry that students who
an event is rare, even a student with an elevated risk
are identified with these indicators will succeed
is unlikely to experience it; the opposite is true for
without intervention.
outcomes that almost all students achieve. An indicator
least a 75 percent likelihood of dropping out. Using
High-yield indicators are useful for targeting inter-
may be able to precisely identify a subset of students
ventions that are expensive or require substantial re-
who are at risk of not achieving an outcomeso that all
sources. Their high degree of precision among students
of the students who are identified will fail to reach the
who are flagged as at risk, however, comes at two poten-
outcome without interventionbut miss the majority
tial costs. The first issue is the sensitivity of the indica-
of students who fail to achieve the outcome because too
torthe percentage of students who actually fail but are
few students were identified as at risk.
not identified with the high-yield indicator. In general,
When deciding which indicators to use and how to
there is a trade-off between minimizing false positives
use them, researchers consider two primary issues. The
(false alarms) and missing true positives (students
first issue is the proportion of students who fail to meet
who really will fail). This is a very substantial issue for
an outcome who would be identified by the indicator as
Chicago, when we try to apply the high-yield indicators
at risk of failure. This is known as the sensitivity of the
from Philadelphia. In Philadelphia, just over half of
indicator; a highly sensitive indicator of dropout would
eventual dropouts were identified using the high-yield
identify a large proportion of the students who drop out.
indicators. 26 But in Chicago, only a small percentage
The second issue is the false-positive or false alarm
of eventual dropouts are identified with middle school
ratethe percentage of students who do not fail who are
high-yield indicators, as discussed in Chapter 3. If all
identified as at risk of failure by the indicator. This is
strategies were based around students with a very high
also known as the false-positive proportion.
likelihood of failure, then the majority of students at
24
If only
60 percent of students flagged by an indicator actually
risk for dropout would be missed in Chicago.
fail, then there is a high rate of false alarms because
The second limitation of high-yield indicators is
40 percent of students would succeed without any in-
that interventions to change the outcome may be very
tervention. Indicator systems with a high false-positive
resource-intensive, and there may not even be any
rate have the potential to waste valuable resources. 25
known intervention that can get the students who are
The false-positive issue was the primary driver of
identified as at risk to eventually succeed. For example,
the indicator work done in Philadelphia by Neild and
if a student has a 90 percent chance of failing to gradu-
Balfanz (2006) and Balfanz, Herzog, and MacIver
ate, an intervention that successfully lowers her risk by
(2007) to identify high-yield indicators of dropout;
30 percentage points still would not be enough to make
24 This is calculated as one minus the specificity, which is one
25 For a description of these issues, see Bowers et al. (2013).

26 Neild and Balfanz (2006) correctly identified 54 percent of
minus the proportion of students who succeeded who were

identified as succeeding.
dropouts with their eighth-grade indicators.
19
her likely to graduate. On the other hand, a student who
interventions if they are to succeed. Thus, indicator
had a 50 percent risk of failing could go either way, and
systems might identify students with different levels
a modest intervention might be enough to make him
of risk to match them with appropriate supports.
succeed. Schools would not want to put an expensive
The chapters that follow use figures that compare
intervention into play for students with a 50 percent
the sensitivity of indicators with the false-positive
risk, because half of the students would succeed with-
rate. 27 Figure 1 provides a description of how to inter-
out intervention, but these students would not be in
pret these charts. Each dot on the chart represents
need of a major intervention.
the ability of an indicator, or a set of indicators, to
As school districts design strategies to address high
accurately predict a later outcome. Indicators, or
school and college success, they can think about strate-
sets of indicators, that are able to identify all of the
gies for students with different levels of risk for failure
students who end up failing have a high sensitivity,
or for not meeting college readiness goals. Similar to
and are represented by dots at the top end of the chart
the Response to Intervention (RTI) approach, interven-
(where the sensitivity gets close to 1.0). Indicators, or
tions with different levels of intensity could be targeted
sets of indicators, that correctly identify students who
to students with different levels of risk. Students with
are at risk of failing without falsely identifying students
moderate risk are in need of modest interventions,
as at risk who will succeed are represented by dots at
while most students at high risk will need very intensive
the far left of the figure.
20
FIGURE 1
Proportion of Students Correctly Identified as Failing to

Reach the Outcome True Positive Proportion (Sensitivity)
Chart for Comparing the Ability of Indicators to Predict Later Outcomes

1.0
Perfect Prediction
Close to Perfect Prediction: Almost all of the students
who failed were identified as at risk by the indicator, and
almost all of the students who were identified as at risk
really did fail. Indicators close to the top left of the chart
are very precise. However, if an indicator is too perfect, it
may suggest little malleability in the outcome, because
there is little doubt about what a students outcome will
be, based on the earlier indicator.
0.9
0.8
0.7
High Sensitivity, High False-Positive: The indicator

identifies almost all of the students who failit doesnt
miss students who are at risk. But, half of the students
identified as at risk are false positives. They would
succeed without intervention.
0.6
0.5
Low Sensitivity, Low-False Positive: The indicator only

identifies half of the students who failmany students
who are at risk are not identified with the indicator.
However, almost all of the students who are identified
as at risk really are at risk. Intervention efforts would
not be wasted on students who would succeed without
intervention.
0.4
0.3
0.2
Ra
0.1
nd
om
Gu
es
Random Guess: The indicator does not tell you anything

more about who is at risk than a random guess. This is
true at any point along the dashed line.
Worse Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Proportion of Students Incorrectly Identified as Failing to Reach the Outcome

False-Positive Proportion (1-Specificity)
27 See Bowers et al. (2013) for more information about using

this type of chart to compare the predictions of different
indicators of high school dropout.
0.9
1.0
How Malleable Are the Indicators?
the extent to which there is evidence that schools can
Another issue to consider is the extent to which the in-
change the indicator to a degree that such efforts could
dicator is malleable and whether there are known strat-
have a substantial impact on the final outcome.
egies for moving it in a school. Indictors do not have to
Chapter 8 begins to get at questions of malleability
be malleable to be useful for identifying students risk;
by examining the degree to which the best indicators
non-malleable indicators may be very useful for target-
of high school outcomes actually show different rates
ing students for interventions if they are highly predic-
of growth across students over the middle grade years.
tive. However, if indicators directly or indirectly affect
If no student shows substantial change on an indica-
the outcome, and if there are known ways to influence
tor over time, or if all students show the same amount
the indicators, they not only allow school practitioners
of growth or decline over time, then there may not be
to identify students risk of failure but also give them an
known strategies to substantially change that indica-
intermediate outcome to work on for moving the later
tor. If so, tracking the indicator may not be an effective
outcome. For example, the ninth-grade early warn-
means of improving the later outcome, compared to
ing indicators used in Chicago around attendance and
other indicators that are movable and also related to
grades have been effective for getting more students on
the later outcome.
the path to graduation because schools have been able to
In the end, the indicators that are likely to have the
develop effective strategies for actually improving stu-
most potential to improve high school outcomes are
dents ninth-grade attendance and course performance.
those that have strong, direct relationships with high
Indicators that are not within the power of prac-
school outcomes, which identify students at risk with a
titioners to change cannot be used to improve school
high likelihood of precision, and which can be changed
practices. For example, students gender, age at entry
through intervention. Some of the most commonly used
into high school, and race might be associated with
indicators that are in use today for decisions about who
their performance in high school, but high school
to target for interventions, and that have been identified
practitioners cannot develop strategies to change these
by research as predicting later outcomes, do not meet
background characteristics. A final question, then, is
these criteria, as discussed in subsequent chapters.
21
22
CHAPTER 2
Changes in Academic Performance

from Eighth to Ninth Grade
Students experiences in high school are often very dif-
This chapter shows how Chicago

students academic performance
changes from eighth to ninth
grade. There are dramatic
decreases in school attendance
and grades when students move
into high school. Testing standards
also change in ninth grade.
ferent from their experiences in middle school. During

the middle grades, student behavior is more closely
monitored by teachers and other adults in the school
than in high school. In some K-8 schools in Chicago, students have the same teacher for most of their classes. In
these schools, there is no need to move from classroom
to classroom, making it difficult to skip class. Even if a
middle school is departmentalized, students are often
accompanied from one class to another by a teacher.
Once students enter high school, they experience much
23
more freedom. They change classes and go to lunch and

the library by themselves, with no adult supervision.
FIGURE 2
They may see each teacher for less than an hour a day,
GPAs Decline from Eighth to Ninth Grade

45
for them throughout the day. As a result, it becomes easi-
40
er for students to skip classes and students are less likely

to have teachers notice when their grades start to slip.
This has implications for their academic performance.28
Students grades and attendance tend to decline
considerably when they move from eighth to ninth
grade. This is a pattern observed in many places across
the country. 29 In Chicago, on average, GPAs go down
Percent of Students
making it less likely they will have an adult looking out
42%
35
31%
30
25
20
18%
16%
10
5
eighth to ninth grade. As shown in Figure 2 , only about
5 percent of students who were in eighth grade in the
5%
0 GPA < 1.0
ninth grade in the 2009-10 school year. In eighth grade,

three-fourths of students had between a 2.0 and 4.0,
1.0 GPA < 2.0 2.0 GPA < 3.0 3.0 GPA < 4.0
Core GPA
2008-09 school year had failing GPAs (less than 1.0).

(18 percent) had GPAs of less than 1.0 when they were in
23%
21%
15
by about half of a point when students move from
One year later, almost one-fifth of these same students
32%
28%
8th Grade
9th Grade
Note: Average GPAs for students entering ninth grade in 2009-10 who were
eighth-graders in 2008-09. The same set of students is represented in each year
(eighth and ninth grade).
Grades decline across all types of studentsboth
which is a C, B, or A average. But by ninth grade, just
boys and girls, across race/ethnicity, and among
over half had at least a C average (2.0) and less than a
students with high as well as low test scores (see
quarter (23 percent) received GPAs of 3.0 or higher in
Table 2). Both boys and girls show a decline in GPAs
ninth grade.
of about half a grade point. The decline also is similar
28 For further information about why students grades and at-
29 See Benner (2011) for a review of literature on the high school
tendance fall when they enter high school, see the UChicago
CCSR research series, Free to Fail or On-Track to College.
transition.
Chapter 2 | Changes in Academic Performance from Eighth to Ninth Grade
TABLE 2
in grades over the high school transition than students
Average Decline in GPA from Eighth to Ninth Grade

by Subgroup
with weak test scores, declining by 0.6 points versus

0.4 points, respectively.
Average Decline in GPA Points
One of the main reasons that students GPAs fall
Girls
-0.5
when they enter high school is the change in students
Boys
-0.5
attendance. Many students stop attending every class
African American
-0.5
Latino
-0.5
every day when they are in high school. 30 In eighth
White
-0.5
Asian
-0.2
Top Quartile ISATs
-0.6
Bottom Quartile ISATs
-0.4
grade, only 5 percent of students had less than 80 percent attendance (see Figure 3); one year later, in ninth
grade, a quarter of these same students had less than
80 percent attendance. A student with 80 percent
attendance is missing, on average, one day of school
a week. In eighth grade, 60 percent of students had
attendance rates of 96 percent or higher. In ninth
Only Asian students show a smaller decline in grades,
grade, only 37 percent of students attended at least
with GPAs falling by 0.2 points instead of half of a point.
96 percent of the time. In just one year, students
One might think that students get poor grades in high
attendance drops dramatically; this occurs when
school because they lack sufficient academic skills to
they transition into high school.
handle high school courses. Students with high test
The main driver of the increase in absence rates
scores, however, actually show slightly larger declines
comes from unexcused absences. Unexcused absences
FIGURE 3
Attendance Declines Dramatically from Eighth to Ninth Grade

30
28%
25%
25
Percent of Students
24
among African American, Latino, and white students.
19%
20
14%
15
12%
9%
10
5%
3%
1%
80%
82%
3%
2%
84%
4%
2%
86%
4%
5%
88%
5%
8%
6%
90%
92%
12%
13%
11%
8%
94%
96%
98%
100%
Average Attendance Rates

8th Grade
9th Grade
Note: Average attendance rates for students entering ninth grade in 2009-10 who were eighth-graders in 2008-09. The same set of students is represented in each
year (eighth and ninth grade).
30 For an analysis of the reasons grades drop from eighth to

ninth grade, see Rosenkranz et al. (2014).
31 For more information, see Rosenkranz et al. (2014).
32 Rosenkranz et al. (2014) used regression equations to
predict students grades in eighth grade and ninth grade
with a series of variables that included test scores, demographic variables, attendance in each grade and students
reports of study habits. Study habits explained 14 percent of

the decline in English grades and 13 percent of the decline in
math grades. Attendance explained an additional 72 percent
of the decline in English grades and 78 percent of the decline
in math grades, beyond study habits. Together, these factors
accounted for 86 percent of the decline in English grades and
91 percent of the decline in math grades.
quadruple from eighth to ninth grade, from five days
One important difference between the two tests is
to 21 days, on average. At the same time, students also
where the benchmark is placed. As shown in Figure 4,
report putting less effort into studying and making sure
the high school test standards (on the PLAN exam)
they get their work done. According to the CPS student
are higher because schools are aiming towards ACTs
connection survey, 27 percent of seventh-graders in
college-readiness benchmark. Eighth-grade standards
spring 2007 said they strongly agree that they try
are based on the ISAT benchmarks set by the state,
hard on schoolwork even if it is boring; two years later,
which are based on grade-level expectations. 33 In
in 2009 when they were ninth-graders, only 18 percent
2009, this same group of students was much less
of the same group of students strongly agreed to the
likely to reach PLAN benchmarks in high school than
same question. There were also declines in the percent-
to meet the state standards in eighth grade, because it
age of students who said they set aside time for home-
is set at a higher level. The benchmark for meeting
work and studying and the percentage of students who
state standards on the eighth-grade ISAT was raised
said they always study for tests. The changes in atten-
in 2013, but it is still below the high school standard.
dance and study habits account for most of the decline
While the average eighth-grade student (the average
31
observed in students grades; statistical models show
math ISAT score is 267) met the eighth-grade math ISAT
that almost all of the gap in GPAs between eighth and
standards (based on 2013 standards as well as previous
ninth grade can be explained by students attendance
standards), the average tenth-grade student failed to meet
and study habits.
the PLAN benchmark at 19 (the average math PLAN score
32
Test Benchmarks Change, Making

It Look Like Test Performance is
Lower in High School
is 16). The ISAT-score benchmarks are provided for each

grade level in Appendix D. The PLAN exam is the middle
test in ACTs EPAS sequence and benchmark scores indicate that students are performing at a level that gives them
It can be difficult to compare students performance
a 50 percent chance of achieving a B in college math, if they
on standardized tests from eighth to ninth grade be-
continue to make normal progress through the remainder
cause the tests that students take in middle school are
of high school. EPAS-benchmark scores are also provided
different from the ones taken in high school. Figure 4
in Appendix D. It is most common for CPS students to
visually compares students performance, showing
have PLAN scores that are four to six points behind the
math scores for the same set of students and comparing
benchmark (in the 13 to 16 point range). Most students
the distribution of eighth-grade scores on the ISAT to
gain just over a point a year on the EPAS system; thus,
the distribution of scores they received on the PLAN
most students PLAN scores are about four to six years
at the beginning of tenth grade. The distributions are
of learning behind ACTs benchmark for the tenth grade
placed above and below each other so that the average
on the math PLAN exam. As is the case with math, few stu-
is at the same place and the range of scores, from the
dents reach the benchmarks in reading and science (with
fifth to the 95th percentile on both tests, is shown in the
benchmark scores of 17 and 21, respectively); it is more
figure. This allows us to compare the general distribu-
likely that students will reach the English benchmark on
tion of scores across all students who take both tests.
the PLAN, where the benchmark score is only 15. 34
33 State standards shown in Figure 4 reflect the previous
34 In September 2013, ACT revised the ACT college readiness
standards and the 2013 standards. Throughout the report

we will focus on the performance levels defined by the cut
scores in 2013, when the Illinois State Board of Education
raised the performance levels. Appendix D shows the range of
ISAT scores for each performance level for all grades before
and after 2013.
benchmarks with more recent data. This has resulted in

changes to the reading and science benchmarks; the reading
benchmark went up a point (from 17 to 18) and the science
benchmark went down a point (from 21 to 20) for all tests in
the EPAS series. For tests taken in 2013 and later, ACT will
apply these revised benchmarks.
25
FIGURE 4
17.0%
Test Score Standards and Score Distributions Are Very Different in Eighth Grade Than in High School
Meeting the Benchmark
15.1%
18
16
11.4%
12
10
11
0.3%
0.1%
0.8%
0.2%
0.8%
0.6%
1.1%
0.8%
2.0%
1.5%
13 14 15 16 17
2.0%
12
2.8%
1.7%
7
1.4%
5.2%
5.2%
1.2%
0.6%
0.1%
26
0.2%
0.0%
0.0%
3.5%
4.1%
2.2%
6.7%
8.4%
10
2.9%
Percent of Students
14
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
PLAN Math Score
Meeting State Standards (246) or Exceeding (288)
0.1%
402-413
390-401 0.0%
0.3%
366-377
378-389 0.0%
0.5%
342-353
354-365 0.0%
1.3%
1.7%
318-329
312-317
294-299
288-293
276-281
282-287
330-341
2.0%
2.5%
306-311
4.6%
4.1%
2.4%
5.9%
270-275
264-269
246-251
240-245
234-239
228-233
0.4%
216-227
204-215
0.1%
192-203
180-191 0.0%
168-179 0.0%
156-167 0.0%
144-155 0.0%
120-131 0.0%
132-143 0.0%
300-305
7.3%
9.2%
258-263
5.4%
4.4%
10
6.6%
9.2%
252-257
Meeting 2013 State Standards (267) or Exceeding (310)
12
3.6%
Percent of Students
14
9.3%
9.1%
16
10.2%
18
ISAT Math Score

Note: Test score distributions for students entering ninth grade in 2009-10 who were eighth-graders in 2008-09 and had data for both test scores. ISAT scores come
from the spring of 2009 when students were in eighth grade. PLAN scores come from the fall of 2010 when students were in tenth grade. The distributions are lined
up so that the average scores are in the same horizontal location as well as the 5th and 95th percentiles. The orange highlighted numbers in the X-axis represent the
average scores for this cohort of students: 16 for PLAN; 267 for ISAT.
In the top panel, no students scored an 8 on the PLAN, therefore there is no bar for that score category.
New ISAT Performance Levels

In 2013, the cut scores that define the ISAT
performance levels were raised to better align with
the more rigorous standards of the Common Core
State Standards. The old cut point defining meets
standards was 231 in reading and 246 in math in
eighth grade, while the exceeds standards was 278
in reading and 288 in math. The new cut points that
define meets standards went up to 248 in reading
and 267 in math. These cut points are very close to
the average scores in the district. The new cut points
that define exceeds standards went down to 271 in
reading and up to 310 for math. Prior to the change
in the cut points for meeting and exceeding state

standards, the exceeds standards point in reading
was much harder to reach compared to the math
exceeds standards point, with very few students
exceeding standards in reading while many more were
able to exceed standards in math in eighth grade.
Throughout this report, we focus on the
performance levels defined by the cut scores in 2013
when the Illinois State Board of Education raised the
performance levels.
Appendix D shows the performance levels before
and after 2013.
Summary
quarter of students end their ninth-grade year with
Students academic performance declines consider-
at least a B average, which means the vast majority
ably between the middle grades and high school. They
are not on-track to be likely to succeed in college. Test
are missing many more days of classwork due to a large
scores are difficult to compare across assessments.
increase in unexcused absences, and they are less likely
Because the standards are much higher in high school,
to exhibit strong study habits. Their grades decline by
however, most students are far away from meeting
half of a GPA point from eighth to ninth grade, and the
the expected scores for their grade level at the start of
decline is exactly what would be predicted based on the
tenth grade, while most were meeting state standards
change in attendance and study habits. Only about a
in eighth grade.
27
28
CHAPTER 3
Middle Grade Indicators of

High School Course Performance
Passing high school classes is an essential step toward
This chapter shows how well

students academic performance
in the middle grades predict their
grades in high school and which
combinations of indicators provide
the best predictions. It examines
predictions of ninth-grade ontrack rates and ninth-grade GPAs,
eleventh-grade progress to graduation and eleventh-grade GPAs,
and grades in ninth-grade English
and math classes.
graduating from high school. When students fail one

or more classes, they miss opportunities to accumulate
credits needed for graduation. Each semester F that
a student receives in ninth grade lowers their probability of graduating by about 20 percentage points
(Allensworth and Easton, 2007). But merely passing
classes is not sufficient for ensuring that students are
prepared to succeed in college. Students with Cs and Ds
in high school are very unlikely to succeed in college.
In fact, only those earning average grades of B or higher
(a 3.0 GPA) have close to a 50/50 chance of getting a
four-year college degree (Roderick et al., 2006; Bowen,
Chingos, and McPherson, 2009).
Unfortunately, many students receive very low
grades in their ninth-grade classes. Around one-third
(32 percent) of first-time ninth-graders in 2009 were
off-track by the end of ninth grade, meaning they failed
one or more core courses and did not accumulate
enough credits to be promoted to tenth grade. Only
Eighth-Grade Core GPA as a

Predictor of High School Outcomes
one-quarter of all first-time ninth-graders (23 percent)
Of all indicators tested, eighth-grade core GPA is the
had a GPA of 3.0 or higher by the end of the ninth grade.
best predictor of being on-track at the end of ninth
This sets the stage for poor grades throughout high
grade and of passing particular subjects. Each poten-
school, as shown later in this report.
tial middle grade indicator was examined to determine
Advocates for improving high school graduation
how well it could predict whether students were on-
and college enrollment rates often ask if it is possible
track in ninth grade, based on its correlation, variance
to intervene before high school and identify students
explained, and correct prediction rate. We also looked
in the middle grade years who are at risk of failure or
to see whether indicators had nonlinear relationships
of earning low grades in high school. In this chapter,
with on-track rates (where the indicator was more
we show the predictiveness of middle grade indicators
predictive at low or high levels), and whether combina-
for being on-track and earning As or Bs in ninth grade
tions of indicators provided much better predictions.
and whether a combination of indicators improve the
Table 3 provides a partial list of the indicators that we
prediction of each outcome. In addition, we look at how
examined; a complete list is provided in Appendix E .
school effects influence the predictiveness of eighth-
Out of a large array of indicators of middle grade
grade indicators for being on-track and earning high
performance that we examined as potential indicators,
gradeswhether the indicators work differently based
the strongest predictor of being on-track at the end
on what school a student attends. Finally, we examine
of ninth grade is students eighth-grade core GPA. 35
the predictiveness of middle grade indicators for being
However, the relationship is only moderately strong,
on-track and earning high grades in eleventh grade.
with a correlation of 0.43 and a pseudo-R 2 of 0.18.
Chapter 3 | Middle Grade Indicators of High School Course Performance
29
What Is the R-Squared Statistic?

The R-squared (R2) statistic describes the proportion
of variance explained in a linear regression model.
Stated differently, it is a measure of how well a predictor, or a set of predictors, explains variation in an outcome. When there is only one predictor, the R-squared
statistic is equal to the square of the correlation
between the predictor and the outcome. In this study,
we use R-squared statistics to tell us how well combinations of middle grade indicators predict high school
outcomes. An R-squared statistic can range from 0 to
1, with higher values indicating a better prediction; a
value of 0 means the indicator does not predict the
outcome at all, while a value of 1 means it produces an
exact prediction. Although there is no strict agreement
on what constitutes a strong or weak R-squared, we
used the following generally accepted guidelines in
this project. R-squares under 0.10 indicate a poor
prediction of the outcome; there is considerable
variation in the outcome that is not explained by the

indicators. This would be equivalent to a correlation of
about 0.3 or less, if just one indicator were used as a
predictor. R-squares between 0.11 and 0.30 indicate a
moderately good prediction for the outcome (equivalent to a correlation of about 0.3 to 0.5 for a single
indicator). R-squares above 0.30 indicate a very good
prediction for the outcome (equivalent to a correlation
of over 0.5 for a single indicator). An R-squared above
0.6 represents an extremely strong prediction (equivalent to a correlation of about 0.8 for a single indicator).
A prediction level that is extremely high means that
schools could very accurately predict students later
outcomes based on the indicators, which might suggest it is useful for targeting interventions. However,
if the prediction is extremely high, it also means that
the outcome might not be sufficiently malleable for
interventions to be effective.
30
Seventh-grade core GPA is similarly as predictive of
learning behaviorsacademic and noncognitive skills
being on-track in ninth grade as eighth-grade core GPA;
rather than skills or interest in a particular subject. A
the correlation is 0.41 and the pseudo-R for is 0.17.

Eighth-grade math and English grades are each
slightly less predictive of ninth-grade on-track status
than core GPA in either seventh or eighth grade (Table
students grade in a single class contains less information

about these learning behaviors and skills than grades
across multiple classes, measured by core GPA.
After grades, eighth-grade attendance is the next most
E1 in Appendix E). It is not surprising that core GPA,
predictive indicator of passing ninth-grade classes (the
a measure of students overall performance in eighth-
pseudo-R 2 is 0.12 and correlation is 0.35) followed by the
grade core classes, is more predictive of on-track, also a
number of core course failures in eighth grade (the pseu-
measure of overall performance, than subject-specific
do-R 2 is 0.10 and the correlation is -0.33), and subject-
GPAs. But, core GPA is also more predictive of whether stu-
specific ISAT scores (the pseudo-R 2 between math ISAT
dents pass their English or math classes and of their overall
and on-track is 0.08; the pseudo-R 2 between reading ISAT
grades in both math and English than their prior grades in
and on-track is 0.05). 36, 37 Combining math and reading
either subject. (Table E1 in Appendix E shows the rela-
ISAT scores does not improve the prediction of on-track
tionship of middle grade indicators with the probability
rates. None of the other test-based metrics, such as growth
of passing particular subjectsEnglish and mathand
in test scores over the middle grade years or relative rank
with earning high grades in those subjects.) This sug-
on test scores compared to school peers, were even moder-
gests that course performance depends more on general
ately predictive of passing ninth-grade classes.
35 Core GPA is calculated from the grades students receive
37 We also examined ISAT subtest scores to determine if particu-
across all core courses (i.e., math, English, science, and

social science).
36 Including a squared term for attendance improved the
prediction of on-track rates slightly; the pseudo-R2 increases
from 0.12 to 13. Including squared terms for GPA, as well as
for reading and math ISAT scores, did not improve the
predictiveness of these indicators.
lar skills in math or reading were especially important to pass

ninth-grade classes. (In math, subtests include algebra, analysis,
geometry, measuring, and number sense. In reading, subtests
include composition, literature, strategy, and vocabulary.)
However, these subtests are no more predictive of ninth-grade
pass rates than their overall test score in each subject. The
overall score provides a more reliable estimate of students
skills as it is determined by a larger number of test items.
TABLE 3
Relationships of On-Track and Earning High Grades in Ninth Grade with Selected Middle Grade Variables
See Appendix E for the relationships of ninth-grade on-track and grades with more middle grade variables
and combinations of variables, and with English and math grades.
On-Track at the End of 9th Grade

Single Indicator
Correlation
Pseudo-R
Earning As or Bs in 9th-Grade Core Classes
% Correct
Correlation
Pseudo-R 2
% Correct
8th-Grade Core GPA

with Squared Term
0.43
0.18
0.18
0.74
0.74
0.44
0.21
0.21
0.81
0.81
8th-Grade Math GPA
0.38
0.14
0.72
0.38
0.15
0.79
8th-Grade English GPA
0.40
0.15
0.73
0.40
0.18
0.80
-0.33
0.10
0.72
-0.22
0.08
0.77
8th-Grade Attendance
with Squared Term
0.35
0.12
0.13
0.73
0.73
0.23
0.08
0.09
0.77
0.77
8th-Grade Math ISAT

with Squared Term
0.27
0.08
0.08
0.68
0.68
0.35
0.11
0.11
0.78
0.78
8th-Grade Reading ISAT

with Squared Term
0.23
0.05
0.06
0.68
0.68
0.29
0.09
0.09
0.78
0.78
8th-Grade Suspensions
-0.24
0.06
0.71
-0.14
0.04
0.77
8th-Grade Misconduct
-0.16
0.03
0.70
-0.09
0.02
0.77
8th-Grade Grit 1
0.08
0.01
0.72
0.06
0.00
0.74
8th-Grade Study Habits1
0.13
0.02
0.72
0.17
0.03
0.74
7th-Grade Core GPA
0.41
0.17
0.73
0.42
0.18
0.80
0.29
0.08
0.71
0.21
0.07
0.77
7th-Grade Math ISAT
0.25
0.07
0.69
0.33
0.10
0.78
0.22
0.05
0.68
0.28
0.08
0.78
8th-Grade Core GPA +

Attendance
0.21
0.75
0.22
0.81

Math ISAT
0.18
0.74
0.21
0.81
0.08
0.68
0.12
0.79
8th-Grade Core
Course Failures
Combining Two 8th-Grade Indicators
8th-Grade Math +
Reading ISAT
Combining Three or More 8th-Grade Indicators

Math ISAT + Reading ISAT
0.18
0.74
0.21
0.81

Attendance + Math ISAT
+ Reading ISAT
0.21
0.75
0.22
0.81
Background
Characteristics 2
0.06
0.70
0.09
0.78
8th-Grade Core
GPA + Attendance +
Math ISAT + Reading
ISAT + Background
Characteristics
0.21
0.76
0.24
0.81

+ Reading ISAT + Course
Failures + Suspensions +
Misconduct + Background
Characteristics
0.21
0.76
0.24
0.81
31
TABLE 3: CONTINUED
Relationships of On-Track and Earning High Grades in Ninth Grade with Selected Middle Grade Variables
See Appendix E for the relationships of ninth-grade on-track and grades with more middle grade variables
and combinations of variables, and with English and math grades.

Single Indicator
Correlation
Pseudo-R 2
% Correct
8th-Grade Core GPA

+ Attendance + Math
ISAT + Reading ISAT +
Seventh-Grade GPA
0.23
All 7th- and 8th-Grade

Indicators
Earning As or Bs in 9th-Grade Core Classes

Correlation
Pseudo-R 2
% Correct
0.76
0.23
0.82
0.22
0.77
0.25
0.82
0.33
0.79
0.34
0.84
Adding in 7th-Grade Indicators
Adding in School Effects

Indicators + School
Effects
32
Notes: 1) Grit and study habits are calculated from students responses to items on UChicago CCSRs annual survey of CPS students. The elementary/middle
grade student survey had a response rate of 59 percent in 2009. Because not all students answer the survey, the sample size for these models is smaller than
the sample sizes for the other models included in this table. The percent of students who are on-track or earn high grades is slightly higher in this smaller
sample resulting in a somewhat higher correct prediction rate for these two variables. 2) Background characteristics include race, gender, special education
status, neighborhood poverty level, and socioeconomic status, free reduced price lunch status, and whether a student was older than 14 when entering high
school. 3) The bolded numbers represent the best indicator or combination of indicators in each group.
Noncognitive skills in eighth grade, including grit

and study habits, are not very predictive of whether
students are on-track at the end of ninth grade. The
2
pseudo-R between study habits and on-track is only

2
into high school when students have different teachers

and a very different environment.
Prior research in Philadelphia found that serious
behavior infractions and suspensions in middle
0.02; the pseudo-R between grit and passing math is
school were related to later high school dropout,
zero. Although study habits and grit have been shown
(Balfanz, Herzog, and MacIver, 2007), but we did
to be predictive of course performance during the same
not find these relationships to be strong in Chicago.
semester,
The relationships were modest whether we examined
38
we do not find that they are very predictive
of course performance one year later. This is consistent
infractions and suspensions in fifth or in eighth grade.
with research in other places showing weaker longitudi-
The differences in prediction power between Chicago
nal relationships between grit and subsequent achieve-
and Philadelphia may be due to differences in disci-
ment than between grit and achievement measured
plinary practices, or record keeping around discipline,
at the same time.
in the different cities. It is possible that there is less
39
This may be because students
demonstration of high levels of perseverance, as well
consistency in disciplinary practices across schools in
as the demonstration of academic behaviors, is often
Chicago than in Philadelphia, leading it to be an unreli-
context specific.
able predictor.
40
Students can show a great deal of
perseverance in one area or context, and very little in
Students background characteristics were much less
another area (e.g., getting homework done but giving up
predictive of ninth grade on-track status than students
on the mile run, showing up on time for one class but
academic performance in the middle grades. Even when
not another). Because the transition from middle school
we combine many background factors together (e.g.,
to high school involves moving from one kind of school
race, gender, special education status, neighborhood
context to a very different kind, it may be that the dem-
poverty level, socioeconomic status, free or reduced
onstration of some noncognitive factors, such as study
priced lunch status, and age when entering high school),
habits and perseverance, do not necessarily carry over
the prediction was less predictive (combined pseudo-R 2
FIGURE 5
Probability of Being On-Track in 9th Grade
On-Track Rates by Eighth-Grade Core GPA and Attendance

100
90
80
70
95.9%
Average Attendance by 8th-Grade Core GPA

N
Low Att
Avg Att High Att
3.0 GPA < 4.0
2.0 GPA < 3.0
1.0 GPA < 2.0
0 GPA < 1.0
6396
8468
4139
960
94%
90%
85%
73%
60
97%
95%
92%
85%
79.7%
100%
100%
99%
98%
85.4%
58.0%
50
40
57.2%
33.7%
30
32.2%
20
10
0
13.8%
0 GPA < 1.0
1.0 GPA < 2.0
2.0 GPA < 3.0
3.0 GPA < 4.0
8th-Grade Core GPA

High Attendance
Average Attendance
Low Attendance
Note: These figures show students predicted probability of being on-track based on their eighth-grade core GPA and either their attendance (see Figure 7) or math
ISAT scores (see Figure 8). The orange squares in each figure show the predicted probability of being on-track for students with the same core GPA who have average
attendance (or average ISAT scores). The gray diamonds show the predicted probability for students with the same eighth-grade core GPA who have above-average
attendance (or ISAT scores). The black diamonds represent the predicted probability for students with below-average attendance (or ISAT scores). The table inside
each figure shows how high, average, and low attendance or ISAT scores were defined for each level of core GPA.
=0.06) than when using any one of the middle grade

academic indicators alone or in combination (eighthgrade GPA, test scores, or attendance).
0.18 to 0.21 when attendance is used in addition to core

GPA to predict ninth-grade on-track. Attendance also
improves the prediction of who will pass their ninthgrade English and math classes than GPA alone (see
Eighth-grade GPA combined with attendance pro-
Appendix E). Figure 5 shows ninth-grade on-track rates
vides a better prediction of who will be on-track than
by students core GPA and attendance rates in eighth
either alone; adding other indicators only marginally
grade. 42, 43 At each GPA level less than 3.0, ninth-grade
improves the prediction. Combining eighth-grade core
GPA with eighth-grade attendance gives a better prediction of who will be on-track at the end of ninth grade
than core GPA alone. 41 The pseudo-R 2 increases from
38 Allensworth and Easton (2007); Duckworth et al. (2007);

Duckworth and Seligman (2005, 2006).
39 The correlations found between grit and students grades in
studies where both were measured concurrently are higher
than those where grit is measured in the prior semester, and
the prior year. See Duckworth et al. (2007); Duckworth and
Seligman (2005, 2006).
40 See Farrington et al. (2012) for a discussion of noncognitive
factors that affect students course grades.
41 The pseudo-R2 columns in Table 3 can be used to compare
predictions using multiple indicators.
42 Figures 5 and 6 are based on statistical models in which
core GPA and either attendance or ISAT scores are used to
model the probability of being on-track at the end of ninth
grade. We used the distribution of students at each level of
core GPA to determine the high (1 standard deviation above
on-track rates are about 20 to 30 percentage points

higher for students with high attendance rates in eighth
grade than students with the same eighth-grade GPA
but low attendance. For example, among students with
mean), average, and low (1 standard deviation below the mean)

attendance or ISAT scores were. Using these values and also
coefficients from the model, we calculated the probability of
passing for three groups of students at each level of core GPA.
43 Because attendance and grades are related to each other,
high and low attendance are not defined in the same way at
each GPA level. No students with high grades have very low
attendance because low attendance interferes with getting
good grades. The inset table in each figure shows what level of
attendance is low, high, or average for students with particular
GPAs. Low attendance is defined as one standard below the
mean for students at each level of GPA; high attendance is one
standard deviation above the mean. For students with above a
3.0 GPA, 94 percent attendance is low. Yet, 92 percent attendance is typical for students with a GPA between 1.0 and 2.0;
for these students, low attendance is defined as 85 percent.
33
FIGURE 6
Probability of Being On-Track in 9th Grade
On-Track Rates by Eighth-Grade Core GPA and Math ISAT Scores

100
90
80
70
94.7%
Average ISAT Math Score by 8th-Grade Core GPA

N
Low ISAT Avg ISAT High ISAT
3.0 GPA < 4.0
2.0 GPA < 3.0
1.0 GPA < 2.0
0 GPA < 1.0
6396
8468
4139
960
257
238
232
229
283
259
250
245
310
280
268
261
71.7%
60
65.2%
48.3%
50
86.6%
40
30
20
25.8%
41.7%
21.5%
10
0
0 GPA < 1.0
1.0 GPA < 2.0
2.0 GPA < 3.0
3.0 GPA < 4.0
8th-Grade Core GPA

High ISAT
34
Average ISAT
Low ISAT
Note: These figures show students predicted probability of being on-track based on their eighth-grade core GPA and either their attendance (see Figure 7) or math
ISAT scores (see Figure 8). The orange squares in each figure show the predicted probability of being on-track for students with the same core GPA who have average
attendance (or average ISAT scores). The gray diamonds show the predicted probability for students with the same eighth-grade core GPA who have above-average
attendance (or ISAT scores). The black diamonds represent the predicted probability for students with below-average attendance (or ISAT scores). The table inside
each figure shows how high, average, and low attendance or ISAT scores were defined for each level of core GPA.
GPAs between 1.0 and 2.0, those with strong attendance
student background characteristics, with eighth-grade
have a 58 percent likelihood of passing their ninth-
core GPA and attendance does not improve the predic-
grade math class, compared to a 32 percent likelihood
tion of on-track rates beyond what is achieved from only
for students with the same eighth-grade GPA but poor
using core GPA and attendance. However, the prediction
eighth-grade attendance.
can be improved somewhat by combining seventh- and
Combining math ISAT scores with core GPA does
eighth-grade GPA and eighth-grade attendance; the
not improve the prediction of who will be on-track at the
pseudo-R 2 increases to 0.23. But the improvement in the
end of ninth grade compared to GPA alone (see Table 3).
prediction that comes from using three indicators is not
As Figure 6 shows, there is very little difference in
that much higher than the prediction that comes from
on-track rates between students with high test scores
using only two (eighth-grade core GPA and attendance).
and students with low test scores at each level of core
Adding in seventh-grade data also requires an extra year
GPA. Furthermore, the differences that do exist disap-
of data. The improvement in prediction may not warrant
pear when we incorporate attendance in the model, or if
the increased complication in the indicator system.
we use more nuanced versions of GPA (rather than 1-point
Figure 7 graphically shows the degree to which
differences). Once we know the grades and attendance of
individual predictors and combinations of predictors
students in middle grade, their test scores provide almost
can identify students who are at risk of being off-track
no additional information about whether they will be on-
by the end of ninth grade (Y-axis), while also showing
track. At the same time, grades and attendance improve
whether predictors are falsely identifying students as
the prediction of who is on-track among students with
off-track (e.g., identifying students who are actually on-
similar test scoressee Appendix C for figures that show
track as being off-track, X-axis). A predictor that is able
the contribution of grades and attendance to predicting
to perfectly identify everyone who is off-track, while
on-track rates among students with similar test scores.
never falsely identifying anyone as being on-track,
Combining other eighth-grade predictors, including
would appear in the upper left corner of the figure. ISAT
FIGURE 7
Sensitivity and Specificity of Eighth-Grade Indicators for Being On-Track at the End of Ninth Grade
Risk of Being Off-Track in Ninth Grade
Proportion of Students Correctly Identified as Off-Track

True Positive Proportion (Sensitivity)
1.0
Perfect
Prediction
0.9
0.8
Ra
nd
om
Gu
es
0.7
Core GPA + Attendance + School Effects

0.6
Core GPA + Attendance + ISAT + School Effects
0.5
Core GPA + Attendance (with and

without ISAT scores)
0.4
Core GPA
English GPA
0.3
Attendance
Math GPA
0.2
Suspensions
ISAT Math and Reading
0.1
Background Characteristics
Grit, Study Habits
0.0
0.0
0.1
0.2
0.3
35
Worse
Prediction
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Proportion of Students Incorrectly Identified as Off-Track False-Positive Proportion (1-Specificity)
math and reading scores are in the bottom left of the
the best predictor of earning As or Bs in specific subjects,
chartthey only correctly identify about 10 percent of
more so than students grade in that particular subject
off-track students, but they do not misidentify many as
(ninth-grade English and math classes; see Table E1 in
off-track who are not. Only a small number of students
Appendix E). Although core GPA is somewhat more
would be identified as at risk of being off-track using
predictive of earning high grades than being on-track, the
test scores alone. Attendance is much better, identify-
relationship between core GPA and high grades is still
ing 30 percent of off-track students; eighth-grade core
only moderately strong. Seventh-grade core GPA is slight-
GPA is the best single predictor, identifying almost
ly less predictive than eighth-grade core GPA of earning
40 percent of off-track students. GPA and attendance
high grades (pseudo-R 2 is 0.18), as are math and English
together identify 44 percent of off-track students,
GPAs (pseudo-R 2 s are 0.15 and 0.18, respectively).
without misidentifying many as off-track who are not.
After middle school grades, ISAT scores in seventh
Adding together all information about students prior to
and eighth grades are the next-best predictors of earn-
high school does not substantially improve the predic-
ing As or Bs in ninth grade. Eighth-grade math ISAT
tion beyond just using core GPA and attendance.
scores are associated with high grades with a pseudoR 2 of 0.11, which is stronger than the relationship with
Eighth-grade core GPA is also the best predictor of
on-track (pseudo-R 2 is 0.08). Combining eighth-grade
earning high grades, followed by test scores and
math and reading ISAT scores slightly improves the
attendance. The best indicators of students readiness
prediction of earning high grades (pseudo-R 2 of 0.12).
to excel in high school classes are similar to those pre-
Seventh-grade math ISAT scores are associated with
dicting the likelihood that students will pass their high
earning high grades with a pseudo-R 2 of 0.10. Math
school classes. Eighth-grade core GPA is the strongest
ISAT scores are a better predictor of high school grades
predictor of whether students earn As or Bs in ninth
than reading ISAT scores; they are much more predic-
grade (pseudo-R is 0.21). Eighth-grade core GPA is also
tive of math course grades and are even slightly better
at predicting English course grades (see Appendix E).
As or Bs than students with 94 percent attendance
Attendance and also the number of courses failed in

eighth grade are somewhat less predictive of earning
(see Figure 8). Among students who earn less than a

2.0, there is no difference between students with high
high grades than ISAT scores, with pseudo-R scores
versus low middle grade attendance in their likelihood
of 0.08 each. Other eighth-grade metrics, including
of earning high grades in ninth grade.
suspensions, grit, and study habits, are either only
Combining eighth-grade math ISAT scores with
weakly or not at all associated with earning high
eighth-grade core GPA also differentiates among
grades in high school.
students who earned a 3.0 or better, but does not

differentiate among students who earned less than
a 2.0 in eighth grade. For example, eighth-graders with
of earning As or Bs beyond core GPA, but only
at least a 3.0 and high ISAT scores (310) are around
among students with strong middle school grades.
20 percentage points more likely to earn As or Bs than
Combinations of indicators only modestly improve the
students with the same grades but low ISAT scores (257)
prediction of who earns As or Bs in high school beyond
(see Figure 9). But among students with less than a 2.0
eighth-grade core GPA. This is because the improve-
GPA in high school, higher ISAT scores increase the
ment in prediction is only evident among high-achiev-
probability of high grades by less than three percentage
ing students. For example, combining eighth-grade
points. While test scores do not strongly differentiate
attendance with core GPA differentiates substantially
who will earn high ninth-grade grades among students
among students who earn a 3.0 or better in eighth grade
with the same middle school GPAs, middle school GPAs
but does not differentiate at all among students who
do strongly differentiate who earns high ninth-grade
earn less than a 2.0. Among eighth-graders with a GPA
grades among students with the same test scores. These
of at least 3.0, students with perfect attendance (100
differences, along with the combination of test scores
percent) are 20 percentage points more likely to earn
with attendance, can be found in Appendix C .
FIGURE 8
Probability of Earning As or Bs by Eighth-Grade Core GPA and Attendance

100
Probability of Earning As or Bs in
9th-Grade Core Class
36
Attendance or ISAT scores improve the prediction
90
80
70
Average Attendance by 8th-Grade Core GPA

N
Low Att
Avg Att High Att
3.0 GPA < 4.0
2.0 GPA < 3.0
1.0 GPA < 2.0
0 GPA < 1.0
6396
8468
4139
960
94%
90%
85%
73%
97%
95%
92%
85%
100%
100%
99%
98%
58.5%
60
50
40
38.1%
30
20.5%
20
6.5%
10
0
9.9%
2.9%
0 GPA < 1.0
1.0 GPA < 2.0
2.0 GPA < 3.0
3.0 GPA < 4.0
8th-Grade Core GPA

High Attendance
Average Attendance
Low Attendance
Note: These figures show students predicted probability of earning high grades based on their eighth-grade core GPA and their attendance or math ISAT scores (see
Figure 9). The orange squares in each figure show the predicted probability of earning high grades for students with the same core GPA who have average attendance
(or average ISAT scores). The gray diamonds show the predicted probability for students with the same eighth-grade core GPA who have above-average attendance
(or ISAT scores). The black diamonds represent the predicted probability for students with below-average attendance (or ISAT scores). The table inside each figure
shows how high, average, and low attendance or ISAT scores were defined for each level of core GPA.
The prediction of who earns high grades can be
of roughly 90 percent. A model using eighth-grade core
improved further if background characteristics or
GPA as a predictor improves that proportion to roughly
seventh-grade GPA are also included as predictors along
60 percent and is the best single indicator to predict the
with core GPA and attendance (pseudo-R increases
likelihood that a student will earn a GPA of less than 3.0
to 0.24 by adding in background and to 0.23 by add-
in ninth grade. Adding eighth-grade attendance rates
ing in seventh-grade core GPA. See Table 3 on p.31.).
and/or ISAT scores, as well as, core GPA reduces the
However, the improvement in prediction may not war-
proportion of false positives slightly. Adding all other
rant the increased difficulty of bringing in data from a
potential indicators only marginally further improves
prior grade. Using all seventh- and eighth-grade indica-
the prediction.
tors together yields only a slight further improvement

in prediction (pseudo-R 2 is 0.25).
Students with the same middle grade performance
Figure 10 graphically compares the predictive ability
have different probabilities of being on-track or earn-
of individual indicators, and their combination. Many
ing high grades in high school, depending on which
of the indicators can correctly identify over 90 percent
schools they attend. Often, there is a perception that the
of students who earn a GPA of less than 3.0, but these
same academic records may indicate different levels of
predictions come with a relatively high rate of false
performance if students come from one school versus an-
positives. The fact that the majority of ninth-graders are
other. For students with middle grade performance that
not earning a GPA of at least 3.0 makes a true-positive
is either very high or very low, middle grade performance
prediction an easy task; the difficulty is in reducing
predicts similar levels of success regardless of where they
the proportion of false-positive predictions (i.e., the
attended the middle grades. Students with particularly
proportion of students incorrectly identified as earn-
weak eighth-grade performance (i.e., GPA of 1.0 or below
ing low grades). A model using eighth-grade attendance
and attendance of 80 percent or below) are unlikely to ei-
as a predictor suffers from a false-positive proportion
ther to be on-track or to earn high grades in ninth grade,
FIGURE 9
Probability of Earning As or Bs by Eighth-Grade Core GPA and Math ISAT Scores
Probability of Earning As or Bs in
9th Grade Core Class
100
90
80
70
Average ISAT Math Score by 8th-Grade Core GPA

N
Low ISAT Avg ISAT High ISAT
3.0 GPA < 4.0
2.0 GPA < 3.0
1.0 GPA < 2.0
0 GPA < 1.0
6396
8468
4139
960
257
238
232
229
283
259
250
245
310
280
268
261
59.1%
60
50
40
37.7%
30
19.4%
20
6.0%
10
0
11.0%
3.4%
0 GPA < 1.0
1.0 GPA < 2.0
2.0 GPA < 3.0
3.0 GPA < 4.0
8th-Grade Core GPA

High ISAT
Average ISAT
Low ISAT
Note: These figures show students predicted probability of earning high grades based on their eighth-grade core GPA and either their attendance (see Figure 8) or
their math ISAT scores. The orange squares in each figure show the predicted probability of earning high grades for students with the same core GPA who have
average attendance (or average ISAT scores). The gray diamonds show the predicted probability for students with the same eighth-grade core GPA who have
above-average attendance (or ISAT scores). The black diamonds represent the predicted probability for students with below-average attendance (or ISAT scores).
The table inside each figure shows how high, average, and low attendance or ISAT scores were defined for each level of core GPA.
37
FIGURE 10
38
Proportion of Students Correctly Identified as Earning a

GPA of Less Than 3.0 True Positive Proportion (Sensitivity)
Sensitivity and Specificity of Eighth-Grade Indicators for Earning As or Bs in Ninth Grade

Risk of Low Grades in Ninth-Grade Core Classes
1.0
Perfect
Prediction
0.9
Core GPA + Attendance + School Effects

Core GPA + Attendance +
ISAT + School Effects
0.8
Math GPA
English GPA
Core GPA + Attendance

(with and without ISAT scores)
Core GPA
0.7
Grit, Study Habits
Attendance
ISAT Math and Reading
0.6
0.5
0.4
0.3
Ra
0.2
nd
om
Gu
es
0.1
Worse
Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Proportion of Students Incorrectly Identified as Earning a GPA of Less Than 3.0

regardless of where they earned those low eighth-grade
For students with a moderate chance of being on-track,
grades. (See Figure 11, left panel; and Figure 12, left
based on eighth-grade indicators, their actual on-track
panel.)
rates can range from 41 to 66 percent, a difference of
44
Students with strong eighth-grade perfor-
mance (i.e., GPA 3.0 and above; attendance of 97 percent
25 percentage points, depending on where they went
or better) are very unlikely to be off-track, regardless of
to middle school, net of any high school effects. For
where they attended middle school (see Figure 11, left
students with a moderate chance of earning high
panel). They are also very likely to earn high grades in
grades, their chances of earning high grades can vary
high school, regardless of which school they attended for
from 47 to 72 percent, which is also a difference of 25
the middle grades. (See Appendix C for the methods of
percentage points, depending on the middle schools
determining differences in outcomes by school.)
they attended.
For everyone else, the likelihood of being on-track
The middle grade schools where students perform
or earning high grades is different depending on which
better than would be expected in ninth grade tend to
middle school they attended, comparing students with
be schools that are generally higher-performing, and
similar middle grade GPAs, attendance, and test scores.
are serving more economically advantaged students.
44 How we calculated school effects: We first calculated each

students overall likelihood of being on-track using a logistic
regression equation in which the outcome was modeled as a
function of students eighth-grade core GPA, attendance, and
ISAT scores. The predicted value from this model was entered
as a student-level predictor into cross-nested models, with
observations simultaneously nested within students middle
school and their high school, predicting whether students
were on-track or earning high grades in ninth grade. These
models produce estimates of school effects on on-track rates
for each middle and high school, net of the effects of the
other school the student attended. The variance components
from these models were used to determine school effects,
net of students individual qualifications and any effect of the
other school the student attended (middle school effects net
of high school effects and vice-versa). Figures 8 and 9 graphs
school effects that were one standard deviation above and
below the mean. The same procedure was used to determine
school differences in the probability of earning As or Bs. More
details on the models are provided in Appendix C.
FIGURE 11
Middle and High School Effects on the Probability of Being On-Track in Ninth Grade
Middle School Variation in On-Track
100
High School Variation in On-Track
95%
98%
Probability of Being On-Track

in 9th Grade (Percent)
90
75%
88%
80
66%
70
77%
60
50
43%
40
34%
30
41%
31%
20
10
0
12%
9%
16%
3%
Very Low
(p < 25%)
3%
Low
Moderate
(25% p < 50%) (50% p < 75%)
High
(p 75%)
Very Low
(p < 25%)
11%
Low
Moderate
(25% p < 50%) (50% p < 75%)
High
(p 75%)
Probability of Being On-Track Based on Middle-Grade Indicators

Above Average School
Average School
39
Below Average School
Note: These figures show the likelihood of being on-track (see Figure 9) and earning high grades (see Figure 10), depending on which elementary/middle school (left
panel) or high
FIGURE
12school (right panel) students attend. The figures are based on cross-nested models in which students are nested in their elementary/middle school and
also in their high school; see Appendix C for additional details. The orange squares in each graph represent the average predicted probability of being on-track or
Middle
and
High
School
Effects
on the
Probability
ofand
Earning
Asscores.
or Bs
Ninth
Grade
earning high
grades
based
on students
eighth-grade
core
GPA, attendance,
math ISAT
Thein
gray
diamonds
above the orange squares represent the
predicted probability for students with similar levels of eighth-grade preparation attending schools with strong positive effects on the probability of being on-track or
earning high grades.
The black
diamonds
below the orange
squares represent
the predicted probability
for students
with similar
of eighth
grade
Middle
School
Variation
in Earning
As or Bs
High School
Variation
inlevels
Earning
As
or preparation
Bs
attending schools with strong negative effects.
Probability of Earning As or Bs
in 9th- Grade Core Classes (Percent)
100
95%
94%
90
77%
80
72%
82%
79%
70
60
51%
50
44%
47%
40
40%
30
20
10
0
21%
5%
6%
2%
Very Low
(p < 25%)
17%
2%
Low
Moderate
(25% p < 50%) (50% p < 75%)
High
(p 75%)
Very Low
(p < 25%)
Low
Moderate
(25% p < 50%) (50% p < 75%)
High
(p 75%)
Probability of Earning As or Bs Based on Middle-Grade Indicators

Above Average School
Average School
Below Average School
Note: These figures show the likelihood of being on-track (see Figure 11) and earning high grades (see Figure 12), depending on which elementary/middle school (left
panel) or high school (right panel) students attend. The figures are based on cross-nested models in which students are nested in their elementary/middle school and
also in their high school; see Appendix C for additional details. The orange squares in each graph represent the average predicted probability of being on-track or
earning high grades based on students eighth-grade core GPA, attendance, and math ISAT scores. The gray diamonds above the orange squares represent the
predicted probability for students with similar levels of eighth-grade preparation attending schools with strong positive effects on the probability of being on-track or
earning high grades. The black diamonds below the orange squares represent the predicted probability for students with similar levels of eighth-grade preparation
attending schools with strong negative effects.
TABLE 4
Correlations Between School Characteristics and School-Level Residuals from Models Predicting Ninth-Grade
On-Track and Earning As or Bs

9th-Grade On-Track
Column A
Column B
Column C
Column D
Column E
Column F
Correlations
with Middle
School
Residual
Incorporating
High School
Effects
Correlations
with High
School
Residual
Incorporating
Middle School
Effects
Correlations
with Middle
School
Residual
without
Incorporating
High School
Effects
Correlations
with Middle
School
Residual
Incorporating
High School
Effects
Correlations
with High
School
Residual
Incorporating
Middle School
Effects
Correlations
with Middle
School
Residual
without
Incorporating
High School
Effects
0.21***
-0.26***
-0.03***
0.27***
-0.65***
0.09***
ISAT Reading
0.23***
-0.26***
-0.01
0.30***
-0.65***
0.09***
% Latino
0.04***
-0.15***
-0.11***
0.16***
0.012
0.22***
% African
American
-0.16***
0.24***
0.08***
-0.31***
0.20***
0.28***
ISAT Math
% White
0.23***
-0.24***
0.34***
-0.49***
0.16***
-0.18***
0.24***
0.04***
-0.23***
0.56***
-0.08***
0.00
0.07***
0.05***
-0.04***
0.49***
-0.02**
Average
Concentration
of Poverty
-0.23***
0.29***
0.05***
-0.29***
0.45***
-0.14***
Average SES
0.09***
-0.03***
0.08***
-0.02*
0.31***
-0.18***
General
-0.13***
0.21***
-0.07***
-0.09***
0.42***
-0.01
Magnet
0.13***
-0.19***
0.05***
0.10***
-0.51***
0.01
Vocational
NA
-0.13***
NA
NA
-0.05***
NA
APC
NA
0.16***
NA
NA
0.12***
NA
% Over Age
40
Earning As and Bs in 9th-Grade Core Classes
Percent Special
Education
-0.01
Note: To generate the correlations above we ran four separate analyses. Two analyses modeled the probability of being on-track and two modeled the probability
of earning high grades. For each outcome, we first modeled the probability of that outcome (either being on-track or earning high grades) using a cross-nested
model in which students were nested in their middle grade school and in their high school. Columns A, B, D, and E are correlations between the school level residuals from these models and school characteristics. We then modeled the probability of each outcome using a hierarchical model in which students were only
nested only in their middle grade school. Columns C and F report the correlations between school level residuals and school characteristics from these two models.
Among students with the same eighth-grade GPAs, atten-
support, social capital, or other types of influence that
dance, and test scores, those who went to a high-achiev-
help students be more successful in their classes the
ing middle school (with higher average ISAT scores) with
next year. It is also possible that high-achieving middle
fewer students living in poverty are more likely to be on-
schools do a better job preparing students for the respon-
track and earn high grades in ninth grade than students
sibilities and demands of high school in ways that are not
who went to a low-achieving middle grade school, with
reflected on tests, such as teaching students how to man-
many students living in poverty (see Table 4, columns
age their workload or write term papers.
A and D). Going to a higher-performing, economically
While students probability of being on-track or of
advantaged middle school seems to confer some benefits
earning high grades depends on which middle school
that are not picked up through grades, attendance, and
they attended, the school effects do not outweigh the
test scores. It is possible that grading standards may be
predictiveness of the indicators themselves. Grades
higher at high-achieving middle schools than low-achiev-
and attendance during the middle grades are important
ing middle schools, so that an A from a high-achieving
regardless of which school students attend; strong
middle school might mean stronger course performance
grades and attendance at a low-performing middle
than an A from a low-achieving middle school. Or it may
school suggest a higher degree of readiness than average
be that peers in high-achieving middle schools provide
grades and attendance at a high-performing school.
Figure 13 shows the extent to which there are
2.2 GPA in ninth grade, net of differences attributable
differences in the expected return from students
to high schools. The difference between a B average
eighth-grade GPAs, in terms of their ninth-grade
at one middle school and a B average at another middle
GPAs, based on which middle school they attended.
school is smaller than the difference between a B
We use the term return to describe the idea that
average and a C average at any school, in terms of
similar students with the same eighth-grade GPAs
students performance in their high school classes. For
from different middle schools can earn different
example, while a student with a B average at a low-
GPAs in ninth-grade; thus one student receives a
return middle school is predicted to earn a 2.2 GPA in
higher return on his eighth-grade GPA than the other
ninth grade, a student with a C average from a high-
student. On average, students with a 3.0 eighth-grade
return middle school is predicted to earn only a 1.9 GPA
GPA (a B average) earn a 2.5 GPA in ninth grade.
in ninth grade. While there may be differences in
However, students with a 3.0 eighth-grade GPA from
exactly what that B means, the fact that a student has
high-return middle schools tend to earn a 2.9 GPA
earned a B averageand not an A or C averageis a
in ninth grade, while students with a 3.0 eighth-grade
better indicator of their success than knowing from
GPA from low-return middle schools tend to earn a
which school the B average was earned. 45
FIGURE 13
Students Predicted Ninth-Grade GPA by Students Eighth-Grade GPA: Differences Attributable to their Middle School
Net of Differences Attributable to High Schools
Predicted 9th-Grade Core GPA
4.0
3.3
3.5
2.9
3.0
2.4
2.5
2.6
1.9
2.0
2.2
1.5
1.3
1.5
1.6
1.0
1.3
1.0
1.0
0.5
0
0.7
0.4
0.5
1.0
1.5
2.0
2.5
3.0
3.5
8th-Grade Core GPA

Middle Schools with:
High-Return GPAs
Average-Return GPAs
Low-Return GPAs
Note: The average-return line represents the average relationship between eighth-grade core GPA and ninth-grade core GPA. For example, students with a 2.5 GPA in eighth
grade earn a 2.0 GPA in ninth grade, on average. Students with the same eighth grade GPA who attend the same high school earn different ninth grade GPAs, on average,
depending on which middle school they attended, as represented by the triangles and diamonds. For example, eighth-graders with a 2.5 GPA from high-return middle schools
will earn ninth grade GPAs of 2.4, on average, while eighth-graders with a 2.5 GPA from low-return middle schools will earn ninth grade GPAs of 1.6, on average.
To calculate these numbers, we ran models which cross-nested students simultaneously in their middle school and high school. A series of dummy variables representing each
of the seven eighth-grade core GPA categories was used to predict ninth-grade core GPA, with students included in a category if their GPA was within 0.25 GPA points of the
category value. The model included within-category continuous variables (one continuous variable for each dummy variable) to capture specific GPA values, calculated as the
difference between the eighth-grade GPA and the midpoint of the category. This was done in order to control for non-uniform distributions of GPA within the category across
schools (e.g., in case students who were at the low or high end of the distribution in the category were more likely to be in undervalued or overvalued schools). Centering at
the midpoint allowed us to avoid problems with multicollinearity, and provides an estimate of the value at the specific mid-point of the category. The model did not include an
intercept, and allowed the dummy variables for each group to vary across middle and high schools. This produced estimates of the variance in middle school effects and high
school effects on high school GPAs for students at each eighth grade GPA point.
45 The middle school differences in on-track rates and probability of earning As or Bs shown in Chapter 3 appear larger
than the school differences in GPAs because the charts in
Chapter 3 are based on benchmarks, rather than averages
(whether students are above or below a particular level of
performance). The use of benchmarks to gauge effects on

performance tends to amplify differences among students
close to the benchmark while suppressing differences
among students far from the benchmark.
41
Where students attend high school makes an even
show middle school effects inclusive of the effects of
bigger difference for their likelihood of being on-track
the high schools where the middle schools students
or earning high grades than where they attended middle
tend to enroll. While the correlation of on-track rates
school (see Figure 11, right panel; Figure 12, right panel
with middle school average math scores is 0.21 net of
on p.39). For example, among students with a moder-
high school effects, the correlation is close to zero (-0.03)
ate probability of passing their ninth-grade math class,
when high school effects are included with middle school
based on eighth-grade indicators, their chances of being
effects. The largest school effects would be observed
on-track varies from 75 percent in some high schools
among students who attend high schools that are atypi-
to 31 percent at other high schools, net of middle school
cal for their middle school. Students who moved from
effects (i.e., comparing similar students from similar
a strong middle school to a weak high school would
middle schools). High school effects on the likelihood
be much more likely to pass in ninth grade than their
of earning high grades are comparable to those of being
middle grade performance would indicate, while
on-track. Among students with a moderate probabil-
students who moved from a weak middle school to a
ity of earning high grades, the probability of doing so
strong high school would be much less likely to pass
ranges from 40 to 77 percent.
than their middle grade performance would suggest.
When we look at the characteristics of the schools
42
In general, knowing which school students attend
where students earn higher or lower grades than their
in the middle grades, and which school they attend in
qualifications would predict, we see the opposite pattern
ninth grade, substantially improves the prediction of
as with the middle grade schools. Students are more
their risk of failure. Sophisticated indicator systems
likely to be on-track or earn high grades if they attend
could consider these school effects when calculating
high schools that serve more low-achieving, disadvan-
students risk of ninth-grade off-track status. As shown
taged peers, compared to students with similar eighth-
in Figure 7 on p.35, by considering school effects, over
grade qualifications who attend higher-achieving high
half of students who are off-track in ninth grade can be
schools (see Table 4, columns B and E). This might
identified using eighth-grade indicators. Using eighth-
result from lower standards at high schools serving
grade GPA together with attendance and school effects
more disadvantaged students with low incoming test
identifies half of students who will be off-track in ninth
scores, or from grading practices that are based on

comparisons among students (e.g., grading on a curve).
grade, without falsely identifying many as at risk of

46
Students might also feel less capable and withdraw ef-
being off-track when they are not.

Knowledge of school effects also improves the predic-
fort if they attend a school with high-achieving peers
tion of which students are at risk for low grades in ninth
where they feel everyone else is smarter than they are,
grade. As shown in Figure 10 on p.38, incorporation of
or exhibit more self-efficacy and engagement if they are
school effects reduces the proportion of students falsely
a strong student relative to their school peers.
identified as at risk of low grades to about 50 percent.
Students from high-achieving middle schools are
This may not be low enough to be useful for targeting
more likely to attend high-achieving high schools
students for intervention, but the information might
than are students from low-achieving middle schools.
be used to have discussions with students and families
Thus, there are two contradictory effects on grades that
about students risk for low grades and what they can do
come from attending a high-achieving middle school
to reduce that risk.
students do better than their grades, attendance, and
students to high schools where students have a lower
Middle Grade Indicators Predict Who

Will Be On-Track and Who Will Earn
High Grades in Eleventh Grade
likelihood of passing their classes and earning high
So far, this chapter has shown how eighth-grade in-
grades. These two school effects cancel each other out.
dicators predict students likelihood of being able to
This can be seen in Table 4, columns C and F, which
pass their classes and earn high grades in ninth grade.
test scores would predict, compared to students from

a low-achieving middle school, but they tend to send
Ultimately, we want to know whether eighth-grade
by the end of eleventh grade is what matters for applica-
indicators can predict whether students drop out or
tions to college.
graduate from high school and earn the high grades
The best indicators of students course performance
they will need to succeed in college. The cohort of
by the end of eleventh grade are the same ones that are
students used in this report had not yet reached twelfth
most predictive of course performance in ninth grade
grade at the time of these analyses, so we focus on their
(see Table 5). Core GPA is the strongest single predictor
performance at the end of eleventh grade (the 2011-12
of eleventh-grade on-track status. Combining attendance
school year). Nearly all students who are on-track by
and eighth-grade core GPA is more predictive of being
the end of eleventh grade graduate at the end of twelfth
on-track at the end of eleventh grade than just eighth-
grade (94 percent do so). Those who are off-track in
grade core GPA alone; it is also more predictive than
eleventh grade either have already dropped out or are
combining eighth-grade core GPA with any other eighth-
very likely to drop outonly 37 percent of off-track
grade indicators, including ISAT scores. Combining
eleventh-graders graduate, among those who have not
seventh-grade GPA with eighth-grade core GPA and
yet dropped out. Moreover, their course performance
attendance slightly improves the prediction, but the
47
TABLE 5
Students Middle Grade Core GPAs Are the Strongest Predictor of Being On-Track and Earning High Grades in
Eleventh Grade


Single Indicator
Earning As or Bs in 11th-Grade
Core Classes
Correlation
Pseudo-R 2
% Correct
Correlation
Pseudo-R 2
% Correct
8th-Grade Core GPA

with Squared Term
0.43
0.18
0.18
0.71
0.71
0.41
0.17
0.18
0.79
0.79
8th-Grade Math GPA
0.38
0.14
0.69
0.35
0.13
0.78
0.41
0.16
0.70
0.37
0.15
0.78
8th-Grade Core
Course Failures
-0.32
0.10
0.69
-0.20
0.06
0.76
with Squared Term
0.37
0.15
0.15
0.71
0.71
0.18
0.04
0.05
0.76
0.76
8th-Grade Math ISAT

with Squared Term
0.30
0.10
0.10
0.66
0.66
0.35
0.11
0.11
0.78
0.78

with Squared Term
0.26
0.07
0.08
0.65
0.66
0.28
0.08
0.08
0.77
0.77
-0.25
0.07
0.68
-0.10
0.02
0.76
8th-Grade Misconduct
-0.18
0.03
0.67
-0.07
0.01
0.76
8th-Grade Grit 1
0.07
0.01
0.68
0.03
0.00
0.74
0.10
0.01
0.68
0.11
0.01
0.74
0.41
0.17
0.71
0.39
0.16
0.79
8th-Grade Study Habits
7th-Grade Core GPA

0.32
0.11
0.70
0.16
0.04
0.76
7th-Grade Math ISAT
0.29
0.09
0.66
0.34
0.11
0.77
0.25
0.07
0.65
0.28
0.08
0.77
46 This is consistent with prior research that showed that student

performance is shaped by students ability levels relative to
their classroom peers; students receive lower grades than other
students with similar skill levels if they are in a classroom where
they have weak skills compared to their classroom peers. See
Kelly (2008); Farkas, Sheehan, and Grobe (1990); Nomi and
Allensworth (2012). However, as noted in Figure 13, observed

inconsistencies in grading are generally about equal to half of
a GPA point, at best.
47 Students who are on-track at the end of eleventh grade
have accumulated at least 17 full-year course credits
required for graduation.
43
TABLE 5: CONTINUED
Students Middle Grade Core GPAs Are the Strongest Predictor of Being On-Track and Earning High Grades in
Eleventh Grade
Single Indicator
Correlation
Pseudo-R 2
% Correct

Attendance
0.23

Math ISAT
8th-Grade Math +
Reading ISAT
Earning As or Bs in 11th-Grade
Core Classes
Correlation
Pseudo-R 2
% Correct
0.74
0.18
0.79
0.19
0.72
0.18
0.80
0.10
0.66
0.11
0.78
Combining Two 8th-Grade Indicators
Combining Three or More Eighth-Grade Indicators
44

Math ISAT + Reading
ISAT
0.19
0.72
0.19
0.79
8th-Grade Core GPA

+ Attendance + Math
ISAT+ Reading ISAT
0.23
0.74
0.19
0.80
0.07
0.67
0.09
0.77
8th-Grade Core
GPA + Attendance +
Math ISAT + Reading
ISAT + Background
Characteristics
0.23
0.74
0.20
0.80

+ Reading ISAT + Course
Failures + Suspensions +
Misconduct + Background
Characteristics
0.24
0.75
0.20
0.80
0.24
0.75
0.20
0.80
0.24
0.75
0.21
0.80
8th-Grade GPA +
Attendance + School
Effects
0.30
0.75
0.26
0.81

Indicators + School
Effects
0.31
0.76
0.29
0.81
Background
Characteristics 2
Adding in 8th-Grade Indicators

+ Reading ISAT +
7th-Grade GPA
Indicators
Adding in School Effects
Note: 1) Grit and study habits are calculated from students responses to items on UChicago CCSRs annual survey of CPS students. The elementary/middle grade
student survey had a response rate of 59 percent in 2009. Because not all students answer the survey, the sample size for these models is smaller than the sample
sizes for the other models included in this table. The percent of students who are on-track or earn high grades is slightly higher in this smaller sample resulting
in a somewhat higher correct prediction rate for these two variables. 2) Background characteristics include race, gender, special education status, neighborhood
poverty level, and socioeconomic status, free reduced price lunch status, and whether a student was older than 14 when entering high school. 3) The bolded
numbers represent the best indicator or combination of indicators in each group.
improvement in prediction from a pseudo-R 2 of 0.23
Eighth-grade test scores do help predict who will
to 0.24 may not warrant a more complicated indicator
get high grades in high school, but only among students
system that comes from using three indicators versus
who already were getting high grades in middle school.
two indicators.
Attendance provides even more information about who
Core GPA is also the best single predictor of whether
will get high grades in high school, among students with
students earn As or Bs in eleventh grade (see Table 5).
high prior grades, than test scores. Neither test scores
Combining eighth-grade core GPA with attendance,
nor attendance provide much information about who
ISAT scores, or seventh-grade GPA only somewhat im-
will earn high grades in high school among students
proves the prediction of who earns As or Bs over using
who had a 2.0 or below in eighth grade, in part, because
eighth-grade core GPA alone, most likely because the
it is so unlikely that students with low grades in middle
combination of predictors only differentiates between
school will earn high grades in high school.
students who have earned very high grades.
Summary
Combinations of three or more indicators do not

considerably improve the prediction of who is on-track
or who earns high grades over predictions based on
Eighth-grade core GPA is the strongest single predic-
only two indicators (grades and attendance), suggesting
tor of on-track status and earning high grades in high
that a complex system incorporating multiple indica-
school. Students grades across all subjects are more
tors will not provide much more information than only
predictive of their grade in a specific class (math or
two indicators. Knowing where students went to middle
English) than their prior grade in the corresponding
and high school, however, does substantially improve
subject. Combining core GPA with attendance, or with
the prediction of whether students will be on-track or
seventh-grade GPA, gives an even better prediction of
earn high grades.
who will be on-track at the end of ninth grade, or earn

high grades, than core GPA alone.
Test scores, either alone or in combination with core
Although middle grade indicators cannot identify

all students who are at risk of being off-track, they
can be used to identify a subset of students who are
GPA, are not all that predictive of who is on-track at the
at high risk of ninth-grade failure. About half of
end of ninth grade. This may seem counter-intuitive
students who are off-track in the ninth grade can be
because test scores are often the primary focus in
identified accurately with middle grade predictors
policy discussions about improving educational attain-
few of the students who are identified as at risk succeed.
ment and they often are used to identify students in
Resources devoted to intervention for these students
need of additional academic support. Often overlooked,
would not be wasted on students who would succeed
attendance rates and GPA in earlier grades are much
without any intervention. The next chapter examines
better for identifying who will fail their classes and be
what middle grade indicators tell us about who is likely
off-track at the end of ninth grade.
to be off-track by the end of ninth grade.
45
46
CHAPTER 4
Who Is at Risk of Being Off-Track

at the End of Ninth Grade?
The previous chapter showed that eighth-grade core
This chapter shows students risk

of failing based on their middle
grade records. This information
could be used by school practitioners to identify students who
are at high risk of failure and to
discuss goals for student performance with both students and
their families.
GPA and attendance together provide the best parsimonious prediction of ninth-grade failure and being offtrack for graduation, of all of the potential indicators
that were examined; other metrics, such as subject-specific grades, test scores, and background characteristics, do not provide more information beyond students
core grades and attendance. The prediction is far from
perfect; students experiences in ninth grade determine
whether they pass their ninth-grade classes and remain
on-track for graduation. But eighth-grade indicators
can identify subsets of students who are at high risk
or very high risk of being off-track by the end of ninth
grade. In this chapter, we used eighth-grade core GPA,
eighth-grade attendance, and eighth-grade ISAT scores
to identify how many students were at risk of being
Students Risk of Being Off-Track by

Middle Grade Indicators
off-track and failing in high school. We included ISAT
Eighth-grade students with less than 80 percent
scores in the prediction to be sure that the risk groups
attendance or GPAs less than 1.0 are at extremely
include students with similar tested skill levels, as well
high risk of being off-track in ninth grade. In
as similar course performance and attendance.
Chicago, about 5 percent of middle grade students
To design intervention systems, schools need to
can be identified as being at very high risk of being
know the magnitude of the level of risk of their stu-
off-track in the ninth grade, based on their eighth-
dentshow many students are at some risk of failing,
grade GPA and attendance (see Figure 14). These are
and how many are at very high risk of failing. Students
students with extremely low grades and attendance in
with different levels of risk may need different inter-
the middle grades.
ventions, and there are cost constraints as to how many

students can be targeted with extra resources. Students
Eighth-grade students with C/D averages and chronic
with very high risk are almost certain to fail without
absence in middle school are at high risk of being
intervention; yet, because they are at very high risk,
off-track in ninth grade. Around 16 percent of students
their outcomes may be very difficult to change unless
can be identified as being at high risk of being off-track
substantial resources are allocated to them. Students
(see Figure 14). These students are more likely to be
with moderate or low risk may succeed without any
off-track than on-track in high school; they have a 50 to
intervention; but they may need only a modest interven-
75 percent likelihood of being off-track. They had very
tion to succeed, and a sizable group will fail without any
poor performance in eighth grade, with an average GPA
support. The decision about how to allocate resources
of 1.5 (a combination of Ds and Cs). Their attendance
hinges on how many students are identified as being at
was better than students at extremely high risk of being
different levels of risk.
off-trackthey came to school 91 percent of the time,

on average. Many of them they were still chronically
absent, however, missing more than five weeks of school
Chapter 4 | Who Is at Risk of Being Off-Track at the End of Ninth Grade?
47
FIGURE 14
The Percent of Students at Very Low, Low, Moderate, High, and Very High Risk of Being Off-Track in Ninth Grade
Based on Students Who Began Ninth Grade in the 2009-10 School Year
40
Percent of Students
35
32.1%
30
26.8%
25
19.7%
20
16.2%
15
10
5.2%
5
0
Very Low (<10%)
Low (10 to < 25%)
Moderate (25 to < 50%)
High (50 to < 75%)
Very High (75%+)
Probability of Being Off-Track in Ninth Grade Based on Eighth-Grade Indicators
48
N: 3936
N: 5364
N: 6423
N: 3234
N: 1041
8th-Gr. Core GPA: 3.6
8th-Gr. Attendance: 98%
8th-Gr. Math ISAT: 295
Note: The five risk groups shown in this chart were created by first running a logistic regression in which the probability of being off-track is modeled as a function of
eighth-grade core GPA, attendance, ISAT math scores, and any significant interaction terms between these three indicators. Using predicted probabilities generated
from the analysis, we then created the five groups, using the cut points described in the parentheses above.
during their eighth-grade year. These students are also
Risk of Being Off-Track in Ninth Grade by

Eighth-Grade Core GPA and Attendance
very unlikely to succeed without extra support, but
Percent of 8th-Grade Attendance
FIGURE 15
39%
98-100
95-98
90-95
80-90
<80
59%
20%
they may be more responsive to intervention that is less

5%
(N=538)
(N=2129)
(N=2737)
47%
27%
9%
(N=153)
(N=1199)
(N=3066)
(N=2390)
74%
56%
37%
16%
(N=243)
(N=1334)
(N=2270)
(N=1043)
83%
67%
55%
30%
(N=284)
88%
(N=821)
82%
(N=880)
(N=216)
(N=259)
(N=135)
0.0 < 1.0
1.0 < 2.0
2.0 < 3.0
As shown in Figure 15, it is the combination of moderate

grades with low attendance or low grades with moderate attendance that puts students at high risk of being
off-track in high school. Students with moderately high
attendance (between 95 and 98 percent) are at high risk
if their GPA is less than 1.0, while students whose GPA is
between 1.0 and 2.0 are at high risk if their attendance
is less than 95 percent. Also at high risk are students
74%
(N=243)
intense than needed for students at extremely high risk.
whose GPA is between 2.0 and 3.0, if their attendance is

3.0 < 4.0
less than 90 percent.
8th-Grade Core GPA

Very Low (<10%) Chance of Being Off-Track
About 60 percent of eighth-grade students are at
Low (10 to <25%) Chance of Being Off-Track
some risk of being off-track in ninth grade, even
Moderate (25 to <50%) Chance of Being Off-Track

High (50 to <75%) Chance of Being Off-Track
though their middle school performance seems
Very High (75%+) Chance of Being Off-Track
satisfactory. Around one-third of first-time ninth-
Note: Based on students who began ninth grade in the 2009-10 school year.
graders are at moderate risk of being off-track (between

a 25 and 50 percent chance of being off-track) and another quarter at low risk of being off-track (between 10
and 25 percent). Students with a moderate risk of being
or math, or their test scores in reading or math, were
off-track had about a C average in eighth grade and 95
not any more predictive of their ninth-grade grades
percent attendance (see Figure 14). Students who were
in either specific subject than their overall GPA and
at low risk of being off-track had a B- GPA and 97 percent
attendance rate. Based on eighth-grade performance,
attendance, on average. While the risk of being off-track
around 11 percent of ninth-grade students are at high or
is not high for these two groups, because of their size,
very high risk of failing math, and 12 percent are at high
these two groups account for just over half of all stu-
or very high risk for failing English (see Figures 16 and
dents who were off-track at the end of ninth grade:
17). Students who are at high or very high risk of failing
37 percent of all students who were off-track in ninth
math or English had about a 1.2 GPA in eighth grade and
grade were at moderate risk, while 15 percent were at
85 percent attendance, on average.
low risk based on their eighth-grade performance.
though these students are not at high risk of being off-
Students Risk of Being Off-Track

in Ninth Grade Carries On to
Eleventh Grade
track, they are at some risk, and modest efforts around
The same middle grade indicators that predict ninth-
attendance improvement could have big pay-offs in the
grade on-track status also predict whether students are
long run because so many students are in these two
on-track to graduate by the end of eleventh grade, where
groups. Such interventions might include conversations
on-track in eleventh grade is defined as having suffi-
with students and parents about their risk of failing in
cient credits to be a twelfth-grader the following year
high school, lowering risk with better attendance and
and still enrolled in school. Thus, middle grade indica-
effort, teaching strategies for help-seeking and support
tors can be used to identify students who are at great
when they are in high school, and discussions about
risk of dropping out before they start high school. In
parental monitoring through the transition to high
fact, students likelihood of being on-track in eleventh
school. These students need to be closely monitored in
grade is similar to their likelihood of being on-track in
the transition to high school; otherwise, many will show
ninth grade, based on their eighth-grade records.
Because of the size of the moderate- and low-risk

groups, costly interventions would be impractical. Even
declining attendance and effort in ninth grade.

Low grades and poor attendance in middle grades
Only about one-fifth of students have very little
indicate students are unlikely to graduate. Students
risk of being off-track in ninth gradestudents
with less than a 1.0 eighth-grade GPA are at high risk of
with a B average or higher and 95 percent or better
being off-track when they finish ninth grade, and also
attendance in middle school. There are some stu-
at the end of eleventh grade (see Figure 18). Only about
dents whose risk of falling off-track to graduation is 10
a quarter (26 percent) of students with an eighth-grade
percent or lower. These are students whose GPA was
GPA that is less than 1.0 are on-track when they finish
a 3.0 or better, and whose attendance was 95 percent
ninth grade and about a quarter (22 percent) are still
or above in eighth grade. Even though being on-track
on-track at the end of eleventh grade. Students with
for graduation is a very low bar for judging high school
higher eighth-grade GPAs have higher likelihoods of
performance, few students leave middle school assured
being on-track in both ninth and eleventh grades, and
of being ready to meet this basic expectation.
the probability is similar in both years because ninthgrade performance is very indicative of performance
The signals for students risk of failing either ninth-
through the remaining high school years.

The same general pattern can be seen in Figure 19,
grade English or math are the same as those for being

off-track. English and math are subjects of primary
which shows ninth- and eleventh-grade on-track rates
concern to educators, as reading and math skills pro-
by eighth-grade attendance. Students with less than 80
vide a base for learning in other subjects. The previous
percent attendance are extremely likely to be off-track
chapter showed that students prior grades in English
for graduation in both ninth grade and eleventh grade;
49
FIGURE 16
The Percent of Students at Very Low, Low, Moderate, High, and Very High Risk of Failing Math
40
34.3%
Percent of Students
35
34.2%
30
25
20.8%
20
15
9.7%
10
5
1.0%
Very Low (<10%)
Low (10 to < 25%)
High (50 to < 75%)
Very High (75%+)
Probability of Passing Ninth-Grade Math Based on Eighth-Grade Indicators
50
N: 4198
N: 6931
N: 6903
N: 1968
N: 211
Note: The five probability groups shown in this chart were created by first running a logistic regression in which the probability of passing ninth-grade math was
regressed on eighth-grade core GPA, attendance, ISAT math scores, and any significant interaction terms between these three indicators. Using predicted probabilities
generated from the analysis, we then created the five groups, using the cut points described in the parentheses above.
FIGURE 17
The Percent of Students at Very Low, Low, Moderate, High, and Very High Risk of Failing English
40
33.7%
Percent of Students
35
29.1%
30
25.2%
25
20
15
10.4%
10
5
1.7%
0
Very Low (<10%)
Low (10 to < 25%)
High (50 to < 75%)
Very High (75%+)
Probability of Failing Ninth-Grade English Based on Eighth-Grade Indicators

N: 5068
N: 6773
N: 5845
N: 2085
N: 349
Note: The five probability groups shown in this chart were created by first running a logistic regression in which the probability of passing ninth-grade math was
regressed on eighth-grade core GPA, attendance, ISAT math scores, and any significant interaction terms between these three indicators. Using predicted probabilities
generated from the analysis, we then created the five groups, using the cut points described in the parentheses above.
FIGURE 18
The Percent of Students Who Are On-Track, Off-Track, or Dropouts at the End of Ninth Grade and
Eleventh Grade on Eighth-Grade Core GPA
9th-Grade On-Track by 8th-Grade Core GPA
100
12%
90
24%
Percent of Students
27%
2%
3%
15%
22%
62%
31%
74%
35%
50
88%
40
96%
87%
38%
76%
64%
52%
20
50%
39%
38%
26%
1.0
95%
76%
64%
30
5%
8%
9%
14%
19%
40%
48%
70
10
4%
36%
80
60
11th-Grade On-Track by 8th-Grade Core GPA
22%
1.0-1.5
1.5-2.0
2.0-2.5
2.5-3.0
3.0-3.5
3.5-4.0
1.0
1.0-1.5
8th-Grade Core GPA
1.5-2.0
2.0-2.5
2.5-3.0
3.0-3.5
3.5-4.0
8th-Grade Core GPA
9th-Grade On-Track
11th-Grade On-Track
9th-Grade Off-Track
11th-Grade Off-Track
51
11th-Grade Dropout
FIGURE 19
The Percent of Students Who Are On-Track, Off-Track, or Dropouts at the End of Ninth Grade and
Eleventh Grade, Based on Eighth-Grade Attendance
9th-Grade On-Track by 8th-Grade Attendance
11th-Grade On-Track by 8th-Grade Attendance
100
90
Percent of Students
80
47%
70
60
68%
40%
30%
23%
14%
27%
20%
16%
25%
59%
7%
4%
10%
16%
38%
56%
11%
20%
28%
81%
31%
50
40
30
44%
20
10
0
53%
60%
70%
77%
35%
86%
60%
86%
41%
27%
19%
< 80%
52%
30%
32%
77%
69%
11%
80-85
85-90
90-92
92-94
94-96
96-98 98-100 < 80%
80-85
85-90
90-92
92-94
94-96
9th-Grade On-Track
11th-Grade On-Track
9th-Grade Off-Track
11th-Grade Dropout
Note: Based on students entering ninth grade in the 2009-10 school year, followed through the 2011-12 school year.
96-98 98-100
FIGURE 20
The Percent of Students Who Are On-Track, Off-Track, or Dropouts at the End of Eleventh Grade
Based on the Number of Course Failures in Eighth-Grade Math and English Classes
11th-Grade On-Track by Number of Semester
Course Failures in 8th-Grade Math Classes
11th-Grade On-Track by Number of Semester

Course Failures in 8th-Grade English Classes
100
11%
90
17%
Percent of Students
80
10%
26%
20%
15%
40%
70
24%
32%
42%
42%
28%
60
33%
36%
33%
50
34%
40
36%
75%
73%
30
42%
53%
41%
20
40%
26%
10
0
22%
Number of Semester Fs in 8th-Grade Math Classes
52
35%
16%
Number of Semester Fs in 8th-Grade English Classes
11th-Grade On-Track
11th-Grade On-Track
11th-Grade Dropout
11th-Grade Dropout
looked for high-yield middle school indicators of drop-
Throughout Middle Grades, GPAs

Are Better Indicators of Risk for High
School Failure Than Other Factors
out in Philadelphia schools. High-yield indicators were
Middle grade practitioners may wonder if there are
defined as those that identified students with a dropout
some levels of performance earlier in the middle
risk of 75 percent or higher. They found that attendance
grades that indicate students are at high risk of
less than 80 percent in middle school was a high-yield
failing when they get to high school. For simplicity, we
indicator of dropping out of high school. They also
focus on single indicators for this comparisoneven
found that failing eighth-grade English or math was a
though they are less predictive than combinations. We
high-yield indicator of dropout (a 75 percent or greater
wondered if there were points at which warning bells
probability).
should go off for middle grade practitioners if they
their risk is greater than 75 percent. This corresponds

closely to the work of Neild and Balfanz (2006) who
48
The same pattern occurs in Chicago. As
shown in Figure 20, students who fail both semesters of
see a student is performing below a particular level.
eighth-grade math, or all four sections of eighth-grade

language arts (reading and writing), have less than a
Students with D averages at any grade are at high
25 percent chance of being on-track for graduation in
risk of failure in high school. As shown in Table 6, a
eleventh grade.
very low GPA, less than a 1.5, prior to eighth grade is
48 Course failures were predictive of on-track in Chicago, but

did not add to the prediction, once we considered students
overall GPAs and attendance.
TABLE 6
Single Indicator Thresholds for Identifying Students at High Risk of Being Off-Track in Grades Five through
Eight, Where the Risk of Being Off-Track Is Greater Than 50 Percent.
Attendance
GPA
5th Grade
6th Grade
7th Grade
8th Grade
Threshhold
84.1%
85.5%
87.8%
87.5%
% of Off-Track Students Identified
6.7%
9.4%
16.9%
19.4%
Threshhold
Math ISAT Scores
1.4
1.3
1.4
1.7
15.3%
21.0%
30.0%
30.8%
Threshhold
176
196
204
227
1.2%
3.2%
3.8%
3.7%
Reading ISAT
Scores
Threshhold
159
176
182
204
1.6%
2.9%
3.9%
5.0%
Number of Days
Suspended
Threshhold
3.7%
4.8%
8.9%
9.9%
the clearest signal that a student is at high risk of being
Summary
off-track in high school. The GPA cut-off for identify-
Many students who perform well in middle school fail
ing students at high risk of being off-track is similar in
classes in ninth grade and are off-track by the end of
fifth through seventh grade, but the percent of off-track
the year. The transition to high school brings students
students who are identified increases as eighth grade
into a very new context, with different peers and differ-
approaches. The district could use this threshold to
ent relationships with teachers and other adults. For
identify students very early on who are almost certain
almost all students, it is important to monitor their
not to graduate without intervention. Students with a
performance in the first year of high school and reach
history of very low grades in the middle grades likely
out to students who start to show signs of withdrawal.
need very intensive interventions if they are to eventu-
Subgroups of students enter high school already at
ally graduate.
high risk, based on their academic performance in
Attendance is the next best indicator in grades five
the middle grades.
through seven for identifying students who are at high
About 5 percent of students can be identified as
risk of being off-track. Students who attend less than
being at extremely high risk of being off-track. These
85 percent of days in any of the middle grades are very
same students are also at extremely high risk of still
likely to fail classes when they arrive in high school and
being off-track in eleventh grade and eventually not
fall off-track for graduation. They already are chronically
graduating from high school. Failing classes in ninth
absent in the middle grades, and they are likely to miss
grade means that they are not accumulating the credits
school at even higher rates when they enter high school.
they need. Unfortunately, it would take substantial
Math and reading ISAT scores and the number of
support to turn around the very low grades and very
days a student is suspended are less predictive for
poor attendance of students at such high risk of failure.
identifying who is at high risk to fail in high school
There are many other students, around 16 percent
than attendance or grades. However, there are small
of first-time ninth-graders, who are at high risk of be-
numbers of students whose test scores are so low, or
ing off-track at the end of ninth grade. These students
suspension rates are so high, that they can be identi-
could go either way, based on their individual experi-
fied as at high risk of high school failure based on these
ences. This is a group for whom interventions may be
single indicators. These are students who are at aca-
more effective. Middle school practitioners, students,
demic warning levels on the ISAT, or who are suspended
and their families need to be aware that students with
for more than a week during the middle grades.
attendance less than 90 percent are at high risk of not
53
graduating. Middle school and high school staff do not
sufficient for ensuring that students are prepared to
need to wait until these students are failing their ninth-
engage in college level work. To be ready for college,
grade classes to intervene. Mentorship programs, such
high school students need to be working hard at a very
as Check and Connect,
high level of performance so that they are learning the
49
may be helpful in improving
their grades and attendance. They could also be identi-
skills and knowledge they need to succeed in college;
fied for support from the start of high school to improve
this means earning As or Bs in high school classes. In
their likelihood of passing ninth-grade classes.
the next chapter, we show which students are likely to
While passing ninth-grade classes is an important

step toward graduating from high school, it is not
earn As or Bs in high school and how many students

have little chance of doing so.
49 Sinclair et al. (2005); Lehr et al. (2004).
54
CHAPTER 5
Who Is at Risk of Earning Less

Than As or Bs in High School?
Passing courses is essential for graduating from high
This chapter shows which

students are likely to earn high
grades in high schoolgrades
that make them eligible for college and likely to succeed, once
there. The information in this
chapter can be used to discuss
goals for student performance
with students and their families.
school, but simply passing is not enough if students are

to have a good chance of succeeding in college; they
need to be engaged in their classes and earning As or Bs.
Students who earn Cs or Ds in high school are unlikely
to graduate from college, while those with a B average (a
3.0 GPA) have about a 50/50 chance of earning a fouryear college degree (see Appendix A). Unfortunately,
few students actually achieve a GPA of at least 3.0 in
ninth grade. In 2009, for example, only 23 percent of
first-time freshmen earned A or B averages in their core
classes. This sets the stage for poor course performance
throughout high school.
50
This chapter identifies which students are likely to

earn As or Bs in high school. While the indicators that
these factors can be seen in Figure 21. The left-hand
are predictive of As or Bs are similar to those for passing
portion of the figure shows that eighth-graders must
classes in high school, the competencies and behaviors
have very strong attendance or very strong test scores,
students must demonstrate in order to earn As or Bs are
in addition to high grades, in order to have even a 50
much higherand their probability of success is much
percent chance of earning As or Bs in ninth grade.
lower. Eighth-graders who plan to eventually attend
Because only students with at least a 3.0 eighth-grade
college need to excel in their courses in order to have
GPA have at least a 22 percent chance of earning As or
a chance of earning sufficiently strong grades in high
Bs in high school, the right-hand portion of the figure
school that put them on the path to college readiness.
takes a finer-grained look at students with high grades
Eighth-Graders with Good Grades,

Attendance, and Test Scores Have
Only a Moderate Chance of Earning
As or Bs in Ninth Grade
by subdividing the 3.0 to 4.0 GPA category. It shows

only students with the highest grades (GPAs above 3.7)
have at least a 50 percent chance of earning As or Bs
in ninth grade. Moreover, of the students with a GPA
greater than 3.7, only those with attendance rates
Chapter 3 showed that students eighth-grade core
greater than 98 percent or ISAT math scores above 310
GPAs are the best predictors of earning high grades
have more than a 75 percent chance of earning As or Bs.
in high school. For students with strong eighth-grade
To put this in context, students with attendance rates
GPAs, considering their attendance rates or ISAT
of 98 percent or higher are missing less than a week of
scores, along with GPA, improves the accuracy of the
school, and those scoring at a 310 or above are
prediction. Students probability of success based on
in the exceeds range on the ISAT.
50 While grades are higher, on average, in eleventh and twelfth

grades than grades nine and 10, this is because students
with low grades tend to drop out. When we compare the
same students over time, we see that grades do not improve

through high school, on average.
Chapter 5 | Who Is at Risk of Earning Less Than As or Bs in High School?
55
FIGURE 21
Students Probability of Earning As or Bs in Ninth Grade by Core Eighth-Grade GPA and

Eighth-Grade Attendance or ISAT Math Score
Percent of 8th-Grade Attendance
GPA by Attendance
10%
98-100
95-98
90-95
80-90
<80
(N=536)
3%
(N=153)
2%
(N=243)
0%
(N=283)
0%
6%
(N=1196)
4%
(N=1332)
2%
(N=818)
2%
22%
(N=2129)
16%
(N=3062)
12%
(N=2264)
6%
(N=878)
57%
(N=2737)
47%
(N=2389)
34%
(N=1043)
22%
(N=214)
98-100
95-98
90-95
80-90
(N=257)
(N=135)
0.0 1.0
1.0 2.0
2.0 3.0
33%
(N=925)
26%
(N=505)
54%
(N=1006)
45%
(N=864)
33%
77%
(N=882)
72%
(N=600)
60%
(N=375)
(N=163)
3.33.7
3.74.0
14%
(N=138)
3.0 3.3
1%
(N=236)
41%
(N=849)
8th-Grade Core GPA

3.0 4.0
8th-Grade Core GPA
GPA by ISAT Math Score
56
22%
8th-Grade ISAT Math
310-410
(N=113)
23%
289-309
(N=571)
8%
267-288
251-266
234-250
120-233
(N=625)
2%
(N=224)
1%
(N=400)
0%
(N=239)
0.0 1.0
5%
(N=1105)
4%
(N=1516)
3%
20%
(N=2194)
14%
(N=2539)
11%
(N=2229)
11%
69%
(N=916)
52%
(N=1553)
46%
(N=2289)
40%
(N=1018)
33%
(N=487)
23%
(N=789)
(N=822)
(N=133)
1.0 2.0
2.0 3.0
3.0 4.0
8th-Grade Core GPA

Very Low (<25%)
310-410
289-309
267-288
251-266
234-250
31%
(N=109)
33%
(N=394)
37%
(N3=954)
32%
(N=568)
27%
52%
(N=263)
45%
(N=605)
47%
(N=887)
84%
(N=544)
73%
(N=554)
62%
(N=448)
47%
(N=364)
44%
(N=305)
(N=151)
3.0 3.3
3.33.7
120-233
3.74.0
8th-Grade Core GPA

Low (25%<50%)
Moderate (50%<75%)
High (75%)
Note: Probabilities are only shown if there are at least 100 students with a specific GPA and attendance rate/ISAT score combination. The percentages are based on
students entering ninth grade in the 2009-10 school year.
Two-Thirds of Students Leave

Middle School with Little Chance
of Earning As or Bs in High School
should be eager to understand why the grades of some
Few students finish eighth grade with the academic
percent probability of earning As or Bs in ninth-grade
records necessary to have a good chance of earning As
core classes. One-fifth of the cohort has between a 25
or Bs in ninth grade. In fact, as shown in Figure 22, only
and 50 percent chance; on average, students in this group
2 percent of students have a high probability of doing so
have an eighth-grade core GPA of 3.13, an attendance rate
(greater than a 75 percent chance). These are exceptional
of 96.8 percent, and an ISAT math score of 275. These
students; on average, they have a core GPA of 3.93 in eighth
students also have strong academic records; yet, this
grade, an attendance rate of 99.3 percent, and a score of
group is unlikely to earn As or Bs in high school (less
324 on the math portion of the eighth-grade ISATwhich
than a 50 percent chance). Nearly two-thirds of students
is higher than the exceeds benchmark of 310.
leaving CPS middle schools (65 percent of students) have
Students with a moderate chance of earning As or Bs
of these students decline in high school.

The vast majority of students have less than a 50
less than a 25 percent chance of earning a GPA of at least
in ninth grade (12 percent of students), who are more
3.0 in ninth grade. Although close to 80 percent of CPS
likely than not to earn high grades, also have strong
students enter high school aspiring to earn a four-year-
eighth-grade records, with an average core GPA of 3.66,
college degree, the majority are simply unprepared to do
attendance rate of 98 percent, and an average ISAT
so. Students performance does not align with their aspi-
math score of 294. Combined with the students who
rations, but neither they nor their teachers may realize
have a high probability, a total of only 15 percent of
it. In order to be on-track for college readiness, students
students have more than just a 50/50 chance of earning
need to be highly engaged in their classes, so much so
As or Bs as ninth-graders. Despite their particularly
that they are earning As or Bs. If this is not happening in
strong eighth-grade academic records, however, none
middle school, it is unlikely to begin in high school when
are certain to be successful in high school. Practitioners
students tend to have less support and monitoring.
FIGURE 22
Students' Probability of Earning As or Bs in Ninth-Grade Core Classes Based on Eighth-Grade Outcomes

70
64.6%
Percent of Students
60
50
40
30
21.0%
20
12.4%
10
2.1%
0
Very Low (<25%)
Low (25 to <50%)
Moderate (50 to <75%)
High (75%+)
Probability of Earning As or Bs in Ninth-Grade Core Classes

8th-Gr. Attendance: 92.7%
9th-Gr. with As or Bs: 9.7%
Note: The percentages are based on students entering ninth grade in the 2009-10 school year. The four probability groups shown in this chart were created by first
running a logistic regression in which the probability of passing ninth-grade math was regressed on eighth-grade core GPA, attendance, ISAT math scores, and any
significant interaction terms between these three indicators. Using predicted probabilities generated from the analysis, we then created the four groups using the cut
points described in the parentheses above.
57
The Likelihood of Earning As or Bs

in Subject-Specific Classes Is Similar
to That Across All Core Classes
Eleventh-grade GPAs are divided into groups, based on
Educators may wonder whether different indicators
to graduate from college and only have access to college
are predictive of high grades in particular classes,
if they have ACT scores of 21 or higher, which is rare.
such as English versus math. We found almost no dif-
Those with a GPA between 2.0 and 3.0 are qualified for
ferences in which indicators of performance predicted
a non-selective or somewhat selective college, if they
high grades in math versus English. Subject-specific
also have an ACT score of at least 18. However, while
indicators, such as grades or test scores in English or
they can enroll in college, their chances of graduat-
math, were less predictive of grades in ninth-grade
ing from college are slim. In order to be qualified for a
English and math than students core GPA and com-
selective college, and to have at least a 50/50 chance of
bined test scores across subjects. Most students have
graduating, students need a GPA above 3.0 and an ACT
low probabilities of earning As or Bs in ninth-grade
score of at least 18. 51 (See Appendix A for a summary
English and math, although their chances are slightly
table that ties college access to students high school
better in English than in math (see Figures 23 and 24).
GPAs and ACT scores.)
About three-fourths of Chicago students leave middle
58
prior research that ties high school grades to college

outcomes: students with a GPA below 2.0 are not likely
Students leaving eighth grade with very high
school with less than a 50/50 chance of earning As or Bs
gradesan A average of 3.5 or higherhave fairly even
in ninth-grade English; 84 percent of students have less
odds of leaving high school with a B average or better
than a 50/50 chance of earning As or Bs in ninth-grade
(3.0 or above). Sixty-one percent of the students with an
math, based on their eighth-grade records. Less than 5
eighth-grade GPA of at least 3.5 earn As or Bs in ninth
percent of students leave eighth grade with a very high
grade; that proportion drops to 54 percent in eleventh
likelihood of earning As or Bs in ninth-grade English
grade. These students have access to selective colleges,
(with at least a 75 percent chance of doing so). Less
if they have an ACT score of at least 18. Another one-
than 2 percent have a very high likelihood of earning
third of the students with the top eighth-grade grades
As or Bs in math in ninth grade, based on their eighth-
earn a GPA between 2.0 and 3.0 in ninth and eleventh
grade records.
grade. While these students have access to somewhat

selective colleges, their chances of graduating from col-
By Eleventh Grade, Even Fewer

Students Earn Bs or Better
lege are less than 40 percent (see Appendix A).
If students are followed an additional two yearsto
eighth grade (with a GPA between 3.0 and 3.5), less than
eleventh gradethe patterns look similar to those seen
a third (29 percent) finish eleventh grade with a GPA
in ninth grade, although grades are somewhat lower in
that signals they are likely to succeed in college (at least
eleventh grade and some students have dropped out of
a 3.0). Another 24 percent finish with grades so low they
school. Grade point averages generally do not improve
are not even qualified for college (less than a 2.0).
as students progress through high school, so low likeli-
Among students who earned a mix of As and Bs in
Eighth-graders who earn low grades have almost no
hoods for earning As or Bs in ninth grade foretell low
chance of graduating from college. More than half of the
GPAs in eleventh grade when students are applying to
students with an eighth-grade GPA between 1.5 and 2.0
college. Put another way, students course performance
are not qualified for college, and 20 percent drop out
in eighth grade has implications for their grades not
by eleventh grade. Students who earn an eighth-grade
only in the following year but also for the entirety of
GPA of less than 1.0 are nearly as likely to drop out (44
their high school careers.
percent) by eleventh grade as not, and the vast majority
Figure 25 shows students ninth-grade GPAs and
eleventh-grade GPAs, based on their eighth-grade GPAs.
who do stay in school perform at levels that leave them

unqualified for college.
51 Roderick et al. (2006)

FIGURE 23
The Percent of Students at Very Low, Low, Moderate, and High Probability of Earning As or Bs in Ninth-Grade
English Classes
70
Percent of Students
60
50
44.6%
40
30.7%
30
20.2%
20
10
0
4.5%
Very Low (<25%)
Low (25 to <50%)
High (75%+)
Probability of Earning As or Bs in Ninth-Grade English

8th-Gr. Reading ISAT: 232
Note: The percentages are based on students entering ninth grade in the 2009-10 school year. The four probability groups shown in this chart were created by first
FIGURE
24 regression in which the probability of passing ninth-grade math was regressed on eighth-grade core GPA, attendance, ISAT math scores, and any
running a logistic
significant interaction terms between these three indicators. Using predicted probabilities generated from the analysis, we then created the four groups using the cut
The
Percent of Students at Very Low, Low, Moderate, and High Probability of Earning As or Bs in Ninth-Grade
points described in the parentheses above.
Math Classes
70
60
Percent of Students
52.6%
50
40
31.0%
30
20
14.8%
10
1.7%
0
Very Low (<25%)
Low (25 to <50%)
High (75%+)
Probability of Earning As or Bs in Ninth-Grade Math

Note for Figures 23 and 24: The percentages are based on students entering ninth grade in the 2009-10 school year. The four probability groups shown in this chart
were created by first running a logistic regression in which the probability of passing ninth-grade math was regressed on eighth-grade core GPA, attendance, ISAT
math scores, and any significant interaction terms between these three indicators. Using predicted probabilities generated from the analysis, we then created the four
groups using the cut points described in the parentheses above.
59
FIGURE 25
Course Performance in Eighth Grade Is Associated with Eleventh-Grade GPA in the Same Way as
Ninth-Grade GPA
9th-Grade Core GPA by 8th-Grade Core GPA
100
9%
90
24%
Percent of Students
80
41%
61%
33%
17%
3%
6%
10%
1.0-1.5
1.5-2.0
2.0-2.5
36%
21%
7%
3.0-3.5
3.5-4.0
1.0
29%
13%
1.0-1.5
8th-Grade Core GPA
60
54%
28%
20%
2.5-3.0
41%
47%
34%
1.0
35%
57%
20
24%
55%
39%
13%
9%
49%
40
25%
5%
38%
41%
30
9%
44%
80%
85%
14%
20%
69%
50
10
28%
31%
57%
70
60
11th-Grade Core GPA by 8th-Grade Core GPA
4%
9%
1.5-2.0
2.0-2.5
17%
2.5-3.0
GPA Greater Than 3.0
GPA Less Than 2.0: Not Qualified for College
Access to Selective Colleges1

More than 40% Graduate from College
Access to Somewhat Selective Colleges1

Fewer than 40% Graduate from College
3.5-4.0
8th-Grade Core GPA
Dropout
GPA 2.0-3.0
3.0-3.5
If students have an ACT score of at least 18.
Strong Test Scores Do Not Inoculate Students

Against Poor High School Outcomes
One of the most common strategies for improving
students educational attainment is to try to increase
their skills in reading and math. However, even
students with very strong test scores can struggle
in high school.
Among students with strong testC scores in
eighth grade:
C Strong test scores are defined as meeting the eighth-grade

2013 Illinois State Standards in both reading and math.
The 2013 standards were raised to be aligned with college
About a third (35 percent) had an average of B or

better in their ninth-grade core classes.
One-fifth (17 percent) had an average lower than C,
making them very unlikely to even enroll in college
and almost certain not to graduate.
Over a fifth (22 percent) missed more than 20 days
of school in ninth grade.
readiness benchmarks. These numbers are based on the

2009 ninth-grade cohort, applying the 2013 state standards
to define the group of high-scoring students in 2009.
Summary
of all but a few students who are still enrolled are so low
While a students eighth-grade record can give some
that they will be unqualified for any four-year college
idea of his likelihood to succeed in ninth grade, many
(less than a 2.0). At the same time, nine out of 10 stu-
Chicago students experience a large decline in perfor-
dents with an eighth-grade GPA greater than 3.5 go on
mance that is not evident until the transition to high
to finish eleventh grade with a GPA that provides access
school. Even students with exemplary eighth-grade
to at least a somewhat selective college (greater than a
records suffer some decline and are unlikely to main-
2.0). Students who aspire to college must be engaged in
tain their eighth-grade levels of performance during
their courses and performing at a high level even prior
high school. Only those eighth-graders with GPAs
to ninth grade.
greater than 3.0 have even a moderate chance of earn-
Most students are unlikely to earn As or Bs in ninth
ing As or Bs in ninth grade. Students with a good chance
grade, despite the fact they may be doing so in eighth
of high grades in high school tend to have eighth-grade
grade. High schools may want to investigate why it
GPAs of 3.5 or higher, along with strong attendance or
is so difficult for students with strong middle grade
test scores.
records to be successful in ninth grade. If high schools
This suggests that course performance in the
could monitor the performance of their students, they
middle grades needs to be very strong if students are
might discover what types of supports are needed to
to be to be ready for college at the end of high school.
keep grades from dropping. To help students at least
Earning a mix of Bs and Cs in eighth grade is simply not
maintain their levels of performance from the previous
good enough. Many students with an eighth-grade GPA
year would yield significant improvements in academic
less than 3.0 drop out by eleventh grade, and the grades
records and their likelihood of succeeding in college.
61
62
CHAPTER 6
Indicators of Whether Students

Will Meet Test Benchmarks
ACT scores, along with students GPA and other factors,
This chapter evaluates potential

middle grade indicators of performance on high school tests
(EPAS) and which combinations
of indicators provide the best prediction. This chapter focuses on
composite scores first, and then
on subject-specific EPAS tests.
determine the type of college students have access to after high school graduation. Students with ACT scores of
24 or higher, in combination with GPAs of 3.0 or higher,
are likely to have access to very selective colleges.
(Table A1 in Appendix A shows the relationships of
students eleventh-grade GPA, ACT scores, and the kind
of colleges students have access to given their qualifications.) Students with the same test scores who attend
colleges with different levels of selectivity have different earnings over their career. This is especially true
63
for minority students, and these differences in earnings based on college selectivity increase over time.
52
Besides being important for college access, ACT scores
better and a 75 percent chance of earning a C or better
are also important when applying for scholarships and
in corresponding college courses. 54 The PLAN college
loans to pay for college.
readiness benchmarks are 15 for English, 17 for reading,
In Chicago, students take three tests as part of the
19 for math, and 21 for science; the composite value is
Educational Planning and Assessment System (EPAS):
18. This chapter examines students average test scores,
the EXPLORE in the fall of eighth and ninth grade,
whether students reach the benchmarks in the different
PLAN in the fall of tenth grade, and ACT in the spring of
subjects, and whether students score a PLAN composite
eleventh grade.
of 18 or above.
53
These tests measure student achieve-
Many Chicago students do not meet the college read-
ment in English, reading, mathematics, and science.

The four subject-area scores are averaged together to
iness benchmarks on the PLAN; 48 percent of first-time
calculate a composite score. This portion of our analysis
freshmen in the fall of 2010, who took PLAN in the fall
focuses on students PLAN scores, because the PLAN is
of 2011, reached the benchmark in English, 30 percent
taken after the ninth-grade year. At the end of the chap-
did in reading, 16 percent in math, and only 6 percent
ter, the focus shifts to ACT scores, which are reported
in science. These subject-specific patterns are simi-
to colleges as part of the application process.
lar to the results of the ACT taken in eleventh grade.
The PLAN test scores range from 1 to 32. For each
While these percentages are lower than the national
subject, ACT has defined a college readiness bench-
numbers, the subject-specific trends mirror the trends
mark; this is the score at which ACT has determined
nationwide. Seventy percent of the 2011-12 PLAN-
that students have a 50 percent chance of earning a B or
tested tenth-graders, nationally, met the benchmark in
52 Dale and Krueger (2012); Hoxby (2001).

53 In the past, CPS students also took the PLAN test in the fall of
their eleventh grade. Some students might also take the ACT
in the fall of their twelfth grade to improve their ACT score
before applying to college.
54 Those are English composition, social science, college
algebra, and biology. Social science college courses
included history, psychology, sociology, political science,
Chapter 6 | Indicators of Whether Students Will Meet Test Benchmarks
and economics classes (Allen and Sconing, 2005). In September 2013, ACT revised the ACT college-readiness benchmarks
with more recent data (Allen, 2013). This has resulted in
changes to the reading and science benchmarks; the reading
benchmark went up a point, and the science benchmark went
down a point for all tests in the EPAS series. For tests taken in
2013 and later, ACT will apply these revised benchmarks.
English, 52 percent in reading, 36 percent in math, and
grade students are on a trajectory that makes it likely
27 percent in science.
they will meet the benchmarks in high school and what
55
Given the low percentages of students reaching the
the goals should be on tests given in the middle grade
benchmarks, many practitioners are looking for ways
years in order to be aligned with later high school tests.
to help students meet these benchmarks through early
Therefore, this chapter shows how middle grade indica-
intervention. Because the EPAS system is not available
tors are related to PLAN test scores and the degree to
prior to eighth grade, it is difficult for school practitio-
which they can predict whether students will reach the
ners, students, and parents to know whether middle
college-readiness benchmarks on the PLAN and ACT.
ACT Benchmarks
64
ACTs college-readiness benchmarks are described

in a report by Allen and Sconing (2005) and then
revised in Allen (2013).
Yet, while they provide meaning to a specific score,

differences in the preparation and later success of
students who meet the benchmark versus those who
The benchmarks were calculated based on data

that came from colleges that participated in ACTs
Course Placement Service. In the 2005 report, the
English benchmark analysis was based on 46 twoyear colleges and 46 four-year colleges with 76,122
students, while the science benchmark analysis
was based on 17 two-year colleges and 14 four-year
colleges with 14,136 students.D Benchmarks were
chosen to be the median value of ACT scores across
colleges that give a 50/50 chance to students to earn
a B or better in a college course. These cut-off scores
varied from college to college. For example, in half
of colleges, students with an ACT score of 18 have
a 50/50 chance of getting a B or better in English
Composition; in another quarter students only need
a 14 to have a 50/50 chance of earning a B or better,
while in 25 percent of colleges students need a score
of 20 in English ACT to have a 50/50 chance at a B or
higher. The variability is lower for algebra and science,
and fewer students were part of those analyses.
ACT benchmarks provide a context from which
one can make meaning out of a score. For example,
on its own, a score of 20 has no meaning. Knowing,
however, that the benchmark score on the math
portion of the ACT represents the point at which a
student has a 50/50 chance of getting a B or higher
in a college algebra class puts a score of 20 in context
as one that is not too far behind the benchmark.
do not are not necessarily meaningful. There are many

factors other than students ACT scores that are more
strongly associated with their college success than
their test score. These factors include attributes of the
college a student attends and the classes in which the
student enrolls at that collegethe quality of instruction, institutional setting, and fit between student and
college. They also include other academic skills and
noncognitive factors that are not measured on the
ACTsuch as students ability to show up to class and
put forth their best effort and creativity, as well as the
supports they receive from family and peers.
Just because students make a benchmark does not
mean that they will do well in their college classes;
students who meet the ACT benchmarks have a 50
percent chance of scoring Bs or better in their college
classes. That means that half of the students who
score just at or above the benchmark do not receive
at least a B. At the same time, many students who
score below the benchmark do perform well. With
ACTs reading benchmark, for example, a score of
21 gives a student a 50 percent chance of a B in a
college social science class, but a score of 16one
that is 4 points lower and equivalent to about three to
four years of growth in high schoolgives a student
a 40 percent chance of earning a B or better.E Thus,
a large difference in scores results in only a slight
difference in the probability of success.
D The 2013 report replicates the same analysis with more
E Based on graph of the relationship between ACT scores
recent data. The number of students varied from 131,000

for social science to 42,000 for biology, while the number
of colleges varied from 136 in the English Composition I
analysis to 90 for biology analysis.
and grades at a typical college in Allen and Sconing

(2005).
55 ACT, Inc. (2012).

PLAN Composite Scores Are

Strongly Predicted by Eighth-Grade
State Test Scores (ISAT Scores)
this, the relationship of past test scores with future test
Eighth-grade test scores on the ISAT are the stron-
to higher scores on PLAN (see Figure 26). Students
gest single predictor of tenth-grade PLAN test scores,
exceeding standards in eighth-grade math have a
compared to all other indicators of middle school
very good chance of scoring 18 or higher on the PLAN
performance. This is not surprisingjust as grades are
composite; more than 94 percent do so. Students at the
the best predictor of future grades, test scores are the
high end of the meets range have a very high chance
best predictor of future test scores. The relationship of
as well; 3 out of 4 students do so. Those students in the
past test scores to future test scores is much stronger,
lower part of the meets range have about a one in
however, than the relationship of past grades to future
four chance of attaining an 18 (28 percent). Students
grades. Students scores on ACTs PLAN are strongly
who do not meet standards in eighth grade have almost
tied to their incoming performance on the State ISAT
no chance of scoring 18 or above in PLAN composite.
tests. The tenth-grade PLAN composite score is cor-
Among students who take the PLAN in tenth grade,
related with students eighth-grade ISAT math score
those not meeting state standards in eighth grade are
at 0.79 and with their eighth-grade ISAT reading score
almost certain to score at chance on the PLAN (below
at 0.74. In general, correlations of 0.80 or higher tend
13). Their skills are not well measured by the test.
scores is very strong.

Higher math test scores in eighth grade correspond
Students with low scores in eighth grade are also less
to be measuring the same underlying construct. Note

that these are correlations between two different tests,
likely to actually take the PLAN test in the fall of their
which are not on the same scale, taken more than a
tenth-grade year. A quarter of students who did not take
year apart, and are testing different subjects. Despite all
the PLAN test left the district before the test took place
FIGURE 26
Percent of Students Not Taking PLAN Tests, Percent of Students Scoring 18 or Above on PLAN Composite,
and the Average PLAN Composite Scores by Eighth-Grade Math ISAT Scores
100%
100
94%
72%
70
20
19
34%
18
16
15
13
12
14
29%
30
12
28%
10
19%
20
14%
10
0
22
16
50
0%
120-233
Academic
Warning
10%
1%
14%
24%
5%
5%
234-250
251-266
Not Meeting
26%
26
24
22
60
40
25
267-288
289-309
Meeting
25%
PLAN Composite Scores
Percent of Students
28
1 Standard
Deviation
80
32
30
90
2%
4
2
310-360
361-410
Exceeding
11%
5%
0.3%
Percent of Students in Each Group
8th-Grade Math ISAT (2013 Student Performance Levels)

No PLAN Data
Percent of Students Scoring 18 or Above in PLAN Composite Scores
Average PLAN Composite Score
Note: This figure shows the relationship between eighth-grade ISAT math scores and the percent of students not taking PLAN in tenth grade (gray bars), the percent
of students scoring 18 or above in PLAN composite scores (orange bars), and the average composite score with a 1 standard deviation to show the variability in
scores (the squares represent the average PLAN composite score and the whiskers represent the 1 standard deviation around the mean). This chart shows the
performance of students who entered ninth grade in the 2009-10 school year, and took the PLAN in fall 2010. Eighth-grade test scores are represented using state
standards in 2013.
65
66
and 7 percent dropped out of school. The rest (66 per-
with tenth-grade test scores, the correlation is much
cent) were still in schoolsome of them were in ninth
weaker than with prior test scoresaround 0.56 (see
grade (25 percent), but a large proportion (41 percent)
Table 7). Combined, students background character-
was enrolled in tenth grade and did not take the test.
istics (e.g., race, gender, special education status, free
The analyses presented in this chapter are based on stu-
and reduced-priced status, neighborhood poverty and
dents who took the PLAN tests their second year of high
socioeconomic status, and whether students are old for
school. Because they represent only those students with
grade) explain a third of the variation in test scores,
test scores who were enrolled in the tenth grade, the av-
similar to core grades (see adjusted-R 2 in Table 7).
erage PLAN scores and percent meeting the benchmark
Special education, whether students are old for grade,
at each level of the ISAT are somewhat biased upward
free and reduced-priced lunch, and race data are the
they would likely be lower if all students who took the
background variables most strongly associated with
eighth-grade ISAT were tested on the PLAN in tenth
test scores. Attendance, misconducts, suspensions,
grade. Even with this upward bias, the probability of
grit, and study habits have very weak relationships with
scoring 18 or above on the PLAN composite for students
tenth-grade test scores, explaining a meager proportion
at the most typical ISAT score levels is very low.
of the variation in test scores.
Test scores in earlier grades are almost as predic-
PLAN scores are better predicted by using more than
tive of PLAN scores as eighth-grade test scores, and
one prior test score, from multiple subjects or grades.
much more predictive than background characteris-
Two middle grade test scores combined from different
tics, grades, or attendance. Table 7 shows the degree
subjects (reading and math) or from different grades
to which middle grade indicators predict students
(seventh and eighth grade) are slightly more predic-
performance on the PLANboth the prediction of the
tive of PLAN scores in any subject than just one test
composite score and the probability of scoring 18 on
by itself. For example, the adjusted-R 2 from a model in
the composite. Not only are tests taken in eighth grade
which eighth-grade math and reading ISAT scores are
highly correlated with PLAN composite scores, but
combined to predict PLAN math scores in tenth grade
tests taken in years prior to eighth grade are almost
improves from 0.63 to 0.68, compared to a model with
as predictive. The correlations between PLAN com-
only eighth-grade math ISAT scores (see Table 7 and
posite scores and ISAT scores in eighth, seventh, and
Table E.2 in Appendix E). This is because any one score
even sixth grade are nearly identical. The correlation
is likely to have measurement error; students have
between the ISAT scores in two consecutive years is
good days and bad days, and multiple measures pro-
also high: in reading it is 0.81 and in math it is 0.88. This
vide a more precise estimate of their true ability than
implies that students who score high one year will tend
one score by itself. This also suggests that each test
to score high the following year and the data from either
measures underlying general skills as much as subject-
yeareighth or seventh gradecan predict how stu-
specific learning. The addition of the information of
dents will do on the PLAN. Even sixth-grade test scores
one more test score helps with the prediction of the
are very good predictors of PLAN composite scores and
PLAN scores mainly for students close to exceeding or
better predictors than other eighth-grade indicators
exceeding in math in eighth grade, where there is more
that are not based on tests. Chapter 8 shows students
variation in their scores. This could be an indication
test trajectories in the middle grades and what these
of a lack of precision in scoring at the high end of the
correlations mean for differences in students test score
ISAT or that school effects are more important for these
growth over the middle grade years.
students, as discussed later in the chapter.
Eighth-grade course performance, other middle
Once we know students reading and math test
grade indicators, and background variables are weaker
scores in eighth grade, adding other information does
predictors of PLAN composite scores than middle grade
not help much with the prediction of high school test
test data. While eighth-grade core GPA is correlated
scores, with the exception of adding school information
TABLE 7
Eighth-Grade ISAT Scores Are Strong Predictors of Tenth-Grade PLAN Composite Scores; Other Middle
Grade Indicators Add Little to the Prediction
Single Indicator
PLAN Composite Scores
Scoring 18 or Above in PLAN Composite
Correlation
Adjusted-R 2
Pseudo-R2
Percent Correct*

Linear Term
Linear & Squared Terms
0.74
0.56
0.58
0.38
0.38
86.5%
86.5%
8th-Grade Math ISAT

Linear Term
0.79
0.63
0.63
0.39
0.39
87.3%
87.4%
0.74
0.55
0.35
85.8%
7th-Grade Math ISAT
0.76
0.61
0.38
87.0%
0.74
0.54
0.35
84.8%
6th-Grade Math ISAT
0.76
0.58
0.36
85.4%
8th-Grade Core GPA
0.56
0.31
0.22
82.5%
0.22
0.05
0.03
77.2%
-0.16
0.02
0.02
77.2%
8th-Grade Misconducts
-0.11
0.01
0.01
77.2%
8th-Grade Grit
0.04
0.00
0.00
74.7%
0.02
0.00
0.00
74.8%
8th-Grade Background
Characteristics
0.32
0.18
80.4%
8th-Grade Reading and

Math ISAT Tests
0.68
0.44
89.0%

Math ISAT Tests + Core GPA
0.69
0.45
89.3%
8th-Grade Reading and Math

ISAT Tests + Background
Characteristics
0.70
0.44
89.2%
All 8th-Grade Student-Level

Indicators
0.71
0.45
89.6%
0.72
0.46
89.8%
School Effects Alone

(Middle and High School
Effects)
0.50

Indicators + School Effects
0.75
Combining Two or More 8th-Grade Indicators
Adding Seventh-Grade Indicators

8th- and 7th-Grade
Reading and Math ISAT Tests
Adding School Effects
Note: See Table E2 in Appendix E for a complete analysis of all middle grade indicators, including gains and growth in test scores. Sample size was kept the same
for most analyses to make comparisons easier, except when data from surveys were analyzed. In those cases the sample sizes get smaller. That is the case when
grit and study habits are part of the analysis. * Given the data for this cohort and analyses, a model with no explanatory variables would be able to correctly
predict 77.2 percent of students whether they score 18 and above versus lower than 18. Given the low variability in the percent of students scoring 18 or above
in the PLAN composite by middle and high school, these models could not be run.
67
or test scores from prior years. Adding core GPA, for
random guess, as indicated by the location of those
example, to eighth-grade reading and math test scores
model statistics on the dotted diagonal line. Other
improves the adjusted-R from 0.68 to 0.69, and adding

2
indicators (e.g., background characteristics, core GPA,
all eighth-grade variables gives an adjusted-R of 0.71.
or middle grade test scores) provide increasingly more
Adding seventh-grade ISAT scores in reading and math
accurate predictions, as indicated by the movement
increases the adjusted-R to 0.72. Given the small
of the dots further to the upper-left of the figure. In
improvements of including extra variables, it may not
particular, using students scores from multiple tests
be worth including more than two test scores in an
(reading and math, seventh and eighth grade) improves
indicator system to predict test scores in high school.
the accuracy of the prediction.
An indicator system based on reading and math

Students test scores in high school depend on
the students at risk of not meeting the benchmark on
where students attend middle school and high
the PLAN composite (see Figure 27). At the same time,
school. Middle school and high school effects explain
30 percent of the students who are identified as at risk
part of the variation we see in PLAN scores for students
will actually have a score of 18 or above. (The y-axis of
with similar middle grade test scores (see Table 7).
Figure 27 shows the proportion of students correctly
Figure 28 shows the degree to which students PLAN
classified as scoring below 18, and the x-axis shows the
scores are systematically different, depending on which
proportion of students incorrectly classified as scoring
middle school and which high school they attend (see
below 18. The different points represent various models
panel A for the impact of attending different middle
for predicting whether students meet a PLAN compos-
schools and panel B for high schools, net of the effects
ite score of 18.) A model using attendance, suspensions,
of the other). Students with the same predicted PLAN
misconducts, grit, or study habits to predict which
scores, based on ISAT scores in the middle grades, can
students score below 18 is no better than making a
have PLAN scores that differ by as much as two points,
FIGURE 27
Correct versus Incorrect Classification of Students Not Scoring 18 on the PLAN Composite
Proportion of Students Correctly Identified as Scoring Below 18
True Positive Proportion (Sensitivity)
68
eighth-grade scores correctly identifies 95 percent of
1.0
Perfect
Prediction
0.9
Core GPA
6th-Grade ISAT Math
8th-Grade ISAT Math
7th-Grade ISAT Math
0.8
All
0.7
8th-Grade ISAT Reading
8th- + 7th-Grade
ISAT Math +
Reading
0.6
8th-Grade ISAT Math + Reading

8th-Grade ISAT Math + Reading
+ Background Characteristics
0.5
Attendance
Also suspensions,
misconducts, grit,
and study habits
8th-Grade ISAT Math +

Reading + Core GPA
0.4
0.3
Ra
0.2
nd
om
Gu
es
0.1
Worse
Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Proportion of Students Incorrectly Identified as Scoring Below 18

0.9
1.0
FIGURE 28
Variation in Predicted PLAN Scores by Predicted PLAN Based on Middle Grade Indicators
Panel A: Elementary School Effects (Controlling for High School Effects)
Predicted PLAN Scores Depending on

Elementary School Effects
32
30
28
26
24
22
1-12
13-15
16-18
19-23
24-27
28-32
30.9
Low
ES Effects
Average
High
-0.3
-0.3
-0.4
-0.4
-0.3
0.1
0.0
0.0
0.0
0.2
0.5
1.1
0.3
0.3
0.5
0.8
1.4
2.2
26.5
20.9
28.8
24.8
20
16.9
18
16
14.2
14
11.7
12
10
19.7
16.0
13.6
11.1
1-12
13-15
16-18
19-23
24-27
28-32
Predicted PLAN Score Based on Middle Grade Test Scores

High ES Effect
Average ES Effect
Low ES Effect
69
Panel B: High School Effects (Controlling for Middle School Effects)
Predicted PLAN Scores Depending on

High School Effects
32
30
28
26
24
22
1-12
13-15
16-18
19-23
24-27
28-32
30.4
Low
HS Effects
Average
High
-0.7
-0.7
-0.6
0.1
1.3
1.9
-0.3
-0.1
0.1
1.1
2.2
2.5
0.2
0.4
0.8
2.1
3.0
3.0
26.5
21.3
29.3
24.8
20
17.1
18
16
14
12
10
14.4
11.9
19.3
15.8
13.3
10.9
1-12
13-15
16-18
19-23
24-27
28-32
Predicted PLAN Score Based on Middle Grade Test Scores

High ES Effect
Average ES Effect
Low ES Effect
Note: These graphs show the variability of students predicted PLAN composite scores depending on which elementary/middle school (panel A) or high school (panel
B) students attend. The graphs are based on cross-nested models in which students are simultaneously nested in their elementary/middle school and also in their high
school. School effects are determined from the variation in school residuals, after controlling for students eighth-grade reading and math ISAT scores; see Appendix
C for additional details on the models. The black squares in each graph represent the average PLAN composite score based on students eighth-grade reading and
math ISAT scores. The diamonds above and below the black squares represent the predicted PLAN scores for students with similar eighth-grade ISAT scores attending
elementary/middle schools with high and low effects (panel A) and attending high schools with high and low value effects (panel B); high and low school effects are
defined as one standard deviation above or below the mean, respectively. The table on each graph shows the value of these school effects as well (i.e., how many PLAN
points the model predicts that a student score based on whether they attend a school where students score higher or lower than typical). Each row represents the
range of school effects for students with similar eighth-grade records; values by columns can be compared to see how different school effects are for students with
different performance in eighth grade.
depending on which middle school they attended, after
generally attend high schools that contribute to their
removing any effects of where they attend high school.
scores by almost two points or more on average. These
Among students with high scores, middle school effects
estimates come from models that compare gains among
are particularly strong. These differences might be
students with the same incoming scores, suggesting
due to variation in the types of skills students learned
that students with high skills leaving the middle grades
in the middle gradesstudents at schools that taught
tend to go to high schools with above-average gains.
specifically to the types of questions on the ISAT, at the
70
Students who attend middle schools with high-
expense of the types of skills and questions asked on
achieving, more advantaged peers are more likely to
ACTs EPAS system (PLAN test), might not score as well
score higher on their PLAN math test than their ISAT
on the PLAN as their ISAT scores would suggest. Some
scores would predict, compared to students with the
middle schools may have better prepared their students
same eighth-grade test scores who attended middle
for the type of work that prepares them to score well on
schools with lower-achieving, less advantaged peers.
the EPAS. Schools at which students perform better on
Table 8 shows the relationships of middle school charac-
high school tests than their middle grade tests would
teristics with middle school effects on test scoreseffects
predict also tend to have students that get better high
that are not reflected in students ISAT scores, and net of
school grades than their middle school grades would
any effects that result from where their students attend
predict; the correlation between middle school effects
high school. The second column shows high school ef-
on grades and tests scores is 0.34, net of high school
fects, net of any effects of where students attend middle
effects and students eighth-grade achievement.
school, and controlling for their eighth-grade ISAT
Variation in PLAN scores for students with the same
scores. The final column shows middle school effects
ISAT scores also depend on the high school they attend.
without controlling for high school effects, allowing for
These differences could represent different rates of
the fact that middle schools tend to send their students
learning in high school or alignment of high school in-
to particular high schools. Students high school PLAN
struction with the PLAN. High school effects tend to be
scores are better than their ISAT scores at predicting
larger for students scoring in the middle of the distribu-
if they come from middle schools serving students who
tion, not the extremes. Two students with same middle
come from neighborhoods with lower levels of poverty
grade ISAT score, who came from the same elementary
(r=-0.34), fewer students who are over-age for their grade
school, can vary in their PLAN math score by as much
(r=-0.34), and more students who are white (r=0.32).
as 2 points, depending on which high school they attend.
Students are also more likely to have higher PLAN
Students gain about a point a year on average, which sug-
composite scores if they attend high schools with
gests that students gain about twice as much each year
higher-achieving and more advantaged peers (see the
on their math scores at some schools than at others.
second column of Table 8), even after removing any
High ISAT scores not only increase students likeli-
effects of their middle school. The absolute size of these
hood of scoring well on the PLAN but also give them
correlations is always larger for high schools than for
access to high schools where students show higher-
middle schools, but these correlations only reflect
than-average gains. Students with low scores in the
school effects that are not reflected in students eighth-
middle grades generally attend high schools where
grade ISAT scores. Attending high schools with high-
students make smaller-than-average gains. As shown in
scoring students is particularly strongly associated with
the table within panel B of Figure 28, among students
students scoring higher on the PLAN than their ISAT
with low test scores, some attend high schools where
scores would predict (r=0.77). This is consistent with
the average school effect is 0.7 points below average,
prior research that suggests students learn more in class-
while others attend schools where the average effect is
es with higher-achieving peers. 56 Their scores are also
about 0.2 above expected; while there is a range among
higher if they attend high schools serving students who
the school effects, none are particularly high and some
come from neighborhoods with lower levels of poverty
are quite low. On the other hand, high-scoring students
(r=-0.42), fewer students who are over-age for their grade
TABLE 8
Relationships of School Characteristics with School Effects on Math PLAN Test Scores

Correlation of Middle
School Effects with School
Variables, Controlling for
High School Attended
ISAT Math
% Latino
% African American
% White
Correlations of High
Variables, Controlling for
Middle School Attended
0.21***
0.77***
Correlation of Middle
Variables, Not Removing
High School Effects
0.40***
0.12**
0.09
0.08~
-0.27***
-0.26**
-0.29***
0.32***
0.46***
0.44***
% Over Age
-0.34***
-0.48***
-0.43***
% Special Education
-0.17**
-0.34***
-0.14***
Average
Concentration
of Poverty
-0.34***
-0.42***
-0.43***
Average Social
Capital
0.15**
0.29**
0.30***
-0.13**
-0.43***
-0.28***
0.58***
0.32***
General
Magnet
0.18***
Vocational
NA
-0.12
NA
APC
NA
-0.18*
NA
Note: Residuals result from 1) a 2-level model where students are nested with in elementary schools and 2) a 2-level cross-nested model where students are
nested with in elementary and high schools. Each model was used to predict the PLAN composite scores based on eighth-grade ISAT reading and math scores
at the student level. No predictors were included at the school level.
(r=-0.48), and more students who are white (r=0.46).

Because students from high-performing, more
the correlations of the PLAN subject-specific scores

with middle grade test scores and grades. More detailed
socially advantaged middle schools are also likely to at-
information on other middle grade predictors can be
tend more advantaged high schools, the relationships of
found in Appendix E . The correlations between the
middle school characteristics with middle school effects
reading and math ISAT tests are lower with the PLAN
are larger if we do not control for high school effects
subject-specific scores than with the composite scores.
when gauging middle school effects (see column three
The ISAT math test by itself is correlated at 0.79 with
of Table 8). Students from socially advantaged middle
the PLAN composite score, which is higher than the
schools (those with fewer students in high-poverty
correlation with the PLAN math test (0.77). Such high
neighborhoods, fewer students old for grade, and serving
correlations suggest that the tests may be measuring
more white students) have higher high school test scores
general academic skills and the ability to take tests,
than their ISAT scores would predict, not only because
at least as much as they measure skills in specific
their middle schools provide direct advantages, but also
subjects. The correlations between the ISAT scores
because these students are more likely to attend high
and the PLAN composite are higher than the correla-
schools where students score higher than expected.
tions with the subject-specific tests because there

is less measurement error when multiple scores are
Any subject-specific ISAT test is strongly predictive
combined (random errors on one test tend to cancel
of any subject-specific PLAN score. Improving PLAN
each other out). The reading and math ISAT tests
composite scores requires improving ISAT scores in
together are more predictive of any subject test than
any or all of the subject-specific tests. Table 9 shows
the subject-specific tests are of later tests of the same
56 Nomi and Allensworth (2013); Ballou (2007); Gamoran (1996).
71
TABLE 9
Correlations of Test Scores and GPA in Eighth Grade with Tenth-Grade Test Scores

Tenth-Grade PLAN
72
English
Reading
Math
Science
Composite
0.70
0.66
0.64
0.60
0.75
8th-Grade ISAT Math
0.71
0.62
0.77
0.66
0.79
ISAT Reading & Math Combined
0.76
0.68
0.76
0.67
0.83
0.49
0.44
0.46
0.41
0.52
8th-Grade Math GPA
0.42
0.37
0.45
0.39
0.47
8th-Grade Core GPA
0.51
0.46
0.50
0.45
0.56
subject (except in the case of math). Likewise, combin-
level of the prior or subsequent test. Two test scores pro-
ing seventh- and eighth-grade scores provides a better
vide a better prediction than one scoreeither by combin-
prediction than scores from just one grade level.
ing the scores across subjects or across grades, since both
Eleventh-Grade ACT Scores Are

Highly Correlated with Scores in
Sixth, Seventh, and Eighth Grade
In eleventh grade, students take the ACT tests. The
scores from these subject tests, along with the compos-
reduce the measurement error associated with one score.

There is no need to develop a complicated prediction
system that includes information about students backgrounds or course performance, since this provides little
additional accuracy to the prediction of later test scores.
Students from particular middle schools perform
ite score, are part of the information that colleges use
better on the high school tests than expectedgiven
for determining admissions. Not surprisingly, the best
their ISAT scoresregardless of the high school they
indicators of ACT composite scores and whether stu-
attend. This suggests that some middle schools may be
dents will score 21 or above are the same indicators that
preparing their students in ways that are more aligned
are most predictive of PLAN scores (see Table 10). Test
with the types of skills tested on the EPAS than other
scores in eighth grade or seventh grade, or even sixth
schools, while other schools are preparing students to
grade, are highly correlated with ACT scores, just as
do well on the ISAT in ways that do not translate to the
they are highly correlated with PLAN scores. The corre-
EPAS. There are also high school effects; students make
lation between the PLAN and ACT composite scores is
higher gains on tests at some high schools than others.
very high (0.89), but still the correlations between
In general, students with higher scores in the middle
middle grade ISAT tests and the ACT tests taken
grades enroll in high schools with higher average gains.
three years later are also very strong (between 0.75
They might have more options available to them than
and 0.80, depending on grade level and subject).
lower-scoring students.
Other information, such as GPA, attendance, suspen-
Most CPS students are at risk of not reaching the
sions, and grit, does not add much to the prediction of
benchmarks on the PLAN or the ACT. Thus, targeted
ACT, once we know students eighth-grade scores.
interventions would be impractical, even though stu-
Summary
dents can be accurately identified. Also, because the

prediction of reaching EPAS benchmarks is so precise,
Students middle grade ISAT scores are highly predictive
it suggests that it is difficult to change students test
of their future scores on the EPAS system. Reading and
score trajectories, as few students achieve outcomes
math ISAT tests strongly predict performance on any of
other than those that were expected. This is discussed
the high school tests, regardless of the subject or grade
more thoroughly in Chapters 7 and 8.
TABLE 10
Relationships of Middle Grade Indicators with ACT Composite Scores

Single Indicator
ACT Composite Scores

Correlation
Adjusted-R
0.76
8th-Grade Math ISAT

Linear
Scoring 21 or Above in ACT Composite

Pseudo-R2
Percent Correct*
0.58
0.58
0.39
0.39
84.1%
84.1%
0.80
0.64
0.64
0.41
0.41
85.7%
85.7%
0.75
0.57
0.36
83.5%
7th-Grade Math ISAT
0.80
0.63
0.40
84.9%
0.75
0.56
0.36
83.1%
6th-Grade Math ISAT
0.77
0.59
0.37
83.6%
8th-Grade Core GPA
0.56
0.31
0.22
78.9%
0.18
0.03
0.02
72.0%
-0.13
0.01
0.02
72.0%

Linear
-0.08
0.01
0.01
72.0%
8th-Grade Grit
0.01
0.00
0.00
69.7%
-0.01
0.00
0.00
69.7%
8th-Grade Background
Characteristics
0.33
0.20
77.1%

Math ISAT Tests
0.70
0.46
87.3%

Math ISAT Tests + Core GPA
0.71
0.46
87.6%
8th-Grade Reading and Math

ISAT Tests + Background
Characteristics
0.73
0.47
88.0%
All 8th-Grade
Student-Level Indicators
0.74
0.47
88.0%
0.73
0.48
88.2%
School Effects Alone

(Middle and High School
Effects)
0.51

Indicators + School Effects
0.79
Combining Two or More 8th-Grade Indicators
Adding Seventh-Grade Indicators

8th- and 7th-Grade Reading
and Math ISAT Tests
Adding School Effects
Note: There are 14,928 students included in these analyses. A model with no explanatory variables would be able to predict correctly 72.0 percent of students
whether they score 21 and above versus lower than 21. Given the low variability in the percent of students scoring 21 or above in ACT composite by middle
and high school, these models could not be run.
73
74
CHAPTER 7
Who Is at Risk of Not Reaching

the PLAN and ACT Benchmarks?
In the previous chapter, we showed that middle grade
test scores strongly predict how students will score on
their tenth- and eleventh-grade tests. In this chapter,
we show who is at risk of not reaching benchmarks and
how middle grade ISAT scores are related to PLAN
scores. Only students with very high test scores in the
middle grades (close to exceeding standards in math
and in reading) have a good chance to reach the bench-
This chapter shows the levels

of performance in the middle
grades that give students a
chance of meeting PLAN and
ACT benchmarks in the tenth
and eleventh grades.
marks in tenth grade. Unfortunately, most students are

scoring far from the high scores (high meets range
or exceeds standards) and have an extremely small
Students are less likely to meet the math and science
chance of reaching those benchmarks. The strong rela-
benchmarks than the reading and English benchmarks.
tionship between ISAT and EPAS scores allows middle
This occurs not because students have lower perfor-
grade practitioners, parents, and students to identify
mance in math and science but because the benchmarks
students likely scores on the PLANwhich could help
are set higher. Based on their middle grade test scores,
them set challenging, but attainable, learning goals that
about one-third of students are at very high risk (less
are appropriate for individual students, as opposed to
than a 1-in-4 chance) of scoring below the English
benchmarks that are universally applied even if unat-
benchmark, and over half of students are at very
tainable by currently known practices.
high risk of scoring below the reading benchmark (see
The Vast Majority of Students Are

at Very High Risk of Not Reaching
the Tenth-Grade PLAN Benchmarks
Figure 29). Over three-fourths of students are at very
high risk of missing the math benchmark, and over 90

percent of students are at very high risk of missing the
science benchmark. Nationally, in 2011-12, only about
More than half of CPS students are not meeting
one-fifth of PLAN takers pass all four benchmarks;
standards, and one-quarter just barely meet standards
70 percent pass English, 52 percent pass reading,
(low range of the meets range) in eighth-grade math.
36 percent pass math, and 27 percent pass science. 58
A student who is just meeting standards or is just below
While many students have extremely low probabili-
standards is likely to score below the benchmark in all
ties of reaching any of the ACT benchmarks, there are
four subjects tested in PLANfrom two points below in
some groups of students who have a fairly good chance
English to six points below in science. In order to meet
of reaching some of the benchmarks. Almost half of
the ACT benchmarks at the end of eleventh grade, stu-
students have a good chance (greater than 50 percent)
dents who are one point behind on the PLAN will need
of meeting the benchmarks in English, and almost a
to make gains that are twice what is typical over the
quarter of students have a good chance of meeting the
tenth- and eleventh-grade years, while students that
benchmarks in reading. On average, these students
are two points behind would need to grow three times
have scores that are at the high end of the meets range
the average rate. 57
on their reading tests in the middle grades. Fourteen
57 Based on a typical growth of just over one point per year.
58 ACT, Inc. (2012).
Chapter 7 | Who Is at Risk of Not Reaching the PLAN and ACT Benchmarks?
75
FIGURE 29
Percent of Students at Different Risks for Not Reaching Benchmarks on the PLAN English, Reading, Math,
and Science Tests
English
90
90
80
80
70
60
50
40
32.4%
29.0%
30
21.1%
17.6%
20
70
60
40
30
Very High
(75%+)
High
Moderate
(50 to <75%) (25 to <50%)
Low
(<25%)
Average
8th-Grade
Math ISAT
224(B)
242(B)
12.6%
High
Moderate
(50 to <75%) (25 to <50%)
Low
(<25%)
Risk of Not Reaching Reading Benchmark of 17
251(M)
267(M)
8th-Grade
Reading ISAT
232(B)
252(M)
261(M)
278(E)
221(B)
246(M)
258(M)
281(E)
8%
37%
64%
87%
241(B)
259(B)
271(M)
297(M)
11%
37%
63%
91%
% Reaching 17
Math
100
100
90
Science
91.0%
90
77.3%
80
Percent of Students
Percent of Students
Very High
(75%+)
7th-Grade
Reading ISAT
% Reaching 15
70
60
50
40
30
20
70
60
50
40
30
20
8.9%
10
0
12.6%
Average
8th-Grade
Reading ISAT
80
19.7%
10
Risk of Not Reaching English Benchmark of 15
76
55.1%
50
20
10
0
Reading
100
Percent of Students
Percent of Students
100
Very High
(75%+)
6.1%
High
Moderate
(50 to <75%) (25 to <50%)
7.7%
Low
(<25%)
10
0
Risk of Not Reaching Math Benchmark of 19

Average
8th-Grade
Math ISAT
7th-Grade
Math ISAT
% Reaching 19
4.8%
Very High
(75%+)
2.5%
High
Moderate
(50 to <75%) (25 to <50%)
1.7%
Low
(<25%)
Risk of Not Reaching Science Benchmark of 21

Average
256(B)
289(M)
300(M)
322(E)
8th-Grade
Reading ISAT
242(B)
273(E)
281(E)
299(E)
8th-Grade
Math ISAT
261(B)
307(M)
322(E)
345(E)
3%
39%
60%
81%
242(B)
278(M)
288(M)
312(E)
4%
36%
63%
88%
% Reaching 21
Note: The first bar on the left of each graph shows the percent of students who are at very high risk of not reaching the benchmark (greater than 75 percent chance).
The next bar represents the group of students at high risk: students with a probability between 50 and 75 percent of not reaching the benchmark. The third bar shows
students with probabilities between 25 and 50 percent, who are at moderate risk. And the last column shows the percent of low-risk students, with probabilities less
than 25 percent of not reaching the benchmark. Each chart is created based on a model with the best two predictors. In the case of English and science, those two
predictors are the eighth-grade ISAT reading and math scores; in the case of reading the two best predictors were the eighth- and seventh-grade ISAT reading and
the in the case of math they were the eighth- and seventh-grade ISAT math. The letters in parenthesis next to the average ISAT scores denotes the 2013 student
performance level.
percent of students have at least a 50 percent chance
more important that students score as high as possible
of meeting the PLAN math benchmark based on their
so that they can access a wider range of colleges.
middle school math scores, and less than 5 percent do
Students with the same ISAT scores can end up
on the science test. These students were exceeding
with very different PLAN scores, although it is unlikely
standards, or close to exceeding standards, on the math
for them to move out of a particular range of scores.
tests in the middle grade years. Only students who are
For example, of the students scoring 241-245 in math
exceeding, or close to exceeding, standards on both
in eighth grade, none reach an 18 on the PLAN; but
the math and reading ISAT in eighth grade have a good
20 percent score 15 or higher, 20 percent score 14,
chance of scoring at least an 18 on the PLAN composite
25 percent score 13, and 34 percent score 12 or below
(see Table 11). Students who just meet state standards
(see Figure 30). A PLAN score of 15 gives students a
on both the math and reading tests in eighth grade have
shot at attaining ACT scores that make them eligible
just a 28 percent chance. Students who are not meeting
to enter a four-year college. For these students, a chal-
state standards on either test have close to a zero prob-
lenging, yet attainable, goal would be to aim for 15 in
ability of reaching an 18 on the PLAN composite.
the PLAN composite. This has implications for the

type of instruction that will lead them to make the
The Benchmarks Are Not

Meaningful for Many Students
largest gains.
Students with ISAT math scores in the meets range
The reality is that most students in CPS come to high
of 286-290 have a 50/50 chance of scoring 18 or above
school with skills that make them very unlikely to reach
on the PLAN and end up with scores that range from
the ACT benchmarks. However, that does not mean that
about 13 to 24. These are students with a good shot at
they will be unprepared for college. The benchmarks
making ACTs benchmarks and being eligible for selec-
provide a context for understanding the scores, but they
tive colleges and scholarships, but their prospects are
are not deterministic in terms of students actual per-
uncertain. Their experiences in high school will shape
formance in college (see ACT Benchmarks, p.64). It is
whether they attain the benchmark goal. High schools
TABLE 11
Percent of Students Scoring 18 or Higher in PLAN Composite by Eighth-Grade Scores

Eighth-Grade Reading ISAT
(2013 Student Performance Levels)
Academic
Warning
Eighth-Grade Math ISAT
(2013 Student
Performance Levels)
Academic Warning
Below
Standards
Meets
Standards
120-217
218-232
233-247
120-233
0%
n=787
0%
n=609
0%
n=293
234-250
0%
n=682
0%
n=1,901
1%
n=1,841
4%
n=367
251-266
0%
n=128
1%
n=870
2%
n=2,526
11%
n=1,273
23%
n=212
4%
n=224
9%
n=1,793
28%
n=2,292
53%
n=949
72%
n=358
34%
n=206
62%
n=814
79%
n=729
90%
n=681
81%
n=154
95%
n=287
99%
n=597
Below Standards
267-288
Meets Standards
289-309
310-360
Exceeds Standards
361-410
Note: Based on students who entered ninth grade in the 2009-10 school year.
248-259
260-270
Exceeds
Standards
271-319
320-364
77
should pay particular attention to students in this
to have access to highly selective colleges, which often
group in order to help them earn eligibility to selective
have higher graduation rates and more financial aid
enrollment colleges.
than less selective colleges, they will need much higher
Students who are exceeding ISAT standards are
scores. This is a group of students who should aim for
likely to score at least an 18 on the PLAN, but in order
test scores in the 20s.
FIGURE 30
Distribution of PLAN Composite Scores by Eighth-Grade Math and Reading Test Scores
Panel A: Math
100%
90%
70%
60%
50%
40%
30%
20%
10%
336-410
296-364
321-325
331-335
Meets
39%
316-320
311-315
306-310
301-305
296-300
291-295
286-290
281-285
276-280
271-275
266-270
261-265
256-260
251-255
246-250
Below
47%
291-295
Academic Warning
8%
241-245
236-240
231-235
226-230
221-225
216-220
211-215
120-210
0%
Exceeds
6%
Panel B: Reading
100%
90%
80%
Percent of Students
70%
60%
50%
40%
30%
20%
10%
Academic Warning
8%
Below
49%
Meets
34%
281-285
276-280
271-275
266-270
261-265
256-260
251-255
246-250
241-245
236-240
231-235
226-230
221-225
216-220
211-215
206-210
201-205
196-200
191-195
186-190
176-180
171-175
0%
120-170
78
Percent of Students
80%
Exceeds
9%
1-12: Not Well Measured by Test
18-20: Likely to Reach ACT Benchmark of 21
13-17: Very Unlikely to Reach ACT Benchmark,
21-24: Already at ACT Benchmark of 21
Except by Random Chance
24+: Probably Eligible for Highly Selective College
Note: Each horizontal white line in the vertical bars indicates a particular score point.
Low Test Scores Mean Most Students

Will Have Access to College Only if
They Have High GPAs
above 21 on the ACT; this represents the largest group of
When we follow students to the end of the eleventh grade,
access only to two-year colleges, where the likelihood
we can see that only students exceeding standards, or
of eventually obtaining a four-year degree is extremely
in the high end of the meets range, on the ISAT have
small. These students need to have GPAs of at least 2.5
a good chance of meeting a score of 21 on the ACT (see
in order to have access to somewhat selective colleges.
CPS students. Furthermore, on average, their core GPA

in eleventh grade is 2.3, which will provide them with
Of students not meeting the old standards in eighth
Figure 31). A score of 21 on the ACT composite is consid-
ered by many to be college-ready, as it is the average of
grade (234-266), hardly any score higher than an 18
the subject-specific benchmarks. Many of the students
on the ACT. With an average eleventh-grade core GPA
exceeding standards in math or in reading in eighth
of 1.8, these students have access to few colleges. If
grade score a 24 or higher on their ACT, and most score at
students not meeting ISAT standards enter high school
least a 21. Almost half have a core GPA greater than 3.0,
with plans to go to college, they need support and mo-
which provides them with access to very selective col-
tivation to ensure their grades are as high as possible,
leges. The other half of students with the top test scores,
earning at least as many As and Bs as Cs. Strong effort
however, do not have a GPA of at least 3.0, and this will
in their classes should also pay off for improving their
limit their access to college despite their high test scores.
test scores. 59 Pushing these students to get higher GPAs
Students who just meet standards on the ISAT math

(in the 267-288 range) are extremely unlikely to score
and improve their academic skills could improve their

chances to attend a four-year college.
FIGURE 31
Eleventh-Grade ACT Composite Scores by Eighth-Grade ISAT Math and Reading Scores
Reading
100
90
90
80
80
Percent of Students
Percent of Students
Math
100
70
60
50
40
30
70
60
50
40
30
20
20
10
10
233
234-250 251-266 267-288 289-309 310-360
360
217
8th-Grade ISAT Math

A.W.
9%
Below
21%/24%
Meets
28%/13%
218-232 233-247 248-259 260-270 271-319
320

Exceeds
6%/0.4%
A.W.
9%
Below
15%/32%
Meets
24%/11%
ACT 24: Access to Selective Colleges if GPA 2.0 and
ACT 18-20: Access to College if GPA > 2.0,
Very Selective Colleges if GPA > 3.0
Access to Selective Colleges if GPA > 3.0
ACT 21-23: Access to College and to Selective Colleges
ACT 17: Access to College Only if GPA Is > 2.5
if GPA 2.5
Dropout
59 High grades in math, science, and English are the strongest

predictors of gains on those subject tests in a given year
(Allensworth, Correa, and Ponisciak, 2008).
Exceeds
9%/0.3%
79
Summary
for most students except by random chance. The goal
Parents, teachers, and students can gauge students
should be to reach the highest attainable scores, even if
likely PLAN and ACT scores based on their ISAT
those scores are below benchmarks. High PLAN scores
performance and develop meaningful goals for learn-
will put students on a better path to reach high ACT
ing based on college expectations before students begin
composite scores; this, in combination with high GPAs,
high school. What those goals are for an individual
will help students to access more selective colleges,
student should depend not only on readiness bench-
even if a student does not score at the benchmark.
marks and standards but also on students current level

of academic skills.
Test scores are useful for parents and teachers to
80
What can be done before students reach high school?

One strategy is to be discerning when advising students
about applying to high school, since the previous
gauge students level of academic skills; they provide
chapter showed that students with the same middle
information that can be used to guide further instruc-
grade scores make very different test score gains at
tion and decisions about coursework. State standards
different high schools. Another obvious answer is to
and college-ready benchmarks provide a context for
assist students to attain higher levels of performance
making meaning from scores. In practice, however,
in the middle grades so that they have a better chance
these benchmarks do not make good goals for the vast
of getting good PLAN and ACT scores. There is a push
majority of students. For students exceeding eighth-
to move all students to the exceeds range on the ISAT,
grade standards in both reading and math, the ACT
and this year standards changed at the state level to
benchmarks provide targets that are too low to provide
more closely align with college expectations. These
a good chance of attaining scholarships and getting
benchmarks are useful only if schools know how to
into highly selective colleges. All other students have a
improve students test scores sufficiently to reach
very slim chance of meeting the ACT benchmark scores,
these levels of performance. Chapter 8 shows how
potentially leading to frustration and disappointment
much students test scores grow in the middle grades;
on the part of students and school staff. Unless schools
variation in growth is much smaller than most people
can figure out how to substantially increase the rates
believe, especially in reading. Currently, there are no
at which the students learngiven the resources they
schools that come close to moving students from just
can make availablethe benchmarks are not attainable
meeting standards to exceeding them.
CHAPTER 8
How Grades, Attendance, and

Test Scores Change
As middle schools work to prepare students for high
school, improving attendance, grades, and test scores
are often their primary goals. This is in perfect alignment with the factors that are most predictive of high
school outcomes. But how much do students grades,
attendance, and test scores change over the course
of middle school? This chapter examines trends in
students grades, attendance, and test scores over the
middle grade years, showing how much they change
and the degree to which differences in growth vary
across schools.
Chapter 3 showed that students grades and attendance in earlier years are predictive of ninth-grade
course performance. Seventh-grade GPA is almost as
This chapter examines the degree

to which students show different amounts of growth and decline in attendance, grades, and
test scores over the middle grade
years. It ends by estimating which
indicators would have the biggest
leverage for increasing students
educational attainment, if they
could attain high levels of growth
in any of these three areas.
predictive as eighth-grade GPA, but seventh-grade

attendance is less predictive. This suggests that stu-
their high school PLAN scores as their eighth-grade
Over the Middle Grades,

Attendance Is Less Constant
Than Grades, Which Are Less
Constant than Test Score Ranks
ISAT scores. If sixth-grade scores are as predictive
Compared to grades or test scores, attendance is the
of high school performance as eighth-grade scores, it
factor that is most likely to change over time. As shown
suggests that there may be little change in students
in Table 12 , the correlation between seventh-grade at-
scoresrelative to other studentsover the middle
tendance and eighth-grade attendance is strong (0.62),
school years. If this is true, what should be the
but not nearly as strong as the correlation of seventh-
expectations for schools around improving students
grade GPA with eighth-grade GPA (0.77) or seventh-
scores in middle school? This chapter shows that
grade ISAT with eighth-grade ISAT (0.81 for reading,
reading scores, in particular, tend to grow at the same
0.88 for math). Eighth-grade attendance is even less
rate for almost all students. There is somewhat more
strongly correlated with attendance in sixth grade, and
variation in math score growth. Grades show more
the correlation of eighth-grade attendance with fifth-
variation in growth over time than test scores, and
grade attendance is just moderate (0.43). In contrast,
attendance changes the most. This suggests that
eighth-grade ISAT scores are very highly correlated
attendance depends more on students specific
with fifth-grade ISAT scores (with correlations of 0.79
experiences in the middle grades than do test scores.
in reading and 0.82 in math). There is much less varia-
dents attendance is less static during the middle

grades years than GPA. Chapter 6 showed that students
sixth-grade ISAT scores are almost as predictive of
tion among students in the degree to which test scores

change over time, relative to other students, than the
degree to which attendance changes.
Chapter 8 | How Grades, Attendance, and Test Scores Change
81
District-Wide, Average GPAs and Attendance Are

Similar from Fifth through Eighth Grade in Chicago
On average in Chicago, attendance rates and grades
are similar throughout the middle school years.
Average attendance rates hover around 95 percent in
fifth through eighth grade (see Figure A). Likewise,
students GPAs are similar across the middle grade
years, hovering between 2.7 and 2.9, on average.
This does not mean that an individual students

attendance and grades are necessarily flat across
the middle grades. It simply means that as many
students show declines in attendance and grades
as show improvements, so that the trends are fairly
flat overall.
FIGURE A
Average Attendance and GPA During the Middle Grades

Average Attendance
82
95.2%
95.0%
95.1%
4.0
94.6%
90
3.5
80
3.0
70
3.5
60
3.0
GPA
Percent of Students
100
Average GPA
50
2.0
30
1.5
20
1.0
10
0.5
5th
Grade
6th
Grade
7th
Grade
8th
Grade
2.9
2.7
2.7
6th
Grade
7th
Grade
2.5
40
2.8
5th
Grade
8th
Grade
TABLE 12
Correlations of Eighth-Grade Attendance, Grades, and Test Scores with Their Attendance, Grades, and
Test Scores in Earlier Years

Correlations
Between
Years
Overall
GPA
ISAT Reading
and Math
Combined
English
Math
Reading
Math
8th-Grade
8th-Grade
8th-Grade
8th-Grade
8th-Grade
8th-Grade
8th-Grade
0.62
0.77
0.90
0.67
0.59
0.81
0.88
6th Grade
0.52
0.66
0.88
0.58
0.48
0.80
0.84
5th Grade
0.43
0.62
0.86
0.55
0.46
0.79
0.82
7th Grade
Attendance
Grades
ISAT
FIGURE 32
Change in Attendance from Fifth Grade through Eighth Grade

100
Percent of Students
95
100%
88th Percentile
97%
50th Percentile
100%
96%
92% 93%
90
88%
10th Percentile
87%
85
80
77%
75
70
Highest Growth (95th Percentile)

Average Growth
Lowest Growth (5th Percentile)
Note: Appendix C describes the methodology for calculating the growth trends. These growth trends are based on HLM models, nesting four observation points
(grades five through eight) within students, and calculating a slope for each student. Variance in the Bayes estimates of the slope coefficients was inflated to match
the model estimate of the true variance in slopes before graphing the distribution of growth trends.
The largest changes in attendance occur among
FIGURE 33
chronically absent students, with some students
Years of Chronic Absenteeism among Chronically Absent

Eighth-Graders in 2011-12
showing large declines in attendance. Although
attendance is more likely to change during the

middle school years than either grades or test scores,
attendance rates do remain about the same for many
students during the middle grades. Students who have
strong attendance in fifth grade are very likely to have
4+ Years
1 Year
28%
32%
strong attendance throughout the middle grade years.

Among students with near-perfect attendance in fifth
grade (99 to 100 percent), the greatest decline is just
a percentage point a yearso that almost all end up
with eighth-grade attendance of at least 97 percent
2-3 Years
40%
(see Figure 32).

Most students who have weak attendance in fifth
grade continue to have weak attendance throughout
the middle grades. An earlier chapter suggested that
chronically absent in fifth grade are likely to be
students with very low attendance in eighth grade
chronically absent in eighth grade. Only a third of
(below 90 percent) could already be identified in
chronically absent eighth-grade students experience
seventh grade as in need of substantial intervention
chronic absenteeism for the first time in eighth grade
efforts in order to be on-track for graduation from high
(see Figure 33). All other chronically absent eighth-
school. The same is true at earlier grade levels. Except
graders have experienced multiple years of chronic
for a group whose attendance improves at the highest
absenteeism during the middle grades, and over a
rate (nearly two points each year), students who are
quarter of these students have been chronically absent
83
ly graduate from high school, and they can be identified
Some Students Show Growth or

Decline in Grades by as Much as
Half of a GPA Point
very early. Even chronically absent fifth-graders whose
For many students, their course grades remain the same
attendance improves considerably over the middle
from year-to-year during the middle grade years. There
grades still have below-average attendance in eighth
are students, however, who show improvements while
grade (93 percent).
others show declines in their GPAs from fifth to eighth
every year since fifth grade. This is a group that is in

need of very substantial support if they are to eventual-
While many students with average attendance in
have the lowest GPAs in fifth grade; their GPAs can grow
dance over the middle grades, there also is a group that
by as much as 0.8 over three years, so that they go from
increasingly misses more days as they go through mid-
having a C- (1.8 GPA) in fifth grade to a C+/B- in eighth
dle school and ends up with attendance rates that put
grade (2.6 GPA; see Figure 34). A 0.8 difference in GPAs
students at risk of poor ninth-grade outcomes (ending
may sound small, but it can make a big difference for
with a 92 percent rate). This is a group for whom more
later student outcomes. As shown in Chapter 3, a one-
modest intervention strategies might be effective to
point difference in GPAs in eighth grade corresponds
keep attendance from falling further. When attendance
to a 20 percentage point difference in the likelihood of
drops below 95 percent at any point in the middle grade
passing ninth-grade math (comparing eighth-grade stu-
years, it is a signal that students may be in need of assis-
dents with a 1.5 GPA to those with a 2.5 GPA). A 2.6 GPA
tance around attendance. Schools might try strategies
in eighth grade, however, is still nearly a full grade point
to support studentsreaching out to families, providing
below where students need to be to have any real chance
mentors, or appointing teachers for special monitoring
of earning a B or better in ninth grade. Even students
and attentionto help these students maintain their
with typical grades in fifth grade (2.8 GPA) who improve
attendance rates or improve them so that they are not
their grades the most over the next three year do not
at high risk of being off-track in ninth grade.
quite make it to the 3.5 threshold (3.3 GPA).
FIGURE 34
Growth in GPA from Fifth Grade through Eighth Grade

4.0
3.7 GPA
90th Percentile
4.0
3.6
3.5
3.0
2.8 GPA
50th Percentile
3.0
3.3
2.8
2.5
GPA
84
grade. Growth in GPA is highest among students who
fifth grade (97 percent) maintain this level of atten-
2.0
2.6
1.8 GPA
10th Percentile
2.0
1.8
1.5
1.2
1.0
0.5
0
5th-Grade GPA
6th-Grade GPA
7th-Grade GPA
8th-Grade GPA

Average Growth
Note: Appendix C describes the methodology for calculating the growth trends, and the rationale for the methods that were used. These growth trends are based on
HLM models, nesting four observation points (grades five through eight) within students, and calculating a slope for each student. Variance in the Bayes estimates of
the slope coefficients was inflated to match the model estimate of the true variance in slopes before graphing the distribution of growth trends.
Differences in Test Score Growth

Are Not Large Enough to Enable
Low-Performing Students to
Eventually Meet ACT Benchmarks
grade gaining 13 and 12 points, respectively, per year.
Almost all students make the same degree of progress
in eighth grade. Most students who start with very high
on the reading ISAT from fifth to eighth grade. While
scores in fifth grade, at the 90th percentile, gain over
there are substantial year-to-year fluctuations, these
the middle grade years around 12 points and end up
tend to even out over time so that students end up with
close to the exceeding range of scores by eighth grade.
about the same percentile rank in eighth grade as they
Low-scoring students, those at the 10th percentile,
had in fifth grade. Figure 35 (top) shows the ISAT read-
gain more, on average, than students who begin middle
ing growth patterns, based on their fifth-grade score,
school with high math scores. The variability in their
for students at the 10th percentile, with average scores,
growth, however, is lower than the students at the 90th
and at the 90th percentile. On average, students grow 10
percentilethere are fewer differences among them in
points per year on the ISAT. There is some variation in
how much they grow. These low-performing students
that growth, but not much.
are not meeting standards in fifth grade and are un-
60
Students who start with
lower scores tend to grow faster than average, averaging
High-scoring students show the most variation in

their growth. Students scoring at the 90th percentile in
fifth grade end up between the 81st and 95th percentile
likely in to meet standards in eighth grade.
12 points a year instead of 10 points a year, while stu-
Similar, but not identical patterns in test score
dents who start with higher reading scores grow eight
growth can were observed in a study on middle grade
points a year, on average. The variation in growth in
students in New York. Variation in growth on tests in
reading scores mostly comes from where students start;
New York was also very small, relative to the variation
there is very little difference in yearly growth rates
in initial scores, and larger in math than in reading.
among students who begin middle school with similar
The standard deviation of growth on the math test was
scores, once multiple years are averaged together.
3 percent the size of the standard deviation of initial
Students at the 10th percentile in reading are well
math scores, and 2 percent of in reading. Thus, students
below meeting standards, and the 12 points per year of
in New York also changed little in their rank order
growth is not enough to make them close to meeting
from year-to-year. Unlike Chicago, though, gains in the
standards by eighth gradeeven among students with
tests in New York were not strongly related to students
the highest growth in this group. In eighth grade, these
initial scoresstudents with lower initial scores made
students score between the seventh and 11th percentile
slightly higher gains in both reading and math (see
among eighth-grade students. Students at the 90th per-
Kieffer and Marinell, 2012).
centile in fifth grade reading grow the least on the read-
Figure 35 (bottom) shows the patterns of math
While Test Growth Trends Are

Similar When Averaged Over
Multiple Years, Year-to-Year Gains
Can Vary Substantially
test scores growth over middle grades. There is simi-
Figure 35 shows students annual test score growth,
lar variation in math growth (standard deviation of
averaging their year-to-year gains over fifth, sixth,
2.3) as in reading, but the variation in growth is not as
seventh, and eighth grades. These average gains are
strongly determined by where students start. On aver-
very different from the yearly gains that take place from
age, students gain 13 points a year. Students at the 10th
one year to the next. Annual gains can reflect random
percentile in fifth grade gain a bit more than average (14
events such as a particularly bad or good testing day on
points), with average and high-scoring students in fifth
either the pre-test or the post-test. Students might have
ing tests. Nevertheless, their relative position remains

very similar, between the 89th and 93th percentiles in
eighth grade.
60 The standard deviation of the growth estimate is 2.0 points.

However, much of this variation can be accounted for by
differences in growth rates for students who start with

different levels of achievement.
85
FIGURE 35
Growth in Reading and Math Test Scores from Fifth Grade to Eighth Grade
Comparing students who start with similar reading test scores in fifth grade
Relative Position
in 5th Grade
Relative Position
in 8th Grade
300
ISAT Reading Scores
280
260
240
Exceeding Standards
93rd Percentile
89th Percentile
Meeting Standards
62nd Percentile
44th Percentile
Exceeding Standards
90th Percentile
Meeting Standards
11th Percentile
220
7th Percentile
Average
200
180
10th Percentile
160
5th-Grade ISAT
86
6th-Grade ISAT
7th-Grade ISAT
8th-Grade ISAT
Comparing students who start with similar math test scores in fifth grade
Relative Position
in 5th Grade
Relative Position
in 8th Grade
320
Exceeding Standards
95th Percentile
ISAT Math Scores
300
280
81st Percentile
Meeting Standards 72nd Percentile
Exceeding Standards
260
40th Percentile
29th Percentile
90th Percentile
240
220
Meeting Standards
5th Percentile
Average
200
10th Percentile
180
5th-Grade ISAT
6th-Grade ISAT
7th-Grade ISAT
8th-Grade ISAT

Average Growth
Note: Appendix C describes the methodology for calculating the growth trends and the rationale for the methods that were used. These growth trends are based on
HLM models, nesting four observation points (grades five through eight) within students, and calculating a slope for each student. Variance in the Bayes estimates of
the slope coefficients was inflated to match the model estimate of the true variance in slopes before graphing the distribution of growth trends. ISAT scores are
vertically scaled so that a score has the same meaning at different grade levels (ISBE, 2012).
Are Certain Types of Students More Likely to Show Improving or

Declining Grades, Attendance, and Test Scores Over the Middle
Grade Years?
Changes over time in students grades, attendance,
and math test scores are not substantially different by
students race, gender, poverty, or disability status. As
shown in Table A, no background factor is correlated
at more than 0.08 with changes in attendance, grades,
and math scores. The strongest relationship is with
gender; boys GPAs are slightly more likely to decline
over the middle grade years than girls GPAs. There
are also some modest correlations of background
factors with changing attendance, with attendance declining slightly less for Asian students than for African
scores show the most growth over the middle grade

years; these may be students who were behind in
basic literacy skills (e.g., decoding, vocabulary), versus
more advanced skills (e.g., comprehension). A review
of the What Works Clearinghouse shows that there are
more successful programs available to schools to work
on basic literacy than advanced literacy. Students
with disabilities, male students, African American
students, and students coming from neighborhoods
with high poverty (more families below the poverty
line and higher male unemployment) and low social
American, white, or Latino students.

Growth over the middle grade years in reading test
scores is correlated with students background characteristics, with students who are generally considered
more disadvantaged showing higher average rates of
growth. This occurs because reading test score growth
in middle grades is highly negatively correlated with
their test scores in fifth grade (at -0.92).F Students
who start the middle grades with very weak reading
status (with lower average levels of education and

lower family incomes) show higher growth in reading
test scores than other students. All of the correlations
between reading ISAT trends with student characteristics shown in Table A are a reflection of how students
score in fifth grade and the negative correlation between initial status in reading and subsequent growth.
They no longer hold if we only compare students with
similar ISAT scores in fifth grade.
TABLE A
Correlations of Change in Attendance, Grades, and ISAT with Background Factors

Attendance
Trend
GPA
Trend
Reading
ISAT Trend
Math
White
0.01
0.05
-0.21
-0.03
Black
-0.06
-0.05
0.24
0.05
Asian
0.05
0.07
-0.15
0.02
Latino
0.04
-0.01
-0.07
-0.03
Male
-0.02
-0.08
0.12
-0.03
Concentration of Poverty
(Students Neighborhoods)
-0.05
-0.04
0.25
0.04
Social Status
(Students Neighborhoods)
0.02
0.05
-0.15
0.01
-0.04
0.01
0.40
0.01
Cognitive Disability
Note: All of the correlations over 0.01 are statistically significant because they are based on 99,300 cases for attendance and 61,791 cases for grades.
The correlations for test scores are based on approximately 22,000 students.
F Math test scores in fifth grade and the growth from fifth to
eighth grade are also negatively correlated; students with
higher initial test scores tend to exhibit lower growth, but
this correlation is much lower (-0.33) than the correlation

of reading scores.
87
TABLE 13
Correlations of Test Score Gains in the Middle Grades in Different Years

Reading (in the top right of the table) and Math (in the bottom left of the table)
Gains
From 5th to 6th grade
Reading -0.39
Math -0.46
Math -0.00

Reading 0.04
Reading -0.55
Math -0.42
different from other years. Students might experience a
Middle Schools Can Affect Whether

Students Grades, Attendance, and
Test Scores Improve or Decline
developmental growth spurt one year that leads them to
About half of the differences in GPA and attendance
do better on a test, or experience a traumatic event that
growth and decline across students during the middle
leads to lower performance. This year-to-year varia-
grades can be attributed to the school that they
tion, however, tends to wash out over time. Students
attend. 61 At some schools, it is typical for students
who have especially strong growth in one year tend to
grades to improve by about a tenth of a GPA point
have especially weak growth in the next. The opposite
each year over the middle grade years, ending up 0.3
is true as well; students who fall behind one year tend
points higher in eighth grade than in fifth grade; at
to make it up the next. Table 13 shows the correlations
other schools, GPAs tend to decline by about a tenth
between gains in test scores from one grade to the next.
of a GPA point each year, ending up 0.3 points lower
Gains from two consecutive years, from fifth to sixth
(comparing schools one standard deviation above the
grade and from sixth to seventh grade, are negatively
mean to schools one standard deviation below, in GPA
correlated; the correlation is -0.39 in reading and -0.46
growth). Thus, students GPAs might end up as much
in math. Good years, in terms of gains, tend to be fol-
as 0.6 points different by the eighth grade, based on
lowed by low gains the following year, with the opposite
which school they attend for the middle grade years.
also true.
That is a sizable difference in terms of students prob-
an exceptionally good or bad teacher one year, showing

especially high or low gains in a given year that are very
88
Because high gains in one year are rarely followed by

high gains in the next, there are few students who make
ability of passing their ninth-grade classes and getting

good grades in high school.
exceptionally high gains over a three-year period. This
At some schools, attendance tends to decrease by
is why it is difficult to move students into the exceeds
about one percentage point each year; at others, atten-
range by eighth grade, unless they were close to exceed-
dance tends to improve by about one percentage point
ing standards in fifth grade. In reading, it seems like an
each year (comparing schools one standard deviation
impossible task. Something dramatically different, or at
above the mean to schools one standard deviation
least very different, from what elementary schools are
below, in attendance growth). This may seem small
doing today, needs to happen in elementary schools to
but, by the end of three years, students at some schools
produce the kind of gains necessary in middle grades to
have improved their attendance by three percentage
put students within reach of the ACT benchmarks. To
points, while others have shown a three percentage
our knowledge, there is no known way to do this on
point decline. Thus, the school that a student attends
a large scale.
may make a difference of six percentage points in their

eighth-grade attendance. As shown in the previous
61 The variance in GPA trends comes from HLM models of students GPAs, with observations (years) nested within students
and nested within their eighth-grade schools. The standard
deviation of the variance in GPA trends at the student level
(within schools) is 0.089, while the standard deviation of the
variance in GPA trends at the school level is 0.096. The variation in attendance trends comes from similar models, with
the standard deviation of variance at the student level (within
schools) at 0.013 and the standard deviation of variance in the
trends between schools at 0.010.
chapter, six percentage points in attendance is very
scores are the strongest predictor. 62 The second factor
substantial in terms of students probability of suc-
depends on having known programmatic effects for
cess in high school; a difference of 10 percentage points
moving the indicators, and evidence that differences
moves a student from perfect attendance (100 percent)
in school practices result in differences in growth on
to chronically absent (90 percent).
the indicator. Variation in growth across students dur-
There are also differences across schools in the size
ing the middle grades provides some indication of the
of students yearly gains on the ISAT; some schools
degree to which there are known strategies to change
show higher gains across the middle grade years than
the indicators during the middle grade years.
others. In reading, half of the differences in students
To put both of these factors together, Figure 36
gains can be attributed to which school they attend.
shows the predicted ninth-grade outcomes for a stu-
Among students at the 50th percentile, for example,
dent that ended fifth grade with average attendance,
students at schools with high growth gain 11 points a
grades, and test scores, but showed very high levels of
year, while those at schools with low growth gain nine
growth on each of the indicators from fifth to eighth
points a year (comparing schools one standard devia-
grade. Figure 36 also shows the predicted ninth-grade
tion above the mean to schools one standard deviation
outcomes for a student that had average performance in
below, in test score growth). In math, school differences
fifth grade, but showed little improvement throughout
account for more than half of the differences across
the middle grade years. While this is simply a simula-
students in yearly growth; students at schools with high
tion, and there are many factors that might influence
growth gain 15 points a year, while those at schools with
students actual performance, it provides an estimate
low growth gain 11 points a year. Despite the differences
of the degree to which students high school outcomes
in what schools can contribute to growth, the gaps in
could be influenced by what happens to their perfor-
test scores are so large from the start of middle school
mance in the middle grades.
that no schools close those gaps in the middle grades.
Middle Grade GPA and Attendance

Have the Greatest Potential for
Improving High School Outcomes
What happens in middle grades does seem to matter considerably for students ninth-grade outcomes,
especially their growth in grades and attendance. While
students with average performance in fifth grade are
unlikely to be at the top of their class in high school,
As schools look for indicators that are likely to have
their probability of being on-track goes from 54 to 95
high leverage for moving student outcomes, they need
percent, depending on whether their grades declined
to consider two factors. First, they need to consider
( from 2.8 to 2.0) or improved ( from 2.8 to 3.3) during
whether moving the indicator would likely move the
the middle grades. Likewise, their predicted ninth-
later outcome they care about (e.g., high school and
grade GPA goes from 1.9 to 2.4, based on whether their
college graduation). Second, they need to consider
grades declined or improved from fifth to eighth grade.
the degree to which they can move the indicator with
Their predicted PLAN scores are also one point higher
known strategies. The first factor depends on the
(16 compared to 15) if their grades improved, rather
degree to which the indicator is strongly and directly
than declined, from fifth to eighth grade.
related to later outcomes. For high school gradua-
Changes in attendance over the middle grade years
tion and college readiness, course grades are the most
are also strongly associated with different ninth-grade
strongly related to later outcomes, followed by atten-
outcomes. Students probability of being on-track goes
dance, followed by ISAT scores; for ACT scores, ISAT
from 66 to 93 percent, depending on whether their
62 We cannot definitively test whether the observed relationships of the indicators with the outcomes are direct, indirect,
or spurious, as there may be unmeasured variables that were
not included in the models employed for Chapter 1 through
Chapter 3. However, the methods employed to compare

among the indicators were designed to determine which
showed the largest direct relationships among the predictors
that were available.
89
FIGURE 36
What Is the Predicted Ninth-Grade Outcome for Students with the Most and Least Improvement in
Attendance, GPAs, and ISAT Scores in the Middle Grades?
Change from Fifth to Eighth Grade
Core GPA
2.8
3.3
95%
2.0
2.4
54%
Attendance
97
99%
93%
93%
2.4
2.1
15
77%
68%
2.2
2.1
15
73%
2.2
16
2.2
16
66%
16
ISAT Math
225
276
253
16
ISAT Reading
215
248
241
75%
0%
20%
40%
60%
80% 100% 0.0
Predicted On-Track Rate
1.0
2.0
3.0
4.0 10
Predicted 9th-Grade Core GPA
95th Percentile of Growth (from Fifth to Eighth Grade)
90
16
15
1.9
12
14
16
18
20
22
Predicted PLAN Composite Score
5th Percentile of Growth (from Fifth to Eighth Grade)
Note: Change from fifth to eighth grade represents improvement that is at the fifth percentile (orange bar) or 95th percentile (gray bar) in terms of change from fifth
to eighth grade in each indicator. The predicted ninth grade outcome is based on the eighth grade indicator level, and controls for GPA, attendance rates, and ISAT
scores in fifth grade (predicting outcomes for a student with average fifth grade achievement: 97 percent attendance, 2.8 GPA, 225 math ISAT, 215 reading ISAT).
attendance declines (from 97 to 93 percent) or improves
the middle grades. Students with the highest growth
(from 97 to 99 percent) during the middle grades.
end up with eighth-grade scores that are only seven
Likewise, their predicted ninth-grade GPA goes from
points higher than students with the lowest growth
2.1 to 2.4, based on whether their attendance declined
(e.g., 241 compared to 248, among students with a score
or improved over time. Improving attendance in the
of 215 in fifth grade). Thus, improving reading scores
middle grades is also associated with higher PLAN
shows a much smaller relationship with ninth-grade
scores, as students who show the most improvements
outcomes than does improving math scores.
in attendance have average scores that are one point
We cannot say whether changing middle school
higher than students with the most decline in atten-
performance would result in better outcomes without
dancecomparing students that had similar attendance
conducting a study that is specifically designed to be
and ISAT scores in fifth grade.
able to make causal claimsshowing that changing an
Larger improvements in students ISAT math scores
earlier outcome results in a change in a later outcome.
in the middle grades are associated with improve-
Figure 35 on p.86, simply shows what happened in high
ments in ninth-grade outcomes, but the differences are
school to students who improved their middle school
smaller than seen with improvements in attendance
performance compared to those who declined. Students
and grades. Students with the most improvement in
with improving performance in the middle grades had
their math scores in the middle grades are nine percent-
better high school outcomes than students who were
age points more likely to be on-track in ninth grade
similar to them in fifth grade but had declining
(77 percent compared to 68 percent) than students
attendance, grades, or math scores; this suggests that
who showed the smallest math score gainscomparing
improving student performance in the middle grades
students with similar test scores in fifth grade. Their
has the potential to improve high school outcomes.
ninth-grade GPAs are also higher by about 0.1 points,
These findings are very similar to the relationships
and their PLAN scores are higher by about one point.
found in New York between declining attendance and
Reading scores show little variation in growth across
test scores and students probability of graduating
high school. Declining attendance in the middle grades
achievement levels in fifth grade, but showed declining
was particularly strongly associated with a lowered
attendance and grades in middle school. There is less
probability of high school graduation. Declining test
variation in growth in test scores across students than
performance was also associated with a lower prob-
in grades or attendance because students who have
ability of graduating in New York, especially in math
high test score gains in one year tend to have low test
(Kieffer and Marinell, 2012).
score gains in the next, and vice-versa. About half of
Summary
the differences in students growth in attendance,

grades, and test scores over the middle grades can be
For most students, attendance, grades, and test score
attributed to their school. Because attendance varies
ranks are fairly stable over the middle grades. However,
the most across the middle grade years, and because it
some students show improvements in grades and at-
is strongly related to ninth-grade course performance,
tendance over the middle grade years; these students
improvements in attendance may have particularly high
ninth-grade outcomes are substantially higher than
leverage for improving high school outcomes, compared
the ninth-grade outcomes of students who had similar
to other indicators.
91
92
CHAPTER 9
Interpretive Summary
School districts across the country are trying to figure out what they
can do to get all of their students to graduate from high school and
be ready for success in college. There is recognition that these efforts
need to begin prior to high school, but there is little guidance on what
it is that schools should target in the middle grades to affect outcomes
at the end of high school, and how to identify who is at risk for not
succeeding in later years.
In this report, we show that students risk of poor per-
performance beyond students grades alone. These
formance in high schooland their level of readiness
two indicators by themselves can be used together in
to succeedcan be identified with just two or three key
a simple indicator chart to show students likelihood
indicators of middle school performance. It is not nec-
of success in their ninth-grade courses, based on their
essary to put together complicated prediction models
middle grade records.
based on multiple indicators. Schools could use infor-
There are other indicators that slightly improve the
mation on students level of risk for different outcomes
prediction of students high school course performance.
to design intervention strategies for individual students
Among students with high grades in middle school, test
and for assessing which problems with readiness are
scores can help predict who is likely to get As and Bs in
most critical in particular schools.
high school. Indicator systems could also incorporate

school effects, since students with similar middle grade
The key indicator of readiness for passing classes,
performance (grades, attendance, and test scores) have
being on-track, and getting good grades in high
different likelihoods of success, based on which middle
school is students overall core GPA in the middle
school they attended and in which high school they
grades. Students GPA across all of their core courses
enroll. While these additional factors improve the
is a better indicator of readiness for passing any one
accuracy of the prediction, there is a trade-off in that
specific class (e.g., algebra, English) than their grades in
they make calculations more complicated.
that specific class in middle school. This is true in fifth
The key indicator of meeting benchmarks on the
grade, as well as in eighth grade. Often people think of
EPAS exams in high school (e.g., the PLAN taken at the
grades as subjective and not good indicators of later
start of tenth grade) is performance on standardized
performance. They may think that GPAs at early grade
tests in middle gradesthe ISAT reading and math
levels are particularly subjective. However, if grades
exams. A students score on either the reading or the
were mostly subjective, they would not predict later
math ISAT is a very good indicator of whether he or she
outcomes. Grades are much better predictors of future
will meet the college-readiness benchmark on any of
performance than indicators that are considered more
the four subject-area tests on the PLAN. Combining
objective, such as test scores. Middle grade attendance
the reading and math ISAT scores together, or combin-
is also predictive of high school course performance
ing ISAT scores from multiple years (e.g., students
and improves the prediction of ninth-grade course
seventh-grade score with their eighth-grade score),
Chapter 9 | Interpretive Summary
93
94
improves the prediction of students PLAN score on any
Philadelphiaattendance less than 80 percent or a final
subject-specific test and on the composite score (the
year grade of F in their math or language arts class puts
average of all of the subject-specific tests). The subject-
students at extremely high risk of failure (with a greater
specific tests are each very predictive of scores in other
than 75 percent chance). 63 This suggests that there are
subjects, almost as predictive as tests within the same
common underlying patterns in the factors that can be
subject. This suggests that both the ISAT and the PLAN
used to identify future dropouts in middle school. One
tests are measuring general knowledge and skills at
key difference between Philadelphia and Chicago is that
least as much as knowledge and skills in any given sub-
a larger proportion of eventual dropouts can be identi-
ject area. In general, the more tests that are used to pre-
fied earlier in Philadelphia than in Chicago. This could
dict the high school test score (e.g., multiple subjects,
be caused by the structure of schools in Chicago versus
multiple years), the more accurate the prediction.
schools in Philadelphiamost elementary schools in
Middle school performance can be used to identify
Chicago serve students in grades K-8 and students do
students at high or very high risk of not performing well
not face a middle school transition. It is possible that
in high schoolfor being off-track, getting low grades,
students fall off-track to graduation earlier when their
or not meeting benchmarks. However, there are many
major school transition occurs in the middle grades
students at moderate risk of receiving low or failing
instead of high school. 64 In Chicago, just focusing on
grades whose performance could go either way. The
students with very high risk of dropout based on middle
change in school context from eighth to ninth grade
school indicators would only capture a very small
makes it difficult to predict accurately the grades these
number of future dropouts. Different cities face differ-
students will receive in high school. For this reason, it is
ent challenges and need to develop strategies with the
critical that student performance be monitored during
indicators that meet their own circumstances.
the ninth-grade year to make sure students are performing up to their potential.
Implications for Addressing

High School Dropout
For students with 80 percent or lower attendance

in the middle grades or a GPA of less than 1.0 in the
middle grades, interventions are strongly warranted
while they are in middle school. These students are
Calls for ending high school dropout often call for early
extremely likely to drop out of school, with a risk
intervention. As shown in this report, early interven-
greater than 75 percent, unless they experience a sub-
tion will not solve all of the problems that lead students
stantial change in the way in which they are engaging
to drop out. It is during ninth grade that many students
with school. Students earning a mix of Cs and Ds or
fall off-track to graduation. In Chicago, most future
below, or who attend less than 90 percent of the time
dropouts do not show strong signs of withdrawal until
in middle school, have less than a 50 percent chance of
ninth grade. However, there are some students that can
being on-track when they get to high school. Moderate
be identified as at high risk very earlystudents who
interventions might be sufficient to get them to succeed
need intervention before high school and across the
in high school.
transitionif they are to have a chance of graduating.

Many students can be identified as at very high risk
In general, students whose middle school attendance

is below 95 percent have not developed strategies to get
of not graduating at least as early as fifth grade, based
to school every day, despite whatever obstacles they
on their attendance and grades. The same levels of
face. Ninety percent attendance is considered chronic
the same indicators identify students with an ex-
absenteeism, and students who are chronically absent
tremely high risk of not graduating in Chicago as in
in the middle grades are more likely to be off-track
63 Neild and Balfanz (2006); Balfanz, Herzog, and MacIver
64 Achievement tends to decline during transitions to middle
(2007).
school and high school (Alspaugh, 1998; Barber and Olsen,

2004; Eccles, Lord, and Midgely, 1991).
supportive high school environment, chronic absence is
Implications for Improving

College Readiness
likely to drop to intermittent attendance, and moderate
Almost all students want to go to college, but the vast
attendance is likely to drop to chronic absence. While
majority of students who leave eighth grade in Chicago
the group of students with less than 95 percent at-
do not even have a 1-in-4 chance of finishing high school
tendance may be too large for intensive interventions,
with the qualifications that indicate they will succeed in
moderate efforts to reach out to families or provide
college. The biggest stumbling block is students grades.
community support to help students address barriers
Despite all of the attention on test scores, high school
to school attendance might actually have considerable
grades are the strongest predictors of college graduation
pay-off later on.
and middle school grades are the strongest predictors of
than on-track when they get to high school. In the less-
In the summer before ninth grade, high schools can

use students grades and attendance from middle school
to identify students for whom it is most critical to estab-
high school grades. Yet, only a quarter of CPS students

earn the grades that indicate they will succeed in college.
Grades are a reflection of students work effort
lish trusting relationships. Students with eighth-grade
coming to class, getting assignments completed,
attendance less than 90 percent or a GPA of less than 2.0
participating, studying, and delivering high-quality
in eighth grade are very likely to need support during
work day after day. These academic behaviors are what
the ninth-grade year. Schools could reach out to these
matter for college success. They are also some of the
students and their parents to establish positive connec-
same behaviors that employers look for in workers.
tions before problems occur. Some schools in CPS have
Preparing for college requires students to work hard in
used Freshman Watchlists for this purpose in the past,
their classes; by putting in effort around challenging
reaching out to particular students during the summer
work, students develop the strategies and skills they
through the districts Freshmen Connection program.
will need at higher levels of education. Students may
Once students are in high school, schools can put
not remember the Pythagorean Theorem down the line,
students on the right path toward graduation by closely
but working hard to understand and apply it helps them
monitoring their attendance and grades. Many schools
build the cognitive frameworks and study habits that
that have used ninth-grade success reports to identify
allow them to conquer math problems later on. Putting
students in need of support have shown substantial
together an excellent term paper may teach students
improvements in their ninth-grade on-track rates and
something new about the world, but it also teaches them
progress toward graduation. 65
how to organize their thinking and put information
Efforts to improve students attendance and grades
together in new ways. It is the process of working hard
could have benefits for improving not only high school
on academic work that gets students ready for college.
graduation rates but also college outcomes. The factors
While students and their parents may know that grades
that are important for high school graduationgrades
and work effort matter for college, they may not realize
and attendanceare also important for college readi-
how important they are or how high they need to be for
ness. Often efforts to improve high school graduation
students to be on a path to college readiness.
are seen as separate from efforts at improving college

readinessone targets low-achieving students, while the
Students need to know that college readiness means
other targets high-achieving students. However, the two
at least B-level work, starting at least in the middle
goals are complementary. Getting more students ready
grades. Grades do not usually get better when work
for college requires substantial improvements in stu-
gets harder and students are given more responsibility.
dents grades and attendance in the middle grade years.
Instead, grades tend to decline as students move into
65 See Allensworth (2013) for a description of the ways in which

high schools used indicators to improve on-track rates.
95
96
and through high school. If students do not have at least
tion and their interactions with students to encourage
a B average in the middle grades, they are extremely
positive mindsets about the work. The fact that non-
unlikely to end high school with at least a B average.
cognitive factors, such as study habits and grit in eighth
Students with lower than a 3.0 high school GPA have a
grade, are not predictive of ninth-grade performance
slim chance of graduating from college, and they will
provides evidence that they are context-specific, rather
be ineligible to attend many colleges or receive most
than just a feature of individuals. When a student is not
scholarships. Families should not have to wait until
putting in effort, a teacher or other adult could find out
the junior year in high school to realize this. Middle
why they are putting in little effortwhat it is about
schools can reach out to families of students who are
the class or about students own experiences and skills
not making high grades to let them know that they are
that is preventing strong performance. Teachers also
not on-track to be ready for college. They can make sure
can design courses so that they intentionally develop
that teachers are keeping up with their grading in the
students learning strategies, such as metacognitive
parent portal and have clear grading policies, so that
skills and study habits, as part of teaching their course
students and parents always know where their grades
subject. Explicitly teaching strategies to do better in
stand and can notice if they slip. For some students,
class can pay off with better success on tests and assign-
this knowledge may be enough to motivate higher work
ments in that class and in future work. Other research
effort. For others, it may take support from teachers,
at UChicago CCSR has found that students grades are
mentors, or support staff to reach out, determine why
higher the more teachers monitor their performance
students grades are low, and then develop strategies to
and provide help as soon as they start to fall behind. 69
support their particular needs.
Teachers can also reach out to students who miss
The ways that teachers structure their classes can
class as soon as they are absent to make sure they start
influence whether students put in strong or weak work
attending regularly and do not fall behind in their work.
effort. Teaching is a complex task. Teachers need to de-
If students miss multiple days, it is important to find
sign methods for engaging students around challenging
out why and figure out strategies to help students get to
academic work, even though students enter their class
school. Students should have close to 100 percent atten-
with different skill levels, different histories of success,
dance to be on a path to be ready for college.
and their own issues and priorities. The ways in which

teachers implement their lessons have implications
Attendance is critical, at least as important as test
for the degree to which their students put in effort. As
performance. It may seem like a low barget students
discussed in the report, Teaching Adolescents to Become
to come to school every day. But if students are not in
Learners,
school, they cannot learn. Even among students with
66
students put more effort into their classes
when they have positive mindsets about the workwhen
high grades and test scores, those with very strong
they feel like they can be successful, that they belong
attendance in middle school (98 percent or higher) get
in the class, that the work has value and will lead them
much better grades in high school than those with mod-
to have stronger skills. A host of factors can undermine
erate attendance (e.g., 94 percent). By the time students
positive mindsets. For example, being at the bottom
are in high school, and in college, they need to know
of the class in skill levels can make students feel they
how to get themselves to class. Efforts aimed at 100 per-
cannot succeed and do not belong.
Lessons without
cent attendance could actually have substantial pay-off
clear goals can keep students from seeing the value in
in students eventual success in college and careers, but
the work they are asked to do. Clear grading standards
problems with attendance are often dismissed as being
and constant feedback can provide motivation to keep
of low importance compared to progress on tests.
up with work.
67
68
Mindsets are not set in stone. They change as stu-
Figuring out how to get to school when other factors

may interferefrom family sickness and transportation
dents move across different contextsdifferent classes,
issues, to the pull of more interesting activitiesis not
different schools. Teachers can modify their instruc-
of secondary importance to improving test scores. This
is more difficult for some schools than others, as schools
currently available to most schools. 71 The good news is
serving many students in poverty will struggle with
that students can have success in high school and col-
issues around transportation, health care, safety, and
lege despite test scores that are well below ACTs bench-
residential instability. But there is no reason to think
marks. In fact, students who score well below ACTs
that these schools would be less successful at improving
benchmark scores do not have a substantially lower
attendance than at improving test scores, which are also
probability of success in college classes than students
hard to address in highly impoverished communities.
who meet the benchmark. For example, while the probability of earning at least a B in a college social science
Schools and the public are concerned about meeting
class for students at the benchmark reading score of 21
ACT benchmarks, but reaching benchmark scores is
is 50 percent, the probability of earning at least a B for
less important for college readiness than maximiz-
students with a score of 16 is 40 percent. 72
ing learning growth and getting good grades. There
Even students who have high test scores and strong
is a strong focus on raising test scores to reach ACT
GPAs in the middle grades do not necessarily perform
benchmark scores. However, while all students scores
well in high school; many students fall off the path to
grow as they move through the middle grades and high
college readiness in high school. In schools serving low-
school, they do not grow at rates that are different
income minority students, it is particularly important
enough to make up for the initial differences in scores
that those students who have a chance of success get
across students. This makes benchmarks irrelevant for
opportunities that will allow them to be competitive
all but a subset of students, because they are too high
with students from more advantaged schools. Students
for some students and too low for others. Setting one
need classroom environments that encourage them to
testing goal for all students sets up many students and
put in strong effort, earn high grades, and show high
schools for failure and does not push students at the
rates of learning growth. If students are coming into
high end to meet their potential. It gives a false sense
high school with strong middle school records and not
of success for schools serving students with high test
performing well, high schools need to find out why.
scores to begin with, while setting impossible standards

for schools serving students with the weakest skills.
All schools have reading and math improvement as
Research has shown that students learn more when

they are in orderly environments with high expectations. 73 Schools can achieve this in multiple ways. For
a priority, and the differences between schools in the
academically strong students, they can run honors
instruction that students receive are not enough to
classes, IB programs, and advanced classes. 74 Or they
make up for the large differences that exist before they
can put sufficient support staff in place in mixed-ability
enter the middle grades. This is not only true in Chicago
classes so that expectations are high for all students,
but it also can be inferred from research in many other
and so that teachers are able to provide differentiated
places. 70 Getting high gains on tests may take more
instruction in an orderly environment.75 They can
resourcessmaller classes, more time, more individual-
make sure that students with low achievement have
ized assistancefor students with weak skills than are
sufficient support, time for learning, and student-
66 Farrington et al. (2012).

67 Nomi and Allensworth (2013).
68 Rosenkranz et al. (2014).
69 Allensworth et al. (2014).
70 As noted in Chapter 4, tests in early grades are extremely
individualized interventions require additional resources for

students.
72 See the figures in Allen and Sconing (2005), the report used
to set the original ACT benchmark scores.
73 See Gates Foundation (2010); Allensworth et al. (2014).
74 In general, high-achieving students have higher test score
gains when in classes that are sorted by skills. See Nomi and
Allensworth (2009, 2012, 2013); Collins and Gan (2013);
Argys, Rees, and Brewer (1996); Loveless (1999).
75 We found that all students achievement improved in algebra
classes where students with below-average skills received an
extra period of instruction simultaneously to their primary
algebra class (Nomi and Allensworth, 2012).
predictive of test scores in later grades, suggesting that there

is little variation in long-term growth on tests taken (e.g., in
California): see Zau and Betts (2008); Kurlaender, Reardon,
and Jackson (2008); New York: Kieffer and Marinell (2012):
or on ACTs EPAS system: ACT, Inc. (2008).
71 For example, evidence suggests that struggling adolescent
readers can show improvements in literacy with intensive,
individualized interventions (Kamil et al., 2008). But
97
centered pedagogy to enable them to be engaged and
many high schools, educators have designed systems for
successful in challenging classes.
reaching out to ninth-grade students whose absences
76
Students tend to
put in more effort and earn higher grades when teach-
are high or grades are low to find out why they are
ers are attuned to their academic needs and provide
struggling and figure out ways to help them perform
support as soon as they start to struggle. Monitoring
better. Schools are setting goals around particular
systems can make it easier for educators to identify
groups of students, identifying them, and tracking the
students who need help so they can reach out as soon
progress of their intervention plans. They are using
as a students performance starts to slip or is below
data on grades and attendance to have difficult, but
their potential.
important, conversations about how they could better

support their students to make progress towards high
98
Monitoring systems could help students get the right
school graduation. Most of these systems have focused
level and kinds of support to keep them on-track for
on getting students on-track to graduation. Similar
high school and college graduation. High schools in
systems could be designed around students who are
Chicago have made extraordinary progress over the
falling off-track to college readiness. The success of
last five years in improving student performance in
data-driven practices around grades and attendance
the ninth grade by using early warning indicators to
in high schools suggests that similar efforts in schools
support student performance in their classes. Ninth-
serving the middle grades might do much to further
grade on-track rates have increased from around 59
improve CPS students educational attainment.
percent to close to 85 percent in just a few years. In
76 When Chicago implemented its double-algebra policy,

it provided support for teachers and twice as much
instructional time for students. The challenge and
instructional quality in algebra classes improved for students

with below-average test scores, and they learned more math
(Nomi and Allensworth, 2009, 2012).
References
ACT, Inc. (2008).
The forgotten middle. Iowa City, IA: ACT. Retrieved

from http://www.act.org/research/policymakers/pdf/
ForgottenMiddle.pdf.
ACT, Inc. (2012).
The condition of college & career readiness.

Iowa City, IA: ACT. Retrieved from http://media.act.org/
documents/CCCR12-NationalReadinessRpt.pdf?_ga=1.256
Allensworth, E., and Easton, J.Q. (2007).
What Matters for Staying On-Track and Graduating in

Chicago Public High Schools: A Close Look at Course
Grades, Failures, and Attendance in the Freshman Year.
Chicago, IL: University of Chicago Consortium on
Chicago School Research.
Allensworth, E., Gwynne, J., Pareja, A.S., Sebastian, J.,
and Stevens, W.D. (2014).
The dropout process in life course perspective: Early

risk factors at home and school. Teachers College Record,
103(5), 760-822.
Free to Fail or On-Track to College: Setting the Stage for

Academic Rigor: Without Classroom Control and Student
Support, a Challenging Curriculum Falls Flat. Chicago,
IL: University of Chicago Consortium on Chicago School
Research.
Allen, J., and Sconing, J. (2005).
Alliance for Excellent Education (2011).
103106.857372413.1414731200.
Alexander, K.L., Entwisle, D.R., and Kabbani, N.S. (2001).
Using ACT assessment scores to set benchmarks for college

readiness. ACT Research Report Series 2005-3. Retrieved
from http://www.act.org/research/researchers/reports/
pdf/ACT_RR2005-3.pdf.
The High Cost of High School Dropouts: What the Nation

Pays for Inadequate High Schools. Washington, DC:
Alliance for Excellent Education.
Alspaugh, J.W. (1998).
Updating the ACT college readiness benchmarks. ACT

Research Report Series 2013 (6). Retrieved from
Achievement loss associated with the transition to

middle school and high school. The Journal of Educational
Research, 92(1), 20-25.
http://www.act.org/research/researchers/reports/pdf/
Argys, L.M., Rees, D.I., and Brewer, D.J. (1996).
Allen, J. (2013).
ACT_RR2013-6.pdf.
Allensworth, E. (2006).
Detracking Americas schools: Equity at zero costs?

Journal of Policy Analysis and Management, 15(4), 623-645.
Update to: From High School to the Future: A First Look

at Chicago Public School Graduates College Enrollment,
College Preparation, and Graduation from Four-Year
Colleges Chicago, IL: University of Chicago Consortium
on Chicago School Research.
Balfanz, R., Herzog, L., and MacIver, D.J. (2007).
Ballou, D. (2007).
Allensworth, E., Correa, M., and Ponisciak, S. (2008).
Baltimore Education Research Consortium (BERC) (2011).
Allensworth, E., and Easton, J.Q. (2005).
Barber, B.K., and Olsen, J.A. (2004).
The use of ninth-grade early warning indicators to

improve Chicago schools. Journal of Education for
Students Placed at Risk, 18(1), 68-83.
From High School to the Future: ACT PreparationToo
Much, Too Late. Chicago, IL: University of Chicago
Consortium on Chicago School Research.
The On-Track Indicator as a Predictor of High School
Graduation. Chicago, IL: University of Chicago
Preventing student disengagement and keeping students

on the graduation path in urban middle-grades schools:
Early identification and effective interventions.
Educational Psychologist, 42(4), 223-235.
Magnet schools and peers: Effects on student achievement
[PowerPoint document]. Retrieved from schoolchoice.
uconn.edu/assets/PPT/4B_Ballou.ppt.
Destination Graduation: Sixth-Grade Early Warning
Indicators for Baltimore City Schools: Their Prevalence
and Impact. Baltimore, MD: BERC.
Assessing the transitions to middle and high school.
Journal of Adolescent Research, 19(1), 3-30.
Benner, A.D. (2011).
The transition to high school: Current knowledge, future

directions. Educational Psychology Review, 23(3), 299-328.
References
99
Bowen, W.G., Chingos, M.M., and McPherson, M.S. (2009).
Crossing the Finish Line: Completing College at Americas Public

Universities. Princeton, NJ: Princeton University Press.
Bowers, A.J., Sprott, R., and Taff, S.A. (2013).
Do we know who will drop out? A review of the predictors

of dropping out of high school: Precision, sensitivity, and
specificity. The High School Journal, 96(2), 77-100.
Collins, C.A., and Gan, L. (2013).
Does sorting students improve scores? An analysis of class

composition. NBER Working Paper No. 18848. Cambridge,
MA: National Bureau of Economic Research. Retrieved
from http://www.nber.org/papers/w18848.
Dale, S., and Krueger, A. (2012).
Estimating the return to college selectivity over the career

using administrative earnings data. NBER Working Paper
No. 17159. Cambridge, MA: National Bureau of Economic
Research. Retrieved from http://www.nber.org/papers/
w17159.
Davis, J.A. (1985).
The Logic of Causal Order. Newbury Park, CA: Sage.
100
Day, J.C., and Newburger, E.C. (2002).
The big payoff: Educational attainment and synthetic

estimates of work-life earnings. Washington, DC:
U.S. Census Bureau (P23-210).
Duckworth, A.L., and Seligman, M.E.P. (2005).
Self-discipline outdoes IQ in predicting academic performance of adolescents. Psychological Science, 16(12),

939-944.
Duckworth, A.L., and Seligman, M.E.P. (2006).
Self-discipline gives girls the edge: Gender in selfdiscipline, grades, and achievement test scores.
Journal of Educational Psychology, 98(1), 198-208.
Duckworth, A.L., Peterson, C., Matthews, M.D., and Kelly,
D.R. (2007).
Grit: Perseverance and passion for long-term goals.

Journal of Personality and Social Psychology, 92(6),
1087-1101.
Dweck, C., Cohen, G., and Walton, G. (2011).
Academic tenacity. An unpublished white paper prepared

for the Bill & Melinda Gates Foundation.
Eccles, J.S., Lord, S., and Midgley, C. (1991).
What are we doing to early adolescents? The impact of

educational contexts on early adolescents. American
Journal of Education, 99(4), 521-542.
Farkas, G., Sheehan, D., and Grobe R.P. (1990).
Coursework mastery and school success: Gender, ethnicity, poverty groups within an urban school district.
American Educational Research Journal, 27(4), 807-827.
Farrington, C.A., Roderick, M., Allensworth, E., Nagaoka,

J., Keyes, T.S., Johnson, D.W., and Beechum, N.O. (2012).
Teaching Adolescents to Become Learners: The Role of

Noncognitive Factors in Shaping School Performance.
Chicago, IL: University of Chicago Consortium on Chicago
School Research.
Gamoran, A. (1996).
Student achievement in public magnet, public comprehensive, and private city high schools. Educational Evaluation
and Policy Analysis, 18(1), 1-18.
Gates Foundation. (2010).
Learning about teaching: Initial findings from the

measures of effective teaching project. Retrieved from
http://www.metproject.org/downloads/Preliminary_
Findings-Research_Paper.pdf.
Geiser, S., and Santelices, V. (2007).
Validity of High School Grades in Predicting Student

Success Beyond the Freshman Year: High-School Record
versus Standardized Tests as Indicators of Four-year
College Outcomes. Berkeley, CA: University of Berkeley
Center for Studies in Higher Education.
Heckman, J.J., and LaFontaine, P.A. (2007).
The American high school graduation rate: Trends and

levels. NBER Working Paper No. 13670. Cambridge, MA:
National Bureau of Economic Research. Retrieved from
http://www.nber.org/papers/w13670.
Heppen, J.B., and Therriault, S.B. (2008).
Developing Early Warning Systems to Identify Potential

High School Dropouts. Washington, DC: National
High School Center. Retrieved from http://www.
betterhighschools.org/pubs/documents/IssueBrief_
EarlyWarningSystemsGuide.pdf.
Hoxby, C.M. (2001).
The return to attending a more selective college: 1960 to

the present. In M. Devlin and J. Meyerson (Eds.), Forum
Futures: Exploring the Future of Higher Education, 2000
Papers (pp. 13-42). Forum Strategy Series, Volume 3.
Oxford, OH: Jossey-Bass.
Illinois State Board of Education. (2008).
Illinois Standards Achievement Test 2008 Technical

Manual. Springfield, IL: Illinois State Board of Education
Assessment Division. Retrieved from http://www.isbe.
net/assessment/pdfs/isat_tech_2008.pdf.
Interpretive Guide 2011 Illinois Standards Achievement

Test. Springfield, IL: Illinois State Board of Education.
Retrieved from http://www.isbe.net/assessment/pdfs/
ISAT_Interpr_Guide_2011.pdf.
Neild, R.C., and Balfanz, R. (2006).
Kamil, M.L., Borman, G.D., Dole, J., Kral, C.C., Salinger, T.,
and Torgesen, J. (2008).
Nomi, T., and Allensworth, E. (2009).
Illinois Standards Achievement Test 2012 Technical

Manual. Springfield, IL: Illinois State Board of Education
Assessment Division. Retrieved from http://www.isbe.net/
assessment/pdfs/isat_tech_2012.pdf.
Improving adolescent literacy: Effective classroom and

intervention practices: A practice guide (NCEE #20084027). Washington, DC: National Center for Education
Evaluation and Regional Assistance, Institute of
Education Sciences, U.S. Department of Education.
Retrieved from http://ies.ed.gov/ncee/wwc/pdf/practice_
guides/adlit_pg_082608.pdf.
Kelly, S. (2008).
What types of students effort are rewarded with high

marks? Sociology of Education, 81(1), 32-52.
Kieffer, M.J., and Marinell, W.H. (2012).
Navigating the Middle Grades: Evidence from New York City.

New York, NY: Research Alliance for New York City Schools.
Koretz, D. (2005).
Alignment, high stakes, and the inflation of test scores.

Yearbook of the National Society for the Study of Education,
104(2), 99-118.
Koretz, D. (2008).
Measuring Up: What Educational Testing Really Tells Us.

Cambridge, MA: Harvard University Press.
Kurlaender, M., Reardon, S.F., and Jackson, J. (2008).
Middle School Predictors of High School Achievement in

Three California School Districts. Santa Barbara, CA:
University of California, California Dropout Research
Project.
Lehr, C.A., Sinclair, M.F., and Christenson, S.L. (2004).
Addressing student engagement and truancy prevention

during the elementary school years: A replication study
of the Check and Connect model. Journal of Education for
Students Placed at Risk (JESPAR), 9(3), 279-301.
Loveless, T. (1999).
The Tracking Wars: State Reform Meets School Policy.

Washington, DC: Brookings Institution Press.
Luppescu, S., Allensworth, E., Moore, P., de la Torre, M.,
and Murphy, J., with Jagesic, S. (2011).
Trends in Chicagos Schools Across Three Eras of Reform.

National Center for Education Statistics. (2007).
The condition of education (NCES 2007-064). U.S.

Department of Education. Washington, DC: National
Center for Education Statistics.
References
Unfulfilled promise: The dimensions and characteristics of

Unfulfilled Promise: The Dimensions and Characteristics of
Philadelphias Dropout Crisis, 2000-05. Philadelphia, PA:
Philadelphia Youth Transitions Collaborative.
Double-dose algebra as an alternative strategy to
remediation: Effects on students academic outcomes.
Journal of Research on Educational Effectiveness, 2(2),
111-148.
Sorting and supporting: Why double-dose algebra led

to better test scores but more course failures. American
Educational Research Journal, 50(4), 1-33.
Skill-Based Sorting in the Era of College Prep for All:

Costs and Benefits. Chicago, IL: University of Chicago
Raudenbush, S., and Bryk, A.S. (2002).
Hierarchical Linear Models: Applications and Data Analysis

Methods. Newbury Park, CA: Sage.
Roderick, M., Coca, V., Moeller, E., and Kelley-Kemple, T.
(2013).
From High School to the Future: The Challenge of Senior

Year in Chicago Public Schools. Chicago, IL: University of
Chicago Consortium on Chicago School Research.
Roderick, M., Nagaoka, J., Allensworth, E., Coca, V.,
Correa, M., and Stoker, G. (2006).
From High School to the Future: A First Look at Chicago

Public School Graduates College Enrollment, College
Preparation, and Graduation from Four-Year Colleges.
Roderick, M., Nagaoka, J., Coca, V., Moeller, E., Roddie,
K., Gilliam, J., and Patton, D. (2008).
From High School to the Future: Potholes on the Road to

College. Chicago, IL: University of Chicago Consortium
on Chicago School Research.
Rosenkranz, T., de la Torre, M., Stevens, W.D., and
Free to Fail or On-Track To College: Why Grades Drop

When Students Enter High School and What Adults
Can Do About It. Chicago, IL: University of Chicago
Rumburger, R., and Lim, S.A. (2008).
Why Students Drop Out of School: A Review of 25 Years

of Research. Santa Barbara, CA: California Dropout
Research Project.
101
Sinclair, M.F., Christenson, S.L., and Thurlow, M.L. (2005).
Promoting school completion of urban secondary youth

with emotional or behavioral disabilities. Exceptional
Children, 71(4), 465-482.
Sum, A., Khatiwada, I., and McLaughlin, J. (2009).
The Consequences of Dropping Out of High School:

Joblessness and Jailing for High School Dropouts and the
High Cost for Taxpayers. Boston, MA: Center for Labor
Market Studies (Northeastern University).
Tough, P. (2011).
What if the secret to success is failure? New York Times.

Retrieved from http://www.nytimes.com/2011/09/18/
magazine/what-if-the-secret-to-success-is-failure.
html?scp=2&sq=the%20character%20test&st=cse.
Zau, A.C., and Betts, J.R. (2008).
Predicting Success, Preventing Failure: An Investigation of

the California High School Exit Exam. San Francisco, CA:
Public Policy Institute of California.
102
103
Appendix A
Students College Access and Graduation
Rates, Based on High School Performance
104
In a 2006 study, UChicago CCSR examined the college-
strongest predictor of college graduation. Figure A.1
going patterns of CPS students, and compared the col-
comes from an update to that report, and shows the
leges students attended with their qualifications upon
relationship between students high school GPAs and
leaving high school, to determine what types of ACT
their college graduation rates, among students who en-
scores and GPAs students needed to have a chance of at-
rolled in a four-year college. Only students with at least
tending colleges of different selectivity levels (Roderick
a 3.0 high school GPA had at least a 50 percent chance
et al., 2006). While colleges often provide guidelines of
of graduating from college, among those who enrolled
qualifications in their application materials, the analy-
in a four-year college. In their study of North Carolina
sis showed that the students were often able to gain
colleges Bowen, Chingos, and McPherson (2009) also
admission with lower qualifications than were stated.
found that students high school GPA was a stronger
Table A.1 shows the types of colleges that students
predictor of college graduation than their SAT score,
potentially could access, based on their qualifications,
particularly after controlling for which college students
given the colleges at which students with such qualifica-
attended. As was found in Chicago, they found that a
tions actually enrolled. The selectivity categories are
3.0 GPA was the point at which students probability at
based on Barrons ratings.
graduating reached 50 percent, among students in their
The same study also examined CPS students college
North Carolina sample. For this reason, this report
graduation rates, based on their high school qualifica-
focuses on a GPA of 3.0 or above as a primary indicator
tions. Students high school GPA turned out to be the
of readiness to succeed in college.
TABLE A .1
Categories for Access to College Types Based on CPS Juniors GPAs and ACT Scores

U nweighted GPA in Core Courses (By the End of Junior Year)
Composite
ACT Score
Less than 2.0
2.0-2.4
2.5-2.9
3.0-3.4
3.5-4.0
Missing ACT
Two-Year
Colleges
Nonselective
Four-Year
Colleges
Somewhat
Selective
Colleges
Selective
Colleges
Selective
Colleges
Less than 18
Two-Year
Colleges
Nonselective
Four-Year
Colleges
Somewhat
Selective
Colleges
Somewhat
Selective
Colleges
Selective
Colleges
18-20
Nonselective
Four-year
Colleges
Somewhat
Selective
Colleges
Somewhat
Selective
Colleges
Selective
Colleges
Selective
Colleges
21-23
Somewhat
Selective
Colleges
Somewhat
Selective
Colleges
Selective
Colleges
Selective
Colleges
Selective
Colleges
24 or Higher
Somewhat
Selective
Colleges
Selective
Colleges
Selective
Colleges
Very Selective
Colleges
Very Selective
Colleges
Note: Roderick et al. (2006)
FIGURE A.1
College Graduation Rates by Unweighted High School GPA

Percent Graduated from a Four-Year
College Within Six Yaears
80
75%
70
60
54%
50
40
36%
30
21%
20
15%
105
10
0
2.0 or Less
502 Students
2.1-2.5
940 Students
2.6-3.0
1,594 Students
3.1-3.5
1,772 Students
3.6+
995 Students
Unweighted High School GPA

Note: These were CPS alumni who enrolled full-time in a four-year college by spring following their high school graduation and enrolled in a college for which we have
graduation information. Allensworth (2006).
Appendix A
Appendix B
Data Definitions
Sample Used for the Indicator
Analysis (Chapters 3 through 7)
since these data just became available in 2007. There-
The indicator analysis is based on the cohort of students
growth over time in grades and attendancestudents
who were first-time freshmen in the fall of 2009. The
who were in fifth grade in 2007-09.
fore, we used later cohorts of students to examine
students in the analytic sample attended either a neighborhood, vocational, magnet school, or an Academic
106
Change in attendance from fifth to eighth grade
Preparatory Center (APC) in ninth grade. Charter
was examined using data from three cohorts of
school students are also included in the analysis of test
students: those who were in fifth grade in the fall of
scores (see Chapters 6 and 7). Students transcripts are
2007, 2008, or 2009. This cohort comprised 99,300
not available for charter schools, which is why they are
students. The samples used in the analyses of at-
not included in the analysis of ninth-grade on-track
tendance and also GPA trends differ from the sample
rates or grades. Alternative school students are not in-
used to validate middle grade indicators and analyze
cluded in the analysis. Alternative schools do not follow
ISAT trends because of limited data from the middle
the same curriculum and graduation requirements as
grade years for students who were first-time ninth-
other schools. Students were included if they enrolled
graders in 2009. The first year in which attendance
for at least 120 days during the prior year (2008-09
and GPA data were available for students in elemen-
school year) out of a school year of 170 days total.
tary and middle schools was 2007-08, meaning there

were only two years of middle grade data for the
In Chapters 3 through 5, the analysis is based on

students with data on ninth-grade GPA and eighthgrade GPA, attendance, and ISAT scores. This cohort
comprised 19,963 students.
In Chapters 6 and 7, the analysis is based on students
cohort of students used in the rest of the report.

Change in GPA from fifth to eighth grade was examined using data from two cohorts of students: those
who were in fifth grade in the fall of 2007 or 2008.
Eighth-grade GPA data were not yet available for the
with data on tenth-grade PLAN and eighth-grade
fifth-graders in 2009 when the analysis was per-
GPA and attendance, and ISAT scores in eighth,
formed. This cohort comprised 61,791 students. Only
seventh, and sixth grade. This cohort comprised
two cohorts of fifth-grade students were used in the
20,356 students (including students in charter high
GPA analysis, instead of the three cohorts used in the
schools).
attendance analysis because at the time the analysis

was done we had not yet received GPA data files from
Sample Used for the Growth

Analysis in Chapter 8
Growth analyses were done in different ways for grades
2012 or beyond when most of the fifth-graders from

2009-10 would have been in seventh grade or higher.
Change in test scores from fifth to eighth grade was
and attendance than for test scores. To examine test
examined using students who were ninth-graders
score growth, we used one cohort of students over four
in the fall of 2009. Students form this cohort with
yearsthe same 2009 ninth-grade cohort used for all
test scores in any of those years are used to calculate
other analyses in the report. For the analysis of grades
test scores growth. This cohort comprised just over
and attendance, we did not have data on grades and
27,300 students.
attendance for this cohort as far back as fifth grade,
Variables Considered as Potential

Study Habits (reliability=0.73) is a measure constructed from students responses to the following questions
on UChicago CCSRs annual survey of CPS students:
Core GPA is the average of grades earned in the following subjects: English, math, science, and social
I set aside time to do my homework and study
science on a 4-point scale where A=4 regardless of
I try to do well on my schoolwork even when it
level (e.g., Honors).
Growth in GPA: the average yearly growth or
If I need to study, I dont go out with my friends
decline in GPA that students experienced
between fifth to eighth grade
Failures is a count of the number of semester

failures in eighth-grade core subjects (English,
math, science, and social science).
Misconducts include infractions of severity 4 or
higher. We did not include lower levels of misconducts because they are less likely to be reported
consistently across schools. Higher-level infractions
are reported more consistently than lower-level
infractions because they are usually accompanied
by a suspension.
Suspensions is a count of the number of days a
student spent out of school due to suspension.
Grit (reliability=0.86) is a measure constructed
isnt interesting to me
I always study for tests

Background Characteristics include indicators of
race/ethnicity, gender, special education status,
neighborhood concentration of poverty and social status, free/reduced-priced lunch status, and
whether a student was older than 14 when entering
high school. Neighborhood concentration of poverty
is based on data from the 2000 U.S. Census on the
census block group in which students lived. Students
home addresses were used to link each student to a
particular block group within the city, which could
then be linked to census data on the economic conditions of the students neighborhood. Two indicators
were used to construct these variables, the log of
the percentage of household above the poverty line
and the log of the percentage of men employed in the
from students responses to the following questions
block group. Neighborhood social status is based on
on UChicago CCSRs annual survey of CPS students.
data from the 2000 U.S. Census on the census block
It is based on a scale developed by Duckworth et al.
group in which students lived. Students home ad-
(2007):
dresses were used to link each student to a particular
I finish whatever I begin

I am a hard worker
I continue steadily towards my goals
I dont give up easily
block group within the city, which could then be

linked to census data on the economic conditions
of the students neighborhood. Two indicators were
used to construct these variables, the average level of
education among adults over age 21 and the log of the
percentage of men in the block group employed as
managers or executives.
Appendix B
107
Attendance is a proportion of the number of days

attended out of the number of days enrolled.
of earning a C or better in corresponding college
Growth in attendance: the average yearly growth
courses. Appendix D lists the benchmarks for the
or decline in attendance rate that students
experienced between fifth and eighth grade.
Illinois Standard Achievement Test (ISAT) measures

individual student achievement relative to the Illinois
Learning Standards. Students in grades three through
eight take the ISAT in reading and math. Information
on the ISAT is available in ISBE (2012).

108
Subscales: scores in the reading test in eighth

grade in the areas of vocabulary development,
reading strategies, reading comprehension,
literature, and extended-response results.
Scores in math test in eighth grade in the areas of
number sense, measurement, algebra, geometry,
data analysis, statistics and probability; and
extended-response results for mathematical
knowledge, strategic knowledge, and explanation
results.

of earning a B or better and a 75 percent chance
Gains over middle grades: gains were calculated

annually over the middle grades (from fifth to
sixth grade, from sixth to seventh grade, and from
seventh to eighth grade) and averaged.
Relative class rank: relative rank in reading/math

score from average of eighth-grade peers in the
same school.
High School Outcomes

PLAN and ACT tests are tests taken by CPS students
as part of the Educational Planning and Assessment
System (EPAS): the EXPLORE in the fall of eighth
and ninth grade; PLAN in the fall of tenth grade;
and ACT in the spring of eleventh grade. These tests
measure student achievement in English, reading,
mathematics, and science. The four subject-area
scores are averaged together to calculate a composite
score. The PLAN test scores range from 1 to 32. For
PLAN and ACT. Information about the EPAS system

tests is available at ACTs website: www.act.org.
Ninth-grade on-track is an indicator of whether
students are making the basic level of progress in
ninth grade to be on-track to graduating within
four years. Ninth-graders are on-track if they
have at least five full-year credits and have failed
no more than one semester in a core course by the
end of their first year of high school. Students who
are on-track at the end of their freshman year are
nearly four times more likely to graduate from high
school than their classmates who are not on-track
(Allensworth and Easton, 2005).
Eleventh-grade on-track is an indicator of whether
students have made sufficient progress to be ontrack to graduate from high school within four years
by the end of their third year of high school. Students
are considered on-track in eleventh grade if they
have at least 17 credits by the end of their third year
of high school. Students are considered off-track in
eleventh grade if they do not have at least 17 credits
by the end of their third year of high school, regardless of whether they are officially designated as
eleventh-graders or not.
GPA is the average of grades earned on a 4-point
scale where A=4 regardless of level (e.g., Honors).
Core GPA takes the average of English, math,
science, and social science; Math GPA is based
on grades from only math classes; and English
GPA is based on grades from only English classes.
Passing Math is an indicator of whether students
passed all semesters of math class. Put another way,
this variable indicates whether or not students received exactly zero semester Fs in math classes.
Passing English is an indicator of whether students
each subject, ACT has defined a college-readiness
passed all semesters of English class. Put another
benchmark; this is the score at which ACT has
way, this variable indicates whether or not students
determined that students have a 50 percent chance
received exactly zero semester Fs in English classes.
Appendix C
Research Methodology
Methods for Comparing Indicators
in Chapters 3 and 6
The process of examining potential indicators of high
school success involved several steps. We examined, in
the following order: 1) which indicators showed strong
relationships with later outcomes, and how predictive
they were individually; 2) the shape of the relationship
among those indicators that were strongly predictive;
3) which indicators were predictive above and beyond
othersadding new information to the prediction; 4)
how the indicators functioned together to predict later
outcomes; and 5) whether predictions varied by school.
These steps were performed separately for each high
school outcome.
We began by gauging the general strength of the
relationship of each potential indicator with each outcome through bivariate correlations. For dichotomous
outcomes (those that can be characterized as success
versus failuresuch as being on-track, passing math,
meeting the benchmarkwe also calculated the sensitivity and specificity of predictions using each potential
indicator as a predictor of each outcome, with 50 percent probability as the point for classifying each case
as a success or failure on the outcome. The bivariate
relationships and prediction statistics of each potential
indicator with high school outcomes are provided in
Tables E.1 and E.2 in Appendix E .
For those indicators that showed moderate to strong

relationships with high school outcomes, we examined the shape of the relationshipwhether there was
a steady improvement in the outcome for each unit of
improvement in the indicator, or whether the indicator
was related to the outcome at just high or low levels, etc.
The relationships were examined through simple charts
that graphed levels of the outcome at different levels of
the predictor. These figures are not included in the report, but the nature of relationships of the variables can
be seen through figures that graph each outcome by the
predictors; for example, Figures 7 and 8 show that the
Appendix C
relationship between eighth-grade GPA and the probability of earning As and Bs in high school is strongest
among students with high eighth-grade GPAs.
We then conducted a series of analyses in which we
examined combinations of potential indicators for each
high school outcome. With each outcome, we started
with the indicator that had the strongest bivariate relationship, and then we added additional predictors one
at a time to determine whether each added new information to improve the prediction. We used regression
models (or logistic regression models for dichotomous
outcomes), comparing the R-square (pseudo-R-square),
percent correct prediction, sensitivity, and specificity
derived from each model. We focused on model statistics, rather than coefficients associated with individual
variables, to discern whether inclusion of each additional potential indicator in the model improved the prediction of the high school outcome beyond the prediction
of a simpler model without that potential indicator. The
model statistics from select combinations of predictorsthose with the greatest potential for improving the
predictionare displayed in the bottom of Tables 3 and
7 in Chapters 4 and 7 and in the tables in Appendix E .
Once we discerned the indictors that provided

unique information to the prediction model, we
examined the ways in which the different indicators together predicted the outcomes; for example, whether an
indicator provided information at all levels of the other
predictor or whether its relationship with the outcome
depended on levels of the other predictor. Figures 7
and 8 provide an example of a case when a predictor
(attendance or ISAT scores) improves the prediction

of the outcome (earning As and Bs) only at particular
levels of another indicator (among students with high
eighth-grade GPAs).
The figures displayed in Chapter 3 show the contribution of attendance and test scores for predicting ontrack rates and high grades among students with similar
middle school GPAs. We chose those displays because
109
middle grade GPAs had the strongest bivariate rela-
scores. In contrast, as shown in the figures in Chapter 3,
tionship with the high school outcomes. An alternative
test scores only modestly add to the prediction of high
approach would be to group students by test scores and
school grades, beyond students middle grade GPAs.
then examine the contribution of grades or attendance

with the same test scores. Figures C.1 through C.4 show
Methods for Calculating School

Effects in Chapters 3 and 6
this. Students are more likely to be on-track in ninth
In studying school effects, we wanted to know if stu-
grade or earn high grades if they have high test scores.
dents who looked like they should have the same levels
However, the probability of being on-track or earning
of success in high school, based on their middle school
high grades depends considerably on attendance and
qualifications, had different levels of success depend-
middle grade GPAs among students with similar test
ing on which middle or high school they attended. To
for predicting high school outcomes among students
FIGURE C.1
95.3
100
88.7
90
80
77.8
71.3
83.4
70
60
64.9
50
40
30
20
44.0
Average Attendance by 8th-Grade ISAT Math Score

N
Low
Average High
33.9
290 < ISAT

264 < ISAT 290
238 < ISAT 264
ISAT 238
10
0
ISAT 238
238 < ISAT 264
3005
5859
7852
3247
94%
91%
88%
83%
264 < ISAT 290
97%
96%
94%
91%
100%
100%
100%
100%
290 < ISAT
8th-Grade ISAT Math Score

High Attendance
Average Attendance
Low Attendance
FIGURE C.2
On-Track Rates by ISAT Math Scores and Eighth-Grade Core GPA

Probablity of Being On-Track in 9th Grade
110
Probablity of Being On-Track in 9th Grade
On-Track Rates by ISAT Math Scores and Eighth-Grade Attendance Rates
97.1
100
91.4
90
80
78.3
71.0
82.1
70
60
61.2
50
40
30
20
42.9
Average GPA by 8th-Grade ISAT Math Score

N
Low
Average High
33.6
290 < ISAT

264 < ISAT 290
238 < ISAT 264
ISAT 238
10
0
ISAT 238
238 < ISAT 264
3005
5859
7852
3247
2.7
2.0
1.5
1.1
264 < ISAT 290

High Attendance
Average Attendance
Low Attendance
3.3
2.8
2.2
1.9
290 < ISAT
3.9
3.5
3.0
2.6
do this, we used two stages of prediction models: one
test scoresthese were the predictor variables most
model to estimate each students likely outcome, based
strongly related to the ninth-outcomes. This predicted
only on their individual qualifications at the end of
value () was entered as a predictor in the school-effects
middle school, and another to determine whether stu-
model to represent all of a students individual qualifica-
dents from each school had actual outcomes that were
tions. Combining all prior records into one variable that
higher or lower than predicted based on their individual
represents a students predicted outcome made it easier
qualifications.
to interpret the school effects in the second stage of
The first stage of analysis determined each students
analysis than a model with multiple student parameters.
predicted outcome (or probability of success for di-
It also attributes all shared variance that may exist
chotomous outcomes), based only on their middle school
between individual characteristics and schools to indi-
performance, using eighth-grade GPA, attendance, and
vidual characteristics, assuring that any school effects
FIGURE C.3
Probability of Earning As or Bs by ISAT Math Scores and Eighth-Grade Attendance
Probablity of Earning As or Bs in
9th-Grade Core Classes
100
90
80
70
Average Attendance by 8th-Grade ISAT Math Score

N
Low
Average High
290 < ISAT

264 < ISAT 290
238 < ISAT 264
ISAT 238
3005
5859
7852
3247
94%
91%
88%
83%
97%
96%
94%
91%
100%
100%
100%
100%
111
60.5
60
50
39.9
40
39.8
30
21.3
20
12.2
18.9
10
6.8
ISAT 238
238 < ISAT 264
264 < ISAT 290
290 < ISAT

High Attendance
Average Attendance
Low Attendance
FIGURE C.4
Probability of Earning As or Bs by ISAT Math Scores and Eighth-Grade Core GPA
Probablity of Earning As or Bs in
9th-Grade Core Classes
100
90
80
70
Average Core GPA by 8th-Grade ISAT Math Score

N
Low
Average High
290 < ISAT

264 < ISAT 290
238 < ISAT 264
ISAT 238
3005
5859
7852
3247
2.7
2.0
1.5
1.1
3.3
2.8
2.2
1.9
3.9
3.5
3.0
2.6
75.6
60
48.0
50
40
26.3
30
20
13.5
21.3
10
0
11.2
ISAT 238
238 < ISAT 264
264 < ISAT 290

High Attendance
Appendix C
Average Attendance
Low Attendance
290 < ISAT
that are discerned do not result simply from the selec-
those four years they were included in the analysis. We
tion of students with particular achievement levels into
included all schools and all students with test data, re-
specific schools. For this reason, we may be underesti-
gardless of whether they were receiving special educa-
mating school effects, but we are certain that any school
tion or bilingual services.
effects that are identified actually can be attributed to

schools rather than to individuals.
School effects were calculated through cross-nested
112
We modeled trends in ISAT scores over time using

a two level HLM model in which annual ISAT scores at
level 1 were nested within students at level 2. Students
models, with students nested simultaneously in their
scores at level 1 were modeled as a function of students
eighth-grade school and their ninth-grade school. For
grade level (centered at grade eight), and also whether a
each outcome, the predicted value of the outcome from
student had been retained in that year or had skipped a
the first stage analysis () controls for students quali-
grade that year. Because there were only four points to
fications (see models below). The extent that students
the trend, and preliminary models had shown that the
from particular middle schools or high schools have out-
quadratic term was very small in these models, we did
comes that are higher or lower than the predicted values
not include a nonlinear term in the model for simplic-
are captured with the school effects (b 00 and c 00 ). Cross-
ity when estimating school effects. We also looked for
nesting students in both schools allows us to capture
anomalies from linear growth at specific grade levels,
middle school and high school effects net of each other.
which are discussed further below. The models for read-
LEVEL-1 MODEL
outcome =
0 + 1*()
LEVEL-2 MODEL
(students nested simultaneously in their middle and

high school)
0 = 0 + b 00 + c 00
1 = 1
After determining the middle and high school
effects, net of each other, we ran hierarchical models
with students nested in their middle grade school only
to determine the middle school effects incorporating
high school effects. These models attribute high school
ing and math ISAT scores, attendance, and grades were

basically the same:
LEVEL-1
ISAT Score = 0
+ 1*(Grade Centered Grade 8) +

2*(Old for Grade) + 3*(Retained) + 4*(Skipped
Grade) + e
LEVEL-2
0 = 00 + u0
1 = 10 + u 1
2 = 20
3 = 30
4 = 40
effects to middle schools, showing the actual effect of
where:
attending a particular middle school, given the sending/
grade is the growth trajectory slope, centered around
receiving patterns that exist between schools.
Methods for Calculating Growth

Over Time in Chapter 8
We begin by discussing the analysis of growth trends
on the ISAT; attendance and grade models were similar,
but had some differences, as discussed below. Growth
trends on ISAT scores were based on the population
of students who were first-time ninth-graders in the
grade 8, so that it is 0 when grade=8, -1 when grade=7,

etc.
old for grade is 1 if student was old for grade at the
first data point they appear in the data
retained is 0 if never retained in grade at that point,
1 if retained that year and for each year after that, 2
the second time retained and 2 after that, and so on
if there are multiple years of grade retention
fall of 2009. For the ISAT trends, data comes from the
skipped grade is 0 if never skipped a grade, 1 if they
tests students took in the spring of 2006, 2007, 2008,
skipped a grade and after that, 2 the second time
and 2009. If students had data at any point in time in
and so on.
The retention variables allow students to have mul-
scores grow about 13 points per year in math, with a
tiple observations for a given grade, accumulating addi-
standard deviation of 2.3 points across students. The
tional growth for the extra year of instruction through
retention and skipped grade variables are about half
the retained variable, rather than through the trend
the size of the yearly growth trend reflecting smaller
variable. The extra year of instruction is accounted for
growth in the retained year than in years when students
in the retained variable for all subsequent observations.
are promoted, and an acceleration in growth in the years
The skipped grade variable has the opposite function,
that students skip grades, both of about half a year.
so that a move of two grade levels does not count as
In reading, the growth parameter and the intercept
two years in the trend calculation. The gap in learning
(ISAT score in eighth grade) are highly negatively cor-
from the year that did not occur is subtracted out of all
related, meaning that students with high scores tend to
subsequent years.
grow less than students with low scores. In math, this
The intercept and slope are allowed to vary randomly at the student level, and residuals from these models
is not true; there is a very small but positive correlation

between growth and the ISAT score in eighth grade.
were used to calculate Empirical Bayes estimates of
Scores at some grade levels do not conform to a strict
growth for each student. The model was also run with
linear trend. Grades six and seven tend to be under-
a set of dummy variables for different grades to see if
predicted by the linear trajectory in reading; grades five
there were systematic differences in particular grades.
and seven tend to be over-predicted by the linear trajec-
These are discussed further below. The ISAT growth
tory in math. Figures C.1 and C.2 show these trends.
trends without school effects are shown in Table C.1.
In reading (see Figure C.5), notice that high-scoring
Students grow around 10 points per year in reading
students improve their scores more from sixth (2007)
(see the grade coefficient), with a 2.0 standard devia-
to seventh (2008) grade than in the previous year, and
tion in trends across students (see u0 for reading). Their
then improve less from seventh to eighth grade. This
TABLE C.1
HLM Analyses for ISAT Growth Trends
Intercept
Grade (Centered
Around Grade 8)
Old for Grade
Reading
Reading With
Grade Dummies
Math
Math With
Grade Dummies
246.7**
245.6**
266.5**
266.2**
10.2**
10.2**
13.1**
12.0**
-16.4**
-16.4**
-18.2**
-18.3**
Retained in Grade
(Cumulative)
6.5**
6.7**
6.3**
6.8**
Skipped Grade
(Cumulative)
-7.1**
-6.8**
-8.4**
-7.6**
Grade 4~
2.0
-2.6
Grade 5
-0.4
-4.7*
Grade 6
2.4*
0.1
Grade 7
2.3**
-1.5*
Variance-Covariance (Standard Deviations) Components

Level-1 (e)
124.0 (11.1)**
120.8 (11.0) **
105.0 (10.2)**
101.4 (10.1)**
378.9 (19.5)**
379.0 (19.5)**
578.7 (24.1)**
580.6 (24.1)**
u1 (grd_c8)
4.0 (2.0)**
4.9 (2.2)**
5.4 (2.3)**
6.1 (2.5)**
Corr (u 0,u1)
-0.70
-0.65
0.11
0.10
98,103/ 27,366
98,103/ 27,366
98,312/27,509
98,312/27,509
u 0 (int)
Observations
Level 1/Level 2
Note: *p<.05 **p<.001

~ Some students were in fourth grade when the tests took placed in the spring of 2006.
Appendix C
113
might be due to the fact that it is seventh-grade scores
LEVEL-2
that count for applications to selective enrollment
0 = 00 + r0
1 = 10 + r1
2 = 20
3 = 30
4 = 40
schools in Chicagohigh-achieving students might put

extra effort into their performance in seventh grade for
this reason. However, the same pattern is not observed
in math (see Figure C.6), so this may be an anomaly.
Students in the not meeting range and low meets range
LEVEL-3
grew the most (also student in exceeds range grew

quite a bit but the number of students in this group is
very small). That is why there is a negative correlation
between the ISAT scores in eighth grade and growth. In
math, the picture is different; the higher the scores the
higher the growth (positive correlation between eighth-
00 = 000+ u00
10 = 100+ u 10
20 = 200
30 = 300
40 = 400
grade score and growth).

Table C.2 shows the estimates of growth trajecto-
ries allowing for variation at the school level. The basic

model is the same as described earlier where we fit a linear growth trajectory where the observations over time
(level 1) are nested within students (level 2) and then
each student is nested within the schools they attended
growth (GRD_C8) in test scores is now divided among

students (r 0 and r1) and schools (u00 and u01 ). The
estimates from reading scores indicate that half of
the variation in test score growth comes from middle
schools (the variance is 2.1 at the student level and 1.9
in the year 2009 (level 3):
at the school level). Schools explain more than half the
LEVEL-1
1.8 at the student level and 3.7 at the school level). This
ISAT Score = 0
+ 1*(GradeCentered Grade 8) +
2*(Old for Grade) + 3*(Retained) + 4*(Skipped
suggests that schools have greater influence on growth
FIGURE C.5
FIGURE C.6
Average ISAT Reading Scores Grouped by

2009 Grade Scores
Average ISAT Math Scores Grouped by

2009 Grade Scores
variation in the math test score growth (the variance is
in math scores than growth in reading.
Grade) + e
320
320
300
300
280
281
255
240
220
200
263
251
241
246
236
226
219
238
205
260
240
220
200
198
2007
280
214
206
2006
253
229
182
180
267
266
260
2008
Not Meeting
Low Meets
High Meets
Exceeds
2009
Middle Meets
306
292
288
ISAT Math Scores
ISAT Reading Scores
114
The variation in scores (intercept) and annual
180
280
279
266
266
255
253
253
237
242
239
234
225
229
213
214
261
222
199
2006
2007
2008
Not Meeting
Low Meets
High Meets
Exceeds
2009
Middle Meets
We used a similar approach for analyzing trends in

attendance and GPA from fifth through eighth grade.
However, the analysis of attendance trends was based
on three cohorts of students who were in fifth grade in
2007-08, 2008-09, or 2009-10. It also included students
who joined one of these cohorts by enrolling in a CPS
school at some point during sixth through eighth grade.
The total sample size is 99,300. The analysis of GPA
trajectories was based on two cohorts of fifth-graders
from 2007-08 and 2008-09 and also includes students
who joined these cohorts between sixth and eighth
grade. The sample size for the GPA analysis is 61,791.
As we described in Appendix B, the samples used in the
analyses of attendance and also GPA trends differ from
the sample used to validate middle grade indicators and
analyze ISAT trends because of limited data from the
middle grade years for students who were first-time
ninth-graders in 2009.
We ran two sets of analyses to look at trajectories
in attendance and GPA. The first analysis used a two
level model in which observations of either GPA or
attendance from fifth through eighth grade are nested
LEVEL-2
0 = 00 + u0
1 = 10 + u 1
2 = 20
where:
TREND is the growth trajectory slope, centered
around grade 5, so that it is 0 when grade=5, 1 when
grade=6, etc.
RETAINED is 0 if never retained in grade at that
point, 1 if retained that year and for each year after
that, 2 the second time retained and 2 after that, and
so on if there are multiple years of grade retention.
Like the analysis of ISAT trends, the RETAINED
variable allows students to have multiple observations for a given grade. It accounts for whether GPA
or attendance is different in retained years, compared to the first time at a grade level.
The intercept and slope (TREND) are allowed to
vary randomly at the student level while RETAINED is
fixed. On average, fifth-grade students have an atten-
within students.
dance rate of 95 percent, and their attendance drops
LEVEL-1
(Table C.3). For students who are retained, attendance
Attendance or GPA = 0+ 1*(TREND) + 2*(RETAINED) + e
is 2.2 percentage points lower than in non-retained
TABLE C. 2
TABLE C.3
HLM Analyses for ISAT Growth Trends Estimating

School Variation
Two-Level HLM Analyses of Attendance and

GPA Trends
only slightly, by 0.002 each year, through eighth grade
Reading
Math
Estimates From a
3-Level Model
Estimates From a
3-Level Model
Intercept
246.2**
265.3**
GRD_C8
10.3**
13.2**
Variance and (Standard Deviations)
FDPOLDFG
-12.8**
-12.9**
Level-1 (e)
CUMRET
6.7**
6.2**
CUMSKIP
-7.9**
-8.9**
Variance-Covariance (Standard Deviations)

r 0 (int)
294.6 (17.2)**
421.6 (20.5)**
2.1 (1.4)**
1.8 (1.3)**
90.9 (9.5)**
176.1 (13.3)**
1.9 (1.4)**
3.7 (1.9)**
Level 1/
96,433 observations
96,539 observations
Level 2/
/26,528 students
/26,635 students
Level 3
/472 schools
/472 schools
r 1 (grd_c8)
u 00 (int)
u10 (grd_c8)
Appendix C
Attendance
GPA
0.952
2.754
TREND
-0.002***
-0.002~
RETAINED
-0.022***
Intercept
0.041***
0.0016 (0.040)
0.133 (0.365)
u 0 (Intercept)
0.0024*** (0.049)
0.393*** (0.627)
u1 (TREND)
0.0002*** (0.014)
0.016*** (0.127)
319,879
observations/
99,300 students
184,407
observations/
61,791 students
Observations
Level 1/Level 2
Note: *p<.05 **p<.001 ***p<.0001
115
years. Average GPA for fifth-graders is around 2.8.
Why we used Bayesian estimates with inflated
Between fifth and eighth grade, GPAs drop only slightly,
variance to demonstrate growth in attendance,
by 0.002 each year. GPAs are nearly half a point higher in
grades, and test scores over time in Chapter 8.
retained years than in non-retained years (Table C.3).
After running the models described above, we then
The second analysis uses a three-level model in

which observations are nested within students who are
grades, and test scores for each student, and inflated
nested within schools. This allows us to determine how
the variance among the estimates to match model-
much variation in growth in attendance and GPA can be
based standard deviations of growth trends. The
attributed to students and how much can be attributed
rationales for these growth measures are described
to school. Schools explain about a third of the variance
below, along with details on the models and resulting
in attendance growth and a little more than half of the
coefficients. The two-level hierarchical linear models
variance in GPA growth (Table C.4).
(observations nested within students) were used to

create the estimates of each students growth when
LEVEL-1
Attendance or GPA = 0 + 1*TREND + 2*(RETAINED) + e
comparing indicators, and used to estimate the growth

trends in Figures 32, 33, and 34 in Chapter 8.
To understand the degree to which students test
LEVEL-2
116
developed Bayesian estimates of growth in attendance,
scores, grades, and attendance change over time, we
0 = 00 + r0
1 = 10 + r1
2 = 20
wanted to be sure that random fluctuation would not

inflate our assessment of the variation to which student
achievement changed over time. At the same time, we
LEVEL-3
did not want to minimize variance in growth/change
00 = 000+ u00
10 = 100+ u 10
20 = 200
that really occurred through overly stringent statistical

adjustments. This issue was a particular concern with
test scores, which are based on just one score obtained
on one day in the year, while grades and attendance
TABLE C.4
Three-Level HLM Analyses of Attendance and GPA Trends
are based on performance over the entire school year.

There is considerable error in any given test score; a
Attendance
GPA
students score on any given test may be high or low
0.947
2.69
TREND
-0.003***
0.001
for random reasonsa good or bad testing day, feel-
RETAINED
-0.019***
0.043
Intercept
Variance and (Standard Deviations)

Level-1 (e)
0.0016 (0.040)
0.133 (0.364)
r 0 (Intercept)
0.0020*** (0.045)
0.260***(0.510)
r 1 (TREND)
0.0002*** (0.013)
0.008*** (0.089)
u 0 (Intercept)
0.0005*** (0.022)
0.141*** (0.376)
u1 (TREND)
0.0001*** (0.009)
0.009*** (.096)
the measurement error associated with any one score

is larger than the average growth that students make
from year-to-year. For example, on the ISAT reading
test, students gain about 10 points a year, on average.
But a typical score has a standard error of around nine
points, which means that a students true ability is very
likely within 18 points of their measured score (two
Observations
Level 1/
Level 2/
Level3
ing ill, lucky guesses, misreading a question. In fact,
319,879
observations/
99,300 students/
184,407
observations/
61,791 students/
554 schools
545 schools
Note: *p<.05 **p<.001 ***p<.0001
77 Standard errors on the ISAT reading test range from eight to

47 points, depending on the score and grade level. For a seventh grade student with a typical score of 246, the standard
error is nine points. On the ISAT math test, standard errors
times the standard error). 77 If a students measured

score was randomly higher by 10 points one year than it
should have been,
it would look like he had gained an entire extra year
range from six to 49 points; for a typical seventh-grade

score of 265 the standard error is seven points (ISAT 2008
Technical Manual, Illinois State Board of Education Division
of Assessment).
worth of skills when he had not. We examined a
accurate estimates of students abilities at the end of
number of potential methods of estimating growth in
eighth grade (OLS growth estimates or Bayes growth
test scores, and the variance in test score growth over
estimates), given where they started in fifth grade. If
time, described below. We then applied the same meth-
the OLS estimate inflated or deflated growth based
ods to examine changes in grades and attendance over
on poor measurement in some years, the eighth-grade
the middle grade years.
score estimated based on the OLS trajectories should be
A simple method for calculating the average growth
incorrectly lower or higher than their true score. If the
that students make from fifth to eighth grade on the
Bayesian estimate reduced true differences in students
ISAT exam is to estimate a growth trajectory (slope)
growth, the eighth-grade score estimated based on the
for each student using OLS regression models fit to
Bayesian trajectories should also be incorrectly lower
each students data. This method gives the average
or higher than their true score. We then compared
gain made over three years (four time points) for each
those estimated scores to students scores on a differ-
student. However, scores that are much higher or lower
ent exam, the EXPLORE, which is taken at the start
than the students real ability can make it look like
of the ninth-grade year. Students performance on the
gains are smaller or larger than they really are. This is
EXPLORE should be strongly predicted by their true
especially true if a score with substantial error occurs
skills at the end of the eighth-grade year.
at the beginning or end of the time series, in fifth or
All things being equal, we expected that, unless they
eighth grade. As a first, simple examination of whether
were substantially better estimates of growth than the
this should be a concern, we calculated the growth
OLS estimates, the Bayesian estimates should have
trajectory for each student through OLS models, and
had a weaker correlation with EXPLORE scores since
plotted students scores against their estimated growth
their variance was constrained. However, this was not
trajectory, for a random sample of students. We could
the case. The eighth-grade scores that were estimated
easily see that this did seem to be a problem for some
based on the Bayesian growth estimates were more
students, as their estimated slope was overly influenced
predictive of ninth-grade EXPLORE test scores, even
by scores that were substantially lower or higher than
though there was less total variation (see Table C.5).
other points.
This convinced us that much of the variance in the OLS
An alternative method of calculating growth was
estimates was noise, not true growth in skills, espe-
to get Empirical Bayes estimates of each students
cially on the reading test where there were much larger
average yearly growth through hierarchical linear
differences between the trends calculated through OLS
models (HLM) that nested observations within stu-
and HLM.
dents (Raudenbush and Bryk, 2002). The Bayesian
However, we were concerned that representing
estimates minimize variance introduced by scores
the Bayes estimates of growth as the true variance
that are exceptionally high or low, compared to a stu-
in growth would not be accurate in terms of the total
dents scores in other years, given the typical growth
variation in growth. Variation in the empirical Bayes
trajectory. If scores are inconsistent relative to typical
estimates is known to be reduced beyond the true vari-
growth, the trend is adjusted to reflect the most likely
ance, as each estimate is pulled towards the group mean
true growth trajectorygiven the patterns observed in
based on the number of observations used to construct
other students. However, this method has the potential
the estimate. 78 In this case, each estimate was based on
to reduce the differences between students in their
only a small number of test observations (four each).
growth trends beyond what is true variation.
Therefore, we inflated the variance among the Bayes
We decided to test which method produced the most
78 Raudenbush and Bryk (2002).
Appendix C
estimates based on the maximum likelihood estimate of
117
TABLE C.5
Prediction of EXPLORE Test Score Based on Different Estimates of Eighth-Grade Scores

R-squares from models predicting EXPLORE scores in ninth grade as a function of fifth-grade ISAT data
growth trend predicted form OLS versus HLM models
Prediction of EXPLORE Test Score in Ninth Grade
Composite
Reading
English
Math
Using estimated eighth-grade reading

scores based on:
OLS slopes
HLM slopes
0.6872
0.6968
0.5373
0.5485
0.6106
0.6309
0.5454
0.5436
Using estimated eighth-grade math

scores based on:
OLS slopes
HLM slopes
0.6970
0.6994
0.4535
0.4557
0.5535
0.5556
0.6769
0.6784
the variance of the true slopes, obtained from the model
at the 90th percentile. For the ISAT math score figure,
statistics. In this way, the variance of the growth trends
there were 396 students at the 10th percentile, 458 at
represented the best estimate of the true variation of
the 50th percentile, and 250 at the 90th percentile.
growth trends, rather than the shrunken variation from
118
We then determined the variation in growth for

students in each group. To do this, we used the Bayesian
the Bayesian estimates.
Methods for Producing Figures 32,

34, and 35 from the HLM Models
estimates of growth that were calculated for each

student from the HLM models, but we modified them
to adjust for the reduction in variance associated with
Figures 32, 34, and 35 in Chapter 8, visually display
Bayesian estimates. Because the variance in Bayesian
the average growth in student performance (i.e., atten-
estimates is smaller than the estimated true variance
dance, GPA, and ISAT scores) from fifth to eighth grade,
that exists in the population, we inflated the variance
and the variation in growth, for students that started
of the growth estimates to match the maximum likeli-
out with low, medium and high achievement in fifth
hood estimate of the variance in growth, obtained from
grade. These figures show the extent to which some
the model statistics. We used the SAS procedure PROC
students might be catching up to students with higher
STANDARD, which allows the standard deviation for a
initial achievement (i.e., attendance, GPA, and ISAT
variable to be specified, forcing the standard deviation
scores), or falling further behind.
of the Bayesian estimates to match the estimated true
For each measure of student achievement (attendance, GPA, ISAT scores), we selected three groups of
standard deviation (tau) from the model.

Using these modified Bayesian estimates of growth,
students: those whose fifth-grade data on that per-
we then determined what low, average, and high growth
formance indicator placed them at the 10th, 50th, and
was within each of the three groups of students that were
90th percentile. For the attendance figure, which used
defined based on their fifth-grade performance (students
multiple cohorts of students, there were 1,007 students
at the 10th, 50th, and 90th percentiles in fifth grade).
at the 10th percentile, 1,328 at the 50th percentile, and
The standard deviation of growth for each of these sets
1,047 at the 88th percentile.
of students was similar to the standard deviation from
79
For the GPA figure, there
were 626 students at the 10th percentile, 628 at the
the model statistics, except for the growth in reading.
50th percentile, and 738 at the 90th percentile. For the
For example, the standard deviation of math growth
ISAT reading score figure, there were 288 students at
from the model across all students was 2.3 (see Table
the 10th percentile, 714 at the 50th percentile, and 558
C.1 on p.113), while the standard deviation of growth was
79 For attendance, the 88th percentile was the highest percentile

group in fifth grade; students in this group had 100 percent
attendance in fifth grade.
2.0 among students at the 10th percentile, 2.3 among
Methods for Calculating Figure 36
students at the 50th percentile, and 2.8 among students
To create Figure 36, we first regressed each ninth-
at the 90th percentile. Because there is a large negative
grade outcome on the corresponding fifth-grade value
covariance between initial reading test score and growth
for that outcome; in other words, ninth-grade GPA was
on the reading tests, the standard deviation of growth
regressed on fifth-grade GPA, ninth-grade on-track
for each set of students with particular reading scores in
was regressed on fifth-grade GPA and tenth-grade math
fifth grade was smaller than the overall variation. This is
PLAN scores were regressed on fifth-grade math ISAT
because much of the variation in growth overall is deter-
scores. The residuals from this regression represent
mined by students initial status. The standard deviation
how different the ninth-grade outcome is conditional
of growth in each group was about one-third the size of
on the fifth-grade values. The residuals from each
the standard deviation for the population.
of those analyses are then regressed on each of the
The figure graphs the growth that corresponded
primary eighth-grade indicators (GPA, attendance,
to the 5th, 50th, and 95th percentiles for each group
ISAT math, and ISAT reading) to discern the prediction
of students. In this way, the figures show the range of
based on eighth-grade data conditional on the values
growth observed for students that started out initially
of fifth grade. Using the coefficients from both set of
with low, medium and high performance. By using this
regressions, we calculated what the value of each ninth-
method we can see the differences in growth based on
grade outcome would be given an eighth-grade GPA of
initial status that could not be discerned by the over-
2.0 or 3.3 for students with a fifth-grade GPA of 2.5;
all estimates. For example, it shows that students that
eighth-grade attendance of 93 percent or 99 percent
started with initially low attendance had the largest
with a fifth-grade attendance of 97 percent; math ISAT
decline in attendance. We did not include quadratic
scores of 253 or 276 and reading ISAT scores of 241 or
terms in the models because there were only four points
248 with fifth-grade test scores of 225 in math and 215
of time used to construct the slopes and the nonlinear
in reading.
coefficients were very small when included.
Appendix C
119
Appendix D
ISAT Standards and EPAS Benchmarks
TABLE D.1
Scale Score Ranges That Defined Student Performance Levels on the ISAT in 2009
Academic Warning
Below Standards
Meets Standards
Exceeds Standards
120-155
156190
191226
227-329
120157
158202
203236
237-341
120160
161214
215246
247-351
120166
167219
220256
257-360
120173
174225
226266
267-369
120179
180230
231277
278-364
120162
163183
184223
224-341
120171
172199
200246
247-355
Reading Grade Level

3
Math Grade Level
120
120179
180213
214270
271-369
120193
194224
225275
276-379
120206
207234
235280
281-392
120220
221245
246287
288-410
Source: Guides to the Illinois State Assessment (ISBE, 2011). Retrieved from http://www.isbe.net/assessment/pdfs/ISAT_Interpr_Guide_2011.pdf.
TABLE D. 2
Scale Score Ranges That Defined Student Performance Levels on the ISAT in 2013
Academic Warning
Below Standards
Meets Standards
Exceeds Standards
120-159
160206
207235
236-329
120174
175216
217248
249-341
Reading Grade Level
120192
193227
228260
261-351
120201
202236
237266
267-360
120202
203238
239270
271-369
120217
218247
248270
271-364
120172
173213
214254
255-341
Math Grade Level

3
4
120190
191223
224266
267-355
120200
201234
235279
280-369
120213
214246
247291
292-379
120220
221256
257301
302-392
120233
234266
267309
310-410
Source: Retrieved from http://www.isbe.state.il.us/assessment/htmls/isat-cut-scores13.htm.
The ACT college-readiness benchmarks are scores

on the ACT test that represent the level at which students have a 50 percent likelihood of earning at least
a B in a corresponding first-year college course. ACT
has also established college readiness benchmarks for
EXPLORE and PLAN, based on a students likelihood of
meeting the ACT benchmarks, given normal progress in
grades nine through eleven.
TABLE D.3
ACTs College-Readiness Benchmarks

Test
EXPLORE
PLAN
ACT
English
14
15
18
Math
18
19
22
Reading
17
18
22
Science
19
20
23
Source: The Forgotten Middle: Ensuring that All Students Are on Target for
College and Career Readiness before High School, ACT (2008).
Appendix D
121
Appendix E
Relationships of All Potential Indicators with
High School Outcomes
Course Performance
TABLE E .1
Relationships of Middle Grade Indicators with Course Performance

Middle Grade
Indicators
9th-Grade
On-Track
(Pseudo-R 2)
Passing
9th-Grade
English
(Pseudo-R 2)
Passing
9th-Grade
Math
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Classes
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
English
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Math
(Pseudo-R 2)
0.05
0.04
0.03
0.09
0.08
0.05
8th-Grade Math ISAT
0.08
0.06
0.06
0.11
0.09
0.10
0.05
0.04
0.03
0.08
0.07
0.05
ISAT Score indicators
122
7th-Grade Math ISAT
0.07
0.05
0.05
0.10
0.08
0.09
0.05
0.04
0.02
0.08
0.07
0.05
6th-Grade Math ISAT
0.06
0.05
0.05
0.10
0.08
0.09
% Correct Vocabulary
Development
(8th-Grade ISAT)
0.03
0.02
0.02
0.04
0.04
0.02
% Correct Reading
Strategies
(8th-Grade ISAT)
0.03
0.03
0.02
0.05
0.04
0.03
% Correct Reading
Comprehension
(8th-Grade ISAT)
0.05
0.04
0.03
0.08
0.08
0.05
% Correct Literature
(8th-Grade ISAT)
0.03
0.03
0.02
0.06
0.05
0.04
Reading Extended
Response Points
(8th-Grade ISAT)
0.03
0.03
0.02
0.05
0.05
0.03
% Correct Number Sense

(8th-Grade ISAT)
0.05
0.03
0.04
0.08
0.06
0.07
% Correct Measurement
(8th-Grade ISAT)
0.06
0.04
0.04
0.08
0.07
0.08
% Correct Algebra
(8th-Grade ISAT)
0.07
0.05
0.05
0.10
0.08
0.09
% Correct Geometry
(8th-Grade ISAT)
0.05
0.03
0.03
0.07
0.05
0.07
% Correct Data Analysis,

Statistics and Probability
(8th-Grade ISAT)
0.06
0.04
0.04
0.08
0.07
0.07
0.03
0.02
0.02
0.04
0.04
0.04
Math Extended Response

PointsMathematical
Knowledge
(8th-Grade ISAT)
TABLE E .1: CONTINUED

Middle Grade
Indicators
9th-Grade
On-Track
(Pseudo-R 2)
Passing
9th-Grade
English
(Pseudo-R 2)
Passing
9th-Grade
Math
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Classes
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
English
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Math
(Pseudo-R 2)

PointsStrategic
Knowledge
(8th-Grade ISAT)
0.03
0.02
0.02
0.04
0.04
0.04

PointsExplanation
(8th-Grade ISAT)
0.04
0.03
0.03
0.06
0.05
0.04
ISAT Reading Gains Over

Middle Grades
0.00
0.00
0.00
0.00
0.00
0.00
ISAT reading Growth

Over Middle Grades
(Removing Influence of
Starting Point)
0.00
0.00
0.00
0.07
0.06
0.04
Relative Class Rank

Based on ISAT Reading
0.04
0.03
0.02
0.06
0.06
0.04
ISAT Math Gains Over

Middle Grades
0.00
0.00
0.00
0.00
0.00
0.00
ISAT reading Growth

Over Middle Grades
(Removing Influence of
Starting Point)
0.00
0.00
0.00
0.00
0.00
0.00
Relative Class Rank

Based on ISAT Math
0.06
0.04
0.05
0.08
0.06
0.08
0.15
0.13
0.10
0.18
0.18
0.12
8th-Grade Math GPA
0.14
0.11
0.10
0.15
0.14
0.13
8th-Grade Core GPA
0.18
0.15
0.12
0.21
0.20
0.15
8th-Grade Number of Fs
0.10
0.09
0.07
0.08
0.08
0.06
7th-Grade Core GPA
0.17
0.14
0.12
0.18
0.18
0.14
0.12
0.09
0.08
0.08
0.08
0.07
0.06
0.04
0.04
0.04
0.04
0.03
0.03
0.02
0.02
0.00
0.00
0.00
8th-Grade Grit
0.01
0.01
0.00
0.00
0.00
0.00
ISAT Score indicators
Grade-Based Indicators
Behavior-Based indicators
Appendix E
123
TABLE E .1: CONTINUED

Middle Grade
Indicators
9th-Grade
On-Track
(Pseudo-R 2)
Passing
9th-Grade
English
(Pseudo-R 2)
Passing
9th-Grade
Math
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Classes
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
English
(Pseudo-R 2)
Earning
As or Bs in
9th-Grade
Math
(Pseudo-R 2)
Behavior-Based indicators
0.02
0.01
0.01
0.03
0.03
0.02
0.08
0.06
0.05
0.08
0.08
0.06
0.06
0.05
0.03
0.09
0.09
0.06
8th-Grade ISAT Math

& Reading + 7th-Grade
ISAT Math & Reading
0.08
0.06
0.06
0.12
0.10
0.11

Attendance
0.21
0.17
0.15
0.22
0.21
0.16

7th-Grade Core GPA
0.20
0.17
0.14
0.22
0.22
0.16

ISAT Math & Reading
0.18
0.15
0.13
0.21
0.20
0.17

Attendance +
7th-Grade Core GPA
0.23
0.19
0.16
0.23
0.23
0.17

Attendance +
ISAT Math & Reading
0.22
0.17
0.16
0.22
0.22
0.18
All Student-Level
Indicators
0.23
0.19
0.17
0.25
0.24
0.20
School Effects
(Fixed Effects of Middle
and High Schools)
0.25
0.32
0.29
0.34
0.22
0.27
All + School Effects

(Fixed Effects)
0.44
0.47
0.45
0.48
0.41
0.42
Race, Gender, Old-forGrade, Neighborhood
Poverty, Neighborhood
Social Status, Special
Education Status
Combinations of Indicators
124
School Effects
Test Scores
prediction of any PLAN scores. For PLAN English and

2
This section shows adjusted-R for models predicting

2
science, the best prediction comes from combining
PLAN test scores and pseudo-R for PLAN benchmarks
eighth-grade reading and math scores; for PLAN read-
for all the potential middle grade indicators used in the
ing the best prediction comes from combining seventh-
analyses. Table E.2 shows the prediction fit for subject-
and eighth-grade reading scores; and for PLAN math,
specific scores, as well as composite scores.
the best prediction comes from combining seventh-
As with composite scores, the PLAN subject-specific
and eighth-grade math scores. Adding core GPA or
scores are predicted almost as well with sixth-grade
students background characteristics only slightly
scores as with eighth-grade scores. In some cases,
improves the fit of the models. Knowing the middle
PLAN English and reading scores are best predicted
school and high school that students attend can help
by middle grade reading scores, while PLAN math and
with the prediction of PLAN subject-specific scores.
science scores are best predicted by middle grade math
The same findings apply to whether students reach
scores. Combining two test scores helps improve the
the benchmarks in each subject.
125
Appendix E
TABLE E . 2
Relationships of Middle Grade Indicators with PLAN Scores

English
Middle Grade
Indicators
Reading
Math
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)

Linear Term
0.50
0.50
0.36
0.36
0.43
0.44
0.33
0.33
0.41
0.41
0.23
0.24
8th-Grade Math ISAT

Linear Term
0.50
0.50
0.34
0.34
0.37
0.37
0.26
0.26
0.61
0.61
0.35
0.35

Linear Term
0.51
0.51
0.35
0.35
0.43
0.44
0.31
0.31
0.40
0.40
0.22
0.23
7th-Grade Math ISAT

Linear Term
0.49
0.49
0.33
0.33
0.36
0.36
0.26
0.26
0.60
0.60
0.34
0.34

Linear Term
0.50
0.51
0.34
0.34
0.43
0.44
0.30
0.31
0.39
0.39
0.22
0.22
6th-Grade Math ISAT

Linear Term
0.47
0.47
0.32
0.32
0.35
0.35
0.25
0.25
0.55
0.55
0.31
0.31
% Correct Vocabulary
Development
(8th-Grade ISAT)
0.31
0.23
0.26
0.21
0.23
0.14
% Correct Reading
Strategies
(8th-Grade ISAT)
0.24
0.18
0.20
0.15
0.20
0.11
% Correct Reading
Comprehension
(8th-Grade ISAT)
0.44
0.33
0.37
0.30
0.35
0.22
% Correct Literature
(8th-Grade ISAT)
0.30
0.22
0.25
0.20
0.24
0.13
Reading Extended
Response Points
(8th-Grade ISAT)
0.17
0.12
0.14
0.10
0.14
0.07
% Correct Number Sense

(8th-Grade ISAT)
0.32
0.21
0.24
0.18
0.39
0.25
% Correct Measurement
(8th-Grade ISAT)
0.37
0.25
0.27
0.20
0.44
0.29
% Correct Algebra
(8th-Grade ISAT)
0.42
0.29
0.32
0.23
0.48
0.30
% Correct Geometry
(8th-Grade ISAT)
0.30
0.20
0.21
0.16
0.38
0.23
% Correct Data Analysis,

Statistics and Probability
(8th-Grade ISAT)
0.36
0.26
0.27
0.21
0.40
0.26

PointsMathematical
Knowledge
(8th-Grade ISAT)
0.23
0.17
0.17
0.14
0.25
0.14
ISAT Score Indicators
126
Science
Composite
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
0.37
0.38
0.15
0.16
0.56
0.58
0.38
0.38
0.43
0.44
0.18
0.18
0.63
0.63
0.39
0.39
0.36
0.38
0.15
0.16
0.56
0.57
0.35
0.35
0.43
0.44
0.18
0.18
0.62
0.62
0.38
0.38
0.35
0.37
0.15
0.16
0.55
0.57
0.35
0.35
0.39
0.40
0.16
0.17
0.58
0.58
0.36
0.36
0.19
0.09
0.33
0.22
0.15
0.07
0.26
0.17
0.29
0.15
0.48
0.35
0.19
0.09
0.33
0.21
0.11
0.04
0.19
0.11
0.26
0.014
0.40
0.27
0.31
0.15
0.46
0.31
0.33
0.16
0.52
0.34
0.25
0.12
0.38
0.25
0.28
0.13
0.44
0.31
0.17
0.07
0.27
0.17
Appendix E
127
TABLE E . 2: CONTINUED

English
Middle Grade
Indicators
Reading
Math
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)

PointsStrategic
Knowledge
(8th-Grade ISAT)
0.23
0.18
0.17
0.15
0.25
0.15

PointsExplanation
(8th-Grade ISAT)
0.19
0.16
0.14
0.12
0.20
0.10
ISAT Reading Gains

Over Middle Grades
0.05
0.04
0.04
0.03
0.02
0.01
Relative Class Rank Based

on 8th-Grade ISAT Reading
0.34
0.26
0.30
0.23
0.26
0.14
ISAT Math Gains Over

Middle Grades
0.00
0.00
0.00
0.00
0.00
0.00
Relative Class Rank Based

on 8th-Grade ISAT Math
0.35
0.25
0.26
0.19
0.42
0.25
ISAT Score Indicators
128
Grade-Based Indicators
0.24
0.17
0.20
0.16
0.21
0.14
8th-Grade Math GPA
0.18
0.13
0.14
0.11
0.21
0.14
8th-Grade Core GPA
0.26
0.19
0.22
0.17
0.25
0.16
Number of Fs in 8th-Grade
0.05
0.04
0.04
0.04
0.05
0.04
0.04
0.03
0.03
0.02
0.05
0.03
0.02
0.03
0.02
0.02
0.02
0.02
0.01
0.01
0.01
0.01
0.01
0.01
8th-Grade Grit
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.27
0.19
0.21
0.15
0.27
0.14
Combination ISAT Tests

7th- and 8th-Grade Math
Math
0.55
0.52
0.57
0.39
0.35
0.40
0.47
0.39
0.46
0.36
0.27
0.34
0.44
0.63
0.62
0.26
0.36
0.35

and Math
0.61
0.41
0.49
0.36
0.64
0.36
Two ISAT Tests* + Core GPA
0.58
0.40
0.48
0.36
0.63
0.37
Two ISAT Tests* +

0.58
0.41
0.49
0.37
0.63
0.37
Behavior-Based Indicators
Race, Gender, Old-forGrade, Neighborhood
Poverty, Neighborhood
Social Status, Special
Education Status
Science
Composite
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
0.16
0.07
0.27
0.19
0.13
0.04
0.22
0.13
0.02
0.00
0.04
0.30
0.23
0.09
0.38
0.24
0.00
0.00
0.00
0.00
0.29
0.12
0.44
0.28
0.17
0.08
0.27
0.19
0.15
0.06
0.23
0.16
0.20
0.09
0.31
0.22
0.04
0.02
0.06
0.05
0.03
0.01
0.05
0.03
0.02
0.01
0.02
0.02
0.01
0.00
0.01
0.01
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.22
0.10
0.32
0.18
0.38
0.44
0.46
0.17
0.19
0.19
0.61
0.66
0.69
0.41
0.41
0.44
0.48
0.21
0.72
0.46
0.45
0.20
0.69
0.45
0.46
0.21
0.70
0.44
Appendix E
129
TABLE E . 2: CONTINUED

English
Middle Grade
Indicators
Reading
Math
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
0.59
0.41
0.50
0.37
0.64
0.37
School Effects
(Fixed Effects of Middle
and High Schools)
0.40
0.33
0.41

(Fixed Effects)
0.63
0.52
0.67
All
(Two ISAT Tests* +
+ Core GPA + Attendance
+ Suspensions + Number of
Fs + Misconducts + Relative
Class Rank)
School Effects
130
Note: Sample size was kept the same for most analyses to make comparisons easier, except when data from surveys were analyzed. In those cases the sample
sizes get smaller. That is the case when grit and study habits are part of the analysis.
* For English and science PLAN test, as well as composite scores, the two tests are eighth-grade reading and math ISAT tests. For Reading PLAN test the two
tests are eighth- and seventh-grade reading ISAT tests. For math PLAN test the two tests are eighth- and seventh-grade math ISAT tests.
** Given the data for this cohort and analyses, a model with no explanatory variables would be able to predict correctly 51.7 percent of students in English by
assigning all of them to not reaching the benchmark, 69.6 percent in reading, 83.8 percent in math, 93.5 percent in science benchmarks, and 77.2 percent for the
composite score of 18 and above.
Given the low variability in whether or not students reach benchmarks by middle and high school, these models could not be run.
As was the case with PLAN composite, it is easier

to correctly predict which students are at risk of not
variables for the four PLAN subject-specific tests.

In general, most subjects can have a large propor-
reaching benchmarks than which students will reach
tion of students misclassified as at risk of not reach-
the benchmarks, because so few students are success-
ing benchmarks when they actually do. This number
ful. But there are big differences among the subject-
is especially high for science, 64 percent of students
specific tests. In English the proportion of student
who do reach the benchmark are identified as at risk of
correctly classified at risk is 80 percent, the lowest;
not doing so using middle grade reading and math test
in reading it goes up to 91 percent, while in math and
scores. Even though the proportion is large, there are
science it is above 96 percent. Figure E.1 shows the pro-
only 6.5 percent of students in our sample who reach the
portion of students correctly classified as not reaching
science benchmarks. In reading and math the propor-
benchmarks in the Y-axis and the proportion of stu-
tion of students identified as at risk when they actually
dents incorrectly classified as not reaching benchmarks
reach the benchmark is 40 percent, and in English it is
in the X-axis for models with different explanatory
23 percent.
Science
Composite
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
PLAN
Scores
(Adjusted-R 2)
Reaching
Benchmark
(Pseudo-R 2)
0.47
0.21
0.71
0.45
0.39
0.50
0.53
0.75
131
Appendix E
FIGURE E.1
Correct versus Incorrect Classification of Students Not Meeting Benchmarks
Perfect
Prediction
0.9
8th-Grade ISAT Math and Reading
0.8
8th-Grade ISAT Math

All
0.7
Core GPA
0.6
Demographics
Ra
nd
om
Gu
es
Attendance
0.5
0.4
0.3
0.2
0.1
Worse
Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Proportion of Students Incorrectly Identified as Scoring Below Benchmark

Risk of Not Meeting the Reading PLAN Benchmark
Proportion of Students Correctly Identified as Scoring Below
Benchmark True Positive Proportion (Sensitivity)
132

Risk of Not Meeting the English PLAN Benchmark

1.0
1.0
Attendance
Perfect
Prediction
0.9
All
Demographics
Core GPA
7th- and 8th-Grade ISAT Reading
0.8
Ra
nd
om
Gu
es
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Worse
Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8

0.9
1.0
FIGURE E.1
Correct versus Incorrect Classification of Students Not Meeting Benchmarks

Risk of Not Meeting the Math PLAN Benchmark

1.0
0.9
Demographics
Perfect
Prediction
All
Attendance
Core GPA
7th-Grade ISAT Math

8th-Grade ISAT Math
0.8
7th- and 8th-Grade ISAT Math

0.7
0.6
0.5
0.4
0.3
0.2
0.1
Ra
nd
om
Gu
es
0.0
0.0
133
Worse
Prediction
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0


Risk of Not Meeting the Science PLAN Benchmark

1.0
Perfect
Prediction
0.9
All
Demographics
8th-Grade ISAT Math and Reading

8th-Grade ISAT Math
0.8
Core GPA
Attendance
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Ra
nd
om
Gu
es
s
Worse
Prediction
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8

Appendix E
0.9
1.0
ABOUT THE AUTHORS

ELAINE M. ALLENSWORTH is the Lewis-Sebring Director
at UChicago CCSR where she has conducted research on
educational policy for the last 15 years. She is best known for
her studies of high school graduation and college readiness,
and also conducts research in the areas of school leadership
and school organization. Her work on early indicators of high
school graduation has been adopted for tracking systems
used in Chicago and other districts across the country. She
is one of the authors of the book Organizing Schools for
Improvement: Lessons from Chicago, which provides a detailed analysis of school practices and community conditions
that promote school improvement. Dr. Allensworth holds a
PhD in Sociology and an MA in Urban Studies from Michigan
State University. She was once a high school Spanish and
science teacher.
134
134
JULIA A. GWYNNE is a Senior Research Analyst at UChicago

CCSR. Her current work focuses on early warning indicators
of high school and college readiness and the use of indicators with groups such as English Language Learners and
students with disabilities. In addition, she has conducted
research on student mobility, school closings, and classroom instructional environments. She received her doctoral
degree in sociology from the University of Chicago.
MARISA DE LA TORRE is the Director for Internal Research

Capacity at the Consortium on Chicago School Research.
She is very familiar with Chicago Public Schools policies
and part of her work involves studying them. She was the
author of two studies on the effects of policies aimed at
the lowest-performing schools in the district. One, Turning
Around Low-Performing Schools in Chicago, investigates the
effects of these policies on whether schools see improvements after the reform. The other report deals with the effect
of school closings on students academic outcomes, When
Schools Close: Effects on Displaced Students in Chicago
Public Schools, which has been widely cited in the press. This
work prompted Chicago Public Schools to create a Student
Bill of Rights for students affected by school closings. She is
also familiar with the high school choice process in Chicago
Public Schools and one of her studies was published in
School Choice and School Improvement, a book edited by
Mark Berends, Marisa Cannata, and Ellen B. Goldring. She
is currently studying the impact that attending higherperforming high schools have on students academic and
non-academic outcomes. Before joining UChicago CCSR,
she worked for the Chicago Public Schools in the Office of
Research, Evaluation, and Accountability. She received a
masters degree in economics from Northwestern University.
PAUL MOORE is a Research Analyst at UChicago CCSR and

is in the process of completing an MA in the social sciences
at the University of Chicago. His research interests include
quantitative modeling and methodology. Moore is studying the effects of attending higher performing schools on
students academic performance and noncognitive skills.
He earned a BS in mathematics and education science from
Vanderbilt University.
This report reflects the interpretation of the authors. Although UChicago CCSRs Steering Committee provided technical advice, no formal endorsement by these individuals, organizations, or the full Consortium
should be assumed.
CONSORTIUM ON CHICAGO SCHOOL RESEARCH

Directors
Steering Committee
ELAINE M. ALLENSWORTH
Lewis-Sebring Director
LILA LEFF
Co-Chair
Umoja Student
Development Corporation
EMILY KRONE
Director for Outreach and
Communication
JENNY NAGAOKA
Deputy Director
MELISSA RODERICK
Senior Director
Hermon Dunlap Smith
Professor
School of Social Service
Administration
PENNY BENDER SEBRING
Founding Director
SUE SPORTE
Director for Research
Operations
W. DAVID STEVENS
Director for Research
Engagement
MARISA DE LA TORRE
Director for Internal
Research Capacity
KATHLEEN ST. LOUIS

CALIENTO
Co-Chair
Spark, Chicago
Ex-Officio Members
Individual Members
VERONICA ANDERSON
Communications Consultant
JOANNA BROWN
Logan Square
Neighborhood Association
ANDREW BROY
Illinois Network of
Charter Schools
TIMOTHY KNOWLES
Urban Education Institute
RAQUEL FARMER-HINTON
University of Wisconsin,
Milwaukee
Institutional Members
REYNA HERNANDEZ
Illinois State Board of
Education
JOHN R. BARKER
Chicago Public Schools
CLARICE BERRY
Chicago Principals and
Administrators Association
AARTI DHUPELIA
CHRISTOPHER KOCH
Illinois State Board of
Education
KAREN G.J. LEWIS
Chicago Teachers Union
SHERRY J. ULERY
CHRIS JONES
Stephen T. Mather
High School
DENNIS LACEWELL
Urban Prep Charter Academy
for Young Men
RUANDA GARTH
MCCULLOUGH
Loyola University, Chicago
LUISIANA MELNDEZ
Erikson Institute
LISA SCRUGGS
Duane Morris LLP
LUIS R. SORIA
BRIAN SPITTLE
DePaul University
MATTHEW STAGNER
Mathematica Policy
Research
AMY TREADWELL
Chicago New Teacher Center
ERIN UNANDER
Al Raby High School
ARIE J. VAN DER PLOEG
American Institutes for
Research (Retired)
KIM ZALENT
Business and Professional
People for the Public Interest
1313 East 60th Street

Chicago, Illinois 60637
T 773-702-3364
F 773-702-2010
ccsr.uchicago.edu
OUR MISSION The University of Chicago Consortium on Chicago

School Research (UChicago CCSR) conducts research of high technical
quality that can inform and assess policy and practice in the Chicago
Public Schools. We seek to expand communication among researchers,
policymakers, and practitioners as we support the search for solutions
to the problems of school reform. UChicago CCSR encourages the use
of research in policy action and improvement of practice, but does not
argue for particular policies or programs. Rather, we help to build capacity
for school reform by identifying what matters for student success and
school improvement, creating critical indicators to chart progress, and
conducting theory-driven evaluation to identify how programs and
policies are working.

Middle Grades Report

Uploaded by

Copyright:

Available Formats

Middle Grades Report

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Middle Grades Report

Uploaded by

Copyright:

Available Formats

RESEARCH REPORT NOVEMBER 2014

Looking Forward to High School and College

Middle Grade Indicators of Readiness

Elaine M. Allensworth, Julia A. Gwynne, Paul Moore, and Marisa de la Torre

17 Issues in Developing and

23 Changes in Academic Performance

29 Middle Grade Indicators of

55 Who Is at Risk of Earning Less

63 Indicators of Whether Students

75 Who Is at Risk of Not Reaching

81 How Grades, Attendance, and

This report was produced by UChicago CCSRs publications and

Graphic Design: Jeff Hall Design

MIDDLE GRADE INDICATORS OF READINESS

Much of the pressure to improve educational attain-

messages that policymakers and practitioners hear

ment is on high schools, but focus has also turned to

about what matters for academic success, these

earlier grades. There is a very large population of

findings are intended to clarify which factors are

students who struggle with the transition from the

most strongly related to students educational

middle grades to high school, raising concerns that

high school failures are partially a function of poor

Which students are likely to be successful in high

including attendance rates, grades in specific classes

readiness. For example, do students need to start

and GPAs, test scores, study habits, grit, discipline

the middle grades with high performance to leave

records, and background characteristicsare most

middle school with high levels of readiness? Or does

predictive of high school success. It also shows the

students performance change considerably while

degree to which high school readiness is a function

they are in the middle grades? These findings are

of the high school students choose or the middle

intended to help practitioners and policymakers set

school they attended. With all of the conflicting

realistic goals for improvement.

Middle Grade Information Can Be

of a GPA point. We often hear that grades are subjec-

Many characteristics of students are associated

to succeed if they came from one school than another. In

with their academic outcomes, from background

fact, students with the same grades, attendance, and test

characteristics to test scores, grades, attendance,

score records upon leaving eighth grade are more likely

and discipline records, to noncognitive factors. With

to succeed in high school if they came from some middle

so many factors associated with later outcomes, it may

schools than from others, net of the effects of which high

seem like practitioners would need complicated models

schools students attend (see Chapter 3). At the same

to gauge students likelihood of success in later years.

time, the differences in success among students with the

However, after taking into account just two or three

same grades from different middle schools are small,

key middle school indicators, other information about

compared to the differences in success among students

students only marginally improves the prediction of