On-Line Assessment of Higher-Order Thinking Skills: A Java-Based Extension To Closed-Form Testing

ICOTS6, 2002: Thomas et al.
ON-LINE ASSESSMENT OF HIGHER-ORDER THINKING SKILLS:

A JAVA-BASED EXTENSION TO CLOSED-FORM TESTING
David A. Thomas, Giray kten, and Paul Buis
Ball State University
USA
WWW-based mathematics and statistics courses frequently incorporate machine-scorable items
(i.e., True-False, Multiple Choice, and Matching) in both formative and summative assessments.
For instance, WebCT and BlackBoard provide interfaces for the development and delivery of
closed-form quizzes and examinations. Using these technologies, it is relatively easy to determine
whether students possess detailed factual knowledge. It is much more difficult, using these
technologies, to assess higher order thinking skills. This paper presents a Java-based extension to
closed-form testing that may be better suited to assessing higher-order thinking skills.
INTRODUCTION
For over a decade, reform movements in mathematics and statistics education have
emphasized teaching for understanding. One outcome of this activity has been a renewed interest
in the assessment of higher-order thinking skills. We are particularly interested in assessment
procedures and technologies that reveal more about what students know, as opposed to what they
don't know, and how students think about mathematical and scientific ideas. Our current research
focuses on the development of strategies and technologies for assessing higher-order thinking
skills in WWW-based mathematics and statistics courses. We believe that, until achievement in
WWW-based courses is assessed as rigorously and reliably as in on-campus courses, WWWbased courses will not be credible with or acceptable to many institutions and individuals. This
paper describes the development and testing of a new, Java-based, closed-form assessment tool
that addresses that goal. We call the item type that is the basis of our approach a grid.
.
SELECTION TASKS
In grid-based assessment, information (e.g., text, formulas, figures, numbers) is presented
in an array of cells called a grid. A task or set of tasks is presented apart from the grid. In
composing a response, the student scans the grid, looking for the cell or cells that collectively
constitute a correct and complete response to the task. Depending on the task, a correct answer
may consist of one or more cells. Figure 1 illustrates a type of grid-based task called a selection.
Task #1
Which cells contain equivalent
fractions, decimals, percents,
or shaded portions of the
whole?
Figure 1. Sample Selection Task.

Selection tasks ask the student to identify sets of grid cells that stand in a particular
relationship defined by the task. For example, the contents of certain grid cells may be related in
that they are numerically or algebraically equivalent, descriptive or illustrative of a particular
1
concept or process, recognizable parts of some whole, or steps in an algorithm. Such associations
are the very stuff of declarative knowledge structures and are the basis of conceptual knowledge
in mathematics and statistics. In the case of Task #1, cells 2, 6, and 9 contain equivalent
representations of the same number. Therefore, a correct response to Task #1 is the set {2, 6, 9}.
Figure 2 shows a planning table used by the authors to design selection Task #1. Planning
tables of this sort are never shown to students, for they reveal both the solution to the task and
traps set to detect specific misconceptions. The left side of the table shows all pair-wise cell
relationships in a correct response. The right side shows pair-wise cell relationships associated
with misconceptions set as error traps by the teacher.
Equivalent Forms (Cells)
fractions decimals percents graphics
fractions
decimals
percents
graphics
6&9
2&6
2&9
fractions
5&6
1&7
2&7 6&8
Figure 2. Planning Table.
Error Traps (Cells)

decimals percents graphics
4&5
3&8
2&4
Selection tasks are scored using a computerized matrix algebra procedure that compares
all of the pair-wise cell relationships in a student response to the pair-wise cell relationships in a
correct response. This comparison is used to generate an item score (i.e., partial credit) for the
task. The same computer program that scores student responses is used to generate formative
feedback messages keyed to "trapped" misconceptions. For instance, students who have difficulty
converting fractions to decimals might believe that cells 5 and 6 are equivalent. In a fully
developed grid-based assessment system, students will receive feedback focused on correcting the
specific misconceptions implicit in their responses. At the ICOTS-6 conference, examples of this
sort of scoring and feedback will be demonstrated as opportunity permits.
SEQUENCING TASKS
Procedural knowledge is an important aspect of many disciplines. In particular, knowing
"what comes next" is critical in many scientific, mathematical, and statistical contexts. A
sequencing task requires the student to order all of the cells in a grid according to some criterion.
For instance, given a grid containing
Names of selected chemical elements, science students might be asked to order the
elements on the basis of their atomic numbers;
A mixed set of integers, fractions, and decimals, mathematics students might be asked to
order them smallest to largest;
Steps in a hypothesis testing procedure, statistics students might be asked to order the
steps.
Like selection tasks, sequencing tasks are scored using a procedure that compares all of
the pair-wise cell relationships in a student response to the pair-wise cell relationships in a correct
response. This approach avoids a distracting and confusing issue: The enormous number (n!) of
permutations possible in a set of n cells. Instead, our scoring scheme focuses on characterizing the
internal "orderliness" of a sequence by inspecting only (n2 - n)/2 pair-wise cell comparisons.
These inspections are easily handled using simple computational matrix algebra.
COMBINATION TASKS
Combination tasks require students to first select and then sequence a subset of grid cells.
For instance, the superintendent of a power plant might want to know whether the plant's
operating engineers can reliably and consistently select a suitable sequence of actions from a
complex set of possible responses when presented with various crises. A less dramatic
circumstance is shown in Task #2, where statistics students are asked to select then sequence a set
of actions in order to decide which of two standardized test scores represents a superior
performance. The format of the cell contents indicates that two sets of calculations are required,
followed by a decision. The correct response to this task is the ordered sequence 1-6-4-8.
Task #2
You wish to compare
scores
from
two
standardized
tests
to
determine which score is
better. Scores from both
tests
are
normally
distributed. List the cell
numbers that contain the
proper steps in the order
that you would perform
them.
1
Raw score(s) minus
mean score(s)
2
Divide by mean(s)
3
Compare raw scores
4
Compare z-scores
5
Mean score(s) minus
raw score(s)
6
Divide by standard
deviation(s)
7
Divide by raw score(s)
8
Select larger
score
9
Select smaller
score
z-
z-
Figure 3. Combined Selection & Sequencing Task.

PRELIMINARY RESULTS
We have developed methods for aggregating scores on individual selection, sequencing,
and combination items to yield an overall examination score. We have also studied the expected
outcome of random guessing relative to each item type and their aggregations. In the course of
these investigations, we have developed closed-form descriptions of the expected outcomes that
agree closely with simulations created using independent methods. On the basis of these findings,
we believe that grid-based assessment may provide a "finer grained picture" of student knowledge
structures and misconceptions than traditional closed-form assessment.
We have also conducted a limited number of informal investigations to determine
whether students perform differently on grid-based items than they do on comparable open-ended
items. For example, a group of 30 students enrolled in the course Mathematics for Elementary
Teachers at Ball State University were given a paper-and-pencil statistics quiz that included Task
#2 above and a related open-ended item. Full or partial credit was assigned to each response.
When these scores were analysed, it was clear that differences in performance on the two item
types were no greater than one might expect between two open-ended items of the same type (r =
0.82). That is, students who did well on traditional open-ended items, also did well on grid-based
items, and so on. While a great deal of work remains to be done on the issues of validity and
reliability of measurement using grid-based methods, we are encouraged to note that average
students appear to adapt quickly to the new item types. Consequently, we are currently planning
additional trials addressing a wide range of technical issues and content areas.
COMPUTER IMPLEMENTATION
Our first step in developing a computer environment for grid-based testing was to create a
spreadsheet model of the scoring procedure. In the spreadsheet, we were able to view all of the
variables and algorithms employed in the model simultaneously. This made it easy to track the
manner in which the model treated student responses and to verify that the model was performing
as intended. While a full discussion of the scoring procedures is beyond the scope of this paper,
the fundamental concept may be illustrated as follows in the case of a selection task. Related
strategies have been developed for analysing responses to sequencing and combined tasks.
1. Given a selection task with n cells and a planning table like that shown in Figure 2, the
instructor creates an n x n matrix, C, called the correct response matrix. In this matrix, each
entry Cij is either a 1 or a 0, depending on whether the contents of cells i and j do or do not
meet the selection criterion of the task. Matrix C characterizes all pair-wise cell relationships
in a correct response.
2. Using the same procedure, a similar matrix R is created that characterizes a student's response
to the item. Matrix R characterizes all pair-wise cell relationships in the student's response.
3. When matrix R is subtracted from matrix C, an error matrix E is obtained like that shown in
Figure 4. In this matrix, a 1 indicates an error of omission, i.e., an essential pair-wise
3
relationship is missing. A -1 indicates an inclusion error, i.e., a false relationship is included

in the response. And a 0 indicates a correct response. Because of the symmetry of this matrix,
only the lower triangular portion need be considered.
4. The overall score for the item is computed as the number of 0's in this lower triangular matrix
divided by the number of elements in that portion of the error matrix. For instance, the
response {5, 6, 9} to Task #1 produces the error matrix seen in Figure 4. Reports of this sort
are meant for the instructor, not the student.
CELL 1
CELL 2
CELL 3
CELL 4
CELL 5
CELL 6
CELL 7
CELL 8
CELL 9
Legend
Score:
CELL 1
CELL 2
CELL 3
CELL 4
CELL 5
CELL 6
CELL 7
CELL 8
0
0
0
0
0
0
0
0
0
0
0
1
0
0
1
0
0
0
0
0
0
0
0
0
0
0
-1
0
0
-1
0
0
0
0
0
-1
0
1
88.9%
5.6%
5.6%
Inclusion Error (An incorrect association is included in the response.)

Correct
Omission Error (A necessary association is missing in the response.)
Of the indicated pair-wise associations are correct (0: 32/36)
Inclusion Errors (-1: 2/36)
Omission Errors ( 1: 2/36)
Figure 4. Error Matrix.

In order to deploy grid-based assessment across TCP-IP networks, we are currently using
Java to implement scoring and reporting procedures. At the moment, we have an alpha level
version of the presentation and analysis modules in testing. At ICOTS-6, we hope to also have
alpha level versions of a task editor available for demonstration. Our goal is to make this tool
available to interested faculty at Ball State University sometime in 2003 for trials in
undergraduate mathematics and statistics courses.
DISCUSSION
Our interest in grid-based assessment was first motivated by Johnstone's (1987, 1979)
work with paper-and-pencil grids in university chemistry courses. What we have added to his
work are powerful response analysis and feedback models and computer-based implementations
in the context of teaching mathematics and statistics. We are also conducting human-factors
studies on the reliability and validity of measurement in grid-based assessment. From preliminary
studies, it is apparent that these measures are likely to depend heavily on the quality of the
individual tasks used in any given assessment. Since this is also true of conventional closed-form
testing, we hope to identify practices that contribute to the quality of conventional tasks that are
transferable to grid-assessment tasks. We believe that a critical factor in the development of gridbased assessment will be the creation and testing of mathematical and statistical assessments by
independent scholars. ICOTS-6 participants interested in participating in our research in this
manner are invited to contact David Thomas via email at [email protected]. Full contact
information is available at the URL http://www.cs.bsu.edu/homepages/dathomas/.
REFERENCES
Johnstone, A.H. (1987). Methods of assessment using grids. Talk given at the Science-Math
Education Centre at Curtin University of Technology, Perth, Western Australia.
Johnstone, A.H., & Mughol, A. (1979). Testing for understanding. School Science Review, 61
(214), 147-150.

On-Line Assessment of Higher-Order Thinking Skills: A Java-Based Extension To Closed-Form Testing

Uploaded by

Copyright:

Available Formats

On-Line Assessment of Higher-Order Thinking Skills: A Java-Based Extension To Closed-Form Testing

Uploaded by

Document Information

Original Description:

Original Title

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

On-Line Assessment of Higher-Order Thinking Skills: A Java-Based Extension To Closed-Form Testing

Uploaded by

Copyright:

Available Formats

ICOTS6, 2002: Thomas et al.