Inequity and Excellence in Academic Performance Ev - Asneku
Inequity and Excellence in Academic Performance Ev - Asneku
Inequity and Excellence in Academic Performance Ev - Asneku
net/publication/324204066
CITATIONS READS
26 326
5 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Theresa Dicke on 17 September 2019.
Abstract
Educational policy makers traditionally perceived there to be a trade-off between
educational excellence and educational equity. With the rise of cross-national
comparative datasets, however, research has begun to suggest such a trade-off does not
exist and indeed, higher variance in achievement and more segregated school systems
may be associated with lower performance. However, such research has tended to focus
on between nation analysis for which important covariates are not controlled (e.g.,
response set differences, latent cultural differences, etc.). Likewise, relatively little
consideration has been given to whether a trade-off may exist for high or low
performing students. Using five cycles of the PISA database, the current research
explores within country trajectories in achievement and inequality measures to test the
hypothesis of an excellence/equity trade-off in academic performance. Rejecting the
trade-off hypothesis, we find a robust negative relationship between performance and
inequality which is of statistical and practical significance. Detailed analysis of
countries with large changes in average achievement from 2000 to 2012 suggest a focus
on low and average performers may be critical to successful policy interventions within
a given country.
Introduction
educational outcomes with ensuring equity both in terms of equal opportunity and in
minimising excessive variation in those outcomes. This has been an ongoing concern
standardized tests (Checchi, 2006). Our paper is primarily concerned with issues
excellence) and the degree of variation in performance within a nation (our measure
test scores will be negatively associated with average educational achievement. In this
way we seek to directly challenge views that countries educational policies must
make implicit trade-offs between educational excellence and equality. To test this
research we focus on a) changes that occur within countries over time; and b) on
where the changes occur in the academic achievement distribution. In the following
EXCELLENCE DOES NOT REQUIRE INEQUALITY 2
sections we first position our research within the broader domain of educational
about this debate and the limitation with the existing evidence base that we seek to
overcome.
Lubienski, 2017), sociology (Jerrim et al., 2016), and psychology (e.g., Parker et al.,
2017). Given such broad concern, it is perhaps not surprising that this domain space
Burger, 2016), and social class, ethnicity, and gender inequality in educational
attainment and opportunities (see Goldthorpe, 2007 for a review). Not only do these
research projects differ in outcomes of interest (e.g., adult income, years of education
obtained, type of education obtained), but also in what purpose they perceive
education playing in the socio-political and economic context (e.g., as human capital
methodology they pursue (e.g., a focus on correlations between parent and child years
critical that we are clear in the definition of the factors we consider here and our
purpose.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 3
prerequisite for social integration”. Note that Pfeffer’s definition implies an aim of
educating students at least up to the point that they reach the threshold required for
social integration. This reflects a socio-political focus that education should develop
citizens (see Walzer, 1983) and that this need is of increasing significance and
(Nussbaum, 1998). Its difficulty comes in defining where this threshold is, whether it
international context. It is worth noting for example that Pfeffer does not define a
measurement and in definition. This approach is promoted by the OECD, who state
tests has an association with increased human capital and productivity as measured by
2010). This is not to deny the other roles that education plays, rather, this is an initial
step that is broadly inclusive of social science foci, does not require the consideration
focused on the relative chances of individuals within a given social class, ethnicity, or
for example, Maximally Maintained Inequality (Raftery, & Hout, 1993), Effectively
Maintained Inequality (Lucas, 2001), Rational Action Theory (Breen & Goldthorpe,
generally used as a critical control or causal mechanism in these theories, where for
concerns in academic ability (e.g., Hung, 2009). It is within these definitions that we
variance as has frequently been implied by educational theory and policy (see Pfeffer,
produce the highest average performance if schools can tailor offerings to different
levels of the underlying talent distribution of the student population (see Hoxby,
2003; Van de Werfhorst & Mijs, 2010 for a review; see Walberg, 2000 for an applied
introduction). Checchi (2006, see also Hoxby, 2003) provides a detailed treatment of
this argument but, put simply, it stipulates that, in the absence of government
interference, families will choose a level and type of education for their children that
will maximise the child’s achievement and, should this occur for most children,
maximise the achievement of the nation as a whole (Friedman, 2002; Hoxby, 2003).
At the core of this idea is that differentiated, stratified, decentralized, and/or private or
privatised education (see Bol et al., 2013; Kerckhoff, 1995; Parker et al., 2016 for a
review) provides a context that prepares children with different underlying talent with
EXCELLENCE DOES NOT REQUIRE INEQUALITY 5
appropriate skills. This may mean less talented children are provided with educational
content specifically focused on vocational skills (see Brunello & Checchi, 2007 for an
overview). For talented children, no longer hampered by the need for teachers to limit
the scope and speed of content for the benefit of less talented children, increased
education system variance will maximise their learning gains (see Van de Werfhorst
Under this model, increased academic excellence for a country will tend to be
associated with greater variance in achievement than equal systems due to selection
effects, signalling, and different educational content (Jakubowski, Patrinos, Porta, &
Wisniewski, 2010; Parker et al., 2016; Pfeffer, 2015). Thus there is a potential
Gans & King, 2014). According to the trade-off position, excellence (i.e., high
performing) comes at the cost of variability in results. But does the empirical
concern is the central tenet that an excellent educational system is, by necessity,
counter to a system with limited variability. State policymakers will thus need to
balance the trade-off between these competing goals. However, this trade-off is
thought to have several parts. First educational differentiation or school choice means
that different children receive different levels or types of education. Second, it may be
that increased variance occurs due to mechanisms unrelated to government policy or,
may come about due to wider social stratification be it by race, ethnicity, or social
EXCELLENCE DOES NOT REQUIRE INEQUALITY 6
class (e.g., Rowe & Lubienski, 2017). Third, there may be barriers that prevent
children from disadvantaged backgrounds gaining access to the type of education best
suited to their underlying talent. Indeed, due to access to economic or other resources,
risk adversity, or poor decision-making, parents might choose a type of education that
is inappropriate for the child and thus require policies that provide such children with
educational chances more in keeping with their ability (Friedman, 2002; Gans &
King, 2014). Often suggestions for policy interventions indicate that, apart from
ensuring talented children are not misplaced, no checks should be placed on the
as they increase the options available to parents and improve overall performance.
Government intervention, on the other hand, should focus only on reducing risks of
excellence and equity argument is elegant. Yet it has been increasingly scrutinized by
empirical evidence derived largely from studies using large-scale international student
assessments (see Van de Werfhorst & Mijs, 2010 for a review). Anecdotally, criticism
of the implied excellence/equity trade-off comes from the observation that high
performing countries like Finland appear to combine both high levels of equality
(including both low barriers to entry and relatively undifferentiated education) with
empirical results that suggests the major sources of variance and stratification led to
educational systems with high academic ability variance may, in reality, have poorer
average performance. In relation to the former, Brunello and Checchi (2007) found
attainment and labor market outcomes, and that the effects are larger for earlier
tracking. Jerrim et al. (2015) found that private schooling in Australia, the UK, and
the US was associated with advantages in both education and labor market outcomes.
Finally, Parker et al. (2016) found that ability stratification was associated with lower
achievement.
found that early tracking increased educational inequality and some evidence that it
was associated with lower mean performance. Micklewright and Schnepf (2007)
showed that the distance between the 95th and 5th percentile in achievement in a
country and their median performance was negatively correlated. Likewise, Checchi
et al. (2014) found no or negative relationships between various forms of variance and
of years of education and the average years of education within a country for rich
countries. Pfeffer (2015) combined both research traditions to show that there is no
Almost all of the research to date, however, has focused on between (cross-
important to note that changes in stratification may or may not occur evenly across
the achievement distribution, with changes in variance at the top or bottom half
potentially have different implications. Micklewright and Schnepf (2007) suggest, for
distribution to the tails) at the bottom end of the distribution may be most important.
policy at such students (Breakspear, 2012). Alternatively, Ryan (2013), focusing only
on Australia, suggests declines in the top half of the distribution account for that
country’s decline in math performance. This suggests that reduction in polarity at the
top end of the distribution (i.e., the highest performers becoming more similar to the
Current Research
The current research makes use of over a decade of the Programme for
between within country inequality and average achievement. We also use the multiple
rounds to consider how within country changes in inequality and achievement are
H1: Trends in inequality from 2000 to 2012 will be zero or negatively related to
H2: Changes in inequality from one PISA round to the next will be zero or negatively
achievement.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 9
H3: When large changes in inequality occur, changes in the top or bottom of the
average achievement.
Measures of Inequality
the literature. These include measures focused on how children of different levels of
ability are sorted into schools such as the between-school achievement variance or
2015). These measures provide an index of the degree to which a country’s education
schools and thus incorporates both formal differentiation (e.g., tracking) and informal
(e.g., social segregation). Other measures focus on the degree of variance in academic
performance between children within the same country. These include absolute or
relative (i.e., scale invariant) measures (see Handcock & Morris, 1999 for a review).
measure of the amount of between-school ability stratification (ICC; see Parker et al.,
2016); b) the distance between the 95th and the 5th percentile in achievement as a
and c) a constructed Gini index of achievement and, where possible, relative polarity
There are few criteria for what indicates large or small variation in these
measures. And this is particularly the case in the context in which we use them, where
we rely on change over time. In the absence of criteria then, we utilise extensive
sensitivity analyses using multiple measure, across multiple academic domains, and
EXCELLENCE DOES NOT REQUIRE INEQUALITY 10
the use of multiple statistical methods. Thus, our focus is on results that show
Method
Participants
All analyses were done at the country level using participant level indicators
of math, science, and reading achievement from all five PISA rounds. We focus upon
OECD countries (based on membership as of 2000) with the exception of Mexico and
random sample of 15 year olds selected from within each school (OECD, 2004). A set
of weights is provided such that the sample is representative of the target population.
In total, participants came from 27 countries with a total sample of 1,026,173 for
analysis related to reading achievement and 957,735 for analysis related to math and
science achievement. The reason for the difference in participant numbers is that all
participants received the reading test in PISA 2000 but a sub-sample received either
the math or the science tests. In all other PISA rounds all participants received
Measures
1 The use of OECD countries excluding Mexico and Turkey is relatively common (e.g.,
Mickelwright & Schepf, 2007; Parker et al., 2016). The reason for this is a) considerable differences
between Mexico and Turkey and the rest of the OECD in GDP and human development indexes and b)
a large number of not at school youth in these countries at the age of interest leading to potential
systematic bias in estimates.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 11
curriculum. Answers from the achievement tests were summarized by the survey
organizers into a single score for each of the three domains using an item-response
model, the intuition being that true skill in each subject is unobserved and must be
estimated from the answers to the test (see OECD, 2004, for further details). Five
plausible values were generated for each pupil, estimating their true proficiency in
each subject. These scores were scaled by the survey organizers to have a mean of
500 points and standard deviation of 100 across OECD countries in the first PISA
round. Country average performance, Gini, and ICC estimates were all estimated with
each of the plausible values separately and then averaged to provide country specific
point estimates.
Gini Index. The Gini index was calculated separately for each academic
domain, country, and PISA round. As with all measures used in the present research
the Gini was calculated using the population weight via the reldist package in R
(Handcock & Aldrich, 2002). The index varies between zero (indicating a uniform
within a school resemble each other—and differ on average from those in other
the degree to which schools are homogenous in the academic achievement. ICCs were
multiplied these by 100 so that they varied from 0 to 100 (see Marks, 2006).
EXCELLENCE DOES NOT REQUIRE INEQUALITY 12
P95 – P5. The distance between the 95th and 5th percentile of achievement was
There was evidence of considerable change in achievement and all inequality indexes
Statistical Analysis
relationship between estimates derived for each country. We focus here on estimates
derived using a series of multilevel models with PISA round estimates nested within
country. As such all analyses are done at the country level, and no individual level
data is modelled in the analysis reported in the results. There are debates about how
appropriate the use of multilevel level models are in the context of country
comparisons. In particular, there are concerns that random effects models remain
common despite the fact that a) countries are rarely sampled randomly from a population
(or in our case include all, or almost all, countries in a relevant population; i.e., the
OECD) and b) country-specific estimates can be biased (due to shrinkage) when there are
few countries (e.g., Byran & Jenkins, 2015). As such we also test the robustness of the
results using country fixed effects models. Models were fit with random effects for
country with inequality and PISA cycle estimated as fixed effects. Detailed
For trajectory models multilevel growth curve models were estimated (H1). In
each case both the intercept (i.e., initial level at year 2000) and slope (i.e., slope of
linear interpolated trajectories from 2000 to 2012) were estimated as random effects
for country. Such models were run separately for academic achievement and
inequality measures. Country specific slope estimates were drawn from the resulting
parameter estimates. We also calculated the simple difference between PISA 2000
EXCELLENCE DOES NOT REQUIRE INEQUALITY 13
and 2012 achievement and inequality measures and looked at the relationship between
these. Growth curve models treat PISA cycle as an ordinal variable and thus
summarize the change across PISA cycles in relation to, for example, achievement as
a linear trend. The benefit of this is that it provides a simple summary measure that
reduces the influence of noise around this trajectory, thus reducing the impact of
outlier cycles (e.g., where a country experiences a notable increase in one but only
It is possible, however, that these results may be biased as they impose a linear
trajectory from PISA 2000 to PISA 2012. We aimed to account for this using change
score models (H2). In this case achievement at round k+1 was regressed on
difference score) and the change score of inequality from round k to k+1. The result
inequality over the same lag. Random effects for country were included.
occurred in the achievement distribution. Using the reldist package (Handcock &
Morris, 1999) we isolated changes in the achievement distribution from 2000 to 2012
approaches to this. First, we explored the relationship between relative polarity (RP,
i.e., degree of movement from the median to the tails of the distribution from one
PISA cycle to the next) and changes in achievement for all countries. Second, we
2000 to 2012 for a more detailed analysis. We use both RP measures as well as plots
changes. All RP indexes vary from -1 to 1, with negative values indicating decreased
polarity or a general movement of values toward the median. The median relative
explore the upper (URP) and lower (LRP) portions of the distribution.
Results
H1: Trajectory
were related to linear trends in inequality. For this we extracted country level trends
from a) a series of random intercepts and slopes models; b) a series of country fixed
effect models; and c) the simple difference between achievement and inequality
measures from PISA 2000 to PISA 2012 (hereafter simple). As can be seen from
Table 1, the relationship between the trend in achievement and the trend in inequality
was negative in all cases (including both Pearson and Spearman correlations). In
support of H1, countries that increased in achievement from 2000 to 2012 tended to
see a decline in inequality measures. Relationships were strongest for Gini and ICC
indexes, with correlations routinely around -.50 and often above -.70. The
relationships were more moderate for P95 – P5, and typically only significant for
science. The correlations were similar for all achievement domains, with Figure 1
derived from the multilevel models, showing the relationship between the linear
from 2000 to 2012. It is possible, however, that these results do not give an accurate
changes in inequality from one PISA wave to the next (Table 2). For all academic
domains, a change in the Gini index from one PISA round to the next was associated
metrics, a one-point increase in Gini (inequality) was associated with a 6 (for science)
to 10 (for math) point decline in achievement. Put another way, a one-point increase
in the Gini coefficient measure if inequality is associated with a 0.06 (science) and
0.10 (math) effect size decline in average performance. Significant relationships were
likewise found for reading and science for ICCs and for reading for P95 – P5. Effect
sizes were moderate for the Gini index and ICCs and small for P95 – P5.
polarity from one PISA wave to the next. In all cases the estimates were negative
suggesting that there was not a trade-off between excellence and equity in either the
higher performing or lower performing students (see Table 2). Supporting H3, the
effects for RPM and RPL were only significant in one case. Overall the relationships
were strongest for the upper half of the achievement distribution and significant or
marginally significant for all domains. This indicates that declines in achievement
maybe more strongly weighted toward increases in inequality in the upper portion of
associated with average performing students falling further behind the highest
performing students; such that the right tail of the distribution became increasingly
elongated (i.e., the highest performing students tend to be protected against declines
in achievement). The difference between RPL and RPU were, however, relatively
EXCELLENCE DOES NOT REQUIRE INEQUALITY 16
be beneficial.
inequality tended to occur for countries that experienced notable changes. Given
space constraints we focus here on Germany, Poland, Sweden, and Iceland as those
Germany and Poland were the only two countries to improve by over 20 achievement
points and decreased in Gini by over one-point for each domain from 2000 to 2012.
Sweden declined by almost 30 points in each domain and increased in Gini by well
over one point in both reading and science (and over half a point in math). Likewise,
achievement by over 20 points in math and reading (and over 17 points in science).
The results indicate significant changes in polarity for each focussed country
in at least two of the three achievement domains (see Table 3). Germany and Poland
declined in the upper portion of the distribution with Poland displaying most change
in the lower portion. However, for reading in Germany and reading and science in
Poland significant declines in polarity occurred in both LRP and URP. This shape
change resulted in fewer individuals in the lower and upper deciles than would have
been the case if changes in achievement from 2000 to 2012 were due to location
changes alone. Sweden and Iceland both significantly increased in RP. In both cases
changes were predominantly located in the upper portion of the distribution. What this
means is that, as Sweden and Iceland declined in average achievement, the most
talented students were partially protected. Thus there were frequently 20 to 30% more
students in the top decile than would be expected if achievement declines were
EXCELLENCE DOES NOT REQUIRE INEQUALITY 17
consistent across the whole distribution. Indeed, for science achievement in Iceland
there were approximately equal numbers of students in the top decile of the reference
distribution at both PISA 2000 and 2012, when there should have been only 60% as
Discussion
Consistent with growing evidence, our results suggest that inequality, indexed
achievement at the country level. Importantly, effect sizes were routinely of a similar
size for both relative Gini (variance) and ICC (stratification) indexes of inequality;
though relationships were smaller, but still negative and often significant, for absolute
within country changes in inequality and its association with within country changes
in average achievement. Not only were within country results consistent with
previous research in showing a lack of evidence for a trade off between excellence
and equity, the current results suggested that inequality maybe associated with
declines in performance.
Nordic countries have often been shown to be among the most equal in between
country studies (e.g., Parker et al., 2016, 2017), when considering within country
estimates Iceland and Sweden had some of the most evident declines in achievement
countries have been shown to be some of the most unequal due to early and extensive
and this has been associated with notable decreases in inequality. Taken together,
EXCELLENCE DOES NOT REQUIRE INEQUALITY 18
inequality – Nordic < Anglophone < Germanic – (see Dupriez & Dumay, 2006),
within-country analysis shows a shifting landscape where these monikers hold less
relevance. This could be taken to suggest that overall the inter-country landscape is
becoming more equal. There were notable increases in ICC PISA 2000 to PISA 2012
and (see supplementary material). Thus, given these average increases in ICCs, it may
be that the trend, for OECD countries at least, is toward greater inequality.
It may be that changes unrelated to direct educational policy are driving these
results. As such, we ran further sensitivity analysis on the country fixed effects
between academic excellence and equity controlling for trajectories across the same
disposable income, and in the percentage of GDP spent on social welfare. As Table 4
A major question that emerges from the current research is why there is so
the amount and type of education that a child should invest in is a decision not made
by the child themselves but rather by parents or guardians. Such parents may not
make decisions that lead to the best possible school placement (Friedman, 2002).
Widespread and systematic inefficiencies in child assignment could account for the
results noted here (see Pfeffer, 2015). Indeed, PISA data suggests misplacement
occurs across the socioeconomic ladder (Parker et al., 2017). For example Maaz et al.
(2008) note that in the Germanic system parents from well-off families often insure
EXCELLENCE DOES NOT REQUIRE INEQUALITY 19
that their children are located in university track systems even when teacher
recommendations are for lower track placements. Parker et al. (2017) note that
children of richer parents pay for poor placement with decreased academic self-
inaccurate school placement but pay in terms of more difficult pathways to university.
This would suggest the problem is not with the idea that a school system
should tailor offerings to different levels of the achievement distribution but rather
with its application in context. However, there may also be inherent problems that
suggest issues may continue to occur even with perfect placement. Indeed educational
psychology evidence points to a natural bias in the way young people form
expectations. Marsh (2006) argues that children in more selective schools have lower
academic self-concepts than they would have had they gone to more comprehensive
effects model [REM]; Marsh, 2006). This effect is larger in countries with more
tracking or higher ICCs (Salchegger, 2015). It is possible that this bias in self-
perceptions may account for some of the reason why more stratified systems do worse
Alternatively, peer composition and its potential negative effect on the self-
concept, and thus, their academic achievement (i.e., REM) of high performing
students provide one mechanism to explain the current results. In addition, non-linear
peer effects in learning quality likely provide equally compelling explanation of these
results for the low end of the achievement distribution. Put simply, high performing
students tend to lose very little from association with poorer performing students but
that poorer performing students gain considerable benefits in terms of motivation and
EXCELLENCE DOES NOT REQUIRE INEQUALITY 20
quality of peer interaction (Checchi, 2006; Hanushek, Kain, Markman, & Rivkin,
2003; Hanushek & Wößmann, 2005). From a policy perspective then, it may be that
each country needs to determine whether students across the achievement distribution
may actually benefits from more integrated classrooms; though always with an eye to
Changes in Inequality
average change in variance for all countries, but also the form of this change.
students and declines of average and low performing students. Taken as a whole,
there was evidence of an effective hollowing out of the middle of the achievement
distribution where there was increasingly polarization between the most talented
students and the rest. More in-depth analysis of countries that changed considerably
(i.e., 20 PISA points and 1-Gini point) provided a more nuanced perspective on this
issue.
Ringarp and Rothland (2010) note that Sweden has moved from one of the
most to one of the least centralized systems with increased school choice and
privatization in the last few decades. Iceland has long had a decentralized school
by policy in 2008 and the implication of decentralization likely increased after the
declines in PISA performance. In contrast, the ‘PISA shock’ of 2000 in Germany led
EXCELLENCE DOES NOT REQUIRE INEQUALITY 21
lower performers and immigrants (Breakspear, 2012). In Poland, there was a strong
2012). Our results suggest that for Germany increases mostly centred on the middle of
achievement distribution such that more of the mass of the achievement distribution
was located around the average. For Poland our results show the strong success of
their focus on the bottom of the achievement distribution. Taking all results together,
a hypothesis emerges that a countries’ educational policy that mainly serves talented
the lower and middle portions of the achievement distribution leads to improvement.
Overall there is a need for future research that focuses not just on changes in
inequality overall but where changes occur and what implications this has for how a
given country should determine its educational policy given its own unique context.
Measures of Inequality
It is worth noting that there were some modest differences in the results
however, it should be emphasised that there was a broad consistency. First, the
direction of the relationship between inequality and performance was always negative
regardless of the measure used or the model used to test the relationship. Second, each
measure of inequality was significantly negative for at least one achievement domain
in each model. Nevertheless there were differences. Primary among them was that the
relative measures of variance (Gini) and stratification (ICC) were similar in size and
routinely larger than the absolute measure (95th – 5th percentile). This may be due to
the relative measures having proportional scale invariance while the absolute
measures do not (Handcock & Morris, 1999). Given this property it maybe that the
EXCELLENCE DOES NOT REQUIRE INEQUALITY 22
relative measures are more clearly comparable across time and context than the
absolute measures.
Our research findings are consistent with a broader set of research (e.g.,
Checchi et al., 2014; Hanushek & Woesmann, 2005; Micklewright & Schnepf, 2007;
Van de Werfhorst & Mijs, 2010) that has questioned the value of educational policies,
at a state or nation level, that promote school differentiation and thus, there is a
private or privatized schooling, and tracking. All these policies promote stratification
by ability and as such do not appear to lead to higher average academic ability.
Indeed, as noted above countries such as Sweden and Iceland have increased
decentralization and school choice and have seen notable declines in performance,
while Germany has moved toward increased centralization and seen an increase in
performance. There are several consideration, however, that should be taken into
account when interpreting what our results suggest for policy in a given country.
performance, and it should be noted the achievement tests on which they are based are
low stakes. Speaking against this is modelling which implies improvements in PISA
scores are linked with real world outcomes such as economic growth (see Hanushek
along with the trade-offs between outcomes. For example, tracking maybe associated
with poorer average achievement, however, retention though the full program of study
EXCELLENCE DOES NOT REQUIRE INEQUALITY 23
is high suggesting that the cost of tracking in terms of average achievement may yet
Likewise it should be noted that policies and social change at other levels of
society may require an increase in decentralization and school choice, or at least make
such policies more appealing. As Friedman (2002) has noted school choice maybe
one of the only, or at least one of the most effective, means of reducing educational
children in very poor regions access to high quality schooling. Indeed, countries like
the US have seen exceptional increase in such segregation over the period of study in
this research (Owens, Reardon, & Jencks, 2016) and thus there is good opportunity to
evidence suggests that school choice in the context of residential segregation may
It should also be noted that while the multi-cohort evidence presented here is a
large step forward over previous cross-sectional evidence, the results should not be
taken as causal. In particular, the causal direction is unclear. For example, our results
show that a country can combine both excellence and equality with great success (see
also Simola, 2005). However, it is not certain that equality leads to better performance
or whether higher performance provides scope for countries to focus more closely on
equality may be a proxy for other factors. In particular, social structure not school
structure could drive these results; though previous research suggests this is unlikely
(Dupriez & Dumay, 2006). More likely inequality in funding between schools or even
regions within countries could account for these results (Owens, et al., 2016); likewise
they compare systems with different ages-of-first selection (Pfeffer, 2015). That is the
age at which students are streamed into different tracks. Thus, for example, PISA
considers students at age 15 and yet a number of OECD countries begin tracking
students at age 16 (Bol et al., 2013; Pfeffer, 2015). This has particular implications for
our interpretation of the results for Poland that, as part of the reform of the education
2010). Thus, a criticism of these results is that systems that do not track before age 15
are merely delaying the inevitable. Indeed, Jakubowski et al. (2010) notes that
track. There are several points to be made here. First, Pfeffer’s results are similar in
conclusion to ours despite focusing on the adult population. Namely, there appears to
terminology; see literature review) and equity even when both are measured in after
systems would still eventually emerge in late tracking systems (Jakubowski et al.,
2010), it is certainly not clear that this means that the achievement advantage that late
tracking has over early tracking will dissipate completely. Again the consistency
between our and Pfeffer’s results would seem to indicate that this fear is unfounded.
and theory needs to explain why this negative relationship exists and under what
social conditions it holds. Furthermore there is clearly a need for research, which
distribution effect achievement. Put simply research needs to determine whether and
when policies directed toward those in the bottom half of the distribution are most
effective. Likewise there is a need to consider what forces are behind changes in
variance over relatively short periods in some countries; noting that the current study
covers only a single decade. In particular, in depth analysis of countries that have
shown clear change are needed to unpack the various structures and polices that lead
linking PISA and PIAAC in a synthetic panel design may prove advantages.
as Pfeffer (2015) notes, large-scale assessment which covers the final year of
References
Bol, T., & Van de Werfhorst, H. G. (2013). Educational systems and the trade-off
between labor market allocation and equality of educational
opportunity. Comparative Education Review, 57, 285-308.
Breakspear, S. (2012). The Policy Impact of PISA: An Exploration of the Normative
Effects of International Benchmarking in School System Performance, OECD
Education Working Papers, No. 71, OECD Publishing.
Breen, R., & Goldthorpe, J. H. (1997). Explaining educational differentials: Towards
a formal rational action theory. Rationality and society, 9, 275-305.
Brunello, G., & Checchi, D. (2007). Does school tracking affect equality of
opportunity? New international evidence. Economic policy, 22, 782-861.
Bryan, M. L., & Jenkins, S. P. (2015). Multilevel modelling of country effects: a
cautionary tale. European Sociological Review, 32, 3-22.
Burger, K. (2016). Intergenerational transmission of education in Europe: Do more
comprehensive education systems reduce social gradients in student
achievement?. Research in Social Stratification and Mobility, 44, 54-67.
Checchi, D. (2006). The economics of education: Human capital, family background,
and inequality. Cambridge, UK: Cambridge University Press.
Checchi, D., van de Werfhorst, H., Braga, M., & Meschi, E. (2014). The Policy
Response: Education. In Salverda, Nolan et al. (Eds). Changing Inequalities
and Societal Impacts in Rich Countries: Analytical and Comparative
Perspectives (pp.294-327) Oxford, UK: Oxford University Press.
Dupriez, V., & Dumay, X. (2006). Inequalities in school systems: effect of school
structure or of society structure?. Comparative education, 42, 243-260.
Friedman, M. (2002). Capitalism and freedom. Chicago, University of Chicago press.
Friedman, M. & Friedman, R. (1980). Free to choose. New York, N.Y.: Harcourt
Gans, J. & King, S. (2014). Finishing the Job: Real world policy solutions in health,
education, and transport. Melbourne, Australia: Melbourne University Press.
Goldthorpe, J. H. (2007) On Sociology, 2nd ed. Stanford, CA: Stanford University
Press.
Handcock, M., & Aldrich, E. M. (2002). Applying relative distribution methods in
R. Center for Statistics and Social Sciences. University of Washington.
Handock, M.S. & Morris, M. (1999). Relative distribution methods in the social
sciences. New York, NY: Springer.
Hanushek, E. A., Kain, J. F., Markman, J. M., & Rivkin, S. G. (2003). Does peer
ability affect student achievement?. Journal of applied econometrics, 18, 527-
544.
Hanushek, E.A. & Wößmann, L. (2005). Does educational tracking affect
performance and inequity? Differences-in-differences evidence across
countries. The Economics Journal, 116, 63-76.
Hanushek, E. A., & Woessmann, L. (2010). The High Cost of Low Educational
Performance: The Long-Run Economic Impact of Improving PISA Outcomes.
OECD Publishing. 2, rue Andre Pascal, F-75775 Paris Cedex 16, France.
Hoxby, C. M. (2003). School choice and school productivity. Could school choice be
a tide that lifts all boats?. In C.M. Hoxby (Ed.) The economics of school
choice (pp. 287-342). University of Chicago Press.
Huang, M-H. (2009). Classroom homogeneity and the distribution of student math
performance: a country-level fixed-effects analysis. Social Science Reseearch,
38, 781–91.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 27
Jakubowski, M., Patrinos, H. A., Porta, E. E., & Wisniewski, J. (2010). The impact of
the 1999 education reform in Poland. Washington, DC: World Bank.
Jerrim, J., Chmielewski, A. K., & Parker, P. (2015). Socioeconomic inequality in
access to high-status colleges: A cross-country comparison. Research in Social
Stratification and Mobility, 42, 20-32.
Kerckhoff, A. C. (1995). Institutional arrangements and stratification processes in
industrial societies. Annual review of sociology, 21(1), 323-347.
Lucas, S. R. (2001). Effectively maintained inequality: Education transitions, track
mobility, and social background effects. American journal of sociology, 106,
1642-1690.
Maaz, K., Trautwein, U., Lüdtke, O., & Baumert, J. (2008). Educational transitions
and differential learning environments: How explicit between‐school tracking
contributes to social inequality in educational outcomes. Child Development
Perspectives, 2, 99-106.
Marks, G. N. (2006). Are between-and within-school differences in student
performance largely due to socio-economic background? Evidence from 30
countries. Educational Research, 48, 21-4.
Marks, G. N. (2013). Education, social background and cognitive ability: The decline
of the social. Routledge.
Marsh, H. W. (2006). Self-concept theory, measurement and research into practice:
The role of self-concept in educational psychology. Vernon-Wall Lecture:
British Psychological Society.
Mickelwright, J. & Schepf, S. (2007). Inequalities in industrialised countries. In S.P.
Jenkins & J. Micklewright (Eds). Inequality and poverty re-examined (pp. 129-
145). Oxford, UK: Oxford University Press.
Ministry of Education, Science and Culture (2014). Review of Policies to Improve the
Effectiveness of Resource Use in Schools Country Background Report Iceland.
Nussbaum, M. C. (1998). Cultivating humanity. Harvard University Press.
Organisation for Economic Co-operation and Development [OECD] (2004). PISA
2003 Technical Report. Paris: Organisation for Economic Co-operation and
Development.
Owens, A., Reardon, S. F., & Jencks, C. (2016). Income Segregation Between
Schools and School Districts. American Educational Research Journal,
0002831216652722.
Parker, P. D., Jerrim, J., Schoon, I., & Marsh, H. W. (2016). A Multination Study of
Socioeconomic Inequality in Expectations for Progression to Higher Education
The Role of Between-School Tracking and Ability Stratification. American
Educational Research Journal, Online First.
Parker, P.D., Marsh, H.W., Guo, J. Anders, J., Shure, N. & Dicke, T. (2017). An
Information Distortion Model of Social Class Differences in Math Self-concept,
Intrinsic Value and Utility Value. Journal of Educational Psychology. Online
First.
Pfeffer, F. T. (2015). Equality and quality in education. A comparative study of 19
countries. Social Science Research, 51, 350-368.
Raftery, A. E., & Hout, M. (1993). Maximally maintained inequality: Expansion,
reform, and opportunity in Irish education, 1921-75. Sociology of education, 1,
41-62.
Ringarp, J., & Rothland, M. (2010). Is the grass always greener? The effect of the
PISA results on education debates in Sweden and Germany. European
Educational Research Journal, 9, 422-430.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 28
Rowe, E. E., & Lubienski, C. (2017). Shopping for schools or shopping for peers:
public schools and catchment area segregation. Journal of Education Policy, 32,
340-356.
Ryan, C. (2013). What is behind the decline in student achievement in Australia?
Economics of Education Review,37, 226-339.
Salchegger, S. (2015). Selective School Systems and Academic Self-Concept: How
Explicit and Implicit School-Level Tracking Relate to the Big-Fish–Little-Pond
Effect Across Cultures. Journal of Educational Psychology, online first.
Saporito, S. (2003). Private choices, public consequences: Magnet school choice and
segregation by race and poverty. Social problems, 50(2), 181-203.
Simola, H. (2005). The Finnish miracle of PISA: Historical and sociological remarks
on teaching and teacher education. Comparative education, 41, 455-470.
Thomas, V., Wang, Y., & Fan, X. (2001). Measuring educational inequality: Gini
coefficents of education. World Bank: Policy Research Paper 2525.
Van de Werfhorst, H. G., & Mijs, J. J. (2010). Achievement inequality and the
institutional structure of educational systems: A comparative
perspective. Annual review of sociology, 36, 407-428.
Walberg, H. J. (2000). Market theory of school choice. Education Week, 19, 46-49.
Walzer, M. (1984). Spheres of justice: A defense of pluralism and equality. Basic
Books.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 29
Table 1
Correlation between Achievement Trajectory and Inequality Trajectory
Random Effect Country Fixed Effect Simple
(Pearson/spearman) (Pearson/spearman) (Pearson/spearman)
Math
Gini -.722***/-.727*** -.681***/-.756*** -.689***/-.725***
ICC -.536**/-.519** -.571**/-.563** -.503**/-.457*
P95 – P5 -.347/-.374 -.313/-.369 -.323/-.439*
Reading
Gini -.613***/-.523** -.567**/-.524** -.542**/-.489***
ICC -.495**/-.485** -.558**/-.552** -.670***/-.694***
P95 – P5 -.255/-.180 -.230/-.227 -.188/-.149
Science
Gini -.704***/-.727*** -.695***/-.662*** -.706***/-767***
ICC -.440*/-.318 -.500**/-.414* -.443*/-.281
P95 – P5 -.371*/-.464* -.395*/-.392* -.391*/-.518**
Notes. * p < .05; ** p < .01; *** p < .001. Random Effect = Correlation of slope with achievement
slope from a multilevel growth curve model. Country Fixed Effect = Correlation of slope with
achievement slope from a country fixed effect model. Simple = correlation of difference from PISA
2000 to 2012 in achievement and inequality measures.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 30
Table 2
Lagged Results
Est SE p
Math
Gini -10.556 1.908 -.369 ***
ICC -0.342 0.357 -.245
P95 – P5 -.138 .075 -.104 ^
RPM -.454 .314 -.151
RPL -.178 .275 -.060
RPU -.562 .284 -.223 ^
Reading
Gini -7.917 1.320 -.413 ***
ICC -0.531 0.178 -.458 **
P95 – P5 -.138 .059 -.165 *
RPM -.454 .314 -.151
RPL -.614 .242 -.250 *
RPU -.419 .252 -.172 ^
Science
Gini -9.252 1.422 -.380 ***
ICC -0.499 0.183 -.380 **
P95 – P5 -.123 .072 -.127 ^
RPM -.706 .292 -.295 *
RPL -.234 .278 -.100
RPU -.824 .242 -.397 ***
Notes. ^p < .10 * p < .05; ** p < .01; *** p < .001. = Estimates taken from a model in which
achievement and stratification are standardized around the grand mean. Gini = Gini estimates of
Achievement. ICC = Intra-class correlation of achievement, P95 – P5 = distance in achievement
between the 95th and 5th percentile. RPM = Relative Polarity Median of achievement RPL = Relative
Polarity Lower of achievement, RPU = Relative Polarity Upper of achievement.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 31
Table 3
Relative Changes in Polarity
Germany Poland Sweden Iceland
Est p Est p Est p Est p
Math
RPM -0.014 -0.059 *** 0.017 0.072 ***
RPL -0.036 -0.117 *** -0.003 0.039
RPU 0.009 -0.002 0.037 0.107 ***
Read
RPM -0.093 *** -0.085 *** 0.075 *** 0.021
RPL -0.087 *** -0.115 *** 0.063 ** 0.017
RPU -0.099 *** -0.054 * 0.086 *** 0.026
Science
RPM -0.066 *** -0.092 *** 0.026 * 0.086 ***
RPL -0.030 -0.095 *** -0.004 0.061 *
RPU -0.101 *** -0.088** 0.055 * 0.111 ***
Notes. * p < .05; ** p < .01; *** p < .001. RPM = Relative Polarity Median of achievement, RPL =
Relative Polarity Lower of achievement, RPU = Relative Polarity Upper of achievement.
Table 4
Country fixed effects controlling for country level covariates
No controls Social Welfare GDP Disposable Income
(Pearson/spearman) (Pearson/spearman) (Pearson/spearman) (Pearson/spearman)1
Math
Gini -.681***/-.756*** -.680***/-.752*** -.694***/-.754*** -.710***/.775***
ICC -.571**/-.563** -.574**/-.557** -.600***/-.562** -.676***/-.656***
P95 – P5 -.313/-.369 -.313/-.381* -.324/-.379 -.368/-.445*
Reading
Gini -.567**/-.524** -.574**/-.536** -.558**/-.551** -.609***/-.519**
ICC -.558**/-.552** -.560**/-.549** -.586**/-.540** -.591**/-.544**
P95 – P5 -.210/-.264 -.210/-.271 -.226/-.281 -.327/-.311
Science
Gini -.695***/-.662*** -.696***/-.654*** -.753***/-.688*** -.785***/-.739***
ICC -.500**/-.414* -.503**/-.397* -.536**/-.427* -.553**/-.453**
P95 – P5 -.395*/-.392* -.394*/-.379 -.496**/-.427* -.558**/-.539**
Notes. * p < .05; ** p < .01; *** p < .001. All covariates were taken from the OECD
(https://data.oecd.org/). 1 These results exclude Luxemburg for whom disposable income data was not
available.
EXCELLENCE DOES NOT REQUIRE INEQUALITY 32