A new twist in the educational tracking debate

Economics of Education Review 22 (2003) 307–315
www.elsevier.com/locate/econedurev
A new twist in the educational tracking debate
Ron Zimmer ∗
RAND Corporation, 1700 Main Street, Santa Monica, CA 90407-2138, USA
Received 5 December 2000; accepted 18 June 2002
Abstract
Recently, the practice of tracking has been receiving more attention by both educators and researchers and some
have questioned the policy merit. One of the strongest arguments against tracking is that it creates homogenous classes
according to ability and, therefore, reduces the positive spillover effect referred to as a peer effect. While peer effects
have been found to be an important input into the production of education no study has specifically examined whether
these effects are more or less prevalent in classes where tracking occurs. Utilizing individual student level data, this
current research examines whether the peer effect occurs in cases in which tracking is present. The results suggest that
the use of tracking diminishes the impact peers have on student achievement for low- and average-ability students
while the peer effect is unaffected by tracking for high-ability students.
 2003 Elsevier Science Ltd. All rights reserved.
JEL classification: I21
Keywords: Educational economics
1. Introduction
Recently, the practice of tracking has been receiving
more attention by both educators and researchers, and
some have questioned the policy’s merit. In the education
arena, those who advocate tracking argue that all
students, regardless of ability, would learn more in a
tracked class relative to a nontracked class (Hallinan,
1994). In tracked classes the teacher can tailor the curriculum
to the ability level of the students, thus creating
the optimal level of educational gains for all students.
However, opponents argue against tracking for three primary
reasons. First, tracking leads to a different set of
resources being allocated to high-tracked versus low-
∗ Corresponding author. Tel.: +1 310 393 0411; fax: +1 310
451 7059.
E-mail address: rzimmer@rand.org (R. Zimmer).
0272-7757/03/$ - see front matter  2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0272-7757(02)00055-9
tracked classes (Oakes, 1990).1 Second, tracking breeds
social inequities as minority and low-income groups are
over-represented in low-track and under-represented in
high-track classes (Braddock & Dawkins, 1993; Gorman,
1987; Oakes, 1985, 1990). Third, tracking creates homogenous
classes according to ability, therefore reducing
the positive spillover effect, referred to as a peer effect
(Betts & Shkolnik, 2000). While further research is
needed to address the first two issues, this current
research examines the effect tracking has on the peer
effect. That is, does tracking diminish the positive spillover
effect from high-ability students to low-ability students?
In addition, the effect from tracking on student
achievement is examined while holding the peer level
constant.
The literature increasingly suggests there are differen-
1 However, Betts and Shkolnik (2000) found that neither
class size nor teacher characteristics vary much whether the
student is placed in a tracked or an untracked class.
308 R. Zimmer / Economics of Education Review 22 (2003) 307–315
tial effects from tracking for students of different abilities
(Argys, Rees & Brewer, 1996; Hoffer, 1992; Kerckhoff,
1986)2. However, this literature has failed to
determine whether this effect is from curriculum and
resource effects or a differential peer effect from the
organizational structure of grouping students. In the past,
researchers have found strong support for a peer effect
in the production of education (Hanushek, Kain, Markman,
& Rivkin, 2001; Henderson, Mieskowski, & Savageau,
1978; Summers & Wolfe, 1977; Zimmer &
Toma, 2000; Hoxby, 2000). However, no study has
examined whether these peer effects are more or less
prevalent in classes where tracking occurs and what
effect they are having on student achievement in tracked
classes3. Zimmer and Toma (2000) interacted the variance
of ability in a classroom with the mean ability
(peer-effect variable) of students and found that greater
variance within a classroom actually reduces the peer
effect. Therefore, it could be the case that peer effects
are less prevalent in schools that are not tracked as
opposed to schools that are. That is, students of different
ability levels who are mixed together in non-tracked
schools would not be able to interact effectively to create
the peer effect.
This current research adds to the literature by examining
the impact tracking has on the peer effect by including
an interaction term. The results suggest that the use
of tracking diminishes the impact peers have on student
achievement for low and average-ability students while
the peer effect is unaffected by tracking for high-ability
students. A secondary result suggests that the institutional
process of tracking, when controlling for the
peer level, has no effect on high-ability students,
2 Kerchoff (1986) examined grouped versus non-grouped
students in British schools and found that high-ability students
learnt more than non-grouped students while low-grouped students
learnt less. Hoffer (1992) examined grouped versus nongrouped
students using the Longitudinal Study of America
(LSAY) and found that tracking had no effect for averageability
students, weak positive effects for high-ability students,
and strong negative effects for low-ability students. Argys et al.
(1996), using the National Longitudinal Study of 1988 (NELS),
suggest that tracking creates educational gains for high-ability
students, whereas low-ability students experience educational
losses. In contrast, Betts and Shkolnik (2000) suggest only weak
differential effects between ability groups, with tracking having
no effect on low-ability students but a small positive effect for
high-ability students and a small negative effect on averageability
students. In a more recent study, Figlio and Page (2002),
using the NELS data, support the finding from Betts and Shkolnik
as they control for the endogeneity of track placement and
finds no evidence that low-ability students are hurt by tracking.
3 White and Kane (1995) suggest that one of the main shortcomings
of the existing literature is that it does “not separate
the effects of course content and instruction from the effects of
homogenous grouping” (p. viii).
whereas, surprisingly, the institutional practice of tracking,
when controlling for the peer level, has a positive
effect on low-and average-ability students. Only if the
reduction in the peer effect is combined with the effect
from tracking does the positive effect from low-ability
students wash out. In other words, the loss of exposure
to more able students offsets the potential of tracking to
improve educational performance for the low-ability students.
2. Data
To examine the impact tracking has on the peer effect,
and ultimately, the impact peers and tracking have on
student achievement, a data set with characteristics of
the individual students is necessary, including whether
or not the student was in a tracked classroom. One such
data set is from the second study (SIMS) from International
Association for the Evaluation of Educational
Achievement (IEA)4. The IEA, through carefully constructed
surveys, collects family, school, classroom, and
peer characteristics of individual students for the purpose
of cross-country comparison, including the US5. This
data set includes characteristics of the students’ families
and teachers, the organizational structure of the schools,
including whether or not the classes are tracked, and preand
post-year mathematics test scores for individual students,
which allow for a value-added estimate6. In the
data, tracking is designated as a dichotomous variable (1
if there is tracking in the classroom and 0 if there is not)
and is interacted with a variable representing the ability
of peers (the mean test score of classmates). If the
resulting coefficient is negative for the interaction term,
then it can be concluded that tracking diminishes the peer
4 The most recent IEA data set was not used because it did
not have the pre-school-year test score to do a value-added
model.
5 In 1981, the IEA carefully selected a sample of schools
and administered a broad array of questionnaires answered by
students, teachers and administrators (Robitalle & Garden,
1989). The schools selected for the survey were designed to
reflect the socioeconomic characteristics of each country. The
students in the analysis were in the eighth grade.. The survey
asked questions in regard to the student’s family, schools, teachers,
and peers within the classrooms (Robitalle & Garden, 1989)
The IEA went to great lengths to ensure the quality of the data
through rigorous quality control steps including creating a manual
for inputting the data and auditing them once they were collected.
6 The mathematics tests, known as SIMS (Second International
Mathematics Study), were developed through the input
of mathematics experts and measurement specialists and the
tests were to reflect the curriculum of each country. The tests
went through a review process that included several pilot tests
before being implemented.
R. Zimmer / Economics of Education Review 22 (2003) 307–315 309
effect, whereas a positive coefficient suggests that the
peer effect is enhanced. Using the US public school portion
of the data, this study examines the impact of tracking
on the peer effect and student achievement for high-,
average-, and low-ability students using a standard education
production model. Together, the information provided
by the surveys and the test scores renders a robustness
to the results.
3. Model
To examine the effect from tracking, the peer effect,
and the interaction of tracking with the peer effect, I followed
a standard education production function model
employed by Betts and Shkolnik (2000). The model
examines the impact of various inputs, including family
and school resources, the student’s peers, and student’s
individual characteristics, including a proxy for ability7.
The value-added model is displayed as follows8:
At,j
f(Ft,j,St,j,Pt,j,At-1). (1)
By using the model represented above, the valueadded
impacts of various inputs are measured as student
achievement for the period t–1 to t9. In this formulation,
the effects of all prior inputs are captured in achievement
from the previous period, or At-1
10. Formally, output is
measured by individual student j’s educational achievement
(At) in the current period as a function of inputs
that include student j’s family characteristics (Ft) over
period At-1 to At, school inputs (St) for student j over
7 Conceptually, the ideal model (as displayed below)
includes family inputs (F), school inputs (S), environmental
inputs of peers (P), and the innate ability of the student (I) in
period t:At,j = f(Ft,j,St,j,Pt,j,It,j).However, it is difficult to gain an
accurate proxy for the innate ability of the students (Hanushek,
1979) and the lack of an ability variable is generally assumed to
bias upward the effects of family background on achievement.
Therefore, researchers often use a value-added model to mitigate
the negative effect of not including such a variable. By
estimating the value-added model, the biases are minimized
because only the growth effect of innate ability is omitted.
8 Argys et al. suggests that “if there are unobservable student
or school characteristics that affect both achievement and track
placement, than any association between achievement and
tracking may simply be due to these characteristics” (1996,
p.624). To control for the selection, Argys et al. model the process
through which students are assigned to a particular track
and then include these selectivity corrections in the main
achievement model. Unfortunately, I was unable to find a suitable
instrument in the data set for this present analysis.
9 Betts and Shkolnik (2000) argue that it is critical to control
for the student’s initial achievement if accurate estimates are to
be obtained.
10 In this case, achievement in all other periods, such as t–1,
is a function of the cumulative inputs in t–1.
period At-1 to At, peer influence (Pt) of student j over
period At-1 to At, and student j’s achievement in the previous
period (At-1)11.
To apply this model I measure achievement, both in
the current and previous period, as the number of questions
answered correctly by student j on a mathematics
test with a possible range of 0 to 40. The achievement
level (At) is the students’ end-of-the-year test score and
is labeled POSTTEST, while the beginning-of-the-year
test (At-1) is labeled PRETEST12. Other inputs include a
vector of family inputs (F) and a vector of school inputs
(S), including the characteristics of the student’s teacher
and school13. (A complete list of variables with their
summary statistics is included in Table A1 in the
appendix.)
Two variables of primary importance in this current
research are the tracking and peer variables14. The tracking
variable is simply a dummy variable represented by
a 1 when tracking is used in the classroom and a 0 when
it is not. The peer variable is defined as the mean test
score of students in a classroom and is represented by
the mean of the scores at the beginning of the school year
of all students in the observed student’s classroom15. As
consistent with the literature, I also square the mean
score to capture nonlinear as well as linear effects on
fellow students16. This peer variable is also interacted
with the tracking variable. If tracking is utilized within
the classroom, then tracking is equal to 1 and the coefficient
of the interaction term represents the impact tracking
has on the peer effect. Because the peer level is also
included, the impact of tracking on the peer effect is
measured while holding the peer level constant. A negative
and significant coefficient suggests that tracking
diminishes the peer effect, while a positive and significant
coefficient suggests that tracking enhances the peer
11 For further information on the benefits of using the valueadded
model, see Boardman and Murnane (1979).
12 It should be noted that students that did not appear in both
the pre- and post-test scores were deleted from our sample.
13 Argys et al. (1996) suggest that it is critical to control
teacher and school characteristics if low-ability tracks and highability
tracks are systematically assigned different resources.
14 It should be noted that the model suffers from endogenity
problems. Part of the purpose of tracking is to put students in
more homogenous classes. Therefore, the peers of a student are
partially a function of whether a student is in a tracked class.
No suitable instrument could be found to perform an IV model
to correct for this problem.
15 This is one of the definitions used by Zimmer and Toma
(2000) for peers.
16 Henderson et. al. (1978) along with Zimmer and Toma
(2000) squared the mean effect to capture a peer effect that is
increasing at a decreasing rate (i.e., the squared term is expected
to be negative).
310 R. Zimmer / Economics of Education Review 22 (2003) 307–315
effect. Finally, an insignificant coefficient suggests that
tracking has no effect on the peer effect.
To examine the effect from tracking, the peer variable,
and the interaction of tracking and the peer variable, four
samples of data from US public schools were utilized.
The first sample uses the full data set of students as the
observations. Therefore, the first model examines the
impact on achievement from tracking, peers, and the
interaction of tracking with the peer effect, along with
the student’s other school, family, and the past inputs.
In addition to using all students in the sample, three
additional samples are used for analysis: one that is
restricted to high-ability students, one that is restricted to
low-ability students, and one that is restricted to averageability
students17. Therefore, the second estimated model
uses all the same inputs, but restricts the data to include
only high-ability students. A student is classified as highability
if he or she is placed in the upper 20th percentile
of all students across the nation on the math test at the
beginning of the year. Again, the peer-effects literature
assumes that a student placed in a class with a higher
mean peer group will experience a positive achievement
effect. A third model uses the same inputs but restricts
the data to include only low-ability students. A student
is classified as low-ability if he or she placed in the lower
20th percentile of all students across the nation on the
same mathematics test. A fourth model uses the same
inputs but restricts the sample to average-ability students.
A student is classified as average-ability if he or she
scored between the upper 20th percentile and the lower
20th percentile on the same mathematics test.
As in Betts and Shkolink (2000) and Zimmer and
Toma (2000), each model includes variables that represent
family and school inputs. The variables that represent
the family socioeconomic characteristics (F) of a
student is the occupation level of the father and the
mother and the education level of the father and the
mother18. The variables that represent school inputs (S)
include the size of the classroom, the general experience
of the teacher, the math experience of the teacher, the
teacher’s pedagogy training, the number of classes of
mathematics training for the teacher, the teacher’s age,
and the teacher’s gender. Finally, the student’s inputs
include the student prior achievement and student’s gender.
The following section will present the results of the
17 The restricting of data in this way is consistent with the
work of Kerckhoff (1986); Hoffer (1992), and Argys et al.
(1996).
18 The variables for the education level of the father and
mother are limited education measures defined as a binary variable
that is equal to one when the father’s or mother’s education
is at least secondary, respectfully. Because of the limitations of
these variables, readers should interpret the results with caution.
model focusing primarily on the interaction term, the
peer effect, and tracking.
4. Results
Using a model specification similar to that of Betts
and Shkolnik (2000), I estimate the effect from tracking,
peers, and the interaction of tracking and peers from four
segmented data sets. Table 1 displays the results. Column
one of the table lists the variables, while columns
two and three display the estimated coefficient and tscores
of each of the variables from the full set of data.
Columns four and five display the estimated coefficient
and t-scores of the model that restricts the data to only
high-ability students. Columns six and seven display the
estimated coefficient and t-scores of the model that
restricts the data to only low-ability students. Finally,
columns eight and nine display the estimated coefficient
and t-scores of the model that restricts the data to only
average-ability students.
In all the models, the tracking variable is a dichotomous
dummy variable, 1 if tracking is present in the
classroom, 0 if it is not. The peer variable is the average
of beginning-of-year test scores of all students in individual
j’s classroom. The interaction term is the interaction
of the tracking variable and the peer variable. Included
in each of the models, in addition to the peer variable,
tracking variable, and interaction term, are family and
school inputs, along with the student’s gender and prior
achievement (as proxy for past inputs). The dependent
variable is the achievement score at the end of the
school year.
For this study, the primary focus is on the coefficient
for the interaction term. Also of great interest are the
coefficients of the tracking and peer variable and the total
effect from tracking, including the mechanism of tracking,
the interaction, and peer effects. First, let us focus
on the interaction term. For the full sample, the averageability
students, and the low-ability students, the coefficients
are negative and significant. However, when the
data are restricted to only high-ability students, the coefficient
is insignificant, suggesting that tracking neither
enhances nor diminishes the peer effect for high-ability
students. Therefore, the results suggest that tracking
diminishes the positive peer effect for low- and averageability
students while neither enhancing nor diminishing
it for high-ability students19.
19 The sensitivity of these results is tested by alternatively
defining high- average- and low-ability students. In the first
alternative, high-ability is defined as students placed in the top
25th percentile, average-ability is defined as students placed in
the middle 50th percentile, and low-ability is defined as students
placed in the lower 25th percentile. In the second alternative,
high-ability is defined as students placed in the top 15th percentile,
average-ability is defined as students placed in the middle
R. Zimmer / Economics of Education Review 22 (2003) 307–315 311
Focusing on the secondary points, let us examine the
results of the peer variable and tracking. To interpret the
magnitude (and significance) of the coefficient for the
peer effect and tracking requires taking into account the
coefficient and joint significance of all variables that
include multiplicatives of the tracking and peer variables.
More formally, our model is
At,j
B0  B1(Pt,j)  B2(P2
t,j)  B3(P∗
t,jTt,j) (2)
 B4(Tt,j)  ....
where At,j is the post-year level of student achievement
for student j in period t and is a function of the student
j’s peers (Pt,j or the mean pre-year test score of classroom
for student j), student j’s peer squared (P2
t,j or the
square of the mean pre-year test score of classroom for
student j), the interaction term (the mean pre-year test
score of the classroom, Pt,j, for student j times the dichotomous
tracking variable, Tt,j, for student j), the tracking
variable (Tt,j or the dichotomous 1/0 dummy variable for
tracking within the classroom for student j), and all other
explanatory variables. The estimated peer effect is the
partial derivative of At,j with respect to the peer variable
(Pt,j) and is ∂A/∂P = B1 + 2B2(Pt,j) + B3(Tt,j), which is
the estimated effect from student j’s peers. The estimated
tracking effect is the partial derivative of At,j with respect
to tracking (Tt,j) and is ∂A/∂T = B3(Pt,j) + B4, which is
the estimated effect from tracking. To test the significance
of the peer effect and tracking, the resulting beta
coefficients from the partial derivatives must be jointly
tested. Therefore, an F-test was conducted on the joint
significance of the peer variable and tracking.
Focusing first on the peer effect, the F-test indicates
significance at the 0.001 level for each of the models,
except the model that restricts the data to high-ability
students. In other words, the higher the mean test scores
of a student’s classmates, the better the average- and
low-ability students will perform20. It is surprising that
this effect does not show up for the high-ability students
as well. However, Argys et al. (1996) suggest that in
many cases, placement in high-tracks is not strictly based
upon ability, but on other factors including parental
influence on the decision process. In other words, highability
students may not always be placed exclusively
with high-ability students and therefore, have less of a
chance of gaining a positive peer effect.
70th percentile, and low-ability is defined as students placed in
the lower 15th percentile. In both cases, the interaction term
remained insignificant for high-ability students and negative
and significant for average- and low-ability students. This suggests
this robustness of the results. These results are available
upon request from the author.
20 These results are also true for the alternative definitions of
high-, average-, and low-ability as specified in footnote 21.
Again, results are available upon request from the author.
As noted above, to interpret the results of the peer
effect, the coefficient of all variables that include the
peer variable must be considered, including the coefficient
of the interaction term when tracking is present
(the interaction term does not have to be considered
when tracking is not present because the variable takes
on a value of zero). For instance, in the full sample, the
peer effect yields an estimated coefficient of 0.22 when
tracking is present and 0.38 when it is not21. The coefficient
implies that, on average, increasing the class mean
on the beginning-of-year scores by one point increases
a representative student’s end-of-year math test score by
approximately 0.22 points when tracking is present and
0.38 when it is not.
As with the peer coefficient, the F-test is employed to
test the significance of the tracking coefficient. In these
models, tracking is significant for the full sample and the
average- and low-ability students, and insignificant for
the models in which the data set was restricted to highability
students22. The results from the full sample of
data and the data restricted to average-ability students
suggest a negative effect from tracking, net of the peer
effect. However, unlike previous research (which measured
the total effect from tracking), the tracking variable,
on the margin, has a positive coefficient for low- and
average-ability students when controlling for the peer
level.
The total tracking effects, including the peer effect and
the effect from the mechanism of tracking (e.g., curriculum,
curriculum pace, etc.) are illustrated in Figs. 1–3.
Each figure compares student test scores in a tracked versus
a non-tracked class at different peer levels. Fig. 1
suggests that high-ability students are better off in a
tracked class at higher peer levels and better in a nontracked
class at lower levels. This fact is due to the offsetting
effects from tracking and peer variables. In
assessing the overall tracking effect, the estimates suggest
that the negative effect from the mechanism of
tracking (–4.05) is countered by the positive effect of
peers (0.38 and 0.2). The figure indicates that at the
approximate mean peer level of 18, as measured by the
pre-test score, the positive peer effect dominates the
21 For the full sample, the total coefficient of the peer effect
is derived by adding the coefficient of the peer variable (0.69)
plus two times the coefficient of the squared peer variable (-
0.01) times the mean value of the peer variable (15.81) plus the
coefficient of the interaction term (-0.15). The coefficient of the
interaction term is added only when tracking is present.
22 The alternative cases have similar results except for highability
students (defined as the top 15th percentile) and averageability
students (defined as the middle 50th percentile). In the
high-ability case, tracking has a negative and significant effect.
In the average-ability case, tracking becomes insignificant when
defined more narrowly. Once again, results are available upon
request from the author.
312 R. Zimmer / Economics of Education Review 22 (2003) 307–315
Table 1
Public school results
Variables Full sample High-ability students only Low-ability students only Average-ability students only
Est. coefficient t-scores Est. coefficient t-scores Est. t-scores Est. coefficient t-scores
coefficient
Tracking 1.98 2.61a –4.05 –1.30 3.56 2.77a 2.36 2.08a
Peer variable 0.69 6.21a 0.38 0.97 0.83 2.83a 0.73 3.85a
Peer variable squared –0.01 –3.14a –0.01 –1.01 –0.02 –1.47 0.01 –1.91b
Interaction –0.15 –2.96a 0.20 1.15 –0.30 –2.96a 0.17 –2.22a
Father’s occupation 0.28 1.51 0.26 0.75 0.93 2.34a 0.12 0.50
Mother’s occupation 0.14 0.74 –0.28 –0.79 0.16 0.42 0.25 0.96
Father’s education 0.37 1.10 –0.65 –0.82 0.91 1.66b 0.28 0.58
Mother’s education 0.46 1.22 1.86 1.92b –0.08 –0.14 0.42 0.77
Pre-Test 0.85 59.66a 0.67 14.66a 0.41 3.87a 0.97 27.96a
Teacher’sexperience 0.04 1.80b –0.04 –1.00 –0.01 –0.06 0.07 2.56a
Teacher’s math –0.01 –0.56 –0.03 –0.95 0.01 0.18 –0.01 –0.49
experience
Teacher’s gender 0.50 2.72 0.28 0.79 0.85 2.19a 0.54 2.11a
Teacher’s age 0.01 1.02 0.05 1.90a 0.02 0.62 0.002 0.08
Teacher’s education –0.002 –0.82 –0.004 –0.95 –0.01 0.28 –0.001 –0.38
Teacher’s math 0.01 0.56 0.03 0.95 0.01 0.03 0.002 0.10
education
Class size –0.03 –2.33a –0.05 –2.06a –0.03 –0.10 –0.03 –1.60
Student’s gender 0.51 2.83a 0.44 1.35 0.07 0.18 0.65 2.62a
Intercept –2.80 –2.67a 7.78 2.02a –2.03 –2.46b –4.94 –2.99a
F-Value for peer effect 27.96a 0.79 8.51a 11.89a
F-value for tracking 6.96a 1.67 7.80a 4.38a
Sample size 3706 763 703 2238
R-square 0.6536 0.2966 0.1711 0.3602
a Indicates significance at the 0.05 level.
b Indicates significance at the 0.10 level.
R. Zimmer / Economics of Education Review 22 (2003) 307–315 313
Fig. 1. High-ability sample.
Fig. 2. Low-ability sample.
Fig. 3. Average-ablility sample.
negative mechanism of the tracking effect and the overall
effect is positive.
Fig. 2 suggests that low-ability students are better off
in a tracked class at lower peer levels and better off in
a non-tracked class at higher peer levels. In assessing
the overall tracking effect, the estimates suggest that the
positive effect of tracking (3.56) is countered by the
negative effect of a reduced peer effect from tracking,
as highlighted by the coefficient of the interaction term
(–0.3). The figure indicates that at an approximate mean
peer level of 12, the reduction of the peer effect dominates
the positive mechanism of tracking and the overall
tracking effect is negative.
Finally, Fig. 3 suggests that average-ability students
are better off in a tracked class at lower peer levels and
better off in a non-tracked class at higher peer levels. In
assessing the overall tracking effect, like the low-ability
sample, the estimates suggest that the positive effect of
tracking (2.36) is countered by the negative effect of a
reduced peer effect from tracking, as highlighted by the
coefficient of the interaction term (–0.17). The figure
indicates that at the approximate mean peer level of 14,
the reduction of the peer effect dominates the positive
mechanism of tracking and the overall tracking effect
is negative.
Therefore, in each of these cases, whether a student
is better off in a tracked versus a non-tracked class
depends upon the level of peers and suggests contradictory
choices. For instance, low-ability students perform
better in tracking classes only when they have lowability
peers, which is more likely when classes are
tracked, and perform better in a non-tracked class when
they have high-ability peers, which is more likely when
classes are not tracked.
Other patterns from the analysis should also be noted.
Father’s occupation and education is significant (and
positive) only for low-ability students, while mother’s
education is significant (and positive) only for highability
students and mother’s occupation is insignificant
in all cases. School inputs of the teacher’s general
experience, math experience, gender, age, general education,
math education, and also the class size have
mixed results. The teacher’s experience is positive and
significant for the full sample as well as for averageability
students, whereas the teacher’s experience in
teaching mathematics, as well as the teacher’s training
in mathematics, is insignificant in all cases. The teacher’s
gender is positive and significant for low- and averageability
students and the teacher’s age is positive and significant
for high-ability students only. For the class-size
variable, the coefficient is negative and significant in two
cases (high-ability students and the full sample). The student’s
gender is positive and significant for low-ability
students only. Finally, the control variable for previous
educational attainment (pre-test) is positive and significant
in each of the cases.
5. Conclusions
In the introduction to the paper, I noted that there has
been a brewing debate over the effects of tracking on
the educational achievement of students. Opponents
argue against tracking for a number of reasons including
a reduced peer effect for low-ability students. In this
study, I found that tracking does indeed reduce the posi314
R. Zimmer / Economics of Education Review 22 (2003) 307–315
tive peer effect for low- and average-ability students.
However, as a secondary point, I also found that the
mechanism of tracking, at the margin, when holding the
peer level constant, has a positive effect on these same
students. In total, however, tracking, including both the
mechanical and peer effect of tracking, can have a positive
effect on low- and average-ability students, but only
when these students have lower-level peers, which is
more likely to occur with tracking. Therefore, it is hard
to advocate tracking when the results suggest that it
reduces the peer effect for low- and average-ability students
and in total has an insignificant effect on highability
students. It also has a positive effect on low- and
average ability students only when these students are surrounded
by lower-level peers, which tracking is more
likely to create.
Acknowledgements
I thank Dominic Brewer, Richard Buddin, Dan Goldhaber,
Eric Eide, Eugenia Toma, and two anonymous
reviewers for their helpful comments.
Appendix
Table A1
Descriptive statement
Variable Description Mean SD
Peer variable Mean beginning-of-the- 15.81 4.70
year test score (out of
possible 40) of students
in a classroom
Tracking Binary variable=1 if the 0.59 0.49
student is in a tracked
class; 0 if not
Father’s Binary variable=1 if 0.39 0.49
occupation father’s occupation is
professional or skilled; 0
if unskilled
Mother’s Binary variable=1 if 0.62 0.49
occupation mother’s occupation is
professional or skilled; 0
if unskilled
Father’s Binary variable=1 if 0.88 0.32
education father’s level of education
is at least secondary; 0
otherwise
Mother’s Binary variable=1 if 0.91 0.28
education mother’s level of
education is at least
secondary; 0 otherwise
Student math score at
beginning of the year.
Pretest 15.81 7.93
Student’s Binary variable=1 if
gender student’s gender is male;
0 if female
Teacher’s Binary variable=1 if 0.49 0.49
gender teacher’s gender is male;
0 if female
Teacher’s Number of years of 14.30 8.11
experience teaching experience.
Teacher’s Number of years of 9.78 8.92
math experience teaching
experience mathematics
Teacher’s age Age of teacher 38.94 11.08
Teacher’s Teacher’s number of 28.68 40.75
education semester units of general
pedagogy in
postsecondary education
Class size Total number of students 26.64 6.99
enrolled in the class
Teacher’s Teacher’s number of 10.59 6.49
math training semester units of math
training
Post-test Student math score at end 19.79 9.18
of year.
R. Zimmer / Economics of Education Review 22 (2003) 307–315 315
References
Argys, L. M., Rees, D. I., & Brewer, D. J. (1996). Detracking
America’s schools: Equity at zero cost? Journal of Policy
Analysis and Management, 15(4), 623–645.
Betts, J. R., & Shkolnik, J. L (2000). Key difficulties in identifying
the effects of ability grouping on student achievement.
Economics of Education Review, 19(1), 21–26.
Boardman, A. E., & Murnane, R. J. (1979). Using panel data
to improve estimates of the determinants of educational
achievements. Sociology of Education, 52, 13–121.
Braddock, J. H., & Dawkins, M. (1993). Ability grouping,
aspirations, and achievement: evidence from the National
Longitudinal Study of 1988. Journal of Negro Education,
62(3),
Figlio, D. N., & Page, M. E. (2002). School choice and the
distributional effects of ability tracking: does separation
increase inequality? Journal of Urban Economics, 51,
497–514.
Gorman, A. (1987). The stratification of high school learning
opportunities. The Sociology of Education, 60, 135–155.
Hallinan, M. (1994). Tracking: from theory to practice. Sociology
of Education, 67(2), 79–84.
Hanushek, E.A., Kain, J.F., Markman, J.M., & Rivkin, S.G.
(2001). Does peer ability affect student achievement? NBER
Working Paper 8502.
Henderson, J. V., Miezkowski, P., & Sauvageau, Y. (1978).
Peer group effects and educational production functions.
Journal of Public Economics, 10, 97–106.
Hoffer, T. B. (1992). Middle school ability grouping and student
achievement in science and mathematics. Educational
Evaluation and Policy Analysis, 14(3), 205–227.
Hoxby, C. (2000). Peer effects in the classroom: learning from
gender and race variation. NBER Working Paper 7867.
Kerckhoff, A. C. (1986). Effects of ability grouping and student
achievement in science and mathematics. American Sociological
Review, 51, 842–858.
Oakes, J. (1985). Keeping track: How schools structure
inequality. New Haven, CT: Yale University Press.
Oakes, J. (1990). Multiplying inequalities: the effects of race,
social class, and tracking on opportunities to learn mathematics
and science. Santa Monica, CA: RAND Corporation.
Robitalle, D. F., & Garden, R. A. (1989). The IEA study of
mathematics II: contest and outcomes of school mathematics.
Oxford: Pergamon Press.
Summers, A. A., & Wolfe, B. L. (1977). Do schools make a
difference? American Economic Review, 67, 639–652.
White, T. G., & Kane, M. B. (1995). Ability grouping, tracking,
and student achievement: a critical review of research support
for policy. Washington, D.C: U.S. Department of Education,
Office of Educational Research and Improvement.
Zimmer, R. W., & Toma, E. F. (2000). Peer effects and educational
vouchers: evidence across countries. Journal of
Policy Analysis and Management, 19(1), 75–79.