Are School Uniforms A Good Fit Results F

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

Educational Policy

http://epx.sagepub.com/

Are School Uniforms a Good Fit? : Results From the ECLS-K and the
NELS
Ryan Yeung
Educational Policy 2009 23: 847
DOI: 10.1177/0895904808330170

The online version of this article can be found at:


http://epx.sagepub.com/content/23/6/847

Published by:

http://www.sagepublications.com

On behalf of:
Politics of Education Association

Additional services and information for Educational Policy can be found at:

Email Alerts: http://epx.sagepub.com/cgi/alerts

Subscriptions: http://epx.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://epx.sagepub.com/content/23/6/847.refs.html

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Educational Policy
Volume 23 Number 6
November 2009 847-874
© 2009 Corwin Press
Are School Uniforms a 10.1177/0895904808330170
http://epx.sagepub.com
Good Fit? hosted at
http://online.sagepub.com

Results From the ECLS-K and


the NELS
Ryan Yeung
Syracuse University, New York

One of the most common proposals put forth for reform of the American sys-
tem of education is to require school uniforms. Proponents argue that uniforms
can make schools safer and also improve school attendance and increase
student achievement. Opponents contend that uniforms have not been proven
to work and may be an infringement on the freedom of speech of young people.
Within an econometric framework, this study examines the effect of school
uniforms on student achievement. It tackles methodological challenges through
the use of a value-added functional form and the use of multiple data sets. The
results do not suggest any significant association between school uniform poli-
cies and achievement. Although the results do not definitely support or reject
either side of the uniform argument, they do strongly intimate that uniforms are
not the solution to all of American education’s ills.

Keywords: achievement; education reform; educational policy; educational


reform; policy; elementary education; secondary education

M ark Twain once remarked that “clothes make the man.” This trans-
formative power of clothing provides the theoretical rationale
behind school uniforms, though acceptance of this particular maxim is far
from universal. On one hand, some people believe that uniforms offer
great promise for the nation’s most struggling schools as a potential tool
to reduce crime, improve attendance, and increase achievement. On the
other hand, some people raise questions about the potential effects of
school uniforms on a child’s individuality and their implications for civil
liberties. Despite the attention uniform policies have received from
schools, parents, and the media, quantitative research on school uniforms
remains surprisingly sparse. This study adds to the literature in this area

Authors’ Note: The author would like to thank John Yinger and an anonymous referee for
their helpful comments.

847
Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010
848 Educational Policy

by focusing on one simple to understand yet important question: Do


school uniforms raise achievement?
Proponents of school uniforms argue that among other benefits, school
uniforms can make schools safer, improve attendance, and raise academic
achievement. In his study of uniforms, Joseph (1986) suggested that
uniforms serve dual purposes. They serve to distinguish members from
nonmembers and also to signal to the actor and the audience that a certain
type of behavior is expected. Serving in this manner, uniforms may make
schools safer by identifying trespassers and setting a tone for serious study,
and also discourage the activities of gangs and the associated tensions that
result from wearing gang attire (Stanley, 1996).
Kaiser (1985) listed several other benefits that school uniform proponents
have historically proposed. These benefits include improving discipline,
increasing respect for the teacher, promoting group spirit, maintaining aca-
demic standards through uniformity, easing the strain on family budgets, de-
emphasizing socioeconomic differences, and eliminating the relationship
between dress and school status.
Opponents present a very different set of arguments. The most common
argument proposed is based on the research, or lack thereof, determining a
causal relationship between uniforms and discipline or uniforms and student
achievement. DeMitchell (2006) suggested that uniform policies may only
serve as a bandage in addressing the problems of schools and may provide
policy makers and practitioners a way to avoid making the difficult decisions
necessary to reform public education. In addition, opponents cite the legality of
such policies (Lumsden, 2001). According to the American Civil Liberties
Union, “School uniform policies do violate [students’] First Amendment
rights . . . and go way beyond just having a reasonable dress code that
promotes safety and decorum in schools” (Dress Codes and School Uniforms,
2006). All in all, the courts have not found this argument to be convincing.
Courts in Arizona, Texas, Louisiana, and California have all upheld school
uniform policies and rejected the constitutional argument (DeMitchell, 2006).
Kohn (1998) took a different approach. His opposition was not based
on the efficacy or legality of uniform policies, but in the philosophical
implications. “Overlooked is the more substantive argument,” wrote Kohn,
“that kids don’t learn much of value in an environment where they are
excluded from decision making” (p. 9). If Kohn’s thesis is correct, uni-
forms may actually undermine the goal they are trying to achieve. The
answer to which side is right is ultimately an empirical one, but the force
with which both sides argue their cases indicates that this issue is not one
that will be resolved without debate.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 849

Although uniforms remain rare in American public schools, they have


long been an institution in many private schools, especially in Catholic
schools. In addition, they are a fixture in public schools outside of the
United States, including Japan and South Korea, two countries with school-
children who routinely outperform American children of the same age on
international standardized assessments (Schleicher, Tamassia, & Ikeda,
2004). Data on the pervasiveness of school uniforms in the United States is
limited; however, a telephone survey by the National Association of
Elementary School Principals (n.d.) found a little less than a quarter of all
public, private, and parochial school principals either had uniform policies
in place, were currently writing one, or had it on their agenda. In addition,
many large urban public school districts—including Baltimore, Cincinnati,
Dayton, Detroit, Los Angeles, Long Beach, Miami, Memphis, Milwaukee,
Nashville, New Orleans, New York, Phoenix, Seattle, and St. Louis—have
schools with either voluntary or mandatory uniform policies (U.S. Depart-
ment of Education, 1996). Although school uniforms are far from ubiqui-
tous in American education, they are also not inconsequential, which is why
a careful evaluation of their effects is imperative.
Estimation of school uniform effects necessarily leads to methodolog-
ical challenges in disentangling the effects of school uniforms from other
variables that may be correlated with the presence of a school uniform
policy and achievement. This study uses two empirical approaches to
overcome the impediments described and hence enhance the internal
validity of the analysis. First, this analysis relies upon a “value-added”
estimation strategy, which has been used frequently in the economics of
education literature to mitigate potential bias resulting from omitted vari-
ables. Second, this study conducts analyses based on two data sources—
the Early Childhood Longitudinal Study–Kindergarten Class (ECLS-K)
as well as the National Education Longitudinal Study (NELS)—to deter-
mine if results are robust across data sets and grade levels, which indeed
is what is found.
After controlling for prior year test score and other variables, the results
are remarkably consistent across data sets, specifications, subject areas, and
grade levels. There is little to no evidence that school uniforms raise
achievement. If anything, the impact appears to be negative. Overall,
children in schools with mandatory school uniform policies appear to per-
form worse than their peers in similar schools without mandatory school
uniform policies. Although there can be much to be gained from studying
and implementing successful education policies of other nations, the case
of school uniforms does not appear to such an example.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


850 Educational Policy

The next section reviews the research on school uniform policies and
discusses the methodology each author used to study this topic. The third
section describes the two data sets used in this study, the ECLS-K and the
NELS. Subsequently, I discuss the empirical strategy used in this analysis.
Results are reported in the fifth section. The final section concludes and
discusses implications for policy.

Literature Review

Empirical research into the effects of public school uniform policies in the
United States almost always begins with the case of Long Beach, California;
though selected schools in Washington, D.C.; Baltimore, Maryland; Miami-
Dade County, Florida; Bridgeport, Connecticut; and Detroit, Michigan, had
already required them as early as the 1980s. The Long Beach case drew
national headlines when President Bill Clinton in 1996 declared,

If it means that the school rooms will be more orderly and more disciplined
and that our young people will learn to evaluate themselves by what they are
on the inside, instead of what they’re wearing on the outside, then our public
schools should be able to require their students to wear school uniforms.
(Quoted in White, 2000)

In 1994, Long Beach, California, became the first large urban school dis-
trict in the United States to require all students from kindergarten to eighth
grade (altogether 58,500 students) to wear uniforms. The results were dra-
matic. In the first year after implementation, overall school crime decreased
by 36%, sex offenses fell by 74%, fights between students dropped by 51%,
assault and battery offenses fell 34%, school suspensions dropped by 32%,
and vandalism decreased by 18%. Perhaps most important, less than 1% of
the students elected to opt out of the policy (U.S. Department of Education,
1996). Principals and teachers also cited significantly higher achievement
in some cases (Polachek, 1996).
However, critics have presented several persuasive arguments that weaken
the internal validity of these findings. Paliokas and Rist (1996) suggested
that other factors associated with the treatment may explain the stark results
seen in Long Beach. These factors include changes in community policing
practices and the introduction of other school security measures. The authors
also propose the phenomena of regression to the mean and Hawthorne
effects that may weaken the validity of the study. Posner (1996) proposed a
variant of the Hawthorn effect hypothesis, suggesting that the major change

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 851

in Long Beach was not an actual change in student attitudes or behavior, but
a change in the way adults perceive uniformed students. If this theory is
correct, then the same beneficial results can probably be achieved without
the implementation of uniforms.
Since the Long Beach case, numerous studies have been published
presenting the results of individual case studies of schools or districts that
have implemented school uniform policies. The majority of the literature
has focused on school violence, often utilizing observational studies or
surveys of the attitudes or perceptions of students, teachers, principals, or
parents.1 However, the link between uniforms and achievement remains
relatively unexplored.
Utilizing information drawn from surveys of 301 principals and teachers
and 368 parents, Stevenson and Chunn (1991) found no direct effect of
uniforms on academic achievement, though their analyses are subject to
the research design issues that plague survey research. Murray (1997)
compared student responses in a school with school uniforms and a similar
school without and found the school with uniforms had a significantly
better school climate. He suggested that school uniforms may improve
student achievement through this mechanism of school climate. Brunsma
(2006a) looked at achievement trends for Mount Carmel Elementary
School (MCE) in Mount Carmel, Pennsylvania, from 1996-1997 to 2001-
2002. He found that in the first year of implementation of a school uniform
policy at MCE, achievement in reading, mathematics, and writing took a
turn for the worse and the average test scores no longer mirrored the general
trends of similar area schools. He was quick to acknowledge, however, that
the results in no way suggest a causal relationship.
Pate (1999) studied the effects of uniforms in two Florida school districts,
one rural and one urban. Using a pre-post test quasi-experimental design,
she found that elementary school students in both the urban and rural school
district experienced significant gains in achievement the first year after the
implementation of a school uniform policy, as measured by number of pro-
motions and reading test scores. Effects on discipline were mixed. Draa
(2005) conducted another quasi-experimental design in Ohio to measure the
impact of school uniforms on academic outcomes. Utilizing the time-series
nature of her data, she made four comparisons: same school comparison
over time, an intradistrict comparison between schools that had school
uniforms and schools that did not over time, an intrastate comparison, and
comparisons between matched schools. Given the number of analyses,
the results are difficult to summarize. Very generally, her results suggest
strong positive effects on attendance, weaker effects on graduation rates, and
inconsistent effects on reading and mathematics achievement.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


852 Educational Policy

Two studies that have used multiple regression techniques to study the
effects of school uniforms on academic outcomes are Brunsma and
Rocquemore (1998) and Brunsma (2004). In the former study, the authors
used the NELS—one of the data sets I used in this study—to test the effects
of school uniforms on school attendance, behavior problems, substance
use, and academic achievement. Controlling for a set of school and student
characteristics, the authors found no direct positive effects of uniforms on
any of the outcomes listed previously. In fact, under certain specifications,
achievement is negatively and significantly associated with the presence of
school uniforms.
Brunsma replicated the 1998 analysis in 2004 using standardized tests
for specific subject matter to better isolate the effects of uniforms on
achievement in specific subject areas. As before, the results indicate that
uniforms have a negative and significant effect on reading achievement and
negative, though not statistically significant, effects in math, science, and
history, holding all else constant for students in the 10th grade. He found a
small but statistically significant negative effect in 8th-grade reading and
virtually no effects on the other exams. In the 2004 study, Brunsma also
conducted analyses using ECLS-K data. Using this set of data, Brunsma
found a significant and positive effect in reading, but not mathematics. The
results are close to zero for both of the exams for kindergarten children.
This study builds on Brunsma and Rocquemore’s (1998) and Brunsma’s
(2004) research in one important way. It controls for a previous year of test
score data within a multiple regression framework, thereby mitigating prob-
lems of selection bias that may result from a simple cross-sectional study.
As with the latter Brunsma study, it uses two data sets to disentangle the
role age plays in successful uniform policies.

Data

ECLS-K
There are two sources of data for this study, both published by the
National Center for Education Statistics (NCES). The first is the ECLS-K,
a nationally representative cohort of children beginning in their kinder-
garten year in the fall of 1998-1999. Altogether, six waves were conducted,
with the final wave taking place in the spring of 2004, when the children
were attending fifth grade. The survey used a multistage probability sample
design to select the schools and students sampled (Tourangeau, Nord, Le,
Pollack, & Atkins-Burnett, 2006). The data used in this particular analysis

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 853

come from the public use version of this data set and are based on 8,867
observations.
One of the key strengths of the ECLS-K is its effort to collect data at
various levels of analysis. In each round of data collection, children are
assessed and information is collected from parents. Teachers, school
administrators, and school office staff were also surveyed at various points
of the study (Tourangeau et al., 2006). Most of the variables used in this
study are from the spring third grade wave, with the exception of the vari-
ables measuring achievement in previous years.
Descriptive statistics for the ECLS-K data can be found in Table 1. As
indicated by Table 1, a little more than a quarter (26.3%) of the sample
attended a school that required a school uniform. This binary variable has
a standard deviation of 0.440. Somewhat surprisingly, an almost equal
number of public school students as Catholic school students attend a
school with a uniform requirement, as shown by Table 2. However, the
shares are significantly different, as 9 out of 10 Catholic school attendees
wear uniforms versus 15.3% of public school attendees.
The two dependent variables from the ECLS-K used in this article are
the third-grade reading and mathematics t scores. The t scores are measures
of ability based on performance on standardized examinations rescaled to
a mean of 50 and standard deviation of 10. A score of 50 indicates that a
child is performing at a level comparable to the national average. In this
sample, the mean reading and math scores are slightly above 50, with the
third-grade reading t score centered at 51.971 and the third-grade mathe-
matics t score centered at 51.813. Standard deviation is below the 10 point
population standard deviation for both scores with a 9.335 standard devia-
tion for the reading t score and a 9.394 standard deviation for the math t
score.
As one would expect with a nationally representative survey, the share
and standard deviation of male children in the sample is approximately
one half. About four fifths of the sample attend a public school, whereas
13.6% of children in the sample attend a Catholic institution. White, non-
Hispanic children represent the majority of the sample, providing more
than 73% of the observations. They are followed by Hispanics and
Black, non-Hispanics, who represent 13.2% and 12.2% of the sample,
respectively.
The data set also includes two linear composites intended to measure
socioeconomic status and parental involvement. The former scale was pro-
vided by the ECLS-K and is derived from questions regarding household
income, parental educational attainment, and parental occupations. The

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


854 Educational Policy

Table 1
ECLS-K Descriptive Statistics
Standard
Variable Mean Deviation Minimum Maximum

Dependent Variable
3rd Grade Reading T-Score 51.971 9.335 14.749 84.398
3rd Grade Math T-Score 51.813 9.394 18.426 84.398
School Uniform Required 0.263 0.440 0.000 1.000
Student and Family Characteristics
Male 0.500 0.500 0.000 1.000
Socioeconomic Composite 0.000 0.566 -5.674 1.902
Below Poverty Level 0.153 0.360 0.000 1.000
Race
White, Non-Hispanic 0.731 0.444 0.000 1.000
Asian 0.065 0.247 0.000 1.000
Pacific Islander 0.019 0.138 0.000 1.000
Hispanic 0.133 0.339 0.000 1.000
Black, Non- Hispanic 0.122 0.327 0.000 1.000
American Indian/Alaskan Native 0.023 0.149 0.000 1.000
Multi-Racial 0.005 0.070 0.000 1.000
Special Education Student 0.065 0.246 0.000 1.000
Parental Involvement Composite 0.000 0.566 -1.902 5.674
Parents’ Highest Education Level
8th Grade or Below 0.015 0.122 0.000 1.000
9th - 12th Grade 0.044 0.205 0.000 1.000
High School Degree or Equivalent 0.199 0.399 0.000 1.000
Vocational/Technical Program 0.053 0.225 0.000 1.000
Some College 0.298 0.458 0.000 1.000
Bachelor’s Degree 0.212 0.408 0.000 1.000
Some Graduate School 0.032 0177 0.000 1.000
Masters Degree or Equivalent 0.091 0.288 0.000 1.000
Ph.D., M.D. or Other 0.055 0.228 0.000 1.000
School Characteristics
School Type
Public 0.795 0.404 0.000 1.000
Catholic 0.136 0.343 0.000 1.000
Other Religious Private 0.055 0.229 0.000 1.000
Private, Non-Religious 0.014 0.117 0.000 1.000
Location Type
Large and Mid-Size City 0.341 0.474 0.000 1.000
Large and Mid-Size Suburb and Town 0.400 0.489 0.000 1.000
Small Town or Rural 0262 0.440 0.000 1.000
Region
Northeast 0.186 0.390 0.000 1.000
Midwest 0.298 0.457 0.000 1.000
South 0.314 0.464 0.000 1.000
West 0.202 0.401 0.000 1.000

(continued)
Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010
Yeung / Are School Uniforms a Good Fit? 855

Table 1 (continued)
Standard
Variable Mean Deviation Minimum Maximum

% of Students Receiving Free Lunch 27.862 26.915 0.000 95.000


School Size
0-149 0.051 0.220 0.000 1.000
150-299 0.196 0.397 0.000 1.000
300-499 0.351 0.477 0.000 1.000
500-749 0.269 0.443 0.000 1.000
750 + 0.133 0.340 0.000 1.000
Previous Years’ Test Score
1st Grade Reading T-Score 51.858 8.881 8.195 80.294
1st Grade Math T-Score 51.866 8.907 2.675 78.095
Kindergarten Reading T-Score 51.598 9.487 18.684 87.725
Kindergarden Math T-Score 52.198 9.274 14.507 85.841

Note: There are 8867 observations.

socioeconomic composite in this particular sample has a mean of 0 and


standard deviation of 0.566. The latter composite was generated by the author
and was derived from questions regarding parental attendance at various
school functions and activities. This composite has the same mean and stan-
dard deviation as the socioeconomic composite. Further details on these
composites can be found in the appendix.
Other variables used in this study are included to control for poverty
status, special education student, parental education level, location type,
region, the affluence of a school, and school size.

NELS
As with the ECLS-K, the 1988 NELS is a nationally representative sam-
ple sponsored by the NCES. Unlike the ECLS-K, the focus of the NELS is
children in secondary school. Twenty-five thousand students were selected
in the base year (spring of 1988) through the use of a clustered, stratified
national probability sampling technique when the children were in the
eighth grade. The fourth and final wave was conducted in the year 2000,
when the subjects were 26 years old and out of high school if not college
or graduate school (Curtin, Ingels, Wu, & Heuer, 2002). There are 7,930
observations used in this analysis.
The variables used in this study with the exception of the base-year lagged
test scores are from the first follow-up in 1990, when most of the children

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


856 Educational Policy

Table 2
School Type by School Uniform Status
ECLS-K

Number of
Observations Total Number of
with Uniform Observations Percentage

Catholic 1063 1205 88.2


Other Religious 143 492 29.1
Other Private 50 124 40.3
Public 1076 7046 15.3
Total 2332 8867 26.3

NELS

Number of
Observations Total Number of
with Uniform Observations Percentage

Catholic 360 602 59.8


Other Religious 57 237 24.1
Other Private 67 334 20.1
Public 46 6757 0.7
Total 530 7930 6.7

were sophomores in high school and are from the public use version of the
data set. The first follow-up was also the last point schools were surveyed as
to their uniform policies. As in the case of the ECLS-K, NELS researchers
interviewed pupils, parents, teachers, principals, and school office staff at
various points in the survey. Pupils were assessed in four subject areas: read-
ing, mathematics, science, and history (Curtin et al., 2002).
Table 3 presents descriptive statistics from the NELS portion of this
study. In this sample of older children, the percentage of children attend-
ing a school requiring a school uniform is considerably smaller than in the
ECLS-K. Only 6.7% of children sampled by the NELS attend a school
with a mandatory school uniform policy. Another difference can be seen
in Table 2. As opposed to the primary level, a very small portion (8.7%)
of the students attending schools requiring school uniforms come from
public schools. Approximately 68% of the school uniform observations are
from Catholic schools in the NELS. Additionally, a smaller share of
Catholic schools (59.8%) requires school uniforms than in the ECLS-K.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 857

Table 3
NELS Descriptive Statistics
Standard
Variable Mean Deviation Minimum Maximum

Dependent Variable
10th Grade Reading T-Score 51.840 9.746 30.600 68.910
10th Grade Math T-Score 52.123 9.864 31.630 71.930
10th Grade Science T-Score 51.747 9.921 31.650 72.540
10th Grade History T-Score 51.689 9.678 28.480 73.260
School Uniform Required 0.067 0.250 0.000 1.000
Student and Family Characteristics
Male 0.470 0.499 0.000 1.000
Socioeconomic Composite 0.032 0.785 –2.933 2.762
Race
White, Non-Hispanic 0.739 0.439 0.000 1.000
Asian/Pacific Islander 0.068 0.251 0.000 1.000
Hispanic 0.097 0.296 0.000 1.000
Black, Non- Hispanic 0.087 0.281 0.000 1.000
American Indian/Alaskan Native 0.010 0.098 0.000 1.000
Parental Involvement Composite 0.000 0.689 –2.120 3.040
Parents’ Highest Education Level
No High School Degree 0.078 0.269 0.000 1.000
High School Degree 0.200 0.400 0.000 1.000
Some College 0.409 0.492 0.000 1.000
College Degree 0.159 0.366 0.000 1.000
Masters Degree or Equivalent 0.092 0.289 0.000 1.000
Ph.D., M.D. or Other 0.057 0.231 0.000 1.000
Unknown 0.005 0.070 0.000 1.000
School Characteristics
School Type
Public School 0.852 0.355 0.000 1.000
Catholic 0.076 0.265 0.000 1.000
Other Religious Private 0.030 0.170 0.000 1.000
Private, Non-Religious 0.042 0.201 0.000 1.000
Region
Northeast 0.202 0.402 0.000 1.000
North Central 0.300 0.458 0.000 1.000
South 0.332 0.471 0.000 1.000
West 0.167 0.373 0.000 1.000
% of Students Receiving Free or
Reduced Lunch
0% 0.139 0.346 0.000 1.000
1-10% 0.369 0.483 0.000 1.000
11-50% 0.423 0.494 0.000 1.000
51-100% 0.068 0.253 0.000 1.000

(continued)

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


858 Educational Policy

Table 3 (continued)
Standard
Variable Mean Deviation Minimum Maximum

School Size
1-399 0.139 0.346 0.000 1.000
400-599 0.137 0.344 0.000 1.000
600-799 0.118 0.322 0.000 1.000
800-999 0.118 0.322 0.000 1.000
1000-1199 0.132 0.339 0.000 1.000
1200-1599 0.154 0.361 0.000 1.000
1600-1999 0.095 0.293 0.000 1.000
2000-2499 0.053 0.225 0.000 1.000
2500 + 0.054 0.225 0.000 1.000
8th Grade Test Score
8th Grade Reading T-Score 52.121 9.941 31.89 70.550
8th Grade Math T-Score 52.273 10.104 34.43 77.200
8th Grade Science T-Score 52.141 9.900 32.01 80.140
8th Grade History T-Score 51.950 9.856 28.42 76.710

Note: There are 7930 observations.

For all of these reasons, one has to be more concerned about issues of sort-
ing in estimates using the NELS.
The dependent variables are in the form of t scores and are standardized
with mean of 50 and standard deviation of 10. The actual sample means,
however, are slightly higher than 50, and the standard deviations are
slightly below 10. Scores approximately range from two standard devia-
tions below the mean to two standard deviations above the mean. The same
characteristics also apply to the base-year test scores.
Racial category statistics roughly parallel those in the ECLS-K, though
specificity is somewhat inferior. There is no multiracial category in the
NELS, and Asians and Pacific Islander students are combined in the same
category. There are a roughly equal portion of boys and girls in the sam-
ple, and a plurality of children have a parent who has attended some college
but not completed a degree. The share of public school students sampled is
higher than in the ECLS-K, with more than 85% of children attending a
public school versus a little less than 80% in the ECLS-K.
As in the case of the ECLS-K, two linear composites are also included
in regression analyses with the NELS, an NCES-provided composite
measuring socioeconomic status and an author-generated composite mea-
suring parental involvement. Both variables were derived through similar
processes as the analogous variables in the ECLS-K and are also described in

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 859

further detail in the appendix. Both scales are centered at zero with standard
deviation of one.
Other independent variables are dummy variables indicating region, the
percentage of students receiving free or reduced lunch in a school, and
school size.

Empirical Strategy

This study draws upon the long line of research on education production
functions, such as that presented by Hanushek (1979, 1986). Education
economists have long utilized education production functions to explain
variation in achievement between and among students. Generally, this
framework models some measure of achievement, a test score, for
example, as a function of student, family, and school inputs as well as an
error component. A linear model of achievement can be found in Bifulco,
Duncombe, and Yinger (2005) and is presented in Equation 1 modified to
address the interests of this study:
t= T –1
YijT = αXijT + βUijT + ∑ λ Τ–1 (αXijt + βUijt) + µi + δj + γT + εijT , (1)
t=1

where Y is a test score for student i in school j in year T; and X is a vector


of explanatory variables including student, family, and school characteris-
tics. The variable of interest in this study is U, which is a dummy variable
that is coded 1 if a student attends a school that has a mandatory school uni-
form policy and 0 otherwise. The summation term reflects the cumulative
nature of education and controls for the effects of explanatory variables
carried over from previous years to year T, degraded at the rate of (1 – λ).
Equation 1 also includes a year fixed effect, γ, and time-invariant student
and school fixed effects, µ and δ, respectively. ε represents a stochastic
error term.
Ordinary least squares estimates of β, the treatment effect, are likely to
be biased and undemonstrative of the true impact of school uniforms on
achievement. This form of omitted variable bias can occur if unobserved
student characteristics correlated with achievement are also correlated with
the selection of a school with a school uniform policy. For example,
students with more involved parents may be more likely to attend a school
with a school uniform policy and may also perform better than their peers.
If the regression model does not adequately control for these parental influ-
ences, it will be virtually impossible to isolate the effects of school uniform
policies from the effects of enthusiastic parents.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


860 Educational Policy

More generally, the problem described is one of selection bias, a prob-


lem that affects much of the research in program evaluation. The treatment
variable (school uniform policy) is not randomly assigned and may be asso-
ciated with unobserved characteristics that are also associated with achieve-
ment. In education research, several techniques have been developed to
address this problem, which can also be called endogeneity. This study
relies upon a “value-added” formulation of an education production func-
tion to mitigate this problem of selection bias.
As described in Bifulco et al. (2005), one approach to addressing this
problem of endogeneity is to assume that individual and school fixed
effects µ and δ equal zero. This formulation allows the test score in the pre-
vious year to account for the effect of explanatory variables in previous
years observed or otherwise. After setting µ and δ equal to zero and sub-
tracting λYijT–1 from YijT in Equation 1, the result is

Yijt = αXijt + βUijt + λYijT–1 + (γT – λγT–1) + (εijT – λεijT–1). (2)

Equation 2 is a value-added model. The year fixed effect γ remains in the


equation and serves to control for systematic differences between years that
affect all students in the same way, such as changes in the difficulty or
nature of the examination or in average score of participating students.
Data limitations prevent the estimation of alternative functional
forms. For instance, the “difference” formulation of a production function
requires 2 years’ data for both the dependent and independent variables in
the regression model and, more important, also requires considerable vari-
ation in the decision of schools to adopt or reject school uniform policies.
As most of the schools in the data this study investigates did not have
changes in their policies regarding school uniforms, U and its estimate β
drop out of the analysis.
A third approach is to use instrumental variables estimation techniques
to isolate the variation in achievement explained solely by school uniform
policies. One common approach to instruments when analyzing school-level
treatments (e.g., school uniform policies) is to utilize proximity to other
schools that have also adopted the same treatment as an instrument.
However, as neither the ECLS-K nor the NELS contains specific geographic
information, such an estimation approach is not possible. In addition, aside
from difficulties with safety and achievement, variables that would make a
school more likely to adopt a school uniform policy are not immediately
apparent. Given that instrumental variables techniques that utilize “weak”
instruments and instruments correlated with the error term are likely to lead
to estimates inferior to estimates generated by simple cross-sectional

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 861

ordinary least squares regression, the decision was made to estimate the
value-added equation.
This discussion is not meant to imply that the value-added model is
superior to either the difference formulation or instrumental variables
regression. Indeed, the value-added model, as with differencing, requires
the debatable assumption that the relationship between observed and
unobserved student and school variables and achievement is constant over
time. However, given the limitations of the data and consequences of
faulty two-stage least squares regression, the value-added model appears
to be the best available solution to the problem of selection bias.
A more subtle form of endogeneity is also present in Equation 2. The
final error term in Equation 2 has the same subscript as the previous year
test score used in the regression model and is likely also to be endogenous.
I handle this potential problem by instrumenting the lagged test score with
a previous year test score. Additionally, Schwartz, Stiefel, and Kim (2004)
have found that models that control for two periods of test scores provide
a more robust measure of prior academic achievement and help to con-
trol for changes in the rate of growth in achievement from period to
period.2 However, 2 years of test score data were only available for the
ECLS-K portion of the study. Regressions with the NELS data only con-
trol for a single year of prior test score performance.
All regressions are weighted using Huber-White standard errors to
allow for clustering within schools. Unfortunately, due to issues of sam-
ple size, I am unable to estimate separate regressions for public versus pri-
vate schools, though I do control for school type within the overall
regression model. I also interact school type with the presence of a school
uniform policy to allow for the possibility of heterogeneous treatment
effects across school types. The results section begins with simple t tests,
which test if the difference between the means of achievement for uniform-
and nonuniform-wearing students is significantly different.

Results

Table 4 presents several difference of means t tests for both the ECLS-K
and the NELS data. The NELS results are marked by their consistency. For
both 10th-graders and 8th-graders, children in schools with required uniform
policies score better than children in schools without. This finding is uniform
across examinations. In both years, the gap is largest in reading, with
pupils in schools requiring school uniforms scoring more than 8 points
higher than pupils in schools without the restriction.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


862
Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010

Table 4
Average T-Score by School Uniform Status
ECLS-K 2nd Grade NELS 10th Grade

School Uniform Required? N Reading Math N Reading Math Science History

Yes 2332 51.917 51.010 530 55.175 55.401 53.450 54.023


No 6535 51.990 52.100 7400 51.601 51.888 51.625 51.522
Difference –0.073 –1.090 3.574 3.513 1.825 2.501
T-Test 0.322 4.814*** 8.190*** 7.951*** 4.096*** 5.758***

ECLS-K 1st Grade NELS 8th Grade

School Uniform Required? N Reading Math N Reading Math Science History

Yes 2332 52.100 51.684 530 55.671 54.822 53.657 54.537


No 6535 51.772 51.931 7400 51.867 52.090 52.033 51.765
Difference 0.328 –0.247 3.804 2.732 1.624 2.772
T-Test 1.530 1.147 8.550*** 6.023*** 3.653*** 6.269***

*** significant at .01 level.


Yeung / Are School Uniforms a Good Fit? 863

However, the findings for the ECLS-K analyses are very different. In all
but the first-grade reading exam, the average achievement for students in
schools requiring uniforms is less than the average achievement for schools
that do not require uniforms. It should be noted though, that only one of the
differences is significant. The difference of 1.090 in second-grade math
between pupils who do not wear uniforms to school and pupils who do is
the only significant t test in the ECLS-K analyses.
Whereas the NELS t test results are especially provocative, they do not
take into account the role other variables may play in explaining the differ-
ences, which the regression analysis attempts to do. This difference is
important to remember, as in Brunsma’s (2004) analysis of the NELS data,
he found very different results when moving from a correlational to regres-
sion analysis.

ECLS-K
Table 5 presents regression results from the ECLS-K portion of this study.
In both reading and mathematics, school uniforms appear to have little to no
effect, and a negative effect if any at all. Because of the interaction terms, uni-
form effects must be interpreted as conditional in nature. In reading, uniforms
are associated with negative levels of achievement in every school type with
the exception of public schools. Public school children who wear uniforms
score more than one-tenth of a point higher than public school children who
do not. In Catholic schools, holding all else constant, Catholic schools with
uniform policies have average reading test scores 0.413 points below
Catholic schools that do not. In mathematics, the same trends hold for other
religious private schools and private, nonreligious, though the effects are
smaller. One difference between the reading and mathematics results how-
ever, can be seen in Catholic schools. On the mathematics examination at
least, uniforms in Catholic schools are associated with a 0.178 point increase
in achievement versus Catholic schools without uniforms. Children in other
religious private schools and private, nonreligious schools with uniforms
also score below their counterparts in these types of schools that do not
require uniforms on both reading and mathematics. Uniforms appear to have
little to no effect on mathematics achievement of public school children.
Nevertheless, it should be emphasized that none of the school uniform
estimates using the ECLS data are significant.
The well-publicized and well-researched racial test score gap is also
evident in these regressions. Black, non-Hispanic children score 2.325
points below their White peers in reading and more than a point below in

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


864 Educational Policy

Table 5
ECLS-K Estimates of Student Achievement, 2nd
Grade T-Scores as Dependent Variable
Reading Mathematics

School Uniform Required 0.134 –0.056


(0.308) (0.326)
Interaction Terms
Catholic * Uniform –0.630 0.234
(0.903) (0.692)
Other Religious Private * Uniform –0.926 –0.456
(0.765) (0.844)
Private, Non-Religious * Uniform –2.937 –0.883
(1.968) (0.777)
Student and Family Characteristics
Male –0.091 1.295***
(0.122) (0.126)
Socioeconomic Composite 0.669*** 0.315
(0.181) (0.192)
Below Poverty Level –0.905*** –0.316
(0.216) (0.225)
Race (White, Non-Hispanic—Omitted Category)
Asian –1.197*** 0.958***
(0.620) (0.290)
Pacific Islander –1.402*** 0.651
(0.510) (0.646)
Hispanic –0.392* –0.021
(0.222) (0.230)
Black, Non-Hispanic –2.325*** –1.070***
(0.257) (0.286)
American Indian/Alaskan Native –1.977*** –1.025*
(0.620) (0.617)
Multi-Racial –0.043 –0.004
(0.910) (0.828)
Special Education Student –2.686*** –0.004
(0.315) (0.005)
Parental Involvement Composite 0.300** 0.179
(0.132) (0.132)
Parents’ Highest Education Level (8th Grade
or Below—Omitted Category)
9th - 12th Grade –0.933 –0.839
(0.608) (0.608)
High School Degree or Equivalent –1.118** –0.260
(0.555) (0.542)
Vocational/Technical Program –0.874 –0.098
(0.624) (0.611)

(continued)

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 865

Table 5 (continued)
Reading Mathematics

Some College –0.787 –0.299


(0.572) (0.574)
Bachelor’s Degree –0.297 0.390
(0.620) (0.620)
Some Graduate School 0.277 –0.400
(0.727) (0.717)
Masters Degree or Equivalent 0.020 0.421
(0.675) (0.689)
Ph.D., M.D. or Other 0.328 0.571
(0.767) (0.784)
School Characteristics
School Type (Public—Omitted Category)
Catholic 0.083 –1.916***
(0.844) (0.588)
Other Religious Private –1.257*** –1.280**
(0.377) (0.518)
Private, Non-Religious 1.849 –0.883
(1.239) (0.777)
Location Type (Small Town or
Rural—Omitted Category)
Large and Mid-Size City 0.513** 0.281
(0.250) (0.259)
Large and Mid-Size Suburb and Town 0.076 –0.135
(0.233) (0.252)
Region (Northeast—Omitted Category)
Midwest 0.437* –0.520*
(0.230) (0.281)
South 0.210 –0.679**
(0.253) (0.281)
West –0.014 –0.414
(0.282) (0.322)
% of Students Receiving Free Lunch –0.017 –0.004
(0.005) (0.005)
School Size (0-149—Omitted Category)
150-299 0.291 0.655
(0.424) (0.438)
300-499 0.206 1.071***
(0.418) (0.415)
500-749 0.333 1.162***
(0.441) (0.436)
750 + –0.149 0.772*
(0.479) (0.463)

(continued)

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


866 Educational Policy

Table 5 (continued)
Reading Mathematics

Previous Years’ Test Score


1st Grade Reading T-Score 0.764***
(0.013)
1st Grade Math T-Score 0.976***
(0.012)
Constant 13.743*** 0.651
(1.006) (0.997)

R-Squared 0.610 0.613

Notes: There are 8867 observations. Second stage least squares estimates using Kindergarten
T-Score as instrument for 1st Grade T-Score. Huber-White Robust standard errors adjusted for
clustering by school are presented in parentheses. *significant at .10; **significant at .05;
***significant at .01.

mathematics. On average, Asian children score about 1.2 points below


White, non-Hispanic students in reading but outperform White children by
about a point in mathematics. These estimates are all significant at the .01
level. Special education students fare particularly badly on the reading
exam, with special education students scoring about a fifth of a standard
deviation below regular education students. However, this result is not
evident in the mathematics exam.
The estimates of the composites are as one would expect. In both reading
and mathematics, socioeconomic status is positively associated with
achievement, though the estimate is only significant on the reading exam.
Most of the estimates of parental educational attainment are insignificant,
perhaps absorbed into the estimate of socioeconomic status. Parental
involvement is also positively associated with achievement on either exam-
ination but, as with the estimate of socioeconomic status, is only significant
in the case of reading.
Holding all else constant, Catholic schools in both specifications
perform worse than public schools, especially in mathematics, where the
gap is –1.682 points in Catholic schools with school uniforms versus public
schools with school uniforms and –1.916 in Catholic schools without
school uniforms versus public schools without school uniforms. Children
in other religious private schools also score lower than children in public
schools, a result that is significant on both exams. Another important
school characteristic when it comes to achievement appears to be location
type. In both mathematics and reading, children in large and midsize

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 867

cities perform better on average than their counterparts in small towns or


rural areas.
The R-square estimates for both regressions are approximately equal
at .61. Whereas the R-square is reasonably high, much of the variance in
the data explained by the model can be attributed to controlling for prior
year test score. In mathematics, there is an almost one-for-one relationship
between first-grade and second-grade test score. In reading, the effect is
smaller, but also highly significant. Both of these results imply little
“regression” between testing periods.

NELS
Table 6 presents the regression results from the NELS portion of this
study. As in the case of the ECLS-K regressions, the NELS results provide
little support for the hypothesis that school uniforms have clear positive
effects on achievement. Public school children only benefit from school
uniforms in 10th-grade reading; children who attend public schools with a
uniform policy score a third of a point higher than children who do not. On
every other examination, public school students who are required to wear
uniforms actually score worse than their counterparts who do not face the
same restriction; the gaps range from 0.356 in science to 0.821 in history.
None of the estimates are significant.
In the case of reading, science, and history, Catholic school children
who are required to wear uniforms to school score below Catholic school
children who are not required to wear uniforms to school. In mathematics,
uniforms are associated with a 0.358 point increase in test score in Catholic
schools with uniforms versus Catholic schools without. None of the
Catholic results are statistically different from zero. Significant at a .10 level
is the interaction term of other religious private school type and school uni-
form status on the NELS mathematics examination. In these schools,
children who wear uniforms to school score about one tenth of one standard
deviation higher than children who do not wear uniforms in other religious
private schools. With the exception of mathematics, uniforms are associated
with positive achievement in private, nonreligious schools, though none of
the estimates are significant. This significant finding is not seen in the
ECLS-K analysis.
On each examination in this analysis, Black, non-Hispanic, American
Indian/Alaskan Native, and Hispanic pupils score on average below White
children. In some cases, the gap is quite large as in the case if 10th-grade
science, where Black and American Indian/Alaskan Native children have

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


868 Educational Policy

Table 6
NELS Estimates of Student Achievement, 10th
Grade T-Scores as Dependent Variable
Reading Mathematics Science History

School Uniform Required 0.341 –0.530 –0.356 –0.821


(0.657) (0.397) (0.780) (0.777)
Interaction Terms
Catholic * Uniform –1.306 0.888 –0.53 0.527
(0.930) (0.569) (0.960) (1.101)
Other Religious Private * Uniform –0.259 1.507* 0.78 –0.351
(1.126) (0.916) (1.213) (1.214)
Private, Non-Religious * Uniform 0.147 0.380 0.99 1.305
(1.015) (0.695) (1.197) (1.169)
Student and Family Characteristics
Male –0.308** 0.192* 1.539*** 0.570***
(0.136) (0.110) (0.157) (0.141)
Socioeconomic Composite 0.690*** 0.624*** 1.075*** 0.648***
(0.177) (0.137) (0.191) (0.205)
Race (Omitted Category—White,
Non-Hispanic)
Asian/Pacific Islander 0.376 0.641*** 0.899** 0.596*
(0.320) (0.228) (0.362) (0.320)
Hispanic –0.817*** –0.347* –1.938*** –0.696**
(0.248) (0.204) (0.282) (0.317)
Black, Non-Hispanic –1.460*** –1.326*** –2.835*** –1.407***
(0.250) (0.207) (0.293) (0.280)
American Indian/Alaskan Native –1.459** –1.035** –2.842*** –0.145
(0.680) (0.409) (0.591) (0.816)
Parental Involvement Composite 0.148 0.184** 0.198 0.295**
(0.099) (0.085) (0.121) (0.119)
Parents’ Highest Education Level
(Omitted Category—No High
School Degree)
High School Degree 0.152 0.227 0.156 0.255
(0.273) (0.235) (0.299) (0.344)
Some College 0.347 0.593** 0.468 0.643*
(0.304) (0.250) (0.328) (0.369)
College Degree 0.849** 0.812** 1.229*** 1.681***
(0.390) (0.330) (0.448) (0.489)
Masters Degree or Equivalent 0.999** 0.564 1.266** 2.580***
(0.459) (0.375) (0.516) (0.551)
Ph.D., M.D. or Other 0.895* 0.260 1.516** 2.347***
(0.538) (0.436) (0.623) (0.642)
Unknown –0.857 1.080 –0.424 –1.088
(0.876) (0.898) (1.191) (1.187)

(continued)

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 869

Table 6 (continued)
Reading Mathematics Science History

School Characteristics
School Type (Omitted
Category—Public School)
Catholic 0.950 0.385 0.179 0.342
(0.590) (0.411) (0.531) (1.058)
Other Religious Private 0.984 –0.234 –0.428 0.008
(0.731) (0.614) (0.696) (0.800)
Private, Non-Religious 0.859 –0.114 1.078 0.570
(0.553) (0.461) (0.759) (0.865)
Location Type (Small Town or
Rural—Omitted Category)
Large and Mid-Size City 0.080 0.100 0.075 –0.214
(0.267) (0.204) (0.305) (0.373)
Large and Mid-Size Suburb and Town 0.081 0.056 –0.219 –0.053
(0.236) (0.178) (0.240) (0.274)
Region (Omitted Category-Northeast)
North Central –0.209 –0.227 –0.689*** –0.703**
(0.255) (0.199) (0.267) (0.331)
South –0.106 –0.138 –1.078*** –0.878***
(0.246) (0.183) (0.265) (0.294)
West –0.228 –0.531** –0.903*** –1.396***
(0.300) (0.228) (0.315) (0.346)
% of Students Receiving Free or
Reduced Lunch (Omitted
Category—0%)
1-10% –0.402 –0.719** –0.422 –0.154
(0.379) (0.296) (0.433) (0.593)
11-50% –0.076 –0.583** –0.453 0.120
(0.407) (0.317) (0.441) (0.626)
51-100% –0.235 –1.074*** –1.040** –0.530
(0.474) (0.413) (0.497) (0.703)
School Size (Omitted
Category—1-399)
400-599 –0.120 –0.082 –0.240 –0.501
(0.342) (0.278) (0.365) (0.398)
600-799 –0.250 –0.028 –0.487 –0.215
(0.367) (0.283) (0.384) (0.452)
800-999 –0.314 0.113 –0.426 –0.691
(0.342) (0.276) (0.435) (0.431)
1000-1199 0.091 0.177 –0.723* –0.422
(0.355) (0.266) (0.386) (0.478)
1200-1599 0.120 –0.021 –0.567 –0.263
(0.336) (0.275) (0.392) (0.434)

(continued)

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


870 Educational Policy

Table 6 (continued)
Reading Mathematics Science History

1600-1999 0.956** 0.350 0.133 0.297


(0.394) (0.310) (0.468) (0.499)
2000-2499 0.581 –0.035 0.636 –0.612
(0.505) (0.369) (0.515) (0.614)
2500 + 0.552 0.268 0.173 -0.008
(0.430) (0.360) (0.509) (0.603)
8th Grade T-Score 0.733*** 0.813*** 0.641*** 0.661***
(0.007) (0.007) (0.010) (0.010)
Constant 13.589*** 9.802*** 18.809*** 17.375***
(0.729) (0.576) (0.791) (0.926)
R-Squared 0.654 0.781 0.580 0.578

Notes: There are 7930 observations. Ordinary least squares estimates with Huber-White
Robust standard errors adjusted for clustering by school presented in parentheses. *significant
at .10; **significant at .05; ***significant at .01.

average test scores that are almost 3 points below those of White children.
Asian children have significantly better achievement in mathematics and
science at a .05 level and in history at a .10 level.
The effects of a college degree are robust across all of the NELS speci-
fications. No matter the exam, versus pupils whose parents’ have not
earned a high school degree, pupils whose parents’ highest education level
is a college degree score better holding all else constant. The estimates
range from 0.812 in mathematics to 1.681 in history. Boys also appear to
do better than girls in each case with the exception of reading, where they
score three tenths of a point lower than girls on average.
The composites behave the same way as they do in the ECLS-K. Higher
socioeconomic status is associated with higher achievement, a result that is
significant at the .01 level in each specification. Parental involvement is
also positively associated with achievement, results that are significant at a
.05 level of significance in mathematics and history.
Overall, students in Catholic schools appear to score better than
children in public schools, though the results are not significant. This find-
ing is in sharp contrast to the ECLS-K and suggests that school level and
age of children may play an important role in explaining differences in
achievement between Catholic and public school children and may indicate
stronger sorting pressures at the secondary level. In other words, higher
achieving students may be more likely to be concentrated in Catholic
schools at the secondary level than they are at the primary level.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 871

Another interesting school characteristic result can be found in the


regional variables. Holding all else constant, pupils who attend schools in
the Northeast appear to perform better than children who attend schools in
the North Central, South, and West parts of the nation. The estimates of
school size are not consistent across specifications and overall patterns are
difficult to distinguish.
As with the ECLS-K regressions, R-square statistics are high and range
from .580 (science) to .781 (math). The high amount of variation in the data
explained by the model can be largely owed to the strong significance of the
lagged test score (8th-grade t score).

Conclusion

There is a general sentiment that American schools are, for the lack of a
better word, “broken.” Various reforms have been proposed over the
decades focusing on districts, schools, and students. One of the most
common “solutions” to this “problem” of American education is the intro-
duction of school uniforms. Yet despite the exuberance of proponents and a
long literature of qualitative research, the quantitative research on school
uniforms has been relatively sparse by comparison.
This study adds to the research on school uniforms by performing
econometric analyses that control for the previous performance and ability
of students, thereby reducing the possibility of endogeneity bias. In addi-
tion, it performs the estimation on two sets of data, one for primary school
students and one for secondary school students. The results are remarkably
consistent across both sets of data and support one general conclusion:
There is little evidence that uniforms improve achievement in schools.
Once controlling for prior year test score and observable student, fam-
ily, and school characteristics, there are no significant effects of school uni-
forms on performance on second grade reading and mathematics
examinations, as well as on 10th-grade reading, mathematics, science, and
history examinations. In addition, in many of the specifications, the results
are actually negative. Though the empirical design is not perfect, the
results strongly suggest that school uniforms are not a panacea. Instead of
accepting the words of Mark Twain, school officials may be better off
heeding the words of George Washington: “Do not conceive that fine
clothes make fine men any more than fine feathers make fine birds.”

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


872 Educational Policy

Appendix
ECLS-K socioeconomic status compositea
Father/male guardian’s education level
Mother/female guardian’s education level
Father/male guardian’s occupation
Mother/female guardian’s occupation
Household income
ECLS-K parental involvement compositeb
Parents attended open house?
Parents attended a PTA meeting?
Parents attended a parent-teacher conference?
Parents attended a school event?
Parents acted as a school volunteer?
Parents participated in fund-raising?
Alpha coefficient = .5755
NELS socioeconomic status compositec
Father/male guardian’s education level
Mother/female guardian’s education level
Father/male guardian’s occupation
Mother/female guardian’s occupation
Family income
NELS parental involvement compositeb
Parents attended open house
Parents attended a PTA meeting
Parents attended a parent-teacher conference
Parents attended a school event
Parents acted as a school volunteer
Parents participated in fund-raising
Alpha coefficient = .6305

Note: ECLS-K = Early Childhood Longitudinal Study–Kindergarten Class; NELS = National


Education Longitudinal Study.
a. Provided by ECLS-K.
b. Author-generated.
c. Provided by NELS.

Notes
1. Brunsma (2004), Brunsma (2006b), Kiley Wade and Stafford (2003), and Stanley
(1996) all presented excellent reviews of the literature in this important subject area.
2. Schwartz, Stiefel, and Kim (2004) actually included both lagged scores as explanatory
variables in the regression model instead of using the less recent test score as an instrument for
the more recent test score, as I do in this study. Ultimately, which approach is superior is not
clear. However, instrumental variables regression has been the traditional tool used to isolate
variation explained only by test score and not error, which is why it is employed in this study.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


Yeung / Are School Uniforms a Good Fit? 873

References
Bifulco, R., Duncombe, W., & Yinger, J. (2005). Does whole school-reform boost school
performance? Journal of Policy Analysis and Management, 24(1), 47-72.
Brunsma, D. L. (2004). The school uniform movement and what it tells us about American
education: A symbolic crusade. Lanham, MD: ScarecrowEducation.
Brunsma, D. L. (2006a). Studying elementary school uniform practices: Promises and
perils. In D. L. Brunsma (Ed.), Uniforms in public schools: A decade of research and
debate (pp. 13-48). Lanham, MD: Rowman & Littlefield Education.
Brunsma, D. L. (Ed.). (2006b). Uniforms in public schools: A decade of research and debate.
Lanham, MD: Rowman & Littlefield Education.
Brunsma, D. L., & Rocquemore, K. A. (1998). Effects of student uniforms on attendance,
behavior problems, substance use, and academic achievement. Journal of Educational
Research, 92(1), 53-62.
Curtin, T. R., Ingels, S. J., Wu, S., & Heuer, R. (2002). National education longitudinal study
of 1988: Base-year to fourth follow-up data file user’s manual (User’s Manual No. NCES
2002-323). Washington, DC: U.S. Department of Education, National Center for Education
Statistics. Retrieved March 10, 2007, from http://nces.ed.gov/pubs2002/2002323.pdf
DeMitchell, T. A. (2006). School uniforms: There is no free lunch [Electronic version]. Teachers
College Record. Retrieved March 9, 2007, from http://www.tcrecord.org/Content.asp?
ContentID=12891
Draa, V. A. B. (2005). School uniforms in urban public high schools. Doctoral dissertation,
Youngstown State University. Retrieved December 15, 2007, from http://www.eric.ed.gov/
ERICWebPortal/contentdelivery/servlet/ERICServlet?accno=ED497409
Dress codes and school uniforms. (2006). Retrieved March 22, 2007, from http://www.aclupa.
org/education/studentsrightsmanual/freedomofexpression/dresscodesandschooluniform.htm
Hanushek, E. (1979). Conceptual and empirical issues in the estimation of educational pro-
duction functions. Journal of Human Resources, 14(3), 351-388.
Hanushek, E. (1986). The economics of schooling: Production and efficiency in public
schools. Journal of Economic Literature, 24(3), 1141-1177.
Joseph, N. (1986). Uniforms and nonuniforms: Communication through clothing. Westport,
CT: Greenwood.
Kaiser, S. B. (1985). The social psychology of clothing and personal adornment. New York:
Macmillan.
Kiley Wade, K., & Stafford, M. E. (2003). Public school uniforms: Effect on perceptions of
gang presence, school climate, and student self-perceptions. Education and Urban
Society, 35(4), 399-420.
Kohn, A. (1998). What to look for in a classroom . . . and other essays. San Francisco:
Josey-Bass.
Lumsden, L. (2001). Uniforms and dress-code policies (No. 148). Eugene, OR: Educational
Resources Information Center. Retrieved March 23, 2007, from http://eric.uoregon.edu/pdf/
digests/ digest148. pdf
Murray, R. K. (1997). The impact of school uniforms on school climate [Electronic version].
NASSP Bulletin, 81(593), 106-112. Retrieved March 23, 2007, from LookSmart:
FindArticles database.
National Association of Elementary School Principals. (n.d.). Information and resources: Public
school uniforms. Retrieved March 6, 2007, from http://www.naesp.org/ContentLoad.do?
contentId=929

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010


874 Educational Policy

Paliokas, K. L., & Rist, R. C. (1996, April 3). School uniforms: Do they reduce violence—Or
just make us feel better? Education Week, 15(28) 36-52.
Pate, S. S. (1999, December). The influence of a mandatory school uniform policy. Paper presented
at the annual meeting of the Association for Career and Technical Education, Orlando, FL.
Retrieved December 15, 2007, from http://www.eric.ed.gov/ERICDocs/data/ericdocs2sql/
content_storage_01/0000019b/80/19/6b/0d.pdf
Polachek, K. (1996, July 5). Uniforms help solve many school problems [Electronic version].
Long Beach Press-Telegram, p. D5. Retrieved March 23, 2007, from http://www.lbusd
.k12.ca.us/uniforms/article_7.asp
Posner, M. (1996). Perception versus reality: School uniforms and the “halo effect.” Harvard
Education Letter, 13(3), 1-2.
Schleicher, A., Tamassia, C., & Ikeda, M. (2004). Learning for tomorrow’s world: First
results from PISA 2003 (No. 53799). Paris: OECD. Retrieved March 6, 2007, from
http://www.oecd.org/dataoecd/1/60/34002216.pdf
Schwartz, A. E., Stiefel, L., & Kim, D. Y. (2004). The impact of school reform on student
performance: Evidence from the New York Network for School Renewal Project. Journal
of Human Resources, 39(2), 500-522.
Stanley, M. S. (1996). School uniforms and safety. Education and Urban Society, 28(4), 424-435.
Stevenson, Z., Jr., & Chunn, E. W. (1991). Uniform policy/dress codes: School staff and
parent perceptions of need and impact (No. ED331933). Eugene, OR: Education
Resources Information Center.
Tourangeau, K., Nord, C., Le, T., Pollack, J. M., & Atkins-Burnett, S. (2006). Combined user’s
manual for the ECLS-K fifth-grade data files and electronic codebooks (User’s Manual No.
NCES 2006-032). Washington, DC: U.S. Department of Education, National Center for
Education Statistics. Retrieved March 8, 2007, from http://nces.ed.gov/pubs2006/2006032.pdf
U.S. Department of Education. (1996). Manual on school uniforms. Retrieved March 6, 2007,
from http://www.ed.gov/updates/uniforms.html
White, K. A. (2000, February). Do school uniforms fit? [Electronic version]. School
Administrator, 57(2) 36-40. Retrieved March 14, 2007, from LookSmart: FindArticles
database.

Ryan Yeung is a doctoral student in public administration at the Maxwell School of


Citizenship and Public Affairs at Syracuse University and a graduate research associate at the
Center for Policy Research, also located in the Maxwell School. His main research areas are
in social policy, with a particular interest in education policy, public finance, public organi-
zation and administration theory, and research design and methods. His website can be
accessed at http://student.maxwell.syr.edu/ryyeung/.

Downloaded from epx.sagepub.com at SYRACUSE UNIV LIBRARY on September 22, 2010

You might also like