Should Sixth Grade

be in Elementary
or Middle School?
An Analysis of Grade
Configuration and
Student Behavior

Philip J. Cook
Robert MacCoun
Clara Muschkin
Jacob Vigdor

Working Papers Series

SAN07-01

February 9, 2007
 Should Sixth Grade be in Elementary or Middle School?
An Analysis of Grade Configuration
and Student Behavior
Philip J. Cook, Duke University & NBER, Robert MacCoun, University of California, Berkeley,
Clara Muschkin, Duke University, and Jacob Vigdor, Duke University & NBER

Previously published as NBER Working Paper No. 12471

Abstract
Using administrative data on public school students in North Carolina, we find that sixth grade students
attending middle schools are much more likely to be cited for discipline problems than those attending
elementary school. That difference remains after adjusting for the socioeconomic and demographic
characteristics of the students and their schools. Furthermore, the higher infraction rates recorded by
sixth graders who are placed in middle school persist at least through ninth grade. An analysis of end-of-
grade test scores provides complementary findings. A plausible explanation is that sixth graders are at an
especially impressionable age; in middle school, the exposure to older peers and the relative freedom
from supervision have deleterious consequences. These findings are relevant to the current debate over
the best school configuration for incorporating the middle grades. Based on our results we suggest that
there is a strong argument for separating sixth graders from older adolescents.

Keywords: Education, peer influence, adolescence

Contact: Jacob Vigdor

Associate Professor of Public Policy
Terry Sanford Institute of Public Policy// Duke University
Durham, NC 27708-0245
(919) 613-9226
[email protected]

-1-
INTRODUCTION

Is there a “best” grade configuration for schools that serve early adolescents? If so, what

is it? Using past policy decisions as a guide, conventional wisdom on the answers to these

questions has changed several times over the past century. At the beginning of the twentieth

century, school configuration in the United States began moving away from an eight-year

primary and four-year secondary model, toward a definition of secondary education as beginning

in the seventh grade. At that time and continuing through mid-century, middle schools known as

“junior high” (grades 7-9 or 7-8) were the norm (Goldin 1999). This arrangement was intended

to create a transitional period between the sheltered elementary school and the more demanding

high school environment (Juvonen et al. 2004).

In recent decades there has been a marked shift away from junior high school, toward the

middle school configuration of grades 6-8, or occasionally 5-8. In the early 1970s, less than one-

quarter of middle schools incorporated sixth grade: by 2000, three-quarters of all middle schools

enrolled sixth grade students (see Figure 1). North Carolina’s public middle schools, which form

the basis for the analysis that follows, have led the national trend of incorporating sixth grade. In

the 1999-2000 school year, more than 90 percent of the state’s 379 middle schools served grades

6-8 (McEwin, Greene and Jenkins 2001). 1

Figure 1

Why is the current generation of sixth graders attending middle school while preceding

generations attended elementary school? The practical problem of dealing with swelling cohorts

of students was a factor in promoting the shift in the 1970s, but there was also support from

1
In 2000-1, 46 North Carolina public schools were structured as K-8, or “elemiddle.” We have
not included these schools in our study, as our goal is to analyze the outcomes associated with
middle school transition.

-2-
educators. In a survey of middle grade school administrators in 2000, 65 percent of respondents

selected the 6-8 grade configuration as the “ideal” form of organization (Valentine et al 2002).

Grade span re-configuration was part of a new paradigm for middle grade education that moved

away from the “bridging” concept, toward focused consideration of the unique challenges faced

by young teens (Juvonen et al 2004; National Middle School Association 1995). The debate

over the proper configuration of grades has heated up again in recent years, with researchers and

practitioners challenging the rationale of a separate middle school. One influential proposal has

been to reduce the number of school transitions through a configuration that combines

elementary and middle grades (Hough 1995; Juvonen et al 2004; Gootman 2007a,b; Zernike

2007).

What has been for the most part lacking in this debate, and what we seek to provide, is

direct evidence concerning what difference the grade configuration is likely to make for

students. 2 Using unique data on disciplinary infractions and end-of-grade (EOG) standardized

test scores for North Carolina public school students, we exploit variation in grade configuration

across and within the state’s school districts. Specifically, we compare the behavioral and

academic outcomes of students who attend different types of schools in sixth grade. Despite

constituting a lower-risk population along several observable dimensions, students who attend

middle school in sixth grade are more than twice as likely to be disciplined relative to their

counterparts in elementary school. These significant differences persist beyond the sixth grade

year. Sixth graders in elementary school also make gains in standardized test scores relative to

2
One exception is a study by Weiss and Knipes (2006) that examines academic outcomes and behaviors in middle
schools, finding few differences among eighth graders in middle and K-8 schools. These results are based on survey
data and refer to a single urban district that has implemented a policy of phasing out traditional middle schools. Of
greater relevance to our project are the results of Bedard and Do, who demonstrate using national data that moving
to a middle-school configuration that includes sixth grade has the effect of reducing on-time high-school completion
rates by approximately 1-3 percent (Bedard & Do 2005).

-3-
their peers in middle school. The results suggest that exposing sixth graders to older peers has

negative and lasting consequences on their academic trajectories.

We begin by providing further background on middle schools. The next section

motivates our analysis and describes several mechanisms by which grade configuration may

influence students. We then characterize the data set and the matching procedure used to select

schools to be included in the analysis. Subsequent sections provide results on infraction rates

and EOG scores, followed by a concluding discussion.

THE MIDDLE SCHOOL DIFFERENCE

The middle school educational environment is different from the elementary school

environment in several ways. A sixth grader in an elementary school will typically be assigned

to one teacher and spend much of the day in that teacher’s classroom with the same group of

students. A sixth grader in middle school will typically be assigned to a team of teachers and

move from classroom to classroom over the course of the school day, with somewhat different

groups of students in each. Middle schools place greater emphasis on discipline and academic

accomplishment (including greater use of between-classroom ability grouping), with less

opportunity for close relationships to specific teachers (National Center for Education Statistics

2000; Mills 1998).

The decision of whether to locate sixth grade in middle school or keep it in elementary

school should take account of the behavioral and academic consequences for the sixth graders

themselves, as well as for the younger grades in elementary school, and the older grades in

middle school. It is a difficult time of life at best. Between the ages of 10 and 14, students

typically must adjust to puberty, as well as to changes in social relationships with peers, family,

-4-
and authority figures (NMSA, 1996; Elias et. al, 1985; Eccles et al. et al., 1993; Rudolph et al.,

2001). Research suggests that difficulties in coping with multiple transitions may underlie some

of the negative effects that many students experience during the transition from elementary to

middle school (Eccles et al. et al., 1993). These effects include a decline in motivation and a loss

of self-esteem, particularly when the transition occurs at younger ages (Simmons and Blythe,

1987; Rudolph et al., 2001); decline in academic achievement (Alspaugh and Harting, 1995;

Alspaugh, 2001; Hanushek, Kain & Rivkin, 2004); strains on interpersonal functioning (Barber

and Olsen, 2004); and in the long term, increased risk of dropping out of school (Alspaugh,

1998; Rumberger, 1995).

Perhaps the most important difference is that a sixth grader in elementary school is

among the oldest students in the school; a sixth grader in middle school is among the youngest,

with daily exposure to older adolescents. In terms of both the developmental changes

experienced by early adolescents, and the social and academic challenges that they face in the

middle school environment, the influence of the peer group on behavior is particularly important.

Research on adolescent delinquency suggests a developmental pattern of delinquent peer

influence: the influence of peers on behavior already is significant in early adolescence, peaks

during middle adolescence, and then begins to decline (Jang 1999). Peer influence may take a

variety of forms, both direct and indirect. Direct influence may include bullying and initiation of

fights, recruitment into delinquent gangs, an enhanced supply of drugs and alcohol, seduction

and sexual importuning, an appreciative audience for rowdy behavior, companionship in truancy,

and so forth. Indirect influence may occur through modeling illicit behavior (Reinke & Walker

2006). 3

3
For an interesting discussion of peer influence in drug use in particular, see Jacobson (2004).

-5-
 School characteristics have been shown to influence peer effects on student behavior. .

The influence of peers on individual substance use tends to be strongest in schools with higher

rates of substance use (Cleveland and Wiebe, 2003). School mobility rates can adversely affect

achievement gains is schools, for students who themselves are new to a school, as well as for

their school peers (Hanushek, Kain & Rivkin, 2004). Other school context characteristics,

including the size of the school population, racial composition, poverty levels, and levels of

parental education also have been linked to peer influences on behavior (Teitler and Weiss,

2000; Ellickson et al., 2003; Hardy, Bukowski & Sippola 2002).

CONCEPTUAL FRAMEWORK

In this study, we estimate the impact of school grade span on EOG test scores and on the

prevalence and incidence of reported disciplinary infractions by North Carolina public school

students. Our main interest is on the infractions data; the data on EOG test scores, while of

intrinsic interest, serve here as a check on the validity of our findings concerning behavior.

Research shows that disciplinary problems in school have a significant impact on academic

achievement (Maguin & Lowber, 1996; South & Messner, 2000) as well as on late adolescent

and adult outcomes (McLeod & Kaiser, 2004; Sampson & Laub, 1992).

We presume that the behavior of student i attending school j in year t can be measured with a

latent variable Bijt. When this variable exceeds some threshold level, which we normalize to

zero, we observe an indicator variable Rijt indicating that a report of a negative behavior has been

filed. Our conceptual model thus takes the form:

(1) Rijt=1 if Bijt = α0 + α1Xi + α2Zj + εijt > 0,

where Xi denotes a vector of student-level characteristics, and Zj a vector of school

-6-
characteristics. Our central hypothesis is that a particular school characteristic, the grade span,

influences behavioral outcomes. The final term, εijt, is a residual reflecting unobserved

individual and school-level characteristics that determine whether a report is filed, plus any true

idiosyncrasies that operate in the report-generating process. There is a possibility that elements

of εijt are correlated with observed school-level characteristics, in which case estimates of the

coefficient vector α2 will be biased. That problem is generic to this sort of non-experimental

analysis, arising most likely because of the possibility that unmeasured qualities of the students

or their schools are important and distributed differently among the two groups of schools. One

way in which we address this problem is by use a pseudo-longitudinal analysis. We discuss this

approach in more detail in the subsequent section.

Why might school grade composition affect student behavioral outcomes? Based on the

discussion in the previous section, we can identify at least three potential causal mechanisms.

a) Social control effects. The overall rate of behavioral problems in a school can be

influenced by various features of the school environment. The degree of freedom

accorded the students, the capacity of the faculty and administration to monitor and

control behavior, and the composition of the student body are all plausibly important.

Elementary and middle schools tend to differ in all these dimensions. 4

b) The transition effect. This mechanism suggests that student behavior will tend to

deteriorate in the first year of exposure to a new school environment, particularly when

4
It should be noted that among the environmental attributes that distinguish the two types of
schools, not all are intrinsic to the form. For example, if sixth-grade teachers tend to prefer an
elementary-school environment to a middle-school environment, then those who have a choice
(including the teachers with most seniority) will tend to concentrate in elementary schools. In
that case the lower infraction rate in elementary schools would in part reflect the superior
classroom-management ability of the teachers there. In that case moving sixth grades into
middle school would not have an effect on the average behavior of students in the state unless
the most able teachers were induced to retire or leave the state as a result.

-7-
 that environment is less closely supervised than previously experienced, simply because

it may take time to learn the rules and stay out of trouble in the new environment. This

mechanism predicts a spike in behavioral problems for sixth grade students entering

middle school, followed by a reversion to the usual age-based trajectory as the student

learns to meet the new expectations. Seventh grade students entering middle school for

the first time should also exhibit a spike in behavioral problems.

c) Deviant peer-influence effects. Sixth graders might also display elevated levels of

behavioral problems if they are unduly influenced by older peers who act in a similar

fashion. This is a true “social” effect (Manski, 1993): placement with a set of well-

behaved older peers would not lead to increased behavioral problems. Deviant contagion

effects could possibly lead to persistently elevated levels of behavior problems, to the

extent that such behaviors are self-reinforcing.

In addition to these hypothesized causal mechanisms, it is possible that any correlation

between school grade span and the measured infraction rate reflects nonrandom sorting of

students. Parents may choose where to live or whether to keep their children in the public

schools based in part on the configuration of grades. That sort of selection process may

influence the characteristics of the student body in ways not necessarily reflected in observed

indicators.

Yet another possibility is that the likelihood that student misbehavior will be reported by

school officials differs between elementary and middle school. In the context of our model,

variation in standards of reporting across schools is one component of the error term εijt. Thus a

finding of higher rates of disciplinary infractions in these schools need not reflect any differences

in actual student behavior during sixth grade. The finding may still be of interest, since at a

-8-
minimum it tells us something about the likelihood that a student will acquire a “record.” This

concern is mainly relevant to the results for sixth grade per se, since by seventh grade all

students in the sample are in middle school.

In addition to the analysis of student infractions, we estimate standard linear models for

the determination of EOG test scores, incorporating the same covariates as in equation (1).

EOG tests are administered and scored the same for sixth graders whether they are in elementary

or middle school. As a result, observed differences in performance in sixth grade are not an

artifact of the measurement system.

DATA AND MATCHING PROCEDURE

Our analysis makes use of an administrative database covering all public schools and

students in the state of North Carolina for a number of years. The data were provided by the

North Carolina Education Research Data Center (NCERDC). The indicators of behavioral

problems are derived from a statewide database of disciplinary infractions recorded during the

2000-1 academic year. Each disciplinary report reflects a decision on the part of a school official

(usually a teacher) of whether to “write up” a student for misbehaving, and then a decision on the

part of the principal of whether to report to the state. (Schools are required to report incidents in

the event that they result in the out-of-school suspension of one or more students, or if the

offense is severe enough to warrant the contact of law enforcement officials, but reporting is

otherwise left to the discretion of school officials.)

In our statistical analysis we work with a sub-sample of all public-school sixth graders for

the 2000-1 school year. We omit charter schools, and drop 11 of North Carolina’s 117 school

districts for which over 25 percent of the infraction reports had incomplete or inaccurate student

-9-
identifiers, and therefore cannot be matched to students included in the NCERDC data base.

(For the remaining districts about 85 percent of infraction reports were identified with a specific

student.) An additional 7 districts are dropped because all of the sixth graders are attending a K-

8 school or some variant thereof, which is outside the scope of our analysis. The remaining 99

districts constitute our district sample. Of these, there are nine in which all sixth graders attend

elementary school, eight in which schools differ with respect to grade configuration, and 82

where all sixth graders attend middle schools. Here is a summary:

Districts
All 6th graders in elementary school 9
6th graders divided between elementary and middle 8
All 6th graders in middle school 82
Subtotal districts in working sample 99

Incomplete data on infractions 11

K-8 grade configuration 7
___
Total public school districts 117

In our statistical work we used a matching procedure to address the concern that sixth

graders are not randomly assigned to elementary or middle schools in North Carolina.. First we

ran a logit regression analysis on the 342 schools that included sixth grade (both middle and

elementary schools) in our district sample to predict the likelihood that the school was a middle

school on the basis of its locale, per-pupil expenditure levels, and student socioeconomic

characteristics. We then excluded schools where the imputed probability was very high (higher

than for any of the elementary schools in the sample) or low. In our subsequent analysis we

experimented with two standards for the matching procedure. The “stringent” matching standard

retained only those schools where the imputed probability was between .6 and .9, leaving just

140 schools in the matched sample. The “loose” matching standard retained schools where the

- 10 -
imputed probability was between .3 and .9, which retained 243 schools and a majority of the

students. As it turns out the two matched samples produce very similar results with respect to

the effects of grade configuration on infractions and EOG scores. The results reported in

subsequent sections are for the “loose” matching standard.

Table 1

Table 1 reports the results of the logit regression analysis of the 342 schools. The

specification includes both school- and district-level variables. The results suggest that the

middle schools tend to have a higher concentration of blacks and Hispanics, and be located in

larger, better funded districts. Rural concentration has little influence. 5 Based on this

regression, Figure 2 depicts the distribution of imputed probabilities for the two categories of

middle school.

Figure 2

The matched sample includes 243 schools with 44,709 sixth graders. Just 11 percent of

the students in the matched sample are in elementary schools. Table 2 compares the sixth

graders in elementary school with those in middle school along a number of individual-, school-,

and district-level variables. The sixth grade students in middle schools are less likely to be poor

(as indicated by qualification for a free lunch), are more likely to have a college-educated parent,

and have somewhat higher EOG scores on average. The matching procedure has the effect of

reducing differences between the two groups with respect to race, per-pupil expenditures, and

size of the district.

5
The schools included in this analysis include a range of locales from rural to mid-size city. The
largest city in North Carolina, Charlotte, is excluded from the analysis due to limitations of the
infractions data for that district. Five schools in locales designated as “large towns” (more than
25,000 population but outside of a Metropolitan Statistical Area) were excluded because they all
had the same configuration.

- 11 -
 Table 2

ANALYSIS OF INFRACTION RATES

A total of almost 20,000 infractions by sixth graders were recorded in the matched

sample during the school year 2000-1. The statistics shown in Table 3 and Figure 3 demonstrate

that while many of the infractions were for minor events or rowdiness, violence played a

prominent role. The incidence for middle school students was .47, or about 1 infraction for

every two students – although in fact infractions were quite concentrated, and only 16.5 percent

of students appear in the infractions database. Most notable for our purposes is that both the

incidence and prevalence rate for every type of infraction were considerably higher for sixth

graders in middle-school than for elementary-school students. The overall incidence was three

times as high for middle school students, and the prevalence rate twice as high.

Table 3 and Figure 3

The large differences in the infraction rate may actually understate the effect of sending

sixth graders to middle school, since in our sample the middle-school sixth graders are more

privileged on average. We used regression analysis in an attempt to adjust for these remaining

post-match differences, and report the results in Table 4. The sample for this analysis consists of

sixth grade students in North Carolina in 2000-1 that are in our matched sample. For columns 1-

3, the results are from logistic regressions where the dependent variable indicates whether (1) the

student appears in the infractions database; (2) the student appears in the database for a violent

infraction; and (3) whether the student appears in the database for a drug-related infraction. For

column 4 the results are from a negative binomial regression where the dependent variable is the

number of infractions of any sort.

Table 4

- 12 -
 The results confirm that attending middle school in sixth grade is associated with greatly

elevated odds of an infraction and of infraction rates. Our point estimates imply that other things

equal, the odds of having at least one infraction in sixth grade are increased by a factor of 2.2 if

in middle school; the odds of a violent infraction are increased by a factor of 2.1, and the odds of

a drug infraction by a factor of 3.8. The results from the negative binomial regression indicate

that the incidence of violations is also greatly elevated.

In these regressions, individual-level control variables include sex, race, parent’s

education and poverty status, old for grade, and preceding year’s EOG scores. Most of these

prove significant and quite influential for the prevalence and number of infractions. It is

noteworthy that Hispanics tend to have lower infraction rates than non-Hispanic whites, other

things equal. Also included are school- and district-level characteristics, although with a few

exceptions they do not prove significant. In particular it is interesting that the number of

students in the sixth grade has a negligible effect on infraction rates.

In results not shown here, we also assessed the possibility that the effect of placing sixth

grade in middle school may have different effects on different demographic groups. We re-ran

the “any infraction” logistic regression with the addition of three indicator variables capturing

interactions between “in middle school” and each of the following: black, Hispanic, and male.

The coefficient estimates of the interaction terms were small relative to standard errors, and

inclusion of these variables had little effect on other coefficient estimates. We conclude that the

effect of middle school placement on sixth grade infraction rates is more or less uniform.

Behavior Before and After Sixth Grade

It would be informative to follow these students over several years of schooling before

- 13 -
and after sixth grade. Infractions in fourth and fifth grade would provide an individualized

baseline on misbehavior. Infraction rates after sixth grade would allow a check on whether the

elevated rate for the middle-school sixth graders is simply the result of problems resulting from a

transition to a new school, or rather sustained over time. As it turns out, we are only able to use

infractions data for the single academic year (2000-1), so it is not possible to follow the behavior

of individual students over time. 6 However, we are able to perform a pseudo-longitudinal

analysis of behavior based on the fact that our database, while only including one year of

infractions data, does include a number of years’ worth of data on other aspects of each student’s

career. In particular we know what sort of school the students who are in fourth or fifth grade in

2000-1 are destined to spend sixth grade, and we know in what sort of school older students in

that year did spend sixth grade. Using this information, we sort all students in grades 4-9 in

2000-1 into two groups, which we identify as 6Es and 6Ms. For example, a ninth grader is a

“6M” if she spent her sixth grade in middle school; a fourth grader is a “6E” if he subsequently

attends sixth grade in an elementary school.

Table 5 and Figure 4

Figure 4 graphs the trajectories for the two groups with respect to probability of an

infraction. These prevalence trajectories are computed for the same set of values for the

regression covariates, shown in Table 5; the difference in trajectories reflects the proportional

effect on the infraction probability estimated from the logistic regression, and the 95 percent

confidence interval represents the uncertainty in that estimate. 7 We see that in the baseline

6
Infractions data are available for later years, but changes in the reporting format of the data render it considerably
more difficult to match these reports to student records.
7
The values assumed for the covariates generally refer to an average male student. Parental
education is specified as high school graduate, and the race variable is 25 percent black (in line
with the sample). Any changes in these or other covariates would only serve to shift both lines

- 14 -
period, grades 4 and 5, 6Es actually have a slightly higher infraction rate than 6Ms. But a large

gap in the other direction opens up in sixth grade, when 6Ms have a much higher infraction rate

than 6Es. The gap narrows a bit through eighth grade, at which point both 6Es and 6Ms are

enrolled in middle school, a statistically significant gap persists as far as the ninth grade. We

found similar patterns in our analyses of violent and drug infractions. 8

These results do not rule out the logical possibility that the observed differences in sixth

grade are partly due to differences in school reporting practices rather than in the actual behavior

of the students. It seems reasonable to suppose that middle schools tend to be more formal and

severe than elementary schools, which might explain the infraction gap between 6Es and 6Ms in

sixth grade. However, it does not explain why that gap persists in seventh, eighth, and ninth

grades, when all the students have moved on past elementary school. Hence we believe that the

observed behavior gap is not an artifact of different school reporting practices.

End-of-Grade Test Scores

We performed a similar pseudo-longitudinal analysis of standardized end-of-grade

(EOG) test scores for math and reading, using the same sample and specification. The results are

of interest in their own right, and serve as a generalized check on the infractions results: In

particular, the EOG scores have the advantage over infraction data of not being influenced by the

standards or operating procedures of the school administration. Observed differences directly

reflect differences in student performance. 9

either up or down by the same proportional amount. The key is the significant difference across
the groups holding all else equal.
8
We do not trace this gap beyond 9th grade because students 16 years of age and older have the option of dropping
out of school. Infraction rates decrease dramatically after 9th grade, presumably because students with the worst
behavioral patterns are most likely to drop out.
9
Of course, this statement presumes that the school administrations do not exercise substantial
discretion in how the tests are administered and do not cheat in the scoring. In fact the

- 15 -
 The EOG tests have been required by the State Board of Education for all public-school

students in grades 3-8 since 1992-3. These multiple choice tests are administered during the

final weeks of the school year. For students in third, fifth, and eighth grades, adequate

performance on the reading and math EOGs is required for promotion.10 For analysis purposes,

we normalize the EOG test scores to have a mean of zero and a standard deviation of one.

As is evident in Table 6, the combination of matching and regression adjustments leaves

significant differences in EOG scores between fourth and fifth graders destined to attend sixth in

middle school, and those destined to attend sixth in elementary school. The former group has

higher scores in both math and reading. We are inclined to interpret these differences as

reflecting differences in characteristics of the students that are not captured by the covariates.

Note that the coefficients on the indicator for sixth grade in middle school drop to near zero in

sixth grade and thereafter. A natural interpretation of this difference-in-difference result is that

the 6Ms as a group tended to have an advantage (even after adjusting for a variety of measured

characteristics) that is evident in elementary school but negated by the “early” move to middle

school. The move to middle school not only leads to behavior problems but also reduced

academic performance.

Table 6

The advantage lost by 6Ms when they move into middle school is about 10% of a

standard deviation. The magnitude of this effect is substantial; the disadvantage associated with

moving to middle school in sixth grade is roughly equivalent to the disadvantage associated with

having an inexperienced rather than experienced teacher for a year (Clotfelter, Ladd and Vigdor

administration of EOG tests is tightly regulated in North Carolina and there is no evidence of
cheating, and has apparently occurred in other jurisdictions (Jacob and Levitt 2003: Cook 2003).
10
For details, see “Assessment Brief: Understanding North Carolina end-of-grade testing”
(March 1, 2004) at www.ncpublicschools.org/accountability/testing.

- 16 -
2006).

The fourth and fifth rows of Table 6 provide some suggestive evidence that the

immediate impact of attending middle school in sixth grade attenuates over time. The estimated

impact of sixth grade middle school attendance on 7th and 8th grade EOG scores is positive in

each instance, though never statistically significant. These coefficient estimates continue to

contrast with the 4th and 5th grade estimates, which are of larger magnitudes and statistically

significant. The difference between them continues to be a significant drop and point to a

conclusion that students who make the transition to middle school in sixth grade suffer long-term

academic as well as behavioral handicaps.

DISCUSSION

The causal mechanisms that account for the inter-grade patterns of infractions and EOG

scores cannot be identified directly from our data. Several differences between elementary and

middle school may be relevant. In comparison with elementary school, middle school provides

students more freedom and lacks the continuity and close connection provided by having one

primary teacher. Most obviously, middle school brings sixth graders into routine contact with

older adolescents who are likely to be a bad influence: older adolescents as a group are more

rebellious and more involved in delinquency, sex, illicit drugs, and other activities that violate

school rules. Of greatest concern is that the negative influence of middle school on sixth graders

appears to linger through ninth grade.

Our results complement the recent finding that school systems that move sixth grade

from elementary to middle school experience a 1-3 percent decline in on-time graduation rates

(Bedard & Do 2005). We conclude that placing sixth grade in middle school increases behavior

problems and reduces academic performance, both in sixth grade and subsequently. It is entirely

- 17 -
plausible that these effects could have the effect for some students of leading to retention in

grade or dropout. Together these findings cast serious doubt on the wisdom of the historic

nationwide shift to the middle school format.

Of course the results reported here are not based on random assignment, which leaves

open the possibility that the true causal process has not been adequately identified. (The

consistency and strength of the findings suggests otherwise.) It should also be noted that the

analysis is based on data that are limited in time and place, and in particular do not include any

large cities.

Decades ago the “middle school” movement was launched on the basis of plausible

speculations concerning potential benefits but without much direct evidence on the effects on

student behavior and performance. As it turns out, moving sixth grade out of elementary school

appears to have had substantial costs. The best school configuration in which to incorporate the

adolescent grades is now being reconsidered by policymakers and experts. Our results suggest

that the middle school configuration that brings seventh and eighth graders into regular contact

with sixth graders is problematic.

The implications of our research for the related debate over K-8 schools are less

certain. As a school moves from a K-5 to K-6 configuration, 6th graders get one more year of a

“childhood” culture. But when a school moves from K-5 to K-8, it exposes all the younger ages

to 7th and 8th graders who are entering adolescence. Whether the benefits to the 6th graders

would be offset by the exposure effects on younger students is an open empirical question.

- 18 -
Acknowledgements

We are grateful to Josh Kinsler for exceptional research assistance, to Camden Cook, Brad

McMillan, and Allison Whitaker for briefing us on the decisionmaking process that produces the

infractions data, and to Katherine Conner and participants of an APPAM conference session and

several university workshops for useful suggestions. This research is based on data from the

North Carolina Education Research Data Center at Duke University, directed by Dr. Elizabeth

Glennie and supported by the Spencer Foundation. We acknowledge the North Carolina

Department of Public Instruction for collecting and providing this information. This research is

supported by grants from the National Institute on Drug Abuse and the William T. Grant

Foundation. Any views expressed in this paper are the authors’ alone and should not be

associated with any affiliated institution.

- 19 -
References

Alspaugh, J.W. (1998). The Relationship of School to School Transitions and School Size to
High School Dropout Rates. The High School Journal, 81, 154-60.

Alspaugh, J.W. (2001). Achievement Loss Associated with the Transition to Middle School and
High School. The Journal of Educational Research, 92, 20-25.

Alspaugh, J.W. & R.D. Harting (1995). Transition Effects of School Grade-Level Organization
on Student Achievement. Journal of Research and Development in Education, 28, 145-49.

Barber, B.K. and J.A. Olsen. (2004). Assessing the Transitions to Middle and High School.
Journal of Adolescent Research, 19, 3-30.

Bedard, K. & C. Do (2005). Are middle schools more effective? The impact of school structure
on student outcomes. The Journal of Human Resources, 40(3), Summer, 660-682.

Cleveland, H.H. and R.P. Wiebe (2003). The Moderation of Adolescent-to-Peer Similarity in
Tobacco and Alcohol Use by School Levels of Substance Use . Child Development, 74, 279-91.

Clotfelter, C.T., H.F. Ladd & J.L. Vigdor (2006). Teacher-Student Matching and the Assessment
of Teacher Effectiveness. Journal of Human Resources 41, 778-820.

Cook, P.J. (2003). Comment on Catching Cheating Teachers. In W.G. Gale & J.R. Pack (Eds.),
Brookings-Wharton Papers on Urban Affairs 2003 (pp. 210-215). Washington, DC: Brookings
Institution Press.

Eccles, J.S., A. Wigfield, C. Midgley, D. Reuman, D. MacIver, & H. Feldlaufer. (1993).

Negative Effects of Traditional Middle Schools on Students’ Motivation. The Elementary School
Journal, 93, 553-74.

Elias, M.J., M. Gara & M. Ubriaco (1985). Sources of Stress and Support in Children’s
Transition to Middle School: An Empirical Analysis. Journal of Clinical Child Psychology, 14,
112-18.

Ellickson, P.L., D.F. McCaffrey, B. Ghosh-Dastidar, & D.L. Longshore (2003). New Inroads in
Preventing Adolescent Drug Use: Results From a Large-Scale Trial of Project ALERT in Middle
Schools. American Journal of Public Health, 93, 1830-36.

Goldin, C. (1999). A brief history of education in the United States. Cambridge, MA: NBER
Historical Paper No. 119.

Gootman, E. (2007a). Trying to find solutions in chaotic middle schools. The New York Times,

- 20 -
Jan. 3rd.

Gootman, E. (2007b). Taking middle schoolers out of the middle. The New York Times, Jan. 24.

Hanushek, E., S. Rivkin, & J. Kain (2004). Disruption versus Tiebout Improvement: The Costs
and Benefits of Switching Schools. Journal of Public Economics, 88, 1721-1746.

Hardy, C., W.M. Bukowski, & L.K. Sippola (2002). Stability and Change in Peer Relationships
During the Transition to Middle Level School. The Journal of Early Adolescence, 22, 117-42.

Hough, D. (1997). A Bona Fide Middle School: Programs, Policy, Practice, and Grade Span
Configurations. In J.L. Irvin (Ed.), What Current Research says to the Middle Level Practitioner
(pp.285-294). Westerville, OH: National Middle School Association.

Jacob, B.A. & S.D. Levitt (2003). Catching Cheating Teachers: The Results of an Unusual
Experiment in Implementing theory. In W.G. Gale & J.R. Pack (Eds.), Brookings-Wharton
Papers on Urban Affairs 2003 (pp. 185-220). Wash., DC: Brookings Institution Press.

Jacobson, M. (2004). Baby booms and drug busts: Trends in youth drug use in the United States,
1975-2000. Quarterly Journal of Economics, 119(4), 1481-1512.

Jang, S.J. (1999). Age-Varying Effects of Family, School and Peers on Delinquency: A
Multilevel Modeling Test of Interactional Theory. Criminology 37, 643-85.

Juvonen, J., V. Le, T. Kaganoff, C. Augustine & L. Constant. (2004). Focus on the Wonder
Years: Challenges Facing the American Middle School. Santa Monica: RAND Corporation.

Manski, C.F. (1993). Identification of Endogenous Social Effects: The Reflection Problem.
Review of Economic Studies, 60, 531-42.

Manquin, E. & R. Loeber (1996). Academic Performance and Delinquency. In M. Tonry (Ed.)
Crime and Justice: A Review of the Research (pp. 145-264). Chicago: University of Chicago
Press.

McEwin, K.C., M.W. Greene, & D.M. Jenkins (2001). Where do North Carolina’s Middle
Schools Stand in the 21st Century? A Status Repot on Programs and Practices, Pinehurst, NC:
NC Middle School Assn.

McLeod, J.D. & K. Kaiser (2004). Childhood Emotional and Behavioural Problems in
Educational Attainment. American Sociological Review, 69, 636-658.

Mills, R. (1998). Grouping students for instruction in middle schools. Eric Digest, Report: EDO-
PS-98-4.

National Middle School Association (1995). This We Believe: Developmentally Responsive

Middle Level Schools, National Middle School Association, Columbus, Ohio.

- 21 -
National Middle School Association (2001). Grade Configuration. NMSA Research Summary
#1 (updated Sept. 2001).

North Carolina Department of Public Instruction (2004). Last Best Chance 2004: Educating
Young Adolescents in the Twenty-First Century. Middle Grades Task Force Report. North
Carolina Department of Public Instruction and State Board of Education.

Reinke, W.M. & H.M. Walker (2006). Deviant peer effects in education in K.A. Dodge, T.J.
Dishion, & J.E. Lansford (Eds.) Deviant Peer Influences in Programs for Youth (pp. 122-140).
New York: The Guilford Press.

Rudolph, K.D., S.F. Lambert, A.G. Clark, & K.D. Kurlakowsky (2001). Negotiating the
Transition to Middle School: The Role of Self-Regulatory Processes. Child Development, 72,
929-46.

Rumberger, R.W. (1995). Dropping Out of Middle School: A Multilevel Analysis of Students
and Schools. American Educational Research Journal, 32, 583-625.

Sampson, R.J., & J.H. Laub. (1992). Crime and Deviance in the Life Course. Annual Review of
Sociology, 18, 63-84.

Simmons, R.O. & D.A. Blythe (1987). Moving into Adolescence. New York: Aldin De Gruyter.

South, S.J. & S.F. Messner. (2000). Crime and Demography: Multiple Linkages, Reciprocal
Relations. Annual Review of Sociology, 26, 83-106.

Teitler, J. & C. Weiss. (2000). Effects of Neighborhood and School Environments on Transitions
to First Sexual Intercourse. Sociology of Education, 73, 112-32.

Valentine, J.W., D. Clark, D. Hackmann, & V. Petzo. (2002). Leadership in middle level schools,
volume I: A national study of middle level leaders and school principals. Reston, VA: National
Association of Secondary School Principals.

Weiss, C.C. & L. Knipes. (2006). Reexamining Middle School Effects: A Comparison of
Middle Grades Students in Middle Schools and K-8 Schools. American Journal of Education,
112, 239-272.

Zernicke, K. (2007). The preteen: Betwixt and bedeviled. The New York Times, Jan. 7.

- 22 -
Table 1
Logistic regression results for matching procedure
Dependent variable: Indicator for 6th grade in Middle School
Coefficient estimates and standard error estimates
Coefficient. SE
School level
% black 2.4263 0.9911

% Hispanic 9.5101 4.2418

% parents without HS diploma -1.8795 2.0366

% students receiving free or reduced lunch -1.0177 1.4583

District level
Number of 6th graders 0.0012 0.0003

5-year growth rate in number of 6th graders -0.2281 1.1842

% rural -0.0789 0.7405

Per-pupil expenditure, local ($000) 0.0026 0.0007

Per-pupil expenditure, federal ($000) 0.0007 0.0013

Constant -3.4368 1.6204

N 342
Pseudo-R2 0.25

- 23 -
Table 2
Summary Statistics for 6th Graders, 2000/2001
North Carolina Public Schools in Matched Sample
Middle School Elementary Difference
School
Number of students 39,596 5,113
Individual level variables
% Male 50.8 50.7 0.1

Race
% White 67.6 65.8 1.8
% Black 25.1 23.5 1.7
% Hispanic 3.9 3.0 1.0
% Asian 1.2 2.1 -1.0
% other 2.2 5.6 -3.5

Parent’s education
% High school grad 46.5 48.6 -2.2
% 2-year college grad 19.0 19.7 -0.7
% 4-year college grad 19.0 15.1 3.9

% Reduced/free lunch 42.9 49.9 -7.0

Avg. math EOG score 159.7 158.9 0.8

Avg. Reading EOG score 155.2 154.9 0.3
School level variables

Number of grades 3.0 6.3 -3.3

Number of 6th graders 257 117 139
District level variables
Per-pupil expenditure local* $1,271 $1,146 $125
Per-pupil expenditure federal $530 $574 -$ 44

% rural 47.5 53.6 -6.1

Number of 6th graders 1045 781 264

5-year growth rate in number (%) 19.2 15.7 3.4
*The bulk of expenditures are state funds

- 24 -
Table 3
Infraction rates for Sixth Graders in Matched Sample, 2000-2001

Middle School Elementary Middle Elementary

Incidence School School school
Incidence Prevalence* Prevalence*
Overall 0.476 0.161 0.175 0.085
(count) (18,833) (824) (6,943) (437)
Violence 0.130 0.057 0.087 0.044

Drug 0.002 0.001 0.002 0.001

Weapon 0.003 0.001 0.003 0.001

possession
Truancy 0.006 0.0002 0.005 0.0002

Rowdy behavior 0.147 0.064 0.066 0.034

Minor 0.170 0.034 0.088 0.026

Property 0.011 0.003 0.010 0.003

Sexual 0.002 0.0004 0.002 0.0004

*At least one infraction during the school year

- 25 -
Table 4
The effect of school configuration on infractions
Matched sample, North Carolina 6th graders, 2000-1
Coefficient estimates (Standard Error estimates)
1. Any 2. Violent 3. Drug 4. Number of
infraction infraction infraction infractions
Logit Logit Logit Negative
binomial
In middle school 0.799 0.730 1.330 0.919
(0.195) (0.189) (0.654) (0.210)
Male 1.122 1.231 0.955 1.247
(0.043) (0.052) (0.279) (0.042)
Race (white omitted)
Black 0.632 0.658 -0.602 0.619
(0.050) (0.070) (0.358) (0.057)

Hispanic -0.431 -0.478 0.045 -0.480

(0.081) (0.103) (0.488) (0.092)

Asian -1.379 -1.155 -1.444

(0.245) (0.278) (0.252)

Other 0.147 0.289 -0.409 0.178

(0.115) (0.150) (0.613) (0.108)
Parent’s education
(High school grad
omitted)
High school dropout 0.318 0.298 0.734 0.306
(0.044) (0.051) (0.309) (0.041)

Trade school -0.199 -0.199 -0.154

(0.085) (0.103) (0.121)

Community college -0.143 -0.132 -0.214 -0.222

(0.059) (0.074) (0.431) (0.065)

4-year college -0.489 -0.640 -0.923 -0.563

(0.067) (0.081) (0.450) (0.067)

Graduate degree -0.865 -1.048 -0.923

(0.156) (0.207) (0.179)
Reduce/free lunch 0.436 0.408 0.505 0.499
(0.043) (0.050) (0.263) (0.044)

Old for grade 0.372 0.329 0.535 0.406

(0.044) (0.054) (0.200) (0.047)

- 26 -
Math EOG score, 5th -0.211 -0.203 -0.312 -0.259
grade (0.029) (0.038) (0.157) (0.032)

Reading EOG score, -0.204 -0.141 -0.158 -0.192

5th grade (0.028) (0.034) (0.143) (0.028)
School-level variables
% reduced/free lunch -0.615 0.050 -1.235 -0.866
(0.632) (0.561) (1.777) (0.695)

% black 0.850 0.831 -1.109 1.160

(0.460) (0.407) (2.076) (0.528)

% Hispanic -0.527 -2.578 -2.774 -0.254

(1.852) (1.584) (5.094) (1.653)

% parents without HS -0.300 0.467 0.453 -0.587

diploma (1.042) (0.984) (3.896) (1.195)

District level
Number of 6th graders -0.053 -0.068 -0.174 -0.023
(0.112) (0.101) (0.305) (0.100)

Per-pupil expenditure, 0.572 0.616 -0.052 0.408

local (0.237) (0.231) (0.602) (0.272)

Per-pupil expenditure, -0.356 -1.066 0.685 -0.532

federal 0.488 (0.541) (1.343) (0.539)
Constant -3.806 -4.589 -7.670 -3.214
0.562 (0.527) (1.513) (0.593)
Sample size 44,709 44,709 40,715 44,709

Note: bold font indicates that the coefficient estimate is significantly different from zero, p<.05.
All standard errors are cluster corrected by school.

- 27 -
Table 5

Pseudo-Longitudinal analysis of school configuration on infractions

Matched sample, North Carolina, 2000-2001
OLS Coefficient estimates (SE estimates)
Any Infraction Violent Infraction Drug Infraction
Logit Logit Logit
Fourth Graders: will -0.446 -0.606 -1.211
attend 6th grade in middle (0.190) (0.177) (0.927)
school
Fifth graders: will attend -0.223 -0.381 -0.415
th
6 grade in middle school (0.156) (0.170) (0.688)
Sixth graders: attending 0.799 0.730 1.330
th
6 grade in middle school (0.195) (0.189) (0.654)
Seventh graders: attended 0.370 0.199 -0.482
th
6 grade in middle school (0.166) (0.139) (0.248)
Eighth graders: attended 0.246 0.078 0.307
th
6 grade in middle school (0.162) (0.142) (0.267)
Ninth graders: attended 6th 0.412 -0.022 0.810
grade in middle school (0.178) (0.204) (0.324)
Note: Each cell includes the coefficient and standard error (cluster corrected by school) from a
separate regression. The coefficient in each cell is the estimated effect of an indicator of whether
the student will attend sixth grade in middle school (for the fourth and fifth graders), is attending
sixth grade in middle school, or did attend sixth grade in middle school (seventh and eighth
grade); in every case the contrast is with attending sixth grade in elementary school.
Note: bold font indicates that the coefficient estimate is significantly different from zero, p<.05.
All standard errors are cluster corrected by school.

- 28 -
Table 6
Pseudo-Longitudinal analysis of school configuration on EOG scores
Matched sample, North Carolina, 2000-1
OLS Coefficient estimates (Standard Error estimates)
EOG math EOG Reading
Fourth Graders: will attend 0.155 0.123
6th grade in middle school (0.031) (0.030)

Fifth graders: will attend 6th 0.107 0.090

grade in middle school (0.035) (0.026)

Sixth graders: attending 6th 0.000 -0.041

grade in middle school (0.041) (0.033)

Seventh graders: attended 0.028 0.020

6th grade in middle school (0.042) (0.038)

Eighth graders: attended 6th 0.082 0.025

grade in middle school (0.049) (0.033)

Note: Each cell includes the coefficient and standard error (cluster corrected by school) from a
separate regression. The regression specifications are the same as those in Table 4, except that
the “EOG scores in 5th grade” are dropped. The coefficient in each cell is the estimated effect of
an indicator of whether the student will attend sixth grade in middle school (for the fourth and
fifth graders), is attending sixth grade in middle school, or did attend sixth grade in middle
school (seventh and eighth grade); in every case the contrast is with attending sixth grade in
elementary school.
Note: bold font indicates that the coefficient estimate is significantly different from zero, p<.05.
All standard errors are cluster corrected by school.

- 29 -
Figure 1. % 6th Graders Attending Middle Schools - NC & National Trends

100%

80%
Percentage

60%

NC
40% National

20%

0%
70

80

90

00
19

19

19

20
Year

- 30 -
Figure 2. Predicted Probability of 6th Grade in Middle School (vs. Elementary School)

0.6

0.5
Proportion of Schools

0.4

0.3
Elementary
Middle
0.2

0.1

0
1 2 3 4 5 6 7 8 9 10

Probability Deciles

Source: Imputed values using equation from Table 1.

- 31 -
Figure 3. Infraction Prevalence - 6th Grade

Sexual

Drug

Weapon Possession

Truancy

Property Elementary
Middle
Rowdy

Violence

Minor

0 0.02 0.04 0.06 0.08 0.1

Prevalence

Source: See Table 3. Note that some students appear in more than one category.

- 32 -
Figure 4. Probability of at Least One Infraction

0.35

0.30

0.25
Imputed Probability

Attended 6th
0.20 in Elementary
Attended 6th
in Middle
0.15
95% CI

0.10

0.05

0.00
4 5 6 7 8 9
Grade

Source: See Table 4. All points are fitted values from equation (1) for a male student with
average characteristics. (See text for additional detail.)

- 33 -