Are you knowledgeabe on Advance Qualitative Research?

S

A a

b

c

a

A R R A A

K S C J

1

c w t a p c c a i o

s l f p S b i w (

N

s

h 0

International Review of Law and Economics 43 (2015) 98–106

Contents lists available at ScienceDirect

International Review of Law and Economics

chool suspension and the school-to-prison pipeline

lison Evans Cuellar a,c,∗, Sara Markowitz b,c

George Mason University, 4400 University Drive, Northeast Module I, MS 1J3, Fairfax, VA 22030, United States Department of Economics, Emory University, 1602 Fishburne Drive, Atlanta, GA 30322-2240, United States NBER, United States

r t i c l e i n f o

rticle history: eceived 4 March 2015 eceived in revised form 26 May 2015 ccepted 9 June 2015 vailable online 17 June 2015

a b s t r a c t

Schools have many available strategies to address problem behavior among students. One option increas- ingly used by schools is to suspend problem youth and remove them for defined periods. The purpose of this paper is to investigate whether this type of disciplinary policy has unintended consequences by giving problem youth greater opportunity to commit crimes outside of school. Previous studies have

eywords: chool discipline rime

uveniles

looked at the “incapacitation” effect of school holidays and teacher strike days, but these studies do not directly address the relevant school policy decisions. The current study relies on administrative data from a school district and a juvenile justice system. The results indicate that out-of-school suspension may increase criminal offending behavior by problem youth, more than doubling the probability of arrest. The effect is particularly large among African American youth, relative to Whites.

. Introduction

Over the past few decades, school districts all across the ountry have adopted “zero tolerance” disciplinary policies as a ay to reduce violence on campus, protect students, and main-

ain environments conducive to learning. Zero tolerance policies utomatically impose punishments on students and mandate sus- ension or expulsion from school for certain offenses, often without onsideration of the circumstances. At their inception, these poli- ies pertained to only the most serious offenses such as bringing

weapon to school, but over time, the polices have expanded to nclude lesser infractions such as alcohol or tobacco use, fighting, r swearing (Kang-Brown et al., 2013).

The trend toward adopting zero tolerance discipline began hortly after the enactment of the 1994 Gun Free Schools Act. The aw requires that every school district receiving federal education unds implement a 1-year mandatory expulsion for students who ossess a firearm on school grounds. Scholars cite the Gun Free chools Act as the catalyst for the widespread adoption of broad ased zero tolerance discipline policies covering a wide variety of

nfractions, with all 50 states adopting some variation of the policy ithin a few years of the enactment of the Gun Free Schools Act

Henault, 2001).

∗ Corresponding author at: George Mason University, 4400 University Drive, ortheast Module I, MS 1J3, Fairfax, VA 22030, United States.

E-mail addresses: [email protected] (A.E. Cuellar), [email protected] (S. Markowitz).

ttp://dx.doi.org/10.1016/j.irle.2015.06.001 144-8188/© 2015 Elsevier Inc. All rights reserved.

© 2015 Elsevier Inc. All rights reserved.

Despite their popularity, zero tolerance policies are not without controversy. Proponents argue they establish expectations around pro-social conduct by specifying rules prohibiting certain behav- ior. Consequently, they may deter undesirable and unsafe behavior by increasing its expected costs relative to its expected bene- fits. The deterrence literature in general, however, underscores that the certainty of punishment has greater deterrent effect than the severity of punishment (Cook, 1980; Nagin, 1998). Expulsion and suspension also may prevent a “contagion” effect on peers if delinquents are removed from the classroom. Teachers report that student behavior problems in class significantly impact their abil- ity to instruct (U.S. Department of Education, 2000); thus, removal of disruptive youth may improve peer outcomes (Lazear, 2001). For example, a study of suspension in middle school found that disruptive behavior reduces academic achievement for the gen- eral student population, implying that suspension might improve overall academic performance (Kinsler, 2011).

Yet other research point to potential downsides: exclusionary discipline may lead to worse educational outcomes for the excluded student, including loss of educational opportunities, poor school performance, and dropping out, which further jeopardizes youth human capital accumulation (American Academy on Pediatrics, 2013). In addition, the heavy reliance on zero tolerance policies have been blamed for contributing to the so-called “school-to- prison pipeline.” This pipeline, “refers to the policies and practices

that push our nation’s school children. . . out of the classrooms and into the juvenile and criminal justice systems” (ACLU, 2014). The most direct pathway is through zero tolerance policies that man- date the referral of students directly to law enforcement authorities

eview

f m v t f i s i p d (

a s A P r fl a i r s S e r e d

c p m s “ u a t r c 8 s s U w s a

i s d b U b l v b d e

2

t c p o

A.E. Cuellar, S. Markowitz / International R

or violation of school policies (Wald and Losen, 2007). Suspension ay also lead to days spent in the community with reduced super-

ision and increased opportunities to commit crimes. Over a longer ime period, indirect pathways are also pertinent. Being removed rom school may adversely impact students’ school connectedness, ncrease alienation, intensify conflict with adults, reduce supervi- ion in the community, and increase youth’s propensity to engage n delinquent behaviors (Skiba et al., 2006). Research has linked sus- ension with the likelihood of dropping out (Cook et al., 2010), and ropping out has been linked with engaging in criminal activities Anderson, 2014).

A number of organizations have been critical of zero toler- nce policies and the criminalization of in-school offenses. Groups uch as the U.S. Department of Education, the American Bar ssociation, the American Academy of Pediatrics, and the American sychological Association have issued policy statements urging eform of the zero tolerance policies to allow for much more exibility in applying punishments. Yet there is surprisingly little cademic research on the effects of zero tolerance policies specif- cally, and school suspension and expulsion more generally. Some esearchers have observed a negative relationship between suspen- ion and student academic achievement (Raffaele Mendez, 2003; kiba et al., 2002; Kinsler, 2013; McFadden and Marsh, 1992). Oth- rs have found that suspension is a predictor of future suspensions, ather than acting as a deterrent (Raffaele Mendez, 2003; Tobin t al., 1996), although in a recent study, Kinsler (2013) shows evi- ence of reduced in-school infractions following suspensions.

In a review of the literature, the American Psychological Asso- iation Zero Tolerance Task force concludes that zero tolerance olicies have not been effective in generating safer school environ- ents, however this conclusion is primarily based on correlational

tudies (Skiba et al., 2006). This same report further concludes, . . .the school-to-prison pipeline has not yet been conclusively doc- mented. While retrospective and some correlational data suggest

relationship between suspension and expulsion and juvenile jus- ice outcomes, it is important to note that available research on elationships between school expulsion and juvenile justice out- omes are at this point primarily descriptive” (Skiba et al., 2006, pp 0). Our paper seeks to fill this gap in the literature. Using data from tudents in middle and high school we examine the effect of out of chool suspension on youth referrals to the juvenile justice system. sing a difference-in-difference technique we find that students ho are suspended are more likely to commit crimes on the actual

uspension days than on non-suspension days, including weekends nd holidays.

Previous studies from the economics literature provides some nsight into the question, while not directly estimating the effects of uspension on criminal activities. Luallen (2006) uses teacher strike ays for identification and finds that total juvenile crime increases y an average of 21.4% on strike days, but violent crime declines. sing teacher in-service days, Jacob and Lefgren (2003) find that eing out of school is associated with greater property crime, but

ower violent crime. Both of these incapacitation studies examine ery short term outcomes pertinent to large student populations, ut do not address the direct policy question around problem stu- ents who face potential disciplinary action. Our study directly xamines the population at-risk for the school-to-prison pipeline.

. Analytical framework and data

Applying Becker’s economic model of crime (1968), youth weigh

he expected payoffs from criminal activity against the expected osts from the probability of being caught and the severity of unishment. There are many possible channels through which out- f-school suspensions may influence decisions of youth, including

of Law and Economics 43 (2015) 98–106 99

directly through time constraints, deterrence, sanctions, and avail- able peers, and indirectly through changes in peer behavior. Suspension as a punishment is quite different from incarcera- tion in that suspension removes students from structured school supervision, a form of incapacitation, and places them under poten- tially less structured parental supervision. Depending on the degree of supervision at home, youth may have more opportunity to engage in crimes in the community. In addition to changing youth’s time constraints, suspension also changes the peer group that is available for a youth to associate with on school days relative to weekends and holidays. The current analysis takes a reduced form approach in that it captures the total net effect of suspension poli- cies on youth crime in the community.

The challenge in comparing outcomes of youth who are sus- pended to outcomes of youth who are not, is that suspension is not likely to be randomly applied. Those who are suspended are more likely to engage in risky problem behavior than those who are not. To help avoid the selection problem, our analysis is limited to youth who are suspended at any point during the study period. Outcomes are compared between periods when students are suspended and periods when they are not. The assumption is that on a regular school day, youth who have been suspended have more opportu- nity to engage in crime than youth who have not been suspended. Further, the opportunity to engage in crime is equal for suspended and non-suspended youth on non-school days, such as weekends and holidays. On school days the difference between in-school and out-of-school suspended youth reflects the increased opportunity that out-of-school suspension provides to youth to engage in crime. The effects, estimated at the individual level, are the impact of being suspended out of school on the probability of offending on that day.

In order to perform detailed analyses of how school suspen- sion interacts with juvenile justice, data were received from two sources, an urban school district and a county juvenile justice sys- tem. The school district is geographically fully contained within the county although the county contains more than one school district. This allows us to measure crimes committed outside the school dis- trict, but not outside the county. Person-level administrative data from both sources are linked using name and date of birth in a procedure described below.

The school data capture all enrolled youth in the urban school district for the 2002 through 2009 academic years, including enroll- ment dates and exit dates (for example, for youth who graduate, drop out, or leave the district). The study is limited to students who are age 13–17 at the beginning of each academic year. Youth are excluded after they turn age 18 because they are no longer handled in the local juvenile court and their offenses would not be recorded in the data. The study includes youth only on days when they are enrolled in school and excludes all days during summer breaks.

Importantly for this study, the school data includes whether a youth received a disciplinary action and if so, the start and end dates and whether the suspension was in-school or out-of school. The analysis takes each day and observes whether a given student was suspended out of school or not. Because the goal is to exam- ine the impact of suspension among similar students, the study is narrowed to the group of students who were ever suspended, either in-school or out-of-school, during the study period. Conse- quently, the study takes all suspended students and compares their outcomes on days when they are suspended out of school to days when they are not suspended out of school. Among 4665 total stu- dents in the study, all were suspended, but the probability of being suspended on any given day during the school year is small, 0.14%.

In addition to suspension status, the school data also indicate a student’s gender, their race or ethnicity, the date they enrolled in school, their primary language, and whether they were identi- fied for English as a Second Language (ESL) instruction or special

1 eview

e a

r a n o b t t i t w T m p i

b c d ( 2 u a n a i

2

a e m t w a t u s r

2

f t a b i t t a e p i (

( A t t a a d

00 A.E. Cuellar, S. Markowitz / International R

ducation. These characteristics are all included as covariates in the nalyses as described below.

The school data were merged with county juvenile justice refer- al data for 2002–2009. A “referral” is the juvenile justice system’s lternative to an adult arrest. Typically, the first step in the juve- ile justice process is for a youth to encounter a law enforcement fficial, although some may be referred by parents, schools, or pro- ation officers (Snyder and Sickmund, 2006). If an arrest is made, he youth is “referred” to the juvenile court for a decision whether he youth should be detained and charged, released, or transferred nto another youth program. When a juvenile court case reaches he juvenile probation department, an intake officer will decide hether to dismiss it, handle it informally, or hear it formally.

hus, the data include all referrals whether they are handled for- ally or informally, however, they do not include encounters with

olice officers that do not lead to a referral. The juvenile justice data nclude referral dates, offense dates, and the offense type.

The school and referral data were matched based on name and irthdate. There are various challenges with name matching that an make it difficult to accurately link data, such as misspellings, ouble entries, inconsistencies with hyphens and middle names Trajtenberg et al., 2006; Feitelson, 2004; Raffo and Lhuillery, 009). As a result we converted names to sound codex groupings sing methods detailed in Trajtenberg et al. (2006) and further ugmented with birth year. The merged analytic data were orga- ized such that each observation presents a person-day. There re 3,913,192 person-day observations in the base model which ncludes weekends and holidays but excludes summer breaks.

.1. Dependent variables

Because the data include detailed enrollment information, we re able to construct daily observations for each day a child was nrolled in school. For each school day it is possible to deter- ine whether or not a youth had committed an offense that lead

o a referral. We constructed two dependent variables. The first as an indicator of whether a given youth committed any offense

s recorded in the referral data and the second was an indica- or of whether a referred youth committed a felony offense. We se definitions provided by the state juvenile offender sentencing tandards. The students in our analysis sample had a total of 2049 eferrals over the 2002–2009 period.

.2. Independent variables

In the analysis the key independent variables include indicators or whether a youth was in suspension status on a given day. Indica- ors were created for whether a youth was suspended out of school nd whether the day was a school day or a weekend or holiday ased on academic calendars provided by the school district. The

ndicators for suspension and school day are interacted to measure he effect of being out of school on a regular school day. Indica- ors also were included for day of week (Monday through Friday) nd weekend and school holidays as prior research finds differ- nces in crime by day of week (Jacob and Lefgren, 2003). Because rior research shows that school characteristics and school policies

nfluence crime rates, the analysis also includes school fixed effects Gottfredson and DiPietro, 2011; Cullen et al., 2006).

Other independent variables include indicators for gender, race White which is the omitted category, African American, Hispanic, sian, and other), primary language (English which is the omit-

ed category, Spanish, and other) and age. The school data include

he youth’s birth month and birth year, but not birth day. Thus, ge is determined assuming the birth day is the 15th of the month nd is calculated as of the first day of the academic year. Indepen- ent variables also include whether a youth received English as

of Law and Economics 43 (2015) 98–106

a Second Language (ESL) services or special education in order to control for the likelihood of school achievement which in turn may reflect differences in individual’s opportunity costs from engaging in crime. From the school and justice referral data it is possible to create measures of past behavior problems. For each youth the number of suspensions and offenses in the prior academic year also is calculated.

A set of models is estimated to measure the individual prob- ability that a youth will commit an offense on a given day as a function of suspension, school, and demographic variables using the following equation:

The basic model utilizes a difference-in-difference (D–D) frame- work. Students are either on out-of-school suspension or not, and they are being observed either on a school day or weekend/holiday. The D–D model tests for the effects on crime of being suspended on a school day as compared to a weekend or holiday, net of the time-of-week-based difference in crime for non-suspended stu- dents. This model is advantageous because it uses a group of similar students and compares their criminal activities under different cir- cumstances. By doing so, we can minimize the influence of some of the underlying trends common to this group of at-risk students and identify a clean effect of being suspended on crime. In other words, this model allows us to attribute the crime observed on school days to the suspension, rather than having it be confounded with a more general pattern of problem behavior that may occur at any time. The model is specified as follows:

Oit = ˇ1 School Dayt + ˇ2 Suspendit + ˇ3 (School Dayt × Suspend Out of Schoolit) + � ′Xit + ı′DOWt + Tt+Ss

In this equation, O is a dichotomous indicator of whether or not a student (i) committed a referred offense on a given day (t). School Day indicates whether the day is a school day as compared to a weekend or holiday, Suspend whether the youth was suspended on a given day. The interaction of School Day and out-of-school suspension gives the differential effect of being suspended out- of-school on a school day. The vector X represents demographics, school services, and past behavior problems measured at the indi- vidual level. The vector DOW represents indicators for days of week (Monday through Friday), while T and S represent academic year and school fixed effects. Models are estimated with a logit regres- sion and standard errors are clustered at the individual level.

One main concern with the estimation of the offense equa- tion is the possible endogeneity of suspension. There are likely unobserved, individual level factors that are correlated with both suspension (or the behavior that prompted the suspension) and the likelihood of committing offenses. For example, troubled chil- dren in general may be more likely to misbehave both in and out of school. We address this concern in a few ways. We try to minimize the unobserved factors in the error term by including indicator vari- ables for the presence of prior offenses and suspensions. We also show some estimates using the Chamberlain (1980) conditional fixed effects logit model. The problem with fixed effects, however, is that these account for time invariant factors, which may not be a useful assumption given the age and situation of our sample of at-risk students. Also, the procedure only uses observations that exhibit variation in the dependent variable. This reduces our sample dramatically from 4665 to 1128 observations.

A second threat to the validity of our study concern regards

closely connected events and/or reverse causality. Disagreements with peers could be closely connected and extend from school to non-school settings. For example, it is possible that a student was arrested for a crime and subsequently suspended or one event could

A.E. Cuellar, S. Markowitz / International Review

Table 1 Descriptive statistics, individual-levela .

Percent of youth

Referred to juvenile justice during observation period 24.2 Referred to juvenile justice for a felony offense during

observation period 7.8

Suspended out of school during observation period 95.5 Male 65.3 Female 34.7 White 36.5 African American 21.8 Hispanic 23.6 Asian 14.2 ESL instruction 12.6 Special education 18.0 Spanish as primary language 14.1 Other primary language 7.6

p T

l d i n p

t w a e d s w p o e r

3

R s a A g E s J A 9 s a s

s s r 1 a s s b e i i

a All youth in the analysis were suspended at some point during the observation eriod. otal sample includes 4665 youth.

ead to both suspension and arrest. Unfortunately, the school data o not capture the reason for each suspension or the date of the

nfraction that lead to suspension, however, the suspension would eed to occur on the same day as the arrest to pose an empirical roblem.

To help alleviate concerns of timing, we show models that use he lags and leads of the suspension days in order to examine hether the days right before or right after are also associated with

changing probability of committing offenses. Results indicating ffects of the lag or lead on the probability of offenses would cast oubt on any causal story linking suspension to crime, while results howing no effects would support a causal relationship. In addition, e limit the analysis to 6 days pre and 6 days postsuspension, com- aring the effect of being in or out of school within a narrow range f days. This reduces the concern that any observed smaller differ- nce on weekends occurs because weekend days tend to be further emoved from the first day of suspension.

. Results

Table 1 provides descriptive statistics for students in the study. ecall that the sample is all students who ever receive a suspen- ion. The sample is predominantly male (65%). Thirty-seven percent re white, followed by Hispanic (24%), African American (22%) and sian (14%). For 14% of youth Spanish was the primary spoken lan- uage. Thirteen percent of the youth received special instruction in nglish as a Second Language and 18% received special education ervices. Of all youth in the study, 24.2% were referred to Juvenile ustice over the study period at least once, 7.8% for a felony offense. ll youth were suspended at least once during the study period and 5% were suspended out of school at least once. The length of their uspensions is skewed. Among the suspensions 60% were for 1 day nd 90% were for 7 days or less (data not shown). Of a total of 14,054 uspension events, 277 were for more than 30 days.

Table 2 investigates the impact of school out-of-school suspen- ion on the probability of committing an offense. The first two pecifications (columns 1 and 2) includes youth ages 13–17. The esults in column (2) use the same empirical model as in column

but also include school fixed effects. The results do not change ppreciably when the school fixed effects are included. The results how that youth who are suspended out of school on days when chool is in session have a statistically significantly higher proba-

ility of committing an offense than youth who are not. Marginal ffects are calculated for the interaction term, and show an increase n the probability of offending on any given day of 0.07% points, ncreasing from a probability of 0.05–0.12%. This is a significantly

of Law and Economics 43 (2015) 98–106 101

larger increase than 13% increase in property offenses reported by Jacob and Lefgren (2003) in their study of teacher in-service days. However, the current sample is by definition highly disruptive and has a high arrest rate overall. The results also show that school days are associated with higher probabilities of offense than on non- school days on average which is consistent with prior literature.

With respect to youth characteristics, the results show that older ages are positively associated with referral although being age 17 has a negative effect relative to younger ages, possibly because not all offenses for 17-year olds appear in the data. As documented in most studies of crime, males have higher probability of offending than females, as do African Americans and Asians relative to whites. Being a primarily Spanish-speaker or other non-English speaker is associated with lower probabilities of offending. This is consis- tent with literature on newer immigrants who may not yet have assimilated. Receiving ESL services is not associated separately with offending. Receipt of special education services is associated with a higher probability of offending. Consistent with other research, the analysis finds that offending is more likely on Fridays than other weekdays but the coefficient is not statistically significant.

Column (3) in Table 2 limits the analysis to youth who are ages 15 and over and shows similarly that out-of-school suspension on school days is associated with increased offending. Among the older youth the marginal effect of being suspended out of school is to increase the probability of offending from 0.05% to 0.12%. As in columns 1 and 2, a higher probability of offending is associated with being male, African American, and Asian, however, the effects of special education are no longer significant. As in columns 1 and 2 being a primary non-English speaker is protective, but the effect of being a Spanish speaker is not significant.

Columns 4 through 6 examine the impact of suspension by race and ethnicity. Column (4) estimates the model for African American youth only. The effect of suspension is larger; the increase in the offense probability from being suspended out of school on a school day rises by 0.11% points, from 0.08% to 0.19%. For Hispanic youth (column 5) and Asian youth (column 6) the effect is smaller and not significant. The effect of other individual characteristics is similar across models.

In Table 3, the rich school and referral data are used to cre- ate additional controls for underlying youth characteristics. The models include two measures of prior behavior, both measured in the previous academic year. The first is the total number of offenses committed in the prior academic year and the second is the total number of suspensions in the prior academic year. This is our preferred model although it, necessarily, reduces the sample size because the analysis does not have data prior to 2002. Col- umn 1 shows the effect for the full sample and column 2 shows the effect for older youth while columns 3 through 5 show results by race and ethnicity. The number of offenses in the prior aca- demic year has a positive and significant effect on offending, while the prior number of suspensions is not significant. Controlling for these prior behaviors, the impact of out-of-school suspensions on school days remains positive and significant for all youth (column 1). The marginal effect of out-of-school suspension on school days increases, increasing the offense probability from 0.05% to 0.14%. For older youth the marginal effect is an increase from 0.05% to 0.13%. For African American youth the effect is even greater. Their offense probability rises to 0.25%. For Hispanic and Asian youth the effect remains not significant.

Table 4 shows the effect of out-of-school suspension on felonies only, controlling for past behavior. Column (1) represents the model with all youth, column (2) includes only older youth, while columns

(3) through (5) show results by race and ethnicity. School fixed effects are included in all models. The results are different from that of all types of offenses, in that being suspended in general is asso- ciated with an increase in the probability of felony offense, but the

102 A.E. Cuellar, S. Markowitz / International Review of Law and Economics 43 (2015) 98–106

Table 2 All juvenile justice referrals.

(1) (2) (3) (4) (5) (6) Age 13–17 Age 13–17 Age 15–17 African American only Hispanic only Asian only

Out-of-school suspension × school day

0.87 0.87 0.89 0.83 0.38 0.58 (0.21) (0.21) (0.30) (0.42) (0.36) (0.58) 0.00 0.00 0.00 0.05 0.30 0.31

Suspension 1.46 1.47 1.34 1.11 2.07 1.86 (0.21) (0.21) (0.30) (0.40) (0.34) (0.53) 0.00 0.00 0.00 0.05 0.00 0.00

School day 0.15 0.15 0.11 0.26 0.38 0.11 (0.08) (0.08) (0.10) (0.13) (0.22) (0.22) 0.05 0.05 0.29 0.04 0.09 0.60

Age 14 0.32 0.52 0.49 0.45 0.78 (0.09) (0.10) (0.17) (0.20) (0.23) 0.00 0.00 0.00 0.02 0.00

Age 15 0.28 0.53 0.43 0.46 0.73 (0.09) (0.11) (0.20) (0.23) (0.26) 0.00 0.00 0.03 0.04 0.00

Age 16 0.17 0.40 −0.14 0.27 0.65 0.55 (0.10) (0.12) (0.09) (0.23) (0.25) (0.29) 0.09 0.00 0.12 0.22 0.0 0.06

Age 17 −0.40 −0.19 −0.72 −0.40 −0.08 −0.47 (0.12) (0.14) (0.11) (0.27) (0.29) (0.37) 0.00 0.18 0.00 0.14 0.79 0.20

ESL −0.002 −0.01 0.14 −0.28 −0.04 0.11 (0.14) (0.14) (0.19) (0.52) (0.19) (0.25) 0.99 0.96 0.48 0.58 0.84 0.65

Special education 0.26 0.20 0.19 0.08 0.54 −0.16 (0.09) (0.09) (0.13) (0.14) (0.26) (0.26) 0.01 0.03 0.13 0.55 0.04 0.54

Male 0.45 0.46 0.71 0.27 0.61 0.52 (0.08) (0.08) (0.10) (0.15) (0.18) (0.20) 0.00 0.00 0.00 0.07 0.00 0.01

Hispanic 0.05 0.002 −0.004 (0.15) (0.15) (0.22) 0.72 0.94 0.98

African American 0.64 0.62 0.70 (0.09) (0.09) (0.11) 0.00 0.00 0.00

Asian 0.23 0.24 0.36 (0.12) (0.12) (0.15) 0.06 0.06 0.02

Spanish −0.29 −0.31 −0.19 −1.22 −0.41 (0.18) (0.18) (0.27) (0.75) (0.17) 0.11 0.09 0.48 0.11 0.02

Other language −0.53 −0.54 −0.47 −0.87 −0.59 −0.27 (0.18) (0.17) (0.21) (0.43) (0.64) (0.22) 0.01 0.03 0.02 0.05 0.36 0.22

Monday −0.03 −0.03 −0.07 0.01 −0.07 0.17 (0.10) (0.10) (0.14) (0.17) (0.29) (0.29) 0.79 0.79 0.62 0.63 0.82 0.56

Tuesday 0.01 0.01 0.06 0.09 −0.12 0.01 (0.10) (0.10) (0.14) (0.18) (0.27) (0.31) 0.91 0.95 0.69 0.63 0.64 0.98

Wednesday −0.04 −0.04 −0.006 0.04 −0.15 −0.02 (0.10) (0.10) (0.13) (0.17) (0.28) (0.27) 0.68 0.68 0.96 0.82 0.59 0.95

Thursday 0.02 0.02 0.06 0.07 −0.13 0.08 (0.10) (0.10) (0.13) (0.18) (0.26) (0.26) 0.80 0.80 0.66 0.68 0.62 0.76

Friday 0.14 0.14 0.16 0.11 −0.19 0.28 (0.10) (0.10) (0.13) (0.17) (0.23) (0.26) 0.14 0.14 0.23 0.54 0.41 0.28

Constant −8.95 −8.63 −7.66 −7.88 −9.73 −10.11 (0.18) (0.56) (0.43) (0.42) (0.82) (0.76) 0.00 0.00 0.00 0.00 0.00 0.00

A.E. Cuellar, S. Markowitz / International Review of Law and Economics 43 (2015) 98–106 103

Table 2 (Continued)

(1) (2) (3) (4) (5) (6) Age 13–17 Age 13–17 Age 15–17 African American only Hispanic only Asian only

School year fixed effects Y Y Y Y Y Y School fixed effects N Y Y Y Y Y N 3,913,192 3,907,019 2,067,077 759,410 851,669 558,569 Log likelihood −16,939 −16,839 −8867 −5154 −2932 −2239

Coefficients/standard errors/P-values. Key marginal effects reported in the text. Standard errors are clustered at the individual level.

Table 3 All juvenile justice referrals, controls for past year behavior.

(1) (2) (3) (4) (5) All youth Age 15–17 African American only Hispanic only Asian only

Out-of-school suspension × school day

1.03 1.01 1.08 0.45 −0.04 (0.28) (0.33) (0.55) (0.52) (0.66) 0.00 0.00 0.05 0.39 0.96

Suspension 1.30 1.18 1.00 2.11 2.23 (0.27) (0.30) (0.54) (0.49) (0.62) 0.00 0.00 0.06 0.00 0.00

School day 0.10 0.09 0.20 0.12 0.25 (0.09) (0.11) (0.16) (0.25) (0.25) 0.32 0.39 0.21 0.63 0.32

Prior year offenses 0.38 0.37 0.30 0.47 0.51 (0.05) (0.04) (0.06) (0.06) (0.07) 0.00 0.00 0.00 0.00 0.00

Prior year suspensions 0.02 0.0003 0.003 0.001 0.002 (0.002) (0.002) (0.003) (0.004) (0.006) 0.21 0.87 0.22 0.74 0.80

Constant −7.53 −7.90 −6.40 −8.46 −6.61 (0.48) (0.43) (0.44) (0.73) (0.61) 0.00 0.00 0.00 0.00 0.00

Academic year fixed effects Y Y Y Y School fixed effects Y Y Y Y N 2,592,722 1,826,822 474,142 547,347 365,180 Log likelihood −11,141 −7533 −3291 −1952 −1576

Coefficients/standard errors/P-values. Key marginal effects reported in the text. S M , speci

i n t c m

s t s l s i o t c

v o m t t s h m f

tandard errors are clustered at the individual level. odels include controls for prior year offenses, prior year suspensions, age, gender

nteraction between out-of-school suspension and a school day is ot statistically significant in the all-age models. This likely reflects he general behavior that leads to both suspension and referral. The oefficient is significant, however, in the model for older youth. The arginal effect is an increase in the probability from 0.01% to 0.12%. Table 5 explores the pattern of any referral where there are

horter suspensions and for particular student subgroups. Because he length of suspension is skewed, column 1 includes only suspen- ions under 30 days at roughly the 98th percentile of suspension engths. Here, too, youth who are expelled out of school days when chool is in session have a statistically significant higher probabil- ty of committing an offense than youth who are not expelled out f school. Coefficients on individual characteristics are generally he same in sign and significance and similar in magnitude to prior olumns.

Some students were suspended multiple times over the obser- ation period. Of the 4665 students, 2035 or 43.6% were suspended nly once whereas others were suspended up to 30 times. To further ake the sample more homogeneous, the next set of specifica-

ions excludes students who were suspended only once. Here, too he result holds that being suspended out of school on days when

chool is in session is associated with a statistically significantly igher probability of committing an offense (column 2, Table 5). The arginal effect for the interaction term is to increase the probability

rom 0.07% to 0.15%.

al education, ESL, primary language, race/ethnicity, and day of the week.

In the main sample, 4% of students were suspended only in- school, while the remainder experienced suspension in-school as well as out-of school or only out-of-school. Column 3 of Table 5 excludes students who only experience in-school suspension. For this subsample the effect of being suspended out of school on a school day is also positive and significant. The marginal effect is to increase the probability from 0.05% to 0.12%.

Out-of-school suspensions result from student misbehavior and do not occur randomly across students. Being suspended out of school may be correlated with other, unobserved student charac- teristics, such as mental health problems which are also correlated with the offense outcome. Although the data constitute a panel of youth, fixed effects models (i.e. the conditional logit model) may help reduce the correlation with suspension and the error term, however fixed effects will not be useful when the unobserved char- acteristics change over time. As an alternative, we estimate models with lags and leads of out-of-school suspension.

Table 6 presents results from the conditional logit models. The sample size is significantly reduced, but the results are consistent with other specifications. Out of school suspension is associated with an increase in referrals to juvenile justice overall, but not

felony referrals specifically.

Finally we tightened our window limiting our analysis to 6 days pre and 6 days post suspension and allowing us to compare the effect of being in or out of school within a narrow range of days. In

104 A.E. Cuellar, S. Markowitz / International Review of Law and Economics 43 (2015) 98–106

Table 4 Felony referrals only, controls for past year behavior.

(1) (2) (3) (4) (5) All youth Age 15–17 African American only Hispanic only Asian only

Out-of-school suspension × school day

0.63 2.10 1.38 0.56 0.18 (0.58) (0.48) (0.96) (0.80) (1.13) 0.28 0.03 0.15 0.49 0.87

Suspension 1.13 −0.48 −0.09 2.59 2.17 (0.58) (0.91) (0.87) (0.67) (1.12) 0.05 0.60 0.92 0.00 0.05

School day 0.35 0.27 0.37 −0.10 0.89 (0.18) (0.22) (0.07) (0.44) (0.50) 0.05 0.22 0.23 0.83 0.08

Prior year offenses 0.42 0.42 0.33 0.55 0.66 (0.05) (0.06) (0.06) (0.01) (0.11) 0.00 0.00 0.00 0.00 0.00

Prior year suspensions 0.003 0.000 0.006 0.003 −0.01 (0.002) (0.002) (0.003) (0.007) (0.01) 0.17 0.96 0.02 0.69 0.53

Constant −8.64 −11.89 −6.52 −12.48 −11.26 (0.60) (0.85) (0.67) (1.41) (0.60) 0.00 0.00 0.00 0.00 0.00

Academic year fixed effects Y Y Y Y Y School fixed effects Y Y Y Y Y N 2,535,884 1,784,617 445,370 493,498 308,181 Log likelihood −3587 −2481 −1277 −548 −583

C S M ace/et

T t t

s I s n B

T A

C S M s

oefficients/standard errors/P-values. Key marginal effects reported in the text. tandard errors are clustered at the individual level. odels include controls for age, gender, special education, ESL, primary language, r

able 7 the effect of out-of-school suspensions on offenses is posi- ive and significant with a marginal effect on the offense probability hat increases from 0.13% to 0.35%.

Table 8 examines the impact of lagged out-of-school suspen- ion, meaning suspension lagged by 1 day, on current offending. f suspension causes criminal behavior due to the lack of school

upervision on a given day, then the lagged variable should have o effect, as long as the current day is not also a suspension day. ecause suspensions can span several days the analysis is limited

able 5 ll juvenile justice referrals, subsamples.

(1) (2) (3) Suspensions under 30 days

Excluding youth with only one suspension

Excluding youth with only in-school suspension

Out-of-school suspension × school day

1.34 0.98 1.03 (0.35) (0.29) (0.27) 0.00 0.00 0.000

Suspension 1.35 1.11 1.28 (0.34) (0.28) (0.27) 0.00 0.00 0.000

School day 0.11 0.07 0.09 (0.09) (0.10) (0.09) 0.25 0.50 0.034

Constant −7.53 −7.17 −7.55 (0.46) (0.27) (0.48) 0.00 0.00 0.00

Academic year fixed effects Y Y Y School fixed effects Y Y Y N 2,573,375 1,517,176 2,491,144 Log likelihood −10,841 −8.948 −11,015

oefficients/standard errors/P-values. Key marginal effects reported in the text. tandard errors are clustered at the individual level. odels include controls for prior year offenses, prior year suspensions, age, gender,

pecial education, ESL, primary language, race/ethnicity, and day of the week.

hnicity, and day of the week.

to suspensions that are 1 day in length and where the current day was not a suspension day. Table 8 (column 1) shows the lagged interaction of school day and out-of-school suspension to have no statistically significant effect on offending.

As an alternative the impact of lead suspensions, that is, suspen- sion on the next day, was also tested. An out-of-school suspension on a school day should have no effect on offenses which occurred in the past. Table 8 (column 2) shows that the lead variable also has no statistically significant impact on offending. In this specification the lead value of being suspended (in school or out of school) is pos-

itive and significant, but only at the 10% level, possibly indicating persistent behavior problems from arrest through school.

Table 6 Juvenile justice referrals, conditional logit models.

(1) (2) All referrals Felony

referrals only

Out-of-school suspension × school day

0.98 0.64 (0.24) (0.46) 0.00 016

Suspension 1.10 0.65 (0.23) (0.44) 0.00 0.14

School day 0.10 0.36 (0.09) (0.19) 0.31 0.06

School year fixed effects Y Y School fixed effects Y Y N 570,289 168,688 Log likelihood −7994 −2323

Coefficients/standard errors/P-values. Standard errors are clustered at the individual level. Models include controls for prior year offenses, prior year suspensions, and day of the week.

A.E. Cuellar, S. Markowitz / International Review

Table 7 Juvenile justice referrals, within 6 days of suspension.

(1) All referrals

Out-of-school suspension × school day

1.00 (0.49) 0.04

Suspension 1.69 (0.43) 0.000

School day −0.14 (0.49) 0.78

Constant −9.38 (0.64) 0.000

School year fixed effects Y School fixed effects Y N 210,831 Log likelihood −3527

Coefficients/standard errors/P-values. Key marginal effects reported in the text. Standard errors are clustered at the individual level. Models include controls for prior year offenses, prior year suspensions, age, gender, special education, ESL, primary language, race/ethnicity, and day of the week.

Table 8 Lag and lead out-of-school suspension, falsification tests.

(1) (2) All youth All youth

Lagged out-of-school suspension × school day

−0.11 (0.74) 0.89

Lead out-of-school suspension × school day

0.58 (1.02) 0.57

Suspension, lag/lead 0.92 1.74 0.71 1.01 (0.19) (0.08)

School day 0.26 0.08 (0.08) (0.09) 0.00 0.31

Constant −8.53 −7.52 (0.56) (0.46) 0.00 0.00

Academic year fixed effects Y Y School fixed effects Y Y N 3844.840 2,543,329 Log likelihood −16,259 −9999

Coefficients/standard errors/P-values. Key marginal effects reported in the text. S M s

4

e o i 1 o A d i a d i

tandard errors are clustered at the individual level. odels include controls for prior year offenses, prior year suspensions, age, gender,

pecial education, ESL, primary language, race/ethnicity, and day of the week.

. Limitations

The study has certain limitations one of which is that youth may ngage in delinquent behavior that does not come to the attention f law enforcement authorities. In longitudinal studies of offend- ng youth, one-quarter to one-third were never arrested (Huizinga, 995). This implies that youth may engage in harmful behavior utside of school, but this would not be measured in the data. nother limitation is that the school data are limited to an urban istrict; the findings may not generalize to rural and suburban sett-

ngs in light of other studies that have found differences across reas in incapacitation effects (Luallen, 2006). Further the offense ata are restricted to a single county and although the county

s significantly larger than the school district, the study cannot

of Law and Economics 43 (2015) 98–106 105

measure offenses that occur in other counties. The findings would be biased downward and effects understated if youth are more likely to travel outside the county to commit crimes on days when they are suspended out of school.

Another possibility is that the same student activity that leads to suspension may also lead to a school referral to juvenile justice. In general, however, studies have found that crimes committed on school property are less likely to be reported to the police than crimes occurring elsewhere, particularly property crimes but also violent crimes (Whitaker and Bastian, 1991; Cook et al., 2010). While youth victimization is as prevalent in school as out of school, youth are much more likely to be arrested for offenses occurring outside of school.

5. Conclusions

Schools face tremendous challenges when designing and imple- menting disciplinary policies that reduce violence on campus, protect students, and maintain environments conducive to learn- ing. The “zero tolerance” policies toward infractions of conduct codes, while broadly used, are controversial and the consequences are not well understood. In this paper, we fill a gap in the liter- ature by evaluating whether school suspension policies increase offending behavior by problem youth. Our paper speaks to the so called “school-to-prison pipeline”, which is a term used to broadly describe policies that push children out of classrooms toward the juvenile justice system. In this paper, we do not examine the poli- cies directly, but rather one of the consequences of such policies, out-of-school suspension.

Specifically, we evaluate whether out-of-school suspension increases referrals to the juvenile justice system among youth with a history of offending behaviors. The evidence presented here points to the conclusion that it does. The results show that among this population, being suspended out-of-school on a school day is associated with a more than doubling of the probability of offense. Further, the study finds that the effect is larger for African Ameri- can youth and is not significant for Hispanic and Asian youth. It is possible that the estimates are biased due to simultaneity if youth behavior, such as aggressive outbursts, that leads to suspension and carries over into behavior that causes arrest. In this case, the lagged and lead values of out-of-school would be positive, but this analy- sis finds that they are not significant. Other robustness checks also confirm our main conclusion.

Schools are not the locus of most youth crime. Most youth crime – 85% of juvenile arrests – derives from offenses that are committed outside of school and those offenses are more seri- ous on average than in-school crimes. Schools, however, need tools to address problems within their institutions. Some options affect problems exclusively within schools and others that can have broader impacts on the community. The U.S. Department of Education promoted several approaches to manage school disci- pline challenges, among them interventions to develop positive school climates and setting clear and high expectations for school behavior. They span a wide range of activities from bullying pre- vention, to school bus behavior, and staff supervision training (U.S. Department of Education, 2014). Our analysis does not inform whether juvenile referral rates would be lower if in-school mis- behavior was dealt with differently. We can also only speculate on whether more aggressive suspension policies deter in-school misbehavior.

This analysis does find that school suspension policies designed

to handle problem behavior in school may contribute to overall crime rates out of school, highlighting a significant potential dis- advantage of using out-of-school suspension as part of a school disciplinary policy. However, this conclusion does not account for

1 eview

t r b F c w s p T t t c t

C

A

i r f

R

A

A

A

A

A

A

B

C

C

C

pipeline. In: Book, S. (Ed.), Invisible Children in The Society and its Schools.

06 A.E. Cuellar, S. Markowitz / International R

he potential positive effects in improving the classroom envi- onment. This remains a question for future research. There is a roader set of consequences that cannot be addressed in this study. or instance, suspension may undermine the individual’s academic areer. As youth fall behind they may be more likely to drop-out, hich could ultimately lead to greater crime in the longer term. The

tudy also does not address spillover effects on peers and the cost to eer achievement and potentially higher crime rates within school. hese too are topics for further research. Nevertheless, the results of his paper provide evidence for the school-to-prison pipeline where he likely mechanism is that suspension lead to days spent in the ommunity with reduced supervision and increased opportunities o commit crimes.

onflict of interest

The authors report no conflicts.

cknowledgements

Funding for this project was provided by the John D. and Cather- ne T. MacArthur Foundation (Cuellar, PI). The content is solely the esponsibility of the authors and does not reflect the views of the under.

eferences

CLU. What Is The School-to-Prison Pipeline? Available at: https://www.aclu.org/ racial-justice/what-school-prison-pipeline (accessed 17.02.14).

merican Academy on Pediatrics, Committee on School Health, 2003. Out-of-school suspension and expulsion. Pediatrics 112, 1206–1209.

merican Academy on Pediatrics, Council on School Health, 2013. Out-of-school suspension and expulsion. Pediatrics 131, e1000–e1007.

merican Bar Association, 2001. School Discipline “Zero Tolerance” Poli- cies. Vera Institute of Justice, New York, retrieved from http:// www.americanbar.org/groups/child law/tools to use/attorneys/school disciplinezerotolerancepolicies.html

merican Psychological Association, 2008. Are zero tolerance policies effective in the schools? Am. Psychol. 63, 852–862.

nderson, D.M., 2014. In school and out of trouble? The minimum dropout age and juvenile crime. Rev. Econ. Stat. 96 (May (2)), 318–331.

ecker, G.S., 1968. Crime and punishment: an economic approach. J. Polit. Econ. 76, 169–217.

hamberlain, G., 1980. Analysis of covariance with qualitative data. Rev. Econ. Stud- ies 47, 225–238.

ook, P.J., 1980. Research in criminal deterrence: laying the groundwork for the

second decade. In: Morris, N., Tonry, M. (Eds.), Crime and Justice: An Annual Review of Research. University of Chicago, Chicago, IL, pp. 211–268.

ook, P.J., Gottfredson, D.C., Na, C., 2010. School crime control and prevention. In: Tonry, M. (Ed.), Crime and Justice: A Review of Research. University of Chicago Press, Chicago.

of Law and Economics 43 (2015) 98–106

Cullen, J.B., Jacob, B.A., Levitt, S., 2006. The effect of school choice on participants: evidence from randomized lotteries. Econometrica 74, 1191–1230.

Feitelson, D.G., 2004. On Identifying Name Equivalencies in Digital Libraries. Information Research 9.4 Web. 22 April 2010, http://informationr.net/ir/9-4/ paper192.html

Gottfredson, D.C., DiPietro, S.M., 2011. School size, social capital, and student vic- timization. Sociol. Educ. 84, 69–89.

Henault, C., 2001. Zero tolerance in schools. J. Law Educ. 30, 547–553. Huizinga, D., Loeber, R., Thornberry, T.P., 1995. Recent Findings from the Program

of Research on the Causes and Correlates of Delinquency NCJ 159042. U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention, Washington, DC.

Jacob, B.A., Lefgren, L., 2003. Are idle hands the devil’s workshop? Incapacitation, concentration, and juvenile crime. Am. Econ. Rev. 93, 1560–1577.

Kang-Brown, J., Trone, J., Fratello, J., Daftary-Kapur, T., 2013. Issue Brief: A Generation Later: What We’ve Learned about Zero Tolerance in Schools. Vera Institute of Jus- tice, New York, http://www.vera.org/sites/default/files/resources/downloads/ zero-tolerance-in-schools-policy-brief.pdf

Kinsler, J., 2011. Understanding the black–white school discipline gap. Econ. Educ. Rev. 30, 1370–1383.

Kinsler, J., 2013. School discipline: a source or salve for the racial achievement gap? Int. Econ. Rev. 54, 355–383.

Lazear, E.P., 2001. Educational production. Q. J. Econ. CXVI, 777–803. Luallen, J., 2006. School’s out. Forever: a study of juvenile crime, at-risk youths and

teacher strikes. J. Urban Econ. 59, 75–106. McFadden, A.C., Marsh, G.E., 1992. A study of racial and gender bias in the punish-

ment of school children. Educ. Treat. Child. 15, 140–146. Nagin, D.S., 1998. Criminal deterrence research at the outset of the twenty-first

century. In: Tonry, M. (Ed.), Crime and Justice: An Annual Review of Research. University of Chicago Press, Chicago, pp. 1–42.

Raffaele Mendez, L.M., 2003. Predictors of suspension and negative outcomes: a longitudinal investigation. New Directions for Youth Development 99, 17–33.

Raffo, J., Lhuillery, S., 2009. How to play the “name games”: patent retrieval com- paring different heuristics. Res. Policy 38.10, 1617–1627.

Skiba, R., Reynolds, C.R., Graham, S., Sheras, P., Conoley, J.C., Garcia-Vazquez, E., 2006. Are zero tolerance policies effective in the schools? An evidentiary review and recommendations. In: A Report by the American Psychological Association Zero Tolerance Task Force, https://www.apa.org/pubs/info/reports/zero-tolerance- report.pdf

Skiba, R.J., Michael, R.S., Nardo, A.C., et al., 2002. The color of discipline: sources of racial and gender disproproationality in school punishment. Urban Rev. 34 (4), 317–342.

Snyder, H.N., Sickmund, M., 2006. Report Juvenile Offenders and Victims: 2006 National Report NCJ 212906. U.S. Department of Justice, Office of Justice Pro- grams, Office of Juvenile Justice and Delinquency Prevention, Washington, DC.

Tobin, X., Sugai, G., Colvin, G., 1996. Patterns in middle school discipline records. J. Emot. Behav. Disord. 4, 82–94.

Trajtenberg, M, Shiff G., Ran M., 2006, August. “The ‘Names Game’: Harnessing Inven- tors’ Patent Data for Economic Research”. NBER Working Paper No. 12479.

U.S. Department of Education, 2000. National Center for Education Statistics, School and Staffing Survey, Public Teacher Questionnaire.

U.S. Department of Education, 2014. Directory of Federal School Climate and Disci- pline Resources, Washington, D.C.

Wald, J., Losen, D., 2007. Out of sight: the journey through the school-to-prison

Lawrence Publishing, Mahwah, NJ. Whitaker, C.J., Bastian, L.D., 1991. Teenage Victims: A National Crime Survey Report

(NCJ-128129). U.S. Department of Justice, Bureau of Justice Statistics, Washing- ton, DC.