Can equitable punishment be mandated? Estimating impacts of sentencing guidelines on disciplinary disparities

International Review of Law and Economics 40 (2014) 51–61

Contents lists available at ScienceDirect

International Review of Law and Economics

Can equitable punishment be mandated? Estimating impacts of sentencing guidelines on disciplinary disparities Anton Bekkerman ∗,1 , Gregory A. Gilpin 1 Department of Agricultural Economics and Economics at Montana State University – Bozeman, P.O. 172920, Bozeman, MT 59717-2920, United States

a r t i c l e

i n f o

Article history: Received 31 March 2014 Received in revised form 19 August 2014 Accepted 16 September 2014 Available online 2 October 2014 JEL classification: K42 I28 I24 Keywords: Disciplinary mandates Marginal deterrence Punishment disparities Racial bias School discipline

a b s t r a c t This study empirically investigates the potentially unintended effects of state laws that seek to improve safety in U.S. public school by mandating standardized student punishment. We estimate the effects of exogenous state-level variation in the quantity and type of such mandates on disciplinary disparities across students who commit serious offenses. Estimation results indicate that more severe punishments are imposed in schools with higher proportions of black or Hispanic students, but such disparities are significantly dampened in states that mandate a higher number of guidelines for serious offenses. However, more guidelines for less severe misconduct tend to increase race-based disciplinary disparities and increase the severity of punishments administered for serious offenses. These outcomes extend the existing sentencing guidelines literature and provide empirical implications for considering marginal deterrence effects when crafting future policies. © 2014 Elsevier Inc. All rights reserved.

1. Introduction The 1994 Goals 2000: Educate America Act established a broad framework for reforming public education in the United States. One of the act’s goals was to “ensure the rights of students to study in a safe and secure environment that is free of drugs, alcohol, and crime” (Goal 7(A)(ii)). The complementary 1994 Gun-Free Schools Act instituted a federally mandated one-year expulsion for any U.S. public school student who knowingly possesses or uses a firearm in a school zone. Between 1995 and 2002, many state governments independently introduced additional legislation that extended the scope of the federally mandated standards. These state-mandated punishments not only applied to a broader range of student offenses—including the use of drugs, alcohol, violence against other students or teachers, and less serious misdemeanors—but also authorized the use of other disciplinary methods.

∗ Corresponding author at: 306 Linfield Hall, DAEE, Montana State University, P.O. Box 172920, Bozeman, MT 59717-2920, United States. E-mail addresses: [email protected] (A. Bekkerman), [email protected] (G.A. Gilpin). 1 Both authors equally contributed to the development of the research and manuscript. Senior authorship is shared. http://dx.doi.org/10.1016/j.irle.2014.09.002 0144-8188/© 2014 Elsevier Inc. All rights reserved.

In an effort to create a safer educational environment, the standardization of student discipline at both the federal and state levels may have also—although unintentionally—affected disciplinary disparities across schools; that is, it may have altered the systematically inconsistent use of punishments across groups of individuals based on demographic and socioeconomic differences. The outcomes of this unintended effect may parallel those of the 1984 Sentencing Reform Act (SRA), which was enacted explicitly to reduce inequitable sentencing decisions across judges in the federal justice system (Spohn, 1990; Fender et al., 2006; Abrams et al., 2013). The SRA instituted standardized punishment guidelines that separated sentencing decisions from judicial discretion and linked them more closely to the offense type, substantially reducing disciplinary disparities (Anderson et al., 1999; Mustard, 2001). The SRA’s effectiveness in reducing punishment inconsistencies in the federal justice system suggests that disciplinary guidelines for public schools could also attenuate persistent disciplinary disparities. This study’s purpose is to empirically quantify the potential disparities across U.S. public high schools in their use of serious student misconduct and associated punishments, and then investigate the role of disciplinary mandates in affecting these disciplinary disparities. We focus on serious student misconducts because they occur relatively infrequently, typically result in a higher degree of disruption to the learning environment, and are often more

52

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

aggressively disciplined by removing students from familiar educational surroundings for an extended period. Disciplinary disparities can lead to long-term, sustained educational and economic inequality for individuals who are more persistently and more severely punished. For example, Bernburg and Krohn (2003) show that discipline resulting in youth’s exclusion from established daily activities can lead to more frequent delinquencies and McCarthy (2000) and Finn (1989) find that disciplinary action that limits access to schoolrelated activities reduces educational attainment and is linked to increased dropout rates. An extensive literature has also shown that school attendance affects future educational attainment, labor market outcomes, individuals’ well-being, and criminality (for example, see Angrist and Krueger, 1991; Acemoglu and Angrist, 2001; LlerasMuney, 2002; Oreopoulos, 2009; Machin et al., 2011). Therefore, understanding whether standardized punishment mandates can dampen or eliminate disciplinary disparities across schools could offer important insights for improving equal access to education. There are significant demographic and socioeconomic gradients in the United States and identifying heterogeneity in schools’ responses to undesirable behavior by students from these gradients, and whether the heterogeneity can be reduced through policy, could be critical to minimizing educational outcome disparities. Identifying the effects of standardized disciplinary mandates using variation in state-level punishment guidelines is advantageous for several reasons. First, disciplinary mandates for public schools were enacted to ensure a safe educational environment, so their impacts on reducing punishment discrepancies are plausibly exogenous. Second, unlike federal mandates, state-level punishment guidelines do not apply uniformly across all educational agencies, resulting in substantial heterogeneity in the quantity and types of guidelines across states. Third, nearly all states have punishment standards that address two types of student offenses: serious student misconduct (such as the possession or use of firearms or non-firearm weapons, use and distribution of illegal drugs or alcohol, and assault and battery offenses) and less harmful behavior (including general misconduct, disobedience, or defiance of authority figures). Therefore, we are able to study the potential links among punishment mandates associated with different offense types. Lastly, Kinsler (2011) shows that punishment disparities within any particular school are minimal and race-based disciplinary inconsistencies that do exist are largely independent of principals’ racial characteristics (see Rocque, 2010, for similar evidence).2 Consequently, disciplinary disparities are most likely to occur between schools, providing an opportunity to investigate how these disparities are affected by state-level guidelines and more clearly understanding factors that could aid in crafting policies that ensure more equitable access to educational opportunities. The empirical analysis uses a unique set of school-level responses from the 2003–2010 School Surveys on Crime and Safety (SSOCS). The responses include detailed information on student misconduct and disciplinary decisions, student body and school attributes, and school administrators’ self-reported measures of crime prevention limitations and misconduct management and prevention programs. These data, along with community characteristics and state and time fixed effects that help capture state-level punishment consistencies between schools, are used to model between-school variation in student misconduct and associated disciplinary actions. Results from an empirical exploratory regression analysis offer evidence of significant between-school race-based disciplinary disparities, indicating that administrators

2

In the federal justice system, sentencing discrepancies have also been shown to exist mostly across judges, rather than across decisions made by any particular judge for similar cases (for example, see Ashenfelter et al., 1995; Schanzenbach, 2005; Iyengar, 2011).

in schools with greater proportions of black or Hispanic students impose a greater number and more severe punishments. Furthermore, we find that for firearm offenses, for which discipline is federally mandated by the 1994 Gun-Free Schools Act, betweenschool inconsistencies were not statistically significant, suggesting that disciplinary mandates could be effective in reducing or eliminating punishment disparities. We then examine the degree to which disciplinary disparities vary with the quantity and type of state-level discipline mandates. The empirical analysis indicates that in states with a below average number of standardized punishment guidelines for serious offenses, race-based inconsistencies across schools’ disciplinary rates exist for the most severe punishments. In states with an above average number of such guidelines, such disparities are significantly lower and these reductions occur without contemporaneous increases in overall punishment rates. This suggests that disciplinary equity could be achieved by reducing the use of severe punishment in schools with higher proportions of black or Hispanic students rather than increasing its use in schools with a larger white student body. However, there is also evidence that schools in these states may be shifting toward using less severe punishments. Punishment guidelines for less harmful behaviors, however, are not associated with similar reductions in race-based disciplinary inequity. Rather, in all states with guidelines for less serious offenses, there are higher rates of severe punishment use for serious student offenses in schools with higher proportions of black or Hispanic students, and the disparities are twice as large across schools in states with an above average number of these punishment mandates. These results offer important implications about the design of sentencing guidelines. They suggest that while guidelines targeting more serious offenses can reduce systematic race-based disciplinary disparities, it may not be possible to easily generalize these effects to extending punishment guidelines for other offense types. This may especially be the case when guidelines mandate similarly severe punishments for both serious and less harmful misconduct, thus reducing marginal deterrence effects in disciplinary disparity. Such guideline design may result in a shift toward the more frequent use of more severe discipline, which could potentially exacerbate inequitable sentencing and contribute to adverse long-run impacts for affected individuals. Consequently, the consideration of specific objectives and possible dependencies among guidelines for different offense types is critical in crafting and implementing effective standardized disciplinary policies. 2. Data description and preliminary insights This research empirically investigates factors that contribute to the variation in school-level punishments of serious student offenses. The main data for this research are responses to the biennial, repeated cross-section School Survey on Crime and Safety (SSOCS) conducted by the National Center for Education Statistics (NCES). The restricted-access data collected during the 2003–04, 2005–06, 2007–08, and 2009–10 school years for 3200 U.S. public high schools contain school-level information describing the number of reported student offenses and associated disciplinary actions.3 We measure schools’ disciplinary outcomes as the ratio of the total number of punishments to the total number of misconduct instances for particular categories of punishment and misconduct.4

3 We limit the responses to those from non-alternative high schools. The exclusion of alternative high schools reduces the sample by less than 5%. Observation counts are rounded to the nearest 10s to comply with data license restrictions. 4 Using a ratio alleviates the need to directly model the endogenous relationship between misconducts and disciplinary outcomes. However, this approach could also

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61 Table 1 Number of student offenses and discipline administered.

Firearms Non-firearm Illegal Alcohol Altercations All misconducts

Students involved

Permanent removals

Prolonged suspensions

Light discipline

0.15 1.31 6.5 2.27 16.62 26.86

0.1 0.57 2.48 0.44 2.04 5.63

0.03 0.52 2.78 1.12 7.17 11.61

0.02 0.23 1.24 0.71 7.41 9.62

Notes: All statistics are per 1000 students per school year and the presented data represent only public high schools with over 500 enrolled students. “Firearms” includes the possession or use of a firearm; “Non-firearm Weapons” includes the possession or use of a non-firearm weapon; “Illegal Drugs” includes the distribution, possession, or use of illegal drugs; “Alcohol” includes the possession or use of alcohol; and “Altercations” includes the actual or intentional touching or striking of another person against his or her will, or the intentional causing of bodily harm to an individual without the use of a weapon. A permanent removal is the discontinuation of educational services for a minimum of one-year due to either an expulsion or transfer to a specialized alternative school. A prolonged suspension is defined as a temporary interruption of school services lasting five or more days, but less than the remainder of the school year. Light discipline includes punishments such as out-of-school suspensions lasting fewer than five days and detention.

The SSOCS data include information about five categories of serious student misconduct committed on school grounds, buildings school buses, or at places that hold school-sponsored events, which we classify as two types. The violent offense type includes the possession or use of a firearm or explosive device and possession or use of a non-firearm weapon. The distribution, possession, or use of illegal drugs; distribution, possession, or use of alcohol; and altercations (defined as actual or intentional touching or striking of another person against his or her will, or the intentional causing of bodily harm to an individual without the use of a weapon) are denoted as harmful offenses.5 Furthermore, there are three types of disciplinary actions: permanent removal from a school, prolonged suspension, or light discipline. A permanent removal is the most severe discipline, discontinuing educational services by a school for a minimum of one year. Permanent removals can include either an expulsion or transfer to a specialized alternative school. Prolonged suspensions are temporary interruption of school services lasting five or more days, but less than the remainder of the school year. The least severe punishment category, light discipline, includes actions such as out-of-school suspensions lasting fewer than five days and detention. Table 1 presents summary statistics describing the number of students involved and disciplinary actions across the five misconduct categories and three punishment types.6 The least prevalent misconduct is students’ involvement in firearm offenses, which, on average, occur to 0.15 students per 1000 and only 11.8% schools experienced at least one incidence of this offense. Unsurprisingly, however, over 68% of students committing this most serious offense were expelled. Non-firearm weapons and alcohol offenses occurred, on average, to 1.31 and 2.27 students per 1000 and with at least one incident in 50.2% and 63.6% of sampled high schools. Disciplinary consequences for these offenses are less severe. Students committing non-firearm weapons offenses are permanently removed in 45.2% of the cases and faced prolonged suspension in 38.9%, and alcohol-related offenses resulted in 22.6% of students being permanently removed and 49.8% being suspended.

introduce division bias. Bekkerman and Gilpin (2014) show that in the case of the SSOCS data, division bias is minimal. 5 These classifications are consistent across a large proportion of U.S. school districts, including major ones such as the Miami-Dade County School Board, New York City Department of Education, and Chicago Public Schools. 6 Statistics are presented for schools with at least 500 students.

53

Table 2 Percentage of students involved in a recorded and disciplined offense in schools with a promant student body race. Predominant student body race

All serious offenses Permanent removal Prolonged suspensions Light discipline Violent offenses Permanent removal Prolonged suspensions Light discipline Harmful offenses Permanent removal Prolonged suspensions Light discipline

Black

Hispanic

White

26% 48% 26%

24% 37% 40%

10% 39% 51%

52% 36% 11%

51% 36% 13%

29% 48% 23%

24% 48% 28%

24% 36% 40%

10% 40% 50%

Notes: A school is categorized as having a predominant student body race if more than 10% of all students in the school are of a particular race. Serious offenses include violent and harmful offenses. The violent offense category includes misconducts related to the possession or use of a firearm or explosive device and possession or use of a non-firearm weapon. The harmful offense category includes misconducts related to the distribution, possession, or use of illegal drugs; distribution, possession, or use of alcohol; and altercations (physical attacks and fights). A permanent removal is results in the discontinuation of educational services for a minimum of one-year due to either an expulsion or transfer to a specialized alternative school. Prolonged suspensions are temporary interruption of school services lasting five or more days, but less than the remainder of the school year. Light discipline includes punishments such as out-of-school suspensions lasting fewer than five days and detention.

The possession of illegal drugs and altercations are most prevalent with 6.50 student incidents and 16.62 student incidents, respectively, per 1000 students (87.9% and 96.6% of schools reported such offenses). Punishment rates for drug possession are similar to those for alcohol possession, but students involved in physical attacks were suspended in 43.0% of the cases and expelled only 13.1%. The descriptive statistics in Table 1 suggest that, on average, higher disciplinary severity is observed more often for misconducts that are perceived to be more serious. Table 2 presents suggestive evidence that there are differences in disciplinary outcomes across student body characteristics, such as demographics. Schools are classified to have a substantial population of a particular demographic if at least 10% of their student body are black, Hispanic, or white.7 The data indicate that across all serious offense categories, schools with higher proportions of black or Hispanic students tend to use permanent removal approximately 12–13 percentage points more frequently than schools with a higher proportion of white students and 16–22 percentage points for violent offenses. Conversely, schools with higher proportion of white students are more likely to observe light discipline for both violent and harmful offenses. These discrepancies suggest that heavier punishments are used in schools with a higher proportion of black or Hispanic students, while schools with predominantly white student bodies are more likely to implement less severe punishments. In addition to information related to student misconduct and discipline, the SSOCS data describe student body characteristics, school administrators’ perceptions and expectations about limitations to their school’s crime prevention capabilities, and existing misconduct management and prevention opportunities. Student body characteristics include the percent of students that

7 For robustness, the racial concentration levels were varied between 10% and 50% and showed no qualitative difference in rankings and minimal quantitative difference.

54

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

scored in the lowest 15th percentile on standardized testing and the proportion eligible for free lunch. These characteristics help control for disciplinary inconsistencies resulting from variation in the socioeconomic status of student bodies. School administrators’ perceptions about the availability of options for alternative placement of disruptive students, fear of student retaliation, inadequate funds for implementing crime prevention methods, fear of district or state reprisal, and the presence of gang activity during the previous school year offer important indicators of expectations about the quality and effectiveness of a school’s ability to manage student misconduct. Administrators’ perception could be positively correlated to the predominant student body race and omitting this information could inappropriately attribute differential treatment to disciplinary action that is not discriminatory.8 There are four measures of schools’ existing misconduct management and prevention opportunities: the existence of a formal process to obtain parental input on school conduct policies, teacher training associated with discipline policies and classroom management, opportunities to provide student mentoring by security staff, and training and technical assistance to parents for dealing with students’ misconduct.9 Lastly, community-level information was collected from several sources and includes four measures: local area unemployment rates (Bureau of Labor Statistics); binary indicators of whether a school is located in a rural, urban, or metropolitan area (U.S. Census Bureau); average county-level per capita personal income in 2005 dollars (U.S. Bureau of Economic Analysis); and the number of adult criminal offenses per 1000 adults (Federal Bureau of Investigation’s Uniform Crime Reports). Table 3 presents the descriptive statistics of these and all other variables used to model school discipline outcomes and Table A1 of Appendix A provides detailed definitions of these variables. Data describing the number of sentencing guidelines are obtained from the Civil Rights Project. States mandate disciplinary actions for both serious offenses (loosely analogous to those that could be tried as felony or criminal cases for adults) and less serious misconducts. We evaluate the mandates based on their content in order to appropriately classify (and quantify) the guidelines based on the offenses’ seriousness to which a guideline pertains (a “felony” or “misconduct”). For classification purposes, we characterize one type of sentencing guidelines as that pertaining to serious “felony” offenses: possession or use of a non-firearm weapon, possession of illegal substances or alcohol, committing physical harm or battery, threatening physical harm or assault, damaging property, theft, membership in a gang, and being officially charged for infractions by law enforcement. The second type of guidelines specifies punishments for less

8

While the SSOCS questionnaire asks administrators to report the number of offenses they perceived to be related to gang activity, we define the variable as a binary indicator that equals 1 if administrators perceived any gang activity during the school year (regardless of frequency) and 0 if no gang activity has been perceived. The decision to specify the variable as a binary measure was made because most responses reported extremely low frequencies of gang-related misconducts, but may signal that organized criminal activity is perceived to impact schools’ disciplinary behaviors. 9 We have also considered controlling for principal characteristics using responses from the Schools and Staffing Survey (SASS) collected by the National Center for Education Statistics. However, relatively few schools in the SSOCS are those for which principal information is available in the SASS. Furthermore, neither the SSOCS school-level nor the SASS principal-level data specify whether the principal or another school administrator is responsible for making disciplinary decisions. Consequently, school-level disciplinary behaviors may be inappropriately modeled if principal information is used. Moreover, Kinsler (2013) shows that differences in principal characteristics do not significantly explain differences in school discipline rates.

Table 3 Descriptive statistics of variables used in the school discipline model. Mean

Std. dev.

Min

Dependent variables Permanent removal 19.3 26.5 0 All serious offenses Firearms 64.2 44.8 0 40.2 43.5 0 Non-firearm weapons Illegal drugs 37.5 41.8 0 20.4 37.3 0 Alcohol 11.8 24.7 0 Altercations Prolonged suspension 41.1 35.8 0 All serious offenses 24.7 39.1 0 Firearms 43.0 43.7 0 Non-firearm weapons 45.0 42.7 0 Illegal drugs 51.3 46.3 0 Alcohol 39.8 41.0 0 Altercations Light discipline 40.5 39.3 0 All serious offenses 11.1 28.6 0 Firearms Non-firearm weapons 16.8 34.1 0 17.5 33.2 0 Illegal drugs Alcohol 28.3 41.8 0 48.4 42.8 0 Altercations Independent variables Student body characteristics 13.1 22.5 0 Percent Black Percent Hispanic 12.7 20.9 0 39.3 24.8 0 Percent free lunch Percent below 15 14.0 14.8 0 Misconduct management and prevention programs Acquire parental input 0.58 0 Mentor students 0.57 0 Train teachers 0.45 0 Assist parents 0.42 0 School administrators’ perceived crime prevention limitations Limited access 0.59 0 0.26 0 Student retaliation 0.67 0 Inadequate funds Fear of reprisal 0.17 0 Gang activity 0.36 0 Community characteristics 6.71 2.79 2.1 Unemployment rate Per capita income (in $1000s of 2005 3.62 1.03 1.4 dollars) Street crime (per 1000 persons) 50.6 29.4 0.0 Urban 0.28 0 Rural 0.53 0 3200 Observations

Max

100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100

100 100 100 99 1 1 1 1 1 1 1 1 1 22.5 13.1 287.9 1 1

Notes: Standard deviations are not provided for binary indicator variables. Dependent variables and student body characteristics are measured in percent. Altercations include physical attacks and fights. Maximum values are truncated to the fourth largest value and observation counts are rounded to the nearest 10s to comply with data license restriction.

harmful “misconduct” offenses: disobedience, persistent absenteeism, unlawful activity, defiance of authority figures, and other vaguely defined behavioral actions School districts typically assure that schools are following these mandates. Statutes are similarly worded across states and heterogeneity is mostly associated with the number of mandates. Furthermore, the categorization of an offense as a ‘felony’ or ‘misconduct’ in these mandates is relatively unambiguous. Table 4 presents a summary of these data, showing that only four states or regions—District of Columbia, Maryland, Montana, and Virginia—did not enact any laws to supplement the firearm mandates in the 1994 Gun-Free Schools Act. Alternatively, New Jersey and West Virginia established an additional 17 and 19 guidelines, respectively. On average, state legislatures instituted nearly six further mandates by 2002 to augment the federal standards and approximately 60% of state disciplinary guidelines are associated

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

55

Table 4 Summary of state punishment laws associated with felony and misconduct offenses. State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri

Overall average

Felony guidelines 4 2 1 8 5 6 6 1 0 7 2 6 1 1 1 3 6 7 4 3 0 10 3 0 3 2

Misconduct guidelines

State

Felony guidelines

1 2 3 2 5 5 1 0 0 3 0 0 1 2 6 0 7 3 5 3 0 0 4 1 0 0

Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

0 10 8 4 11 1 1 2 1 1 0 6 1 1 1 6 6 1 9 2 0 4 11 1 1

Felony guidelines

Misconduct guidelines

3.5

2.3

Misconduct guidelines 0 4 0 3 6 0 2 3 3 1 2 8 2 1 6 2 4 0 3 2 0 0 8 4 0

Source: Table created by authors using data from The Civil Rights Project (2000). Notes: Mandates are evaluated based on their content in order to appropriately classify (and quantify) the guidelines based on the offenses’ seriousness to which a guideline pertains (a felony or misconduct). Statutes are similarly worded across states and heterogeneity is mostly associated with the number of mandates. Furthermore, the categorization of an offense as a felony or misconduct in these mandates is relatively unambiguous. Felony guidelines provide sentencing requirements for offenses committed using a weapon other than a firearm, associated with the possession or use of illegal drugs or alcohol, using physical harm or battery and threats of physical harm or assault, resulting in property damage or robbery, associated with gang membership, and being formally charged with a felony or delinquency charge. Misconduct guidelines provide sentencing requirements for offenses that include disobedience, absenteeism, unlawful activity, misconduct, and authority defiance. Sentencing guidelines for felony offenses associated with the possession or use of a firearm are excluded, because all states receiving federal education funds must comply with regulations of the 1994 Gun-Free Schools Act to permanently remove students committing this offense.

with guidelines for felony offenses and 40% with misconduct.10 Fig. 2 shows the distribution of mandates associated with each offense type across the United States. The figure makes evident that there is no clear spatial correlation among states that mandate a higher or lower number of mandates.

3. Modeling strategy Differences in standardized disciplinary guidelines across states provide variation in the number of legislated guidelines, the types of offenses that the regulations address, and the mandated punishments. This offers an opportunity to evaluate guideline effectiveness based on their intended use and provide important insights for developing future policies. A general model of school discipline outcomes that tests the relationship between

10 Despite the existence of state-based punishment guidelines, many states allowed school districts to apply some disciplinary discretion when composing student conduct protocols. However, state guidelines act as disciplinary floors such that administrators could potentially impose additional (or more severe) punishment on a case-by-base basis. During 2003–2009, amendments to states’ guidelines were proposed to reiterate the existence of established disciplinary mandates while emphasizing the increased role of discretion in categorizing committed offenses (for example, see Alabama HB 843). In almost all cases, however, these changes were not made into law, resulting in minimal variation in established state guidelines over time.

punishment severity and student body racial composition after controlling for student, school, and community characteristics is k dijt = ˇ0 + Dit ˇ1 + S it ˇ2 + M it ˇ3 + ıs + ıt + ijt

(1)

k represents the percentage of discipline j used in school where dijt i during school year t for offense type k, Dit represents a vector of the demographic and socioeconomic characteristics of the student body, S it is a vector of school characteristics, M it is a vector of characteristics describing the community associated with school i, ıs and ıt are state and time fixed effects, and ijt is the idiosyncratic error term. State fixed effects control for disciplinary consistency across schools in the same state as well as factors (e.g., political leanings of state legislatures and educators) that could be correlated with the explanatory variables, including state discipline guidelines. Similarly, time fixed effects help control for unobservable nationwide time-varying influences. A school’s use of discipline j is bounded by 0% and 100%, where 0% indicates that the discipline type was not at all used and 100% indicates that all students committing offense type k received discipline j. Fig. 1 shows examples of these bounds in our data, resulting in substantial mass at either endpoint. The figure presents histograms describing the ratio between the number of administered punishments and the total number of misconducts for three punishment types. For all three punishment types, there is substantial mass at the 0% and 100% bounds. This is likely due to administrators’ consistent use the same disciplines for certain offenses, such that for any particular offense students are homogeneously assigned a

56

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

Fig. 1. Histogram of administered punishment. Source: School-level responses from the restricted-access School Survey on Crime and Safety, 2003–04, 2005–06, 2007–08, 2009–10.

Fig. 2. Number of guidelines by offense type. Notes: Felony guidelines provide sentencing requirements for offenses committed using a weapon other than a firearm, associated with the possession or use of illegal drugs or alcohol, using physical harm or battery and threats of physical harm or assault, resulting in property damage or robbery, associated with gang membership, and being formally charged with a felony or delinquency charge. Misconduct guidelines provide sentencing requirements for offenses that include disobedience, absenteeism, unlawful activity, misconduct, and authority defiance. Sentencing guidelines for felony offenses associated with the possession or use of a firearm are excluded, because all states receiving federal education funds must comply with regulations of the 1994 Gun-Free Schools Act to permanently remove students committing this offense. Source: Figures created by authors using data from The Civil Rights Project (2000).

punishment type j. For example, a school may have a 100% permanent removal rate for students who possess or use a firearm if all students are permanently removed from the school for committing this offense . Regression analyses that appropriately account for substantial probability mass in the tails of the dependent variable distribution yield consistent estimates. Let X it = { Dit , S it , M it } and ˇ = {ˇ0 . . . ˇ3 } from Eq. (1) and dit∗ represent the actual percentage of a particular punishment type j.11 Assuming that dit∗ |X it ∼N(X it ˇ,  2 ), where  2 is the variance, then the three observed outcomes are represented as dit = 0 dit =

dit∗

dit = 100

11

if 0 <

E[dit |X] = 0 · Pr[dit = 0|X] + E[dit |X, 0 < dit < 100] · Pr[0 < dit < 100|X] + 100 · Pr[dit = 100|X].

(3)

Following Wooldridge (2002), we specify a log-likelihood function for modeling equation (2) as a three-part sum it = I{dit = 0} · log(˚[(−X it ˇ)/] + I{0 < dit < 100}



if dit∗ ≤ 0 dit∗

The conditional expectation of a disciplinary action is, therefore, dependent on the probability of each of the three outcomes and cannot be estimated using ordinary least squares because / X␤; that is, E[dit |X] =

< 100

if dit∗ ≥ 100

The subscript j and k are omitted for clarity.

(2)

· log

[(dit − X it ˇ)/] 



· log(˚[(X it ˇ − 100)/])

+ I{dit = 100} (4)

where I{ · } represents an indicator function for each of the three possible outcomes, and ˚(·) and (·) are the standard normal cumulative and probability density functions, respectively. This

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

log-likelihood function can be appropriately estimated using a tobit model. First-order conditions of the log-likelihood function in Eq. (4) with respect to the lth independent variable, Xtl , show that although OLS estimation of Eq. (1) yields correctly signed marginal effects, the magnitudes are biased and must be appropriately scaled (e.g., using the delta method) to account for the likelihood of an outcome being outside of a corner solution. That is, ∂E[dit |X]/∂Xtl = (˚[(X␤ − 100)/] − ˚[(− X␤)/])ˇl . After making the scaling adjustments, estimated coefficients in Eq. (1) can be interpreted as the marginal effects on disciplinary rates of a particular punishment type. For example, estimates of the marginal effects for the parameters ˇ1,k would indicate differences in the use of a particular disciplinary action attributable to an increase or decrease in the relative proportion of students characterized by the kth demographic or socioeconomic characteristic, capturing between-school disciplinary discrepancies related to schools’ student body compositions. Lastly, maximum likelihood estimators (MLEs) for non-linear specifications have been shown to be inconsistent in the presence of fixed effects (Neyman and Scott, 1948; Lancaster, 2000). Although there is ambiguity about the appropriateness of using fixed effects in binary probit and logit models (see Greene, 2004, for an overview of relevant literature), less controversy exists on implementing fixed effects in tobit regressions. Greene (2004) shows that parameters estimated by tobit models and associated marginal effects are largely unaffected by the incidental parameters problem.12 The general school discipline outcome model presented in Eq. (1) can be altered to estimate differences in disciplinary discrepancies at schools in states with above or below a guideline threshold (e.g., above or below the average number of guidelines for each punishment type—permanent removal, prolonged suspension, and light discipline). It is reasonable to assume that state-level mandates would be uncorrelated with any one school’s disciplinary decisions or other variables in the model. For example, it is more likely that district-level policies may be influenced by observed disciplinary outcomes at one or a few schools, but any correlation with state policies is likely to be weak.13 State guidelines can affect the slopes associated with a number of independent variables and we, therefore, interact indicator variables that classify whether schools are in a state with an above or below average number of guidelines, I{ · }, with all regressors in the school discipline outcome model; that is,

k dijt = I{ˇ0 + Dit ˇ1 + S it ˇ2 + M it ˇ3 + ıs + ıt } + ijt .

(5)

This specification includes both state fixed effects and controls for states with a low or high number of guidelines. These variables account for unobservable time-constant differences among schools both within and between states, respectively. Therefore, any differences identified by the interaction terms can be attributed to the disciplinary mandates.

12 Another potential issue is the use of a single-equation method rather than a systems approach. A systems method can lead to inconsistent estimates because any specification error in one equation is propagated throughout the system (Greene, 2004)). Single-equation methods confine the error to the particular equation in which it appears. For example, Shepherd (2002) implements this method to study the effects of truth-in-sentencing laws on the behavior of police, prosecutors, and criminals. 13 We find that empirical correlations between the number of state-level mandates and the independent variables in the model are statistically trivial.

57

4. Estimation results of the school discipline outcomes model We first estimate the general school discipline model (without the mandate interaction effects) to determine whether disciplinary disparities exist between schools, similarly to the inter-judge inconsistencies that have been shown to exist in the U.S. justice system. Empirical evidence of such disparities from this exploratory regression analysis may suggest that sentencing guidelines could be effective in reducing the inequities.14 We then empirically test this hypothesis and obtain a more detailed understanding of sentencing guideline impacts. Regression models are estimated for the permanent removal, prolonged suspension, and light disciplinary punishment outcomes to identify discipline inconsistencies between schools. Each of the three models is estimated using SSOCS sampling weights in a tobit specification with the lower and upper bounds of the dependent variable distribution set at 0% and 100%. Estimation results of the exploratory regression model (Table B1 in Appendix B) show that discipline rates are affected by a school’s student body racial composition, with higher permanent removal rates in schools with a greater proportion of black or Hispanic students. That is, students in these schools have a higher likelihood of being permanently removed from familiar educational environments and have restricted access to educational resources. The results also suggest that schools may trade off punishments of different severity, as increased permanent removal rates in schools with higher proportions of black students are nearly identically offset by reductions in the use of light discipline, and light discipline use is traded off for the most severe punishment in schools with higher proportion of Hispanic students. When the analysis is further separated across the five misconduct types—possession or use of a firearm, non-firearm weapon, illegal drugs, alcohol, or involvement in an altercation—the estimation results (Table B2 in Appendix B) indicate that racial discipline disparities exist in nearly every category of violent and harmful offenses in schools with a higher proportion of black students. However, disciplinary inconsistencies are absent in cases firearm possession or use. The apparent lack of between-school inconsistencies for firearm offenses could be a result of strict federal mandates such as the 1994 Gun-Free Schools Act (GFSA), providing some initial evidence that disciplinary mandates may be effective in reducing or eliminating punishment disparities. These results provide insights about the potential barriers to educational opportunities in schools with larger percentages of black or Hispanic students and about possible policy-facilitated instruments for reducing those barriers. The data provide initial evidence to support the notion that there may be differences in disciplinary behavior at schools where states mandate a higher or lower number of guidelines. For example, Table 5 shows that unconditional mean comparisons of permanent removal rates indicate that these rates are more than 5 percentage points lower in schools whose states have more sanctions for serious offenses and light disciplines are nearly 5 percentage lower in states with a higher number of guidelines for misconducts. Table 5 also indicates that differences for other variables across the subsamples of states with an above or below average number of guidelines are not economically significant. We use the school discipline outcome model in Eq. 5 to formally test the hypothesis that disciplinary discrepancies differ at schools where administrators’ are more or less constrained to make

14 It is necessary to note that the relationships estimated by the exploratory regression analysis may be induced by other factors that are correlated with the student body racial composition and are, therefore, not the primary focus of the study and are presented in Appendix B.

58

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

Table 5 Comparison of means by number and type of state punishment guidelines. Number of guidelines for serious felony offenses

Number of guidelines less harmful misconduct offenses

Below mean

Below mean

Above mean

Dependent variables 21.70 16.20*** 19.60 Permanent removal 43.10*** 38.70 Prolonged 39.60 suspension Light discipline 38.70 40.70 41.7 Independent variables Student body characteristics 15.60 11.92*** 14.94 Percent Black 12.75 13.82 12.01 Percent Hispanic 40.10 39.51 38.94 Percent free lunch 13.99 14.42 13.72 Percent below 15 Misconduct management and prevention program 0.58 0.59 0.59 Acquire parental input Train teachers 0.44 0.45 0.42 0.55 0.59** 0.53 Mentor students Assist parents 0.39 0.44*** 0.39 School administrators’ perceived crime prevention limitations Limited access 0.57 0.63*** 0.57 0.28 0.27 0.29 Student retaliation 0.65 0.71*** 0.64 Inadequate funds 0.20 0.17* 0.19 Fear of reprisal 0.39 0.39 0.37 Gang activity Community characteristics 6.60 6.98*** 6.34 Unemployment rate Per capita income 3.54 3.75*** 3.65 51.86 48.90*** 47.98 Street crime 0.25 0.32 0.27 Urban 0.55 0.46 0.53 Rural Observations

2010

1600

1950

Above mean 19.10 44.00*** 36.90***

12.87*** 14.67*** 40.94** 14.73** 0.58 0.48*** 0.61*** 0.45*** 0.64*** 0.27 0.72*** 0.19 0.42*** 7.27*** 3.61 54.00*** 0.29 0.49 1660

Notes: ***, **, and * denote a statistically significant difference between mean values in states that have a below and above average number of state punishment guidelines at the 1%, 5%, and 10% levels, respectively. School offense rates are the total offenses per student per school year. Observation counts are rounded to the nearest 10s to comply with data license restriction. Weights provided by the SSOCS were used.

disciplinary decisions. That is, we estimate differences in disciplinary discrepancies at schools in states with more (or fewer) than the average number of guidelines (3.5 for felony offense guidelines and 2.3 for misconduct guidelines) for each punishment type—permanent removal, prolonged suspension, and light discipline. Table 6 presents the regression results, with the upper panel corresponding to the analysis of guidelines targeting serious felony offenses and the bottom panel describing the effects of differences in guidelines intended for less harmful misconducts. In states with fewer than 3.5 guidelines targeting felony offenses, regression results offer statistically significant evidence that schools with a higher proportion of black or Hispanic students tend to have a higher rate of permanent removals. Specifically, schools with a 1% increase in the percentage of black or Hispanic students contributes to an average 0.15 and 0.14 percentage point increase in permanent removal rates, respectively. For schools in states with an above-average number of felony sentencing guidelines, however, there is no statistical evidence of race-based punishment disparity for permanent removals. The implications of these results are similar to those in the literature demonstrating that strict sentencing guidelines can reduce disciplinary discrepancies in the federal justice system. That is, increasing the number of sentencing guidelines targeting serious offenses (i.e., specifying standardized punishments for a larger set of felony offenses) reduces race-based punishment inconsistencies between schools for the most severe discipline category.

Table 6 Estimation results by type and number of state punishment guidelines. Permanent removal

Prolonged suspension

Light discipline

Interaction regressions with below/above the mean number of guidelines for serious felony offenses Percent black × below mean 0.151*** −0.023 −0.140** (0.051) (0.058) (0.037) 0.021 0.192*** −0.169** Percent black × above mean (0.067) (0.073) (0.038) −0.063 −0.103 0.140*** Percent Hispanic × below mean (0.042) (0.066) (0.063) Percent Hispanic × above mean 0.069 −0.112 −0.117 (0.044) (0.078) (0.085) Interaction regressions with below/above the mean number of guidelines for less harmful misconduct offenses −0.011 −0.070 0.073** Percent black × below mean (0.036) (0.053) (0.060) 0.128*** 0.113* −0.245*** Percent black × above mean (0.065) (0.071) (0.041) −0.036 −0.083 0.077* Percent Hispanic × below mean (0.042) (0.072) (0.068) 0.141*** −0.158** −0.109 Percent Hispanic × above mean (0.070) (0.076) (0.045) Notes: Estimated marginal effects are displayed, with standard errors provided in parentheses and ***, **, and * denoting statistical significance at the 1%, 5%, and 10% levels, respectively. All regressions include state and time fixed effects, misconduct management and prevention program characteristics (parental input, student mentoring, teacher training, and parental assistance), school administrators’ perceived crime prevention limitations (limited access to alternative placement, student retaliation, inadequate funds, fear of state reprisal, and gang activity), and community characteristics (local area unemployment rate, urbanization indicators, per capita personal income, and county-level crime rate). Weights provided by the SSOCS were used.

The results also show that the inequity reductions are not associated with increases in the overall use of severe punishment, which is indicated by schools’ lower use of permanent removals in states that have an above average number of guidelines. This suggests that disciplinary equity is achieved by reducing the use of severe punishment in schools with higher proportions of black or Hispanic students rather than increasing its use in schools with a larger white student body. However, there is also evidence that schools in these states may be shifting toward using less severe punishments. This dynamic relationship offers novel insights for policy makers seeking to reduce race-based disciplinary inequity. Objectives to minimize inconsistencies for the most severe punishment types—reducing potentially the largest economic consequences of punishment disparity—could be attained by introducing more disciplinary guideline. Reducing disparity rates across all punishment categories is likely more complex, however, requiring specific analyses of a policy’s effects on net disparity gains and losses. Regression results presented in Table 6 also demonstrate the effects of disciplinary mandates in states that are below or above the mean number of guidelines for less harmful misconducts.15 The estimation results in the lower panel of Table 6 indicate that increases in these types of disciplinary guidelines can contribute to inadvertent, perverse impacts on race-based disparities. That is, although these guidelines intend to establish disciplinary standards for lesser misconducts, they contribute to increases in the disparity of punishment for serious offenses. Table 6 shows that in states with either below or above average number of misconduct punishment guidelines, increases in the proportion of black or Hispanic students are associated with higher rates of permanent removal. Moreover,

15 The SSOCS data do not contain disciplinary information for students committing misconduct offenses and we cannot, therefore, directly evaluate the role of behavioral misconduct guidelines in affecting disciplinary disparity for the misconduct offenses.

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

in states with above average number of sentencing guidelines for behavioral misconducts, the marginal effects of a one percentage increase in black or Hispanic students are nearly twice as large as in states with below average number of misconduct mandates. These seemingly unexpected outcomes may be suggestive of absent marginal deterrence effects associated with a gradient of offense seriousness.16 Stigler (1970) shows that marginal costs for committing offenses must be proportional to the degree of seriousness of an offense in order to ensure that there are appropriate disincentives to commit the more serious offenses; these concepts are further advanced by Wilde (1992), Friedman and Sjostrom (1993), and Mookherjee and Png (1994), among others. One of the primary implications of these studies is that if similarly aggressive punishments were applied for both less serious and more serious crimes, then individuals would have a greater incentive to commit the more serious offense because the expected utility from committing the graver crime would exceed that of committing a lesser crime. To borrow an example from Stigler, 1970, pg 527), “[i]f the thief has his hand cut off for taking five dollars, he had just as well take $5000.” In the context of this study, state-level disciplinary statutes that mandate similar levels of punishment for both misconducts and serious offenses could increase incentives for schools with disparate disciplinary behaviors to exacerbate their actions by using a greater amount of more severe punishments. That is, because sentencing guidelines for behavioral misconducts mandate punishments of similar severity (permanent removal and prolonged suspensions) to those used for punishing felony offenses, schools are less likely to use lighter punishment for students committing serious offenses. In effect, discipline guidelines for lesser misconducts may establish punishment severity floors across the ranking of disciplinary alternatives across offense types, which could inadvertently exacerbate race-based disparities. Conversely, if mandated punishments were more dissimilar for different levels of misconduct severity, schools with inconsistent disciplinary behaviors would be less likely to apply that behavior in a way that unequally increases barriers to black and Hispanic students’ educational opportunities. These results suggest important policy implications. Specifically, that it is important to recognize and evaluate the potential dependencies and marginal deterrence effects among different offense levels and sentencing guidelines such that the absolute level of discipline is appropriate for the offense and that sentencing guidelines for lesser offenses do not affect the severity of punishment for more serious ones. An inability or failure to do so can result in ineffective guidelines or worse, unintended outcomes that amplify disparities. 5. Conclusions and policy implication Strict sentencing guidelines have been shown to be important tools for reducing disciplinary disparities. This study investigates conditions whether standardized punishment guidelines for

59

punishing U.S. public high school student’s also aids in reducing race-based punishment inconsistencies across schools. We use data describing misconduct and punishment outcomes in U.S. public high schools to estimate the impacts of student body, school, and community attributes on between-school disciplinary variation. The estimation results offer robust evidence that racial characteristics of a school’s student body affect disciplinary inconsistencies for both violent and harmful offenses, especially in the use of the most severe punishment types. However, we find evidence that such disparities (and potential adverse economic impacts of the disparities) are reduced in states that have a higher number of sentencing guidelines mandating standardized punishment for serious offenses. These results are largely consistent with studies evaluating the impacts of the 1984 Sentencing Reform Act and other research, such as Frakes (2013) who shows that health treatment consistency across physicians increases for those physicians who follow national medical standards. Our study provides specific inferences about discipline disparity in public education systems. For example, while schools’ zero-tolerance policies have been scrutinized as practices that inequitably target individuals, we provide suggestive evidence that for serious offense types, disciplinary guidelines can dampen inconsistencies. Moreover, reductions in race-based punishment disparities are not achieved by increasing punishment rates and severity for other students, which is in contrast to criticisms of sentencing guidelines that claim these policies unnecessarily increase punishment levels for some offenders (for example, see Lott, 1990; Waldfogel, 1994; Kessler and Piehl, 1998). We also show that strict guidelines may not be universally effective and could have unintentional, counter-productive consequences when the policies reduce or eliminate marginal deterrence effects. When sentencing guidelines mandate similar discipline for different offenses, a shift toward using more severe punishments could occur. We show that schools in states with more guidelines for punishing misconduct offenses have higher race-based disparities in the level and severity of discipline for serious offenses, possibly because the guidelines for less harmful offenses establish punishment severity floors (Kessler and Piehl, 1998, show similar behavior in criminal prosecution cases). Policy makers should consider both the overall objective of sentencing guidelines (e.g., whether to reduce the overall level of disparity or to lower inconsistencies for a particular offense type) and possible dependencies among mandates. Policy makers may also need to recognize crime transfer effects, which may cause offenses to be committed in other locations or increases in the rates of less serious crimes (for example, Shepherd, 2002, observes such substitution as a result of truth-in-sentencing legislation). Carefully designed and appropriately implemented disciplinary guidelines can reduce punishment disparities and, in the case of education, reduce barriers to learning opportunities. Appendix A. Variable definitions and sources Table A1 Variable labels, definitions, and sources.

Label

Definition

Source

Firearm Weapon Non-firearm Weapon Illegal Drugs Alcohol Physical Altercations Prolonged Suspension Expulsion Transfer

Use/possession of a firearm with the intent to threaten, injure, or kill. Use/possession of any instrument (other than firearms) used with the intent to threaten, injure, or kill. Distribution, possession, or use of illegal drugs. Distribution, possession, or use of alcohol An actual and intentional touching or striking of another person against his or her will. Out-of-school suspensions lasting 5 or more days, but less than the remainder of the school year. Removal with no continuing school services. Transferred to specialized delinquent student school due to disciplinary reasons.

SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS

16 We would like to thank an anonymous reviewer for bringing our attention to this concept.

60

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

Table A1 (Continued) Label

Definition

Source

Permanent Removal Black Hispanic Free Lunch Below 15 Limited Access

Expulsions and transfers. Percentage of student body racial composition that is black. Percentage of student body racial composition that is Hispanic. Percentage of students eligible for free or reduced-price lunch. Percentage of students below the 15th percentile on standardized tests. Binary indicator to whether the principal or disciplinarian perceives that the lack of or inadequate alternative placement/programs for disruptive students limits the efforts to reduce or prevent misconduct. Binary indicator to whether the principal or disciplinarian perceives that teachers’ fear of student retaliation limits the efforts to reduce or prevent misconduct. Binary indicator to whether the principal or disciplinarian perceives that inadequate funds limit the efforts to reduce or prevent misconduct. Binary indicator to whether the principal or disciplinarian perceives that fear of district or state reprisal limits the efforts to reduce or prevent misconduct. Binary indicator to whether a formal process exists to obtain parental input on misconduct and discipline policies. Binary indicator to whether teachers and staff are trained by security staff on safety or misconduct prevention Binary indicator to whether a student-security staff mentoring program exists. Binary indicator to whether training or technical assistance is provided to parents in dealing with students’ problem behavior. Binary indicator to whether any gang activity has occurred at the school. Local area unemployment rate (in percent). Per capital personal income (in 10,000 of 2009 dollars) Rate of criminal offenses per 1000 persons in the schools’ zip code (in percent). Binary indicator to whether the school is located in a metropolitan area. Binary indicator to whether the school is located in an urban area. Binary indicator to whether the school is located in a rural area.

SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS

Retaliation Inadequate Funds Fear of Reprisal Parental Input Train Teachers Mentor Students Assist Parents Gang Activity Unemployment Rate Per capita Income Street Crime Metro Urban Rural

SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS SSOCS BLS BEA FBI SSOCS SSOCS SSOCS

Appendix B. Supplementary estimation results

Table B1 Estimation Results of the School Discipline Outcomes Model Permanent removal Black Hispanic Free lunch Below 15 Acquire parental input Train teachers Mentor students Assist parents Unemployment rate Per capita income Street crime Gang activity Limited access Student retaliation Inadequate funds Fear of reprisal

0.112*** (0.040) 0.126*** (0.039) −0.065* (0.035) 0.115*** (0.039) 0.541 (0.859) 1.069 (0.973) 1.182 (1.441) −0.919 (1.463) −0.394 (0.348) −1.267 (0.808) 0.011 (0.031)

0.095** (0.039) 0.100** (0.040) −0.067* (0.034) 0.106*** (0.039) 0.606 (0.853) 0.750 (0.983) 0.811 (1.428) −0.939 (1.487) −0.369 (0.345) −1.261 (0.812) 0.010 (0.033) 4.353*** (0.954)

Prolonged suspension 0.094** (0.039) 0.102*** (0.039) −0.067** (0.033) 0.099** (0.039) 0.565 (0.844) 0.722 (0.992) 0.846 (1.406) −0.833 (1.428) −0.385 (0.339) −1.208 (0.839) 0.011 (0.032) 3.876*** (0.955) 0.092 (1.238) 2.580** (1.084) 0.828 (1.225) −0.317 (2.380)

0.071 (0.057) −0.052 (0.065) −0.020 (0.058) −0.013 (0.054) 1.084 (1.333) 0.509 (1.557) 5.970*** (1.875) −1.168 (1.247) 0.721 (0.598) 1.056 (1.046) 0.057 (0.042)

0.047 (0.059) −0.093 (0.068) −0.025 (0.056) −0.028 (0.056) 1.190 (1.313) 0.005 (1.582) 5.511*** (1.800) −1.169 (1.276) 0.798 (0.569) 1.122 (1.025) 0.053 (0.040) 6.359*** (1.665)

Light discipline 0.046 (0.058) −0.089 (0.067) −0.025 (0.056) −0.036 (0.056) 1.206 (1.300) −0.175 (1.564) 5.732*** (1.786) −0.899 (1.296) 0.849 (0.571) 1.090 (1.036) 0.053 (0.040) 6.012*** (1.732) 1.939 (1.872) 2.740 (2.184) 0.541 (2.060) −3.650** (1.590)

−0.192*** (0.059) −0.182** (0.071) 0.128* (0.072) −0.078 (0.058) −1.513 (1.757) −3.573** (1.635) −10.63*** (2.504) 2.609 (1.894) −0.139 (0.622) 0.158 (0.836) −0.078 (0.050)

−0.142** (0.059) −0.099 (0.078) 0.133** (0.067) −0.053 (0.058) −1.759 (1.718) −2.545 (1.617) −9.335*** (2.316) 2.578 (1.951) −0.311 (0.578) 0.032 (0.791) −0.074 (0.050) −13.26*** (1.849)

−0.139** (0.059) −0.103 (0.077) 0.133** (0.067) −0.039 (0.057) −1.649 (1.619) −2.331 (1.615) −9.511*** (2.305) 2.203 (1.915) −0.348 (0.584) 0.008 (0.803) −0.075 (0.048) −12.410*** (1.968) −1.976 (1.970) −4.721** (2.327) −1.984 (2.496) 3.683 (2.752)

Notes: Estimated marginal effects are displayed, with standard errors provided in parentheses and ***, **, and * denoting statistical significance at the 1%, 5%, and 10% levels, respectively. All regressions include state and time fixed effects and urbanization indicators. Weights provided by the SSOCS were used. The estimated relationships may be induced by other factors that are correlated with the student body racial composition and are, therefore, used only as an initial empirical exploratory regression analysis. To demonstrate the potential impacts of excluding variables characterizing administrators’ perception about possible constraints to managing student offenses, a baseline model (with no administrator perception variables) is presented for comparison with a model with only the perceived gang activity variable and a model with all administrator perception variables.

A. Bekkerman, G.A. Gilpin / International Review of Law and Economics 40 (2014) 51–61

61

Table B2 Estimation results of the school discipline outcomes model, by offense type. Violent offenses

Permanent removal Black Hispanic Prolonged suspension Black Hispanic Light discipline Black Hispanic Observations

Harmful offenses

Firearms

Non-firearm weapons

Illegal drugs

Alcohol

Altercations

0.210 (0.193) −0.305 (0.210)

0.239*** (0.080) 0.131 (0.097)

0.162*** (0.059) 0.074 (0.067)

0.082** (0.041) −0.018 (0.043)

0.077*** (0.023) 0.015 (0.026)

−0.084 (0.106) 0.086 (0.114)

−0.234*** (0.083) −0.135 (0.102)

−0.144** (0.062) −0.160** (0.073)

0.041 (0.085) −0.043 (0.092)

0.173*** (0.050) −0.055 (0.062)

−0.005 (0.006) −0.001 (0.002)

0.012 (0.057) 0.069 (0.063)

−0.001 (0.047) 0.016 (0.047)

−0.096 (0.072) −0.015 (0.072)

−0.204*** (0.054) −0.075 (0.060)

350

1500

2620

1940

3000

Notes: Estimated marginal effects are displayed, with standard errors provided in parentheses and ***, **, and * denoting statistical significance at the 1%, 5%, and 10% levels, respectively. All regressions include state and time fixed effects, misconduct management and prevention program characteristics (parental input, student mentoring, teacher training, and parental assistance), school administrators’ perceived crime prevention limitations (limited access to alternative placement, student retaliation, inadequate funds, fear of state reprisal, and gang activity), and community characteristics (local area unemployment rate, urbanization indicators, per capita personal income, and county-level crime rate). Observation counts are rounded to the nearest 10s to comply with data license restriction. Weights provided by the SSOCS were used. The estimated relationships may be induced by other factors that are correlated with the student body racial composition and are, therefore, used only as an initial empirical exploratory regression analysis.

References Abrams, B.M., David, S., Mullainathan, S., 2013. Do Judges Vary in Their Treatment of Race?, Working paper. Acemoglu, D., Angrist, J., 2001. How large are human-capital externalities? Evidence from compulsory schooling laws. NBER Macroecon. Annu. 15, 9–59. Anderson, J.K., James, Stith, K., 1999. Measuring interjudge sentencing disparity: before and after the federal sentencing guidelines. J. Law Econ. 42, 271–308. Angrist, J.D., Krueger, A.B., 1991. Does compulsory school attendance affect schooling and earnings? Q. J. Econ. 106, 979–1014. Ashenfelter, O., Eisenberg, T., Schwab, S.J., 1995. Politics and the judiciary: the influence of judicial background on case outcomes. J. Leg. Stud. 24, 257–281. Bekkerman, A., Gilpin, G., 2014. Assessing Disciplinary Disparities in Schools: Do Student Body Characteristics Matter?, Unpublished working paper, Bozeman, MT. Bernburg, J., Krohn, M., 2003. Labeling, life chances, and adult crime: the direct and indirect effects of official intervention in adolescence on crime in early adulthood. Criminology 41, 1287–1318. Fender, B.F., Brooking, C., McNair, L., Mehta, D., Eastlack, J., Pierce, B., 2006. The impact of race and gender on sentencing for Mississippi’s indigent population. J. Econ. 32, 57–73. Finn, J., 1989. Withdrawing from school. Rev. Educ. Res. 59, 117–142. Frakes, M., 2013. The impact of medical liability standards on regional variations in physician behavior: evidence from the adoption of national-standard rules. Am. Econ. Rev. 103 (1), 257–276. Friedman, D., Sjostrom, W., 1993. Hanged for a sheep: the economics of marginal deterrence. J. Leg. Stud. 22, 345–366. Greene, W., 2004. The behaviour of the maximum likelihood estimator of limited dependent variable models in the presence of fixed effects. Econ. J. 7, 98–119. Iyengar, R., 2011. Who’s the fairest in the land? Analysis of judge and jury death penalty decisions. J. Law Econ. 54, 693–722. Kessler, D., Piehl, A.M., 1998. The role of discretion in the criminal justice system. J. Law Econ. Org. 14, 256–276. Kinsler, J., 2013. School discipline: a source or salve for the racial achievement gap? Int. J. Econ. Rev. 54, 355–383. Kinsler, J., 2011. Understanding the Black–White school discipline gap. Econ. Educ. Rev. 30, 1370–1383. Lancaster, T., 2000. The incidental parameter problem since 1948. J. Econom. 95, 391–413. Lleras-Muney, A., 2002. Were compulsory attendance and child labor laws effective? An analysis from 1915 to 1939. J. Law Econ. 45, 401–435. Lott, J.J., 1990. The effect of conviction on the legitimate income of criminals. Econ. Lett. 34, 381–385. ´ S., 2011. The crime reducing effect of education. Econ. J. Machin, S., Marie, O., Vujic, 121, 463–484. McCarthy, K., 2000. The Effects of Student Activity Participation, Gender, Ethnicity,and Socio-Economic Level on High School Student Grade Point Averages and Attendance. Education Resources Information Center.

Mookherjee, D., Png, I.P., 1994. Marginal deterrence in enforcement of law. J. Polit. Econ. 102, 1039–1066. Mustard, D.B., 2001. Racial, ethnic, and gender disparities in sentencing: evidence from the U.S. Federal Courts. J. Law Econ. 44, 285–314. Neyman, J., Scott, E., 1948. Consistent estimates based on partially consistent observations. Econometrica 16, 1–32. Oreopoulos, P., 2009. Would More Compulsory Schooling Help Disadvantaged Youth? Evidence from Recent Changes to School-Leaving Laws. University of Chicago Press, pp. 85–112 (Chapter 3). Rocque, M., 2010. Office discipline and student behavior: does race matter? Am. J. Educ. 116, 557–581. Schanzenbach, M., 2005. Racial and sex disparities in prison sentences: the effect of district-level judicial demographics. J. Leg. Stud. 34, 57–92. School Survey on Crime and Safety, 2004. U.S. Department of Education, National Center for Education Statistics. School Survey on Crime and Safety, 2006. U.S. Department of Education, National Center for Education Statistics. School Survey on Crime and Safety, 2008. U.S. Department of Education, National Center for Education Statistics. School Survey on Crime and Safety, 2010. U.S. Department of Education, National Center for Education Statistics. Shepherd, J., 2002. Police, prosecutors, criminals, and determinate sentencing: the truth about truth-in-sentencing laws. J. Law Econ. 45, 509–534. Spohn, C., 1990. The sentencing decisions of Black and White judges: expected and unexpected similarities. Law Soc. Rev. 24, 1197–1216. Stigler, G.J., 1970. The optimum enforcement of laws. J. Polit. Econ. 78, 526–536. The Civil Rights Project, 2000, June. Opportunities Suspended: The Devastating Consequences of Zero Tolerance and School Discipline. Harvard University, Working paper. Uniform Crime Reports: Arrests by Age, Sex, and Race, 2003. U.S. Department of Justice, Bureau of Justice Statistics [Computer file]. Inter-university Consortium for Political and Social Research [distributor], Ann Arbor, MI. Uniform Crime Reports: Arrests by Age, Sex, and Race, 2005. U.S. Department of Justice, Bureau of Justice Statistics [Computer file]. Inter-university Consortium for Political and Social Research [distributor], Ann Arbor, MI. Uniform Crime Reports: Arrests by Age, Sex, and Race, 2007. U.S. Department of Justice, Bureau of Justice Statistics [Computer file]. Inter-university Consortium for Political and Social Research [distributor], Ann Arbor, MI. Uniform Crime Reports: Arrests by Age, Sex, and Race, 2009. U.S. Department of Justice, Bureau of Justice Statistics [Computer file]. Inter-university Consortium for Political and Social Research [distributor], Ann Arbor, MI. Waldfogel, J., 1994. The effect of criminal conviction on income and the trust “reposed in the workmen”. J. Hum. Resour. 29, 62–81. Wilde, L.L., 1992. Criminal choice, nonmonetary sanctions and marginal deterrence: a normative analysis. Int. Rev. Law Econ. 12, 333–344. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press, Cambridge, MA.