Please read the following article which was provided in the “Reading & Study” folder for this week: 

Sabia, J. J., & Bass, B. (2017). Do anti-bullying laws work? New evidence on school safety and youth violence. Journal of Population Economics, 30(2), 473-502.  

Then answer the following questions:

1. What does the data tell us?
2. If you were a lawmaker or policy maker, what changes would you make on the data? 
3. Do you feel the methodology is sound, or do you feel there are questions unanswered?

ORIGINAL PAPER

Do anti-bullying laws work? New evidence on school
safety and youth violence

Joseph J. Sabia1,2,3 & Brittany Bass4

Received: 28 March 2016 /Accepted: 5 October 2016 /Published online: 22 November 2016
# Springer-Verlag Berlin Heidelberg 2016

Abstract This study is the first to comprehensively examine the effect of state anti-
bullying laws (ABLs) on school safety and youth violence. Using existing data from
the Youth Risk Behavior Surveys and the Uniform Crime Reports, and newly-collected
data on school shootings, we find little evidence that the typical state ABL is effective
in improving school safety and student well-being. However, this null finding masks
substantial policy heterogeneity. State mandates that require school districts to imple-
ment strong, comprehensive anti-bullying policies are associated with a 7 to 13 %
reduction in school violence and an 8 to 12 % reduction in bullying. In addition, our
results show that strong anti-bullying policy mandates are associated with a reduction
in minor teen school shooting deaths and violent crime arrests, suggesting potentially
important spillover effects.

Keywords Bullying . Youth violence . Anti-bullying laws . School shootings

JEL Classification I28 . I18 . K0

J Popul Econ (2017) 30:473–502
DOI 10.1007/s00148-016-0622-z

Responsible editor: Erdal Tekin

* Joseph J. Sabia
joseph.sabia@unh.edu

Brittany Bass
bassb@uci.edu

1 San Diego State University, San Diego, CA, USA
2 University of New Hampshire, Durham, NH, USA
3 Institute for the Study of Labor (IZA), Bonn, Germany
4 University of California–Irvine, Irvine, CA, USA

http://crossmark.crossref.org/dialog/?doi=10.1007/s00148-016-0622-z&domain=pdf

1 Introduction

“[Following] the Columbine High School massacre, school shootings have gone
from a rare, shocking aberration to a frequent, yet still shocking, tragedy. [And]
even if they’re not getting beat up or shot at, students routinely endure verbal
assaults and torment from other students—often as school faculty look the other
way, or worse, join in. For many children and teens across the country, school
feels like a hostile, oppressive, and dangerous place.” (Klein 2012).

Bullying is defined by the U.S. Department of Health and Human Services
as “unwanted, aggressive behavior among school-aged children that involves a
real or perceived power imbalance, and the behavior is repeated or has the
potential to be repeated” (U.S. Department of Health and Human Services
2014a). It can take the form of verbal abuse or physical violence, including
behavior that crosses criminal thresholds, and may be transitory or permanent in
nature (American Psychological Association 2014). In 2011, nearly 7 million
youths, or approximately 28 % of those ages 12–18, were bullied in the United
States (National Center for Education Statistics 2012). While bullying has been
documented across all demographic groups, historically marginalized groups,
including lesbian/gay/bisexual/transgender (LGBT) youths (Friedman et al.
2006; Daley et al. 2008; Kosciw et al. 2009), youths with disabilities (Blake
et al. 2014; Turner et al. 2011; Cappadocia et al. 2012), females (Faris and
Felmlee 2011; Kumpulainen et al. 1999; Craig 1998), and racial minorities
(Langdon and Preble 2007; Fox and Stallworth 2005; Carlyle and Steinman
2007) are disproportionately affected.

A wide interdisciplinary literature across social psychology, public health, and
economics has found that bullying victimization is negatively related to students’ health
and human capital acquisition. Bullying victimization is linked to diminished psycho-
logical well-being (Hansen and Lang 2014; Juvonen et al. 2010; Duncan 1999; Seals
and Young 2003; Bond et al. 2001), reduced student engagement (Nansel et al. 2001),
poor academic performance (Eriksen et al. 2014; Konishi et al. 2010), and less social
connectedness (Nansel et al. 2001; Nansel et al. 2004). In addition, repeated bullying is
associated with attempted or completed suicides (Kaltiala-Heino et al. 1999; Carney
2000) and increased risk of school shootings (Klein 2012).

In response to increased public awareness of bullying and its potentially adverse
health and human capital consequences, many western nations have adopted anti-
bullying laws. For example, in the United Kingdom and Belgium, all state-funded
schools are legally required to implement anti-bullying programs (U.K. Department
for Education 2013; Stevens et al. 2000); in Finland, 95 % of state-funded schools
have adopted a common anti-bullying curriculum (Finland Ministry of Education
and Culture 2014); and in Sweden, school districts are required to take necessary
precautions to protect students from schoolyard bullying or face civil liability
(Sweden Ministry of Education and Research 2016). In the United States, most
anti-bullying reforms enacted have been at the state and local levels (U.S.
Department of Education 2011).

474 J.J. Sabia, B. Bass

All 50 U.S. states and the District of Columbia have enacted anti-bullying laws
(ABLs) that impose mandates on local school districts to adopt anti-bullying policies.
Despite their universal adoption, there is substantial heterogeneity across states in the
comprehensiveness and strictness of these mandates. Some ABLs provide non-binding
recommendations for school bullying policies, while others include comprehensive
provisions requiring school districts to (1) keep written records of students’ reporting of
bullying to school officials, as well as each incident’s resolution; (2) adopt a graduated
range of sanctions for acts of bullying; and (3) provide training for school staff on
preventing, identifying, and responding to bullying. Other ABLs require schools to
report incident-by-incident school responses to bullying to both the state Department of
Education and parents.

There are a number of channels through which ABLs could reduce student violence.
In a rational crime (or “rational bullying”) framework (Becker 1968), potential bullies
weigh the expected costs and benefits of bullying. ABLs may raise the expected costs
of bulling by increasing the probability of punishment via increased school monitoring
and by reducing victims’ reporting costs. ABLs may also increase the severity of
punishments through graduated sanctions for bullying. In addition, the educational
components of ABLs may change students’ tastes for bullying, reducing potential
bullies’ expected benefits. Finally, provisions of ABLs that require schools to make
bullying policies, and school responses to violations of those policies, publicly avail-
able may incentivize reputation-minded schools to more efficiently allocate resources
to deter bullying. Note that ABLs may be effective in deterring bullying even if
students are unaware of their state implementing an ABL; its bindingness only requires
that school districts, including teachers and administrators, are aware of the policy and
react to it, a supposition for which there is support.1

On the other hand, there may be unintended consequences of anti-bullying legisla-
tion. By mandating that schools devote additional resources to combat bullying, often
without additional state dollars, ABLs may require schools to change their mix of
inputs to produce education. This could result in substitution away from other inputs
(e.g., extracurricular activities, athletics, and teacher quality) that foster social connect-
edness and increase potential bullies’ opportunity costs of time spent bullying. Addi-
tionally, because ABLs only raise the costs of bullying on school property, these laws
may simply shift bullying off of school property rather than reduce the overall
prevalence of bullying. Finally, the provision of public information on school bullying
prevention policies may increase the likelihood that potential bullies are able to avoid
detection.

Using data from the Youth Risk Behavior Surveys, the Uniform Crime Reports, and
newly collected school shooting data, we provide the first comprehensive study of the

1 To take one example, thousands of school employees in the state of New Jersey attended training sessions on
the state’s new anti-bullying law (Hu 2011); over 200 districts purchased anti-bullying policy manuals, DVD
materials, and held staff training sessions. To take another example, schools in Broward County, Florida
require the school principal (or appropriate administrator) to provide information on the process for reporting
incidents at the beginning of each school year. In addition, principals provide information on the investigatory
and appellate processes via the Student Code of Conduct, Employee Handbooks, assembly meetings, and the
school website (Broward County Public Schools. Anti-Bullying Policy 5.9 & Procedural 2010). Finally, in
several districts in Washington state, school staff receive annual training on anti-bullying policies and
procedures, including staff roles and responsibilities; anti-bullying strategies and expectations are incorporated
into the counseling and guidance curriculum (Washington Clover Park School District 2016).

Do anti-bullying laws work? 475

effects of state ABLs on school safety, bullying, and violence. Difference-in-difference
estimates show that the typical state ABL has little effect on student safety or school
violence. However, the implementation of comprehensive, strong school district policy
mandates is associated with a 7 to 13 % reduction in high school violence and an 8 to
12 % reduction in bullying. Moreover, we find that strong ABLs are associated with
reductions in minor teen school shooting deaths and minor teen violent crime arrest
rates, suggesting additional important social benefits of anti-bullying policies.

2 Background

According to a 2013 national survey, 92 % of parents with minor children believe that
bullying contributes to violence in the United States (Peters 2013). Moreover, 78 % of
adults believe that bullying prevention programs should be part of school curricula
(Bushaw and Lopez 2012). Reflecting this concern, public health and education
agencies have taken a more high profile role to combat bullying. In 2011, the
Department of Health and Human Services (DHHS), in conjunction with the Depart-
ment of Education’s Federal Partners in Bullying Prevention Steering Committee
(FPBPSC), launched Stopbullying.gov to provide information to parents, school offi-
cials, and students on how to identify, prevent, and respond to bullying (U.S.
Department of Health and Human Services 2014b).

A number of private not-for-profit firms have also taken action to deter school
bullying. For example, the Parent Advocacy Coalition for Educational Rights’
(PACER) National Bullying Prevention Center was founded in 2006 to

“…actively lead social change, so that bullying is no longer considered an
accepted childhood rite of passage. PACER provides innovative resources for
students, parents, educators, and others, and recognizes bullying as a serious
community issue that impacts education, physical and emotional health, and the
safety and well-being of students.” (PACER 2014)2

Studies have shown that bullying victimization is negatively related to health and
academic performance (Rothon et al. 2011; Wolke et al. 2013; Wilkins‐Shurmer et al.
2003; Hepburn et al. 2012; Bond et al. 2001; Glew et al. 2005; Kim Young et al. 2005;
Gini and Pozzoli 2009). Victims of bullying have been found to be more emotionally
distressed (Gladstone et al. 2006; O’Brennan et al. 2009; Duncan 1999; Nansel et al.
2001), less socially connected (O’Brennan et al. 2009; Eisenberg et al. 2003; Juvonen
et al. 2003), and less academically prepared (Eisenberg et al. 2003; Strøm et al. 2013;
Juvonen et al. 2010) than their non-bullied counterparts. Interestingly, perpetrators of
bullying have been found to be in worse mental health (Undheim and Sund 2010; Ng
Josephine and Sandra Km 2008; Seals and Young 2003; Saluja et al. 2004; Arseneault
et al. 2006; Bender and Friedrich 2011), more likely to abuse alcohol and drugs (Radliff
et al. 2012; Tharp-Taylor et al. 2009), and more likely to carry weapons (Nansel et al.
2004) than those who do not bully. Bullying has also been linked to school shootings,

2 To take another example, Bully Police USA, a high-profile private watchdog group, advocates for the
adoption of strict state and local anti-bullying legislation.

476 J.J. Sabia, B. Bass

either by bullies carrying out acts of violence against perpetrators or victims of bullying
seeking revenge on their tormenters (see, for example, Klein 2012).3

An important limitation of this literature, however, is that prior studies have
treated victimization as exogenously determined. This assumption may be
problematic if difficult-to-measure characteristics of bullies and their targets,
such as discount rates, personality, or family background characteristics, are
related to both the probability of victimization and student outcomes. Recent
work by Eriksen et al. (2014) attempts to disentangle the human capital effects
of bullying from selection and finds evidence of a causal link.

A number of recent studies have examined the relationship between anti-
bullying policies and student well-being. However, these studies have been either
case studies of particular school policies (Jeong and Lee 2013; Salmivalli et al.
2011) or focused on one or two states (Green et al. 2014). Those that have
examined state policies more broadly have used a simple before-after estimator
(Fekkes et al. 2006) or relied on cross-state variation in policies for identification
(Hatzenbuehler et al. 2015).4 The results of these studies have been mixed. A
study of Texas schools found higher bullying prevalence in schools with anti-
bullying policies as compared to those without such policies (Jeong and Lee
2013); a comparison of high school bullying rates in Delaware and Illinois found
lower bullying rates in Delaware, which has a stronger anti-bullying statute
(West 2014); and a study of Finnish schools found that students in schools with
an anti-bullying program faced 1.32 to 1.94 times less bullying than students in
schools without such programs (Salmivalli et al. 2011). The study most similar to
ours, Hatzenbuehler et al. (2015), uses a single wave of the National Youth Risk
Behavior Survey and cross-state policy variation for identification. These authors
find some evidence that states with stricter anti-bullying laws have lower rates of
bullying and cyberbullying victimization in high schools.

Our study contributes to the existing literature in several ways. This study is
the first to use within-state policy variation to comprehensively examine the
effect of state ABLs on school safety, bullying, and youth violence, including
school shootings. We undertake a number of empirical strategies to provide
support for the common trends assumption of our difference-in-difference
models, including falsification tests on older young adults for whom state ABLs
should not bind. Second, given considerable differences across states in anti-
bullying statutes (U.S. Department of Education 2011), we carefully explore
heterogeneity in policy effects by type of ABL. Finally, we examine whether
the effects of state ABLs extend to behavior that crosses the criminal threshold,
including school shootings.

3 To take three examples from Klein (2012), (1) on February 12, 2007 in E.O. Green Junior High School in
Oxnard California, Brandon McInerney (age 14) shot and killed classmate Larry King because King was gay
and McInerney was “disgusted” by King’s “flamboyant behavior”; (2) on March 21, 2005, Jeff Weise (age 16)
shot five students and one adult staff member, and committed suicide, in part, “because he was teased because
he was heavy and wore ‘Goth’ clothing”; and (3) on February 2, 1996, Barry Loukatis (age 14) shot and killed
two students (including the student who reportedly bullied him) because he was a repeated victim of sexual
orientation–based taunting and teasing.
4 Due et al. (2005) conduct cross-national comparisons of anti-bullying laws.

Do anti-bullying laws work? 477

3 Data and measures

3.1 YRBS data

Our initial analysis uses data drawn from repeated cross-sections of both the National
and State Youth Risk Behavior Surveys (YRBS) from 1993 to 2013. The National
YRBS is conducted biennially by the Centers for Disease Control and Prevention
(CDC) and, when weighted, is representative of the population of U.S. high school
students. The State YRBS surveys are also administered to high school students and
contain most of the questions in the National YRBS (NYRBS). While the state surveys
are coordinated by the CDC, they are usually conducted by state education and health
agencies.5 The augmentation of National with State YRBS data has been employed in a
number of recent studies examining the effects of a number of state-level public
policies on risky behaviors.6 The YRBS is well suited for this study because it contains
data on several measures of school safety, violent behavior, and bullying.

Using the YRBS data, we identify four key measures of school safety. First, we
measure whether the respondent avoided school because of concerns about safety
issues using answers to the following questionnaire item:

During the past 30 days, on how many days did you not go to school because you
felt you would be unsafe at school or on your way to or from school?

We generate a dichotomous variable, Unsafe, set equal to 1 if the student reported a
positive number of days not attending school and equal to 0 otherwise. In the YRBS,
6.3 % of respondents in our sample reported not attending school at least one day in the
last 30 days because they felt unsafe (see Table 1). In addition to the above coding, we
also experiment with creating a continuous measure of this outcome.

Next, respondents were asked whether they had been in a physical altercation on
school property during the previous year:

During the past 12 months, how many times were you in a physical fight on
school property?

School Fight is coded equal to 1 if the student reported being in a physical fight on
school property at least once during the past 12 months and 0 otherwise. We find that
12.2 % of respondents reported having been in a physical fight on school property.
Note that because this survey item asks respondents about safety in the prior year rather
than the prior 30 days, we experiment with lagging our ABL policy variable to better
align the timing of the safety outcome with the effective date of the policy.

As noted above, ABLs could incentivize bullies to change the location of bullying
rather than reduce total bullying. Therefore, we also generate a dichotomous variable,

5 Estimates from the state YRBS are designed to be representative at the state level, but recent research with
these data has utilized Census population estimates to introduce weights that will make these data represen-
tative at the national level as well (Anderson and Elsea 2015; Sabia and Anderson 2016; Sabia et al. 2016).
6 Included among these policies are cigarette taxes (Hansen et al. 2016), medical marijuana laws (Anderson
et al. 2015), parental involvement laws for abortion (Sabia and Anderson 2016), and minimum wages (Sabia
et al. 2016).

478 J.J. Sabia, B. Bass

Table 1 Means of outcomes and key control variables, by data source

Mean (SD) [N] Source Years

Dependent variables

Unsafe 0.063 (0.243)
[1,105,255]

YRBS

1993–2013

School Fight 0.122 (0.327)
[1,054,461]

YRBS 1993–2013

All Fight 0.312 (0.463)
[1,031,970]

YRBS 1993–2013

Threat 0.080 (0.271)
[1,070,208]

YRBS 1993–2013

Bullied 0.201 (0.401)
[412,666]

YRBS 2009–2013

Not attending high school 0.146 (0.353)
[1,415,700]

CPS 1993–2013

Violent crimea (ages 13–17) 325.24 (232.45) [979]

FBI Uniform Crime Reports 1993–2012

Violent crimea (ages 20–24) 418.76 (234.45) [980] FBI Uniform Crime Reports 1993–2012

Property crimea (ages 13–17) 1999.90 (1060.77)
[981]

FBI Uniform Crime Reports 1993–2012

Property crimea (ages 20–24) 1163.62 (451.96)
[981]

FBI Uniform Crime Reports 1993–2012

Death (minor teen) 0.104 (0.305) [1020] Anderson and Sabia (2016) 1993–2012

Death (older student) 0.024 (0.152) [1020] Anderson and Sabia (2016) 1993–2012

Homicide (minor teen) 0.060 (0.237) [1020] Anderson and Sabia (2016) 1993–2012

Homicide (older student) 0.017 (0.128) [1020] Anderson and Sabia (2016) 1993–2012

Anti-bullying laws

ABL 0.365 (0.472) Department of Education 1993–2013

Strong ABL 0.139 (0.344) Department of Education 1993–2013

Moderate ABL 0.172 (0.370) Department of Education 1993–2013

Weak ABL 0.054 (0.221) Department of Education 1993–2013

Strong district policy 0.155 (0.357) Department of Education 1993–2013

Moderate district policy 0.151 (0.354) Department of Education 1993–2013

Weak district policy 0.059 (0.230) Department of Education 1993–2013

Written records and reporting 0.051 (0.216) Department of Education 1993–2013

Investigations 0.105 (0.306) Department of Education 1993–2013

Consequences 0.203 (0.395) Department of Education 1993–2013

Communications and
transparency

0.081 (0.271) Department of Education 1993–2013

Legal definitions 0.204 (0.397) Department of Education 1993–2013

Multiple components 0.041 (0.197) Department of Education 1993–2013

Few components 0.166 (0.368) Department of Education 1993–2013

One or no components 0.157 (0.357) Department of Education 1993–2013

Demographic controls

Age 15.96 (1.259) YRBS 1993–2013

Male 0.491 (0.500) YRBS 1993–2013

White 0.590 (0.492) YRBS 1993–2013

Do anti-bullying laws work? 479

All Fight, set equal to 1 if the student reported being in any physical fights during the
past 12 months and 0 otherwise.7 In our sample, 31.2 % of students reported any
fighting.

Finally, students were asked about weapons-related threats in school. Specifically,
respondents were asked:

During the past 12 months, how many times has someone threatened or injured
you with a weapon such as a gun, knife, or club on school property?

7 The questionnaire item in the YRBS about total physical fights was, “During the past 12 months, how many
times were you in a physical fight?”

Table 1 (continued)

Mean (SD) [N] Source Years

Black 0.152 (0.359) YRBS 1993–2013

Grade in school 10.35 (1.122) YRBS 1993–2013

State-specific education controls

National school
lunch participation rate

0.098 (0.019) US Department of
Agriculture

1993–2013

Average student-teacher ratio 14.62 (2.870) NCES Digest of Education
Statistics

1993–2013

Average teacher salary
(in thousands of 2013$)

56.47 (9.196) NCES Digest of Education
Statistics

1993–2013

Share of population with
bachelor’s degree

0.281 (0.061) Current Population Survey 1993–2013

Zero tolerance school violence law 0.932 (0.239) Education Commission
of the States

1993–2013

State-specific economic and
policy controls

Police expenditures per 1000
population

243.96 (95.94) Bureau of Justice Statistics 1993–2013

Law enforcement employees
per 1000 population

2.260 (0.062) Bureau of Justice Statistics 1993–2013

CAP law 0.464 (0.497) Law Center to Prevent
Gun Violence

1993–2013

Shall issue law 0.632 (0.482) Law Center to Prevent
Gun Violence

1993–2013

Zero tolerance drunk driving laws 0.910 (0.275) Updated from Anderson
et al. (2013)

1993–2013

Cigarette taxes (2013$) 1.164 (0.928) Tax Burden on Tobacco 1993–2013

Beer taxes (2013$) 0.300 (0.228) Beer Institute 1993–2013

Unemployment rates 0.061 (0.020) Bureau of Labor Statistics 1993–2013

Per capita income
(in thousands of 2013$)

41.13 (7.347) US Census Bureau 1993–2013

N 1,105,255

Standard deviations are in parentheses and sample sizes are in brackets
a Crime rates are arrest rates per 100,000 population of the appropriate ages

480 J.J. Sabia, B. Bass

Threat is coded equal to 1 if the student reported being threatened or injured
at least once during the past 12 months and coded as 0 if the student had not
been threatened. We find that 8.0 % of the sample reported being threatened or
injured during the past 12 months.

In addition to the four main measures of school safety defined above, we
measure a broader measure of bullying that could include non-physical bul-
lying such as taunting or teasing using responses to the following YRBS
survey item:

During the past 12 months, have you ever been bullied on school property?

We code Bullied equal to 1 if the student responded “yes” and 0 otherwise.
The advantage of this measure is that it captures bullying behavior more
comprehensively; the chief disadvantage is that it is only available in the
final three waves of the YRBS, in 2009, 2011, and 2013. However, as noted
below, this is a period during which we observe many state law changes. In
our sample, 20.1 % of the sample reported being the victim of bullying on
school property, comparable to recent NCES data reports (National Center for
Education Statistics 2012).

It is important to recognize that each of the above measures of bullying or
school safety is self-reported in nature. If ABLs induce more students to be
willing to admit (and report) bullying, then estimated effects of ABLs may be
biased upward. To supplement our self-reported measures of school safety, we
augment our YRBS analysis with objective data on violent behavior in
schools.

3.2 School shooting and crime data

First, we exploit use of a unique dataset constructed by Anderson and Sabia (2016) on
school shooting deaths. Their primary source for data on school shootings comes from
the National School Safety Center’s (NSSC) report on School Associated Violent
Deaths that covers the period 1993 through 2010.8 To supplement the NSSC’s report
and ensure a comprehensive coverage of school shootings, they used the following
additional data sources: Columbine-angels.com (2015), Doll (forthcoming),
Everytown.org (2015a), Van Fleet and Van Fleet (2010), Klein (2012), Lieberman
(2008), National School Safety and Security Services (2010), and Stoptheshootings.org
(2013). These sources, in addition to our Internet searches of newspaper archives,
allowed us to extend our coverage from 1993 to 2012.

We generate state-by-year data indicating whether a school shooting occurred
in each state-year, the circumstances surrounding the shooting, the age(s) of the
shooter(s), and whether the shooting occurred on an elementary school, middle
school, high school, or college/university campus. We define a minor teen

8 The NSSC report can be found at http://www.schoolsafety.us/media-resources/school-associated-violent-
deaths

Do anti-bullying laws work? 481

http://www.schoolsafety.us/media-resources/school-associated-violent-deaths

http://www.schoolsafety.us/media-resources/school-associated-violent-deaths

school shooting death as one that occurred on a high school, middle school, or
elementary school campus; was committed by a shooter ages 13–17; and
involved a death. Death is coded as 1 if the teen school shooting resulted in
a death (homicide, suicide, or both). Homicide is coded as 1 if the teen school
shooting involved a homicide. During the 1993 to 2012 period, we identify 135
school shootings, 73 of which involved a homicide.

Our second source of criminal data is drawn from the Federal Bureau of
Investigation’s Uniform Crime Reports (UCR) from 1993 to 2012 to measure
property and violent crime arrests for minor teens. During this period, the
average number of property crime arrests for minor teens was 1999.9 per
100,000 and the average violent crime arrest rate was 325.2 per 100,000.

3.3 Anti-bullying laws

We begin by generating a dichotomous measure, ABL, which is an indicator for
whether the state mandates that school districts adopt anti-bullying policies.
Effective dates for the implementation of school district anti-bullying policies
are determined in the following manner. Thirty-two (32) states included the
effective date by which school districts were required to adopt anti-bullying
policies in the text of the anti-bullying law statute.9 For the remaining states,
the school district anti-bullying policy mandate is coded as being enacted on
the effective date of the anti-bullying statute. Table 2 presents the effective
dates for each state’s anti-bullying school district policy mandate. Every state
except Montana enacted an anti-bullying law during the 2001 to 2013 period,
with Colorado enacting the first mandate in August 2001, and New York and
Virginia most recently implementing a law in July 2013. In April 2015,
Montana became the final state to adopt an ABL.

In an effort to categorize the breadth and strictness of state ABLs, the DOE,
following a joint DOE and DHHS-hosted Federal Partners in Bullying Preven-
tion Summit, commissioned a legal evaluation of the comprehensiveness,
strength of enforcement, and reporting strictness of 16 components of each
state law.10 Importantly, these DOE evaluations were not based on case studies
of policy impacts, but by legal interpretation of the statutes. The DOE assigns a
score of 0 to 2 to each component, measuring the overall expansiveness of each
provision, and creates an aggregate “intensity rating” based on these scores (see
U.S. Department of Education 2011 for a detailed discussion of these ratings).

9 Information on effective dates from these 32 states were collected from summaries provided by the U.S.
Department of Education’s Office of Planning, Evaluation and Policy Development (U.S. Department of
Education 2011, Exhibit 11) and our own investigation of each state’s anti-bullying statute.
10 Specifically, the DOE examined the extent to which each state legislation addressed four specific areas: (1)
Definitions of Terms, (2) District Policy Development and Review, (3) School District Policy Components
(Written Records and Anonymous Student Reporting Policies, Bullying Definitions, Investigation Policies,
Consequences/Sanctions Policies, and Post-Bullying Mental Health Services), and (4) Additional District
Policy Components (Parental Communications/Staff Training/Transparency, and Legal Remedies).

482 J.J. Sabia, B. Bass

Table 2 State anti-bullying laws (ABL), 1993–2013

State Effective date
of
school district
policy

Overall
DOE
rating

Presence
of strong
components

State Effective date
of school district
policy

Overall
DOE
Rating

Presence of strong
components

AL 07/01/2010 20 IN, LG MTb No law – –

AK 07/01/2007 10 NE 07/01/2009 06

AZ 08/12/2005 13 WR NV 07/01/2005 19 LG

AR 07/16/2003 21 LG NH 01/01/2011 27 IN, CS, TT, LG

CA 01/01/2004 17 CS NJ 09/01/2011 30 WR, IN, CS, TT,
LG

CO 08/08/2001 11 CS NM 04/01/2007 16

CT 02/01/2009 22 WR, IN, CS,
TT

NYc 07/01/2013 20 CS, LG

DC 06/22/2012 22 WR, IN, CS NC 12/31/2009 20 LG

DE 01/01/2008 22 CS, TT, LG ND 07/01/2012 20 IN, LG

FL 12/01/2008 24 IN, CS, LG OH 09/29/2010 18 IN, LG

GA 08/01/2011 13 CS OK 11/01/2002 14

HI 07/11/2011 13 WR, IN, CS OR 01/01/2004 21 IN

ID 07/01/2006 06 PA 01/01/2009 13

ILa 06/28/2010 16 CS RI 09/01/2004 14

IN 07/01/2005 08 SC 01/01/2007 19 IN, LG

IA 09/01/2007 19 LG SD 07/01/2012 07 CS

KS 07/01/2008 06 TN 01/01/2006 14 IN

KY 11/30/2008 15 IN, CS TX 06/17/2011 06 CS

LA 08/01/2001 17 UT 09/01/2012 13 LG

ME 09/01/2006 20 CS, LG VT 01/15/2007 22 WR, IN, CS, LG

MD 07/01/2009 28 CS, TT, LG VA 07/01/2013 18 CS

MA 12/31/2010 23 IN, CS WA 08/01/2011 30 WR, IN, CS

MI 06/07/2012 18 WR, IN, CS,
TT

WV 12/01/2001 23 IN, LG

MN 08/01/2007 03 WI 08/15/2010 09

MS 12/31/2010 11 WY 12/31/2009 19 IN, LG

MO 09/01/2007 10

Component labels WR, IN, CS, TT, and LG indicate strong school district policy mandates for written records
and reporting, investigations, consequences, training and transparency, and legal definitions, respectively
a While Illinois’s 2006 Senate Bill 2630 encouraged school district to adopt anti-bullying policies, the state’s
2010 Senate Bill 3266 was the first anti-bullying law to require school district anti-bullying policies
b Montana passed an ABL in April 2015
c New York’s Dignity for All Students Act was first enacted in 2010 and became effective in 2012. However,
the mandates on school district policies were specified in details in the Act’s Amendment which took effect in
2013

Do anti-bullying laws work? 483

Components rated a “0” by the DOE were usually those components not present in a
state ABL. Components rated as a “2” (most expansive) were more inclusive in nature,
more prescriptive, used less discretionary language, and established clearer measures of
accountability (see U.S. Department of Education 2011 for the complete list of rating
criteria for each component).11 For example, a state ABL’s Purpose component, which
addressed the purpose of laws and policies and prohibitions against bullying, was rated
as a “0” if the state ABL did not contain a prohibition against bullying; a “1” if the state
law explicitly specified a prohibition against bullying, but did not contain language
articulating the purpose or intent; and a “2” if the state law explicitly specified both a
prohibition against bullying and language articulating the purpose or intent. The score
assigned to each component was then summed to generate a composite score ranging
from 0 to 32. Table 2 shows the DOE intensity rating for each state. Washington
received the highest score from the DOE with 30 points, and Minnesota and Texas have
the lowest scores with 3 and 5 points, respectively.

We categorize laws by the interquartile range of composite intensity scores: Strong ABL,
set equal to 1 if the state has an ABL with a DOE intensity rating in the upper 25th
percentile of ratings (corresponding to composite scores between 21 and 32), and 0
otherwise; Moderate ABL set equal to 1 if the state has an ABL with a DOE intensity
rating composite score in the 25th to 75th percentile of ratings (corresponding to composite
scores between 13 and 20) and 0 otherwise; and Weak ABL set equal to 1 if the state has an
ABL with a DOE intensity rating composite score in the lower 25th percentile of ratings
(corresponding to composite scores between 0 and 12) and 0 otherwise. We experiment
with other cutoffs corresponding to quartiles and quintiles of the distribution and find a
similar pattern of results to those reported in the main tables. In our sample, 13.9 % of
respondents lived in states enforcing ABLs with intensity ratings in the top 25th percentile.

As noted above, two of the four areas into which the DOE categorizes state ABLs
relate specifically to school district policies. We restrict attention to these policy
components and examine the effect of district policy-specific intensity ratings on our
outcomes. Analogously coded as the above, Strong District Policy set equal to 1 if the
state has an ABL with a district policy intensity rating in the upper 25th percentile of
ratings and 0 otherwise; Moderate District Policy set equal to 1 if the state has an ABL
with a district policy intensity rating composite score in the 25th to 75th percentile of
ratings and 0 otherwise; and Weak District Policy set equal to 1 if the state has an ABL
with a district policy intensity rating composite score in the lower 25th percentile of
ratings and 0 otherwise.

11 According to the DOE:

Expansiveness was interpreted differently across components; however,
components in law that were rated as more expansive are generally: a)
more inclusive (e.g., defined prohibited behavior broadly without any
limiting conditions, or extended school jurisdiction to cover off-campus
conduct); b) are more prescriptive (e.g., used concrete directives to convey
policy expectations); c) use less discretionary language (e.g., used the term
“shall” instead of “may”); or d) establish stronger measures of account-
ability.” (DOE 2011)

484 J.J. Sabia, B. Bass

Next, among DOE-identified school district policy components, we examine five
mandates most likely to affect marginal bullying decisions. The first component,
Written Records & Reporting, mandates schools to maintain written records of all
incidents of bullying reported to school officials, as well as a procedure that would
allow students, staff, or parents to anonymously report suspected bullying. For exam-
ple, California’s educational code requires the state DOE to assess whether local
schools have maintained documentation of complaints and their resolution for a
minimum of one review cycle. These policies are expected to increase the costs of
bullying by increasing the probability of detecting and punishing potential bullies.

Second, an Investigations mandate requires that schools strictly enforce a procedure for
promptly investigating and responding to any reported incidents of bullying on school
grounds. For instance, Massachusetts’s Investigations mandate requires that “[u]pon
receipt of such a report, the school principal or a designee shall promptly conduct an
investigation” (Massachusetts Advocates for Children Legislation 2010).

Third, a Consequences provision requires school districts to provide a detailed range
of consequences and sanctions for bullying occurrences. For example, under the
California ABL, students who engage in an act of bullying may be suspended from
school or recommended for expulsion (California Education Code 2003). Investiga-
tions and Consequences mandates are expected to raise the expected costs of bullying
by increasing the probability of detection and punishments if caught.

Fourth, a Training & Transparency mandate requires (1) communications to students,
their families, and school personnel of policies and consequences related to bullying and (2)
training for school staff and faculty on preventing, identifying, and responding to bullying.
The Connecticut ABL, for example, requires schools to (1) provide teachers and staff with a
bullying prevention and intervention plan, (2) notify and meet with parents or guardians of
the bullies and the victims of verified bullying incidents, and (3) report all verified incidents
to the Department of Education annually (Connecticut General 2010).

Finally, a Legal Definitions mandate requires definitions of bullying adopted by school
districts to conform to those written in the state legislation. For instance, Oregon law
requires school district bullying policies to include definitions of “harassment,” “intimi-
dation,” or “bullying” and “cyberbullying” that are consistent with the state’s statues
(Oregon Revised 2009). Stricter bullying definitions are likely to result in less ambiguity
about whether bullying has occurred and decrease inappropriate bullying behavior.

Together, these components are hypothesized to raise the expected costs of bullying
behavior to potential bullies by raising the probability of detection (via better detection
methods and lowering the costs of reporting to victims) and mandating harsher punish-
ments if detected.

Using the above five school district policy components, we first estimate the effect
of each strong school district policy mandate (that is receiving a score of “2”),
controlling for each of the other components. However, given the high degree of
collinearity between components12—and the possibility that interactive effects of these
individual policy mandates are important—we next estimate the effect of states having

12 For example, among eight states with high intensity Written Records & Reporting mandates, half have
ABLs with three additional high intensity district policy mandates, three states have two additional high
intensity components, and only one has no other high intensity component. We find that approximately half of
the identifying variation available in each high intensity component is eliminated by the inclusion of controls
for the others.

Do anti-bullying laws work? 485

multiple strong district policy components. For instance, five states, including Con-
necticut, Michigan, New Hampshire, New Jersey, and Vermont, are examples of states
implementing four or more strong district policy components. Fourteen states have
enforced two or three strong district policy components. For instance, Connecticut,
Hawaii, Michigan, New Jersey, Vermont, and Washington have enforced ABLs with
the Written Records & Reporting, Investigations and Consequences mandates. Finally,
Arizona, Nevada, Oregon, and New Mexico are examples of states implementing less
than two components.

Figure 1 shows an event study of student fighting across states that implemented
multiple strong school district components as compared to zero or one components.
The pattern suggests that in the short run, states that implemented stronger ABLs saw a
modestly steeper decline in student fighting relative to states with weaker ABLs.
However, given that pre-treatment trends appeared to differ somewhat, particularly
for our outcome of All Fight, we next turn to a regression approach to account for
spurious differential time trends.

4 Empirical approach

We begin by pooling data from repeated cross-sections of the 1993–2013 National and
State YRBS and, for our dichotomous outcomes, estimate the following difference-in-
difference model via probit:

Y*ist ¼ β0 þ β1ABLst þ β2’Zit þ β3’Est þ β4’X st þ αs þ πt þ αs*t þ εist ð1Þ

where Y*ist is a latent variable measuring safety of student i residing in state s at year t;
ABLst is an indicator for whether a state mandate for school district anti-bullying
policies is in effect in state s in year t (or a set of indicators indicating the strength of

Fig. 1 Event study: student fighting

486 J.J. Sabia, B. Bass

that law or the components of that law); Zit is a vector of demographic controls
including gender, age, grade, and race/ethnicity; Est is a vector of state-specific time-
varying education controls, including whether the state is enforcing a zero tolerance
school violence policy, average pupil-teacher ratio, state average teacher salary, Na-
tional School Lunch Program (NSLP) participation rates, and the share of population
with a Bachelor’s degree; Xst is a vector of state-specific time-varying economic and
policy controls, including alcohol policies (beer taxes and zero tolerance drunk driving
laws), cigarette taxes, per capita expenditures on police, per capita number of law
enforcement employees, child access prevention (CAP) gun laws, concealed carry
(“shall issue”) laws, the state unemployment rate, and per capita income; αs is a
time-invariant state effect; πt is a state-invariant time effect; αs*t is a state-specific
linear time trend; and εist ∼N (0,1).

Identification of β1 comes from within-state variation in ABLs during the 1993–
2013 period. As noted above, 49 states and the District of Columbia enacted bullying
laws during the period under study (Table 2). To produce unbiased estimates of β1, the
parallel trends assumption of difference-in-difference models must be satisfied. This
may be violated if, for example, (1) states enact ABLs in response to school bullying
trends or (2) if there are time-varying state characteristics not captured in state-specific
time-varying education controls or economic and policy controls that are associated
with both the adoption of ABLs and with the outcomes under study.

We pursue a number of strategies to address the possibility of policy endogeneity.
First, we add state-specific linear time trends to the right-hand side of Eq. (1) to control
for unmeasured state trends unfolding linearly. Second, to examine whether results are
driven by student well-being trending differently prior to the implementation of state
bullying laws, we test the robustness of our estimates to the inclusion of policy leads.
Finally, we conduct falsification tests on a set of outcomes similar to those under study
for young adults in their 20s, for whom ABLs should not bind, except through longer-
run effects.

Our key estimation results are shown in Tables 3 through 7. All tables present
marginal effects from probit (estimated at the mean of the right-hand-side variables) or
OLS models. For ease of presentation, we focus on estimates of β1, but estimated
coefficients on the controls are available upon request. Standard errors corrected for
clustering on the state are in parentheses (Bertrand et al. 2004).

5 Main results

5.1 School safety

In Panel I of Table 3, we present baseline difference-in-difference estimates of the
relationship between enforcement of a state ABL and school safety, including only
controls for state and year fixed effects. Our estimates show little evidence that
state ABLs are associated with economically or statistically significant changes in
the probability of not attending school due to an unsafe environment (column 1),
physical altercations on or off school property (columns 2 and 3), or weapons-
related threats (column 4). Panel II adds individual-level demographic controls
and state-level economic controls, and Panel III adds state-specific education-

Do anti-bullying laws work? 487

related controls and public policy controls. The findings in these panels are
qualitatively and quantitatively similar to those presented in Panel I. The precision
of our estimates in Panel III is such that we can rule out, with 95 % confidence,

Table 3 Difference-in-difference estimates of relationship between ABLs and school safety

Unsafe (1) School Fight (2) All Fight (3) Threat (4)

Panel I: Baseline difference-in-difference

ABL 0.001 0.002 0.003 0.0004

(0.004) (0.003) (0.005) (0.003)

Panel II: Panel I + individual and state economic controls

ABL 0.0003 −0.0003 0.0003 −0.0002
(0.003) (0.003) (0.005) (0.002)

Panel III: Panel II + state policy and education controls

ABL 0.0004 0.0004 0.0002 −0.0001
(0.003) (0.003) (0.004) (0.002)

Panel IV: Leads and lags of ABL

3 years before −0.004 −0.009*** −0.016** −0.004
(0.004) (0.003) (0.008) (0.003)

2 years before −0.004 0.006** 0.000 0.003
(0.004) (0.003) (0.004) (0.003)

1 year before 0.001 −0.003 −0.007 −0.001
(0.005) (0.004) (0.006) (0.004)

Year of law change −0.004 0.002 0.001 0.002
(0.004) (0.003) (0.005) (0.004)

1 year after 0.001 −0.004 −0.016 −0.001
(0.005) (0.006) (0.011) (0.005)

2 years after −0.002 0.004 0.007 0.001
(0.004) (0.005) (0.007) (0.004)

3+ years after −0.001 −0.003 −0.013 −0.002
(0.005) (0.007) (0.008) (0.004)

χ2 of ∑(βleads) = 0 (p value) 0.61 (0.43) 0.78 (0.38) 3.04 (0.08) 0.07 (0.79)
χ2 of ∑(βyrchange, βlags) = 0 (p value) 0.18 (0.67) 0.00 (0.96) 0.81 (0.37) 0.00 (0.98)

Panel V: Panel III + state linear trends

ABL 0.001 0.001 0.003 −0.001
(0.003) (0.003) (0.004) (0.003)

State FE? Yes Yes Yes Yes

Year FE? Yes Yes Yes Yes

Controls? Yes Yes Yes Yes

N 1,105,255 1,054,461 1,031,970 1,070,208

Estimated marginal effects are obtained using unweighted probit models using data from the 1993–2013
YRBS. Standard errors corrected for clustering on the state are in parentheses. Controls include state and year
fixed effects and state-specific linear time trends as well as those controls listed in Table 1

***Significant at 1 % level, **at 5 % level, *at 10 % level

488 J.J. Sabia, B. Bass

student safety improvements of greater than 0.4 to 0.8 percentage points, or 2.5 to
9.3 % (relative to the means of the outcomes).13

Could the estimated association we observe in Panels I through III be biased toward
zero? This could occur if states that adopt ABLs are experiencing declining trends in
student safety and these trends are confounding true beneficial effects of these laws.
Moreover, the effects of these laws may be small initially, but could take time to unfold.
We explore each possibility in Panel IVof Table 3, where we include 3 years of policy
leads and 3 years of lagged effects.

Our results provide little support for the hypothesis that student safety was trending
differently in the years prior to the adoption of ABLs, both in individual and joint tests
of the policy leads. Moreover, we find no evidence that the typical state ABL is
associated with significant changes in student safety in the years following the year
of enforcement. In Panel V, we test the sensitivity of our estimates to the inclusion of
controls for state-specific linear time trends. The results are quantitatively similar.

While the typical state ABL appears to have little or no effect on school safety, in
Table 4 we explore whether there might be heterogeneous effects of ABLs by the type
of law adopted. In Panel I of Table 4, we find that strong state ABLs—those with
composite scores in the top 25th percentile—are associated with a 1.1 percentage point,
or 9.0 % reduction in the probability of fighting on school property (column 2). The
beneficial in-school fighting effect of strong state ABLs is larger than for moderate
(composite score in middle 50th percentile) or weak (composite scores in lower 25th
percentile) ABLs (see χ2 tests in the final rows of Panel I). The findings in column (2)
of Panel II suggest that these strong ABL effects are driven largely by strong school
district policy mandates. The results also suggest that the effect of strong ABLs on
overall fighting remains negative (though insignificant), but is also smaller in magni-
tude. This could suggest some degree of substitution of fighting off of school property.

In Panel III, we examine five major components of strong school district policy
mandates. Our results suggest that strictly enforced written records and student
reporting requirements are associated with substantial improvements in school safety.
We find that the enforcement of ABLs with strict student reporting requirements is
associated with a 22.2 % reduction in the probability of safety-related school absences,
a 13.1 % reduction in the probability of fighting on school property, a 3.8 % decrease in
the probability of overall fighting, and a 25.0 % decline in the probability of weapons-
related threats. These results are consistent with the hypothesis that policies that reduce
the costs of student reporting and increase expected punishments from bullying can
reduce school violence.

While none of the other district policy components are statistically distinguishable
from zero at conventional levels, the degree of collinearity between high intensity
mandates (see footnote 11)—as well as the possibility that enforcement of multiple high
intensity school district policy components may affect school safety—motivates our
next set of estimates. In Panel IV, we examine interactive effects of our five key school
district policy components of ABLs (Written Records & Reporting, Investigations,
Consequences, Training & Transparency, and Definitions) on school safety. We find
that ABLs with multiple strong school district policy components (those with four or

13 The lower bound of the 95 % confidence interval is calculated as the ratio of the point estimates subtracting
the product of 1.96 and the standard error to the mean of the relevant dependent variable.

Do anti-bullying laws work? 489

Table 4 Estimated effects of heterogeneous types of ABLs

Unsafe (1) School Fight (2) All Fight (3) Threat (4)

Panel I: Strength of ABL

Strong ABL −0.001 −0.011** −0.006 −0.003
(0.006) (0.005) (0.006) (0.006)

Moderate ABL 0.0004 0.004 0.007 0.0003

(0.004) (0.004) (0.006) (0.003)

Weak ABL 0.004 0.009 0.005 −0.001
(0.005) (0.006) (0.008) (0.005)

χ2 of βStrong = βModerate (p value) 0.02 (0.89) 5.76 (0.02) 3.00 (0.08) 0.24 (0.63)

χ2 of βStrong = βWeak (p value) 0.26 (0.61) 6.28 (0.01) 1.24 (0.26) 0.09 (0.77)

Panel II: Strength of school district mandate

Strong district policy mandate −0.000 −0.008* −0.002 −0.001
(0.005) (0.004) (0.005) (0.005)

Moderate district policy mandate 0.0001 0.007 0.009 −0.0002
(0.004) (0.004) (0.007) (0.003)

Weak district policy 0.004 0.008 0.003 −0.001
(0.005) (0.006) (0.008) (0.005)

χ2 of βStrong = βModerate (p value) 0.00 (0.96) 5.91 (0.02) 1.78 (0.18) 0.03 (0.87)

χ2 of βStrong = βWeak (p value) 0.45 (0.50) 4.75 (0.03) 0.26 (0.61) 0.00 (0.99)

Panel III: Strong district policy components

Written records and reporting −0.014* −0.016** −0.012 −0.020***
(0.008) (0.007) (0.010) (0.005)

Investigations 0.008 −0.007 0.002 −0.0003
(0.005) (0.005) (0.008) (0.005)

Consequences 0.002 0.001 0.001 −0.00004
(0.005) (0.005) (0.009) (0.004)

Training and transparency 0.006 0.010 −0.004 0.019***
(0.010) (0.007) (0.008) (0.004)

Legal definitions −0.007 0.001 0.002 0.0003
(0.005) (0.004) (0.007) (0.004)

Panel IV: Number of strong district policy components

Multiple components −0.008 −0.016*** −0.011** −0.012*
(0.007) (0.005) (0.005) (0.007)

Few components 0.003 −0.0002 0.004 0.005
(0.005) (0.004) (0.006) (0.004)

Zero or one component 0.002 0.007* 0.007 −0.001
(0.003) (0.004) (0.006) (0.003)

χ2 of βMultiple = βFew (p value) 1.71 (0.19) 6.01 (0.01) 4.16 (0.04) 4.83 (0.03)

χ2 of βMultiple = βZero or One (p value) 1.8 (0.18) 11.13 (0.00) 5.53 (0.02) 1.99 (0.16)

N 1,105,255 1,054,461 1,031,970 1,070,208
Estimated marginal effects are obtained using unweighted probit models using data from the 1993–2013
YRBS. Standard errors corrected for clustering on the state are in parentheses. Controls include state and year
fixed effects and state-specific linear time trends as well as those controls listed in Table 1
***Significant at 1 % level, **at 5 % level, *at 10 % level

490 J.J. Sabia, B. Bass

more components strictly enforced) is associated with a 13.1 % reduction in the
probability of fighting on school property, a 3.8 % reduction in the probability of
overall fighting, a 15.0 % reduction in the probability of weapons-related threats, and a
(statistically insignificant) 12.7 % decline in the probability of safety-related school
absences. Moreover, it also appears as though the effects of enforcing more compo-
nents of ABLs are greater than enforcing fewer components.14 15

To ensure that the findings in Table 4 on strong ABLs were not driven by differential
state trends in school safety in the years prior to the enforcement of these laws, we re-
estimate all models in Table 4, but also include policy leads for 3 years prior to the
implementation of an ABL. The pattern of results in Table 5 suggests that trends prior
to the enactment of strictly enforced ABLs cannot explain the school safety effects we
estimate in Table 4.16 17

5.2 Bullying

While the above results examine more severe, violent forms of bullying vic-
timization, we next turn to bullying behavior that includes non-physical victim-
ization such as taunting or teasing. As noted above, data on self-reported
bullying is only available in the final three waves of the YRBS between
2009 and 2013. In Table 6, we examine the impact of state ABLs on the
probability of being bullied on school property. Because we have only three
data points for each state, we begin with a more parsimonious regression
without a wide set of controls. Column (1) includes controls for state and year
fixed effects, column (2) adds individual demographic controls and state-
specific economic controls, and column (3) adds state-specific education and
policy controls.18

Difference-in-difference estimates show that the enactment of the typical state
ABL is associated with a 3.5 to 4.5 % reduction in the probability of being
bullied on school property (Panel I). This effect is driven by strong ABLs
(Panel II), though the effect does not differ by strength of the district policy

14 Using a 1-year lag of ABLs produces a similar pattern of results, available upon request
15 To get a sense of the magnitudes of our estimated ABL effects, we compare them to the estimated effects of
other policies that have been found to improve school safety. Anderson and Sabia (2016) find that state child
access prevention gun control laws, which impose criminal liability on adult gun owners who allow minors
unsupervised access to firearms, are associated with a 15 to 20 % reduction in weapons-related threats, and
Markowitz (2001) finds that a 1 % increase in beer taxes is associated with a 5 % reduction in physical fighting
by students. Finally, the Boston Gun Project, a program designed to reduce youth violence, particularly gang
violence, is associated with an approximately 50 % decline in youth gun assaults (Piehl et al. 2000).
16 An examination of the intensive margin of these outcomes (e.g., frequency of behaviors among those
reporting them) suggests that the effects if ABLs are largely driven along the extensive margin.
17 We also explore whether there were heterogeneous effects of ABLs on bullying on school safety by gender,
age (under 16 and over 16), and region of the country (South versus non-Southern states). The results,
available upon request, show that written records and reporting components have larger effects for male
students as compared to female students, but there is no gender difference across strong versus weaker ABLs.
Estimated ABL effects do not differ by student age, but the impacts of ABLs also appear concentrated in non-
southern states.
18 Given that there are only 3 years of data available for this outcome, the inclusion of state-specific time
trends eliminates much of the identifying variation. An auxiliary regression of state ABLs on controls shows
that the inclusion of state-specific linear time trends increases our estimate of R2 from 80 to 94 %.

Do anti-bullying laws work? 491

component (Panel III). When examining individual district policy components
(Panel IV), we find that the enforcement of Investigations and Consequences
components are particularly important. Our results show that these mandates are
associated with a 9.5 to 12.4 % reduction in the probability of being bullied.
These results are consistent with the hypothesis that school district policies that
increase expected punishments from bullying decrease its likelihood. Thus, it
appears that strict mandates for written records of students’ bullying reports are
most important for deterring more severe types of bullying, while investigations
and consequences are more important for deterring other forms of bullying.

5.3 Minor teen school shootings and crime

ABLs could affect criminal behavior via both direct and indirect channels. First, some
types of bullying behavior—physical violence, threats, and theft—cross the criminal
threshold and ABLs may directly affect them. Second, if non-criminal bullying is a
“gateway” to more severe criminal behavior, ABLs may indirectly affect crime. Third,
those who are victims of bullying sometimes seek out revenge against those they
perceive as tormenting them (Klein 2012).

In Table 7, we present estimates of the relationship between state ABLs and (1)
minor teen school shootings involving deaths (columns 1–4), and (2) minor teen
property and violent crime arrest rates (columns 5–8). In addition to difference-in-
difference estimates (DD), we also present triple-difference (DDD) estimates using
(1) school shootings committed on college or university campuses by those under
age 30, and (2) crime rates of those ages 20–24, as within-state control groups.

Table 5 Robustness of estimates to inclusion of controls for policy leads

Unsafe (1) School Fight (2) All Fight (3) Threat (4)
Panel I: Strength of ABL

Strong ABL −0.008 −0.017*** −0.019** −0.010
(0.006) (0.005) (0.009) (0.009)

Panel II: Strength of school district policy mandate

Strong district policy −0.005 −0.009** −0.009 −0.003
(0.005) (0.004) (0.008) (0.007)

Panel III: Strong district policy component

Written records and reporting −0.013 −0.018** −0.012 −0.030**
(0.009) (0.009) (0.022) (0.013)

Panel IV: Number of strong district policy components

Multiple components −0.011 −0.020*** −0.025** −0.021**
(0.007) (0.007) (0.010) (0.010)

N 1,105,255 1,054,461 1,031,970 1,070,208
Estimated marginal effects are obtained using unweighted probit models using data from the 1993–2013
YRBS. Standard errors corrected for clustering on the state are in parentheses. Controls include state and year
fixed effects and state-specific linear time trends as well as those controls listed in Table 1
***Significant at 1 % level, **at 5 % level, *at 10 % level

492 J.J. Sabia, B. Bass

Table 6 Estimates of relationship between ABLs and Bullied

(1) (2) (3)

Panel I: Any ABL

ABL −0.009* −0.009 −0.008
(0.005) (0.007) (0.006)

Panel II: Strength of ABL

Strong ABL −0.018* −0.017* −0.019*
(0.010) (0.010) (0.010)

Moderate ABL −0.007 −0.007 −0.006
(0.008) (0.012) (0.010)

Weak ABL −0.008 −0.008 −0.007
(0.006) (0.006) (0.007)

χ2 of βStrong = βModerate (p value) 0.75 (0.39) 0.40 (0.53) 0.70 (0.40)

χ2 of βStrong = βWeak (p value) 0.84 (0.36) 0.67 (0.41) 1.01 (0.32)

Panel III: Strength of school district policy mandate

Strong district policy mandate −0.006 −0.007 −0.006
(0.006) (0.008) (0.007)

Moderate district policy mandate −0.017 −0.013 −0.015
(0.017) (0.021) (0.018)

Weak district policy −0.008 −0.008 −0.007
(0.006) (0.005) (0.007)

χ2 of βStrong = βModerate (p value) 0.41 (0.52) 0.06 (0.87) 0.22 (0.64)

χ2 of βStrong = βWeak (p value) 0.08 (0.78) 0.00 (0.96) 0.01 (0.92)

Panel IV: Individual strong district policy components

Written records and reporting 0.008 0.017 0.013

(0.020) (0.022) (0.017)

Investigations −0.013 −0.025** −0.018
(0.010) (0.013) (0.012)

Consequences −0.022* −0.019 −0.019*
(0.012) (0.013) (0.012)

Training and transparency 0.018 0.017 0.014

(0.024) (0.025) (0.021)

Legal definitions 0.012 0.017* 0.015

(0.009) (0.011) (0.012)

Panel V: Number of strong district policy components

Multiple components −0.003 −0.005 −0.005
(0.005) (0.004) (0.005)

Few components −0.007 −0.010 −0.008
(0.009) (0.013) (0.012)

Zero or one component −0.013 −0.010 −0.010
(0.009) (0.011) (0.008)

χ2 of βMultiple = βFew (p value) 0.15 (0.70) 0.19 (0.67) 0.05 (0.82)

χ2 of βMultiple = βZero or One (p value) 0.95 (0.33) 0.31 (0.58) 0.31 (0.58)

Individual and economic controls? No Yes Yes

State policy and education controls? No No Yes

N 412,666 412,666 412,666

Estimated marginal effects are obtained using unweighted probit models using data from the 1993–2013
YRBS. Standard errors corrected for clustering on the state are in parentheses. All models include controls for
state fixed effects and year fixed effects

***Significant at 1 % level, **at 5 % level, *at 10 % level

Do anti-bullying laws work? 493

T
ab

le
7

E
st
im

at
es

of
ef
fe
ct
s
of

A
B
L

s
on

sc
h
oo
l
sh
o
ot
in
gs

an
d
cr
im

in
al
ar
re
st
ra
te
s

S
ch
oo
l
sh
oo
ti
ng
s

A
rr
es
t
ra
te
s

D
ea
th
s

H
o
m

ic
id
e

P
ro
pe
rt
y

V
io
le
n
t

D
D

D
D
D

D
D
D
D
D
D
D
D
D
D
D
D
D
D
D

(1
)

(2
)

(3
)

(4
)

(5
)

(6
)

(7
)

(8
)

P
an
el
I:
A
ny

A
B
L
A
B
L

−0
.0
31

(0
.0
59
)

−0
.0
36

(0
.0
89
)

−0
.0
63

(0
.0
3
8)

−0
.1
2
4*
*
(0
.0
49
)

−0
.1
13

(0
.0
7
5)

−0
.0
1
7
(0
.0
31
)

−0
.1
42
**

(0
.0
70
)

−0
.0
08

(0
.0
65
)

P
an
el
II
:
S
tr
en
gt
h
of

A
B
L

S
tr
on
g
A
B
L

−0
.1
95
*
(0
.1
09
)

−0
.1
69

(0
.1
21
)

−0
.2
35
**

(0
.0
9
9)

−0
.2
3
6*
*
(0
.1
03
)

−0
.0
98

(0
.0
9
3)

−0
.0
98
**

(0
.0
48
)

−0
.1
15

(0
.1
20
)

−0
.1
1
4*
*
(0
.0
44
)

M
od
er
at
e
A
B
L

0.
0
13

(0
.0
83
)

−0
.0
32

(0
.0
99
)

0.
01
0

(0
.0
68
)

−0
.0
8
4

(0
.0
93
)

−0
.1
80

(0
.1
4
7)

−0
.0
0
5
(0
.0
49
)

−0
.2
05

(0
.1
27
)

0.
03
4
(0
.1
20
)

W
ea
k
A
B
L

0.
0
26

(

0
.0
78

)

0
.1
00

(0
.1
32
)

−0
.0
74

(0
.1
18
)

−0
.1
11

(0
.1
1
9)

0.
0
35

(0
.0
59
)

0
.0
28

(0
.0
3
8)

−0
.0
18

(0
.1
23
)

−0
.0
13

(0
.0
43
)

χ
2
of

β
S
tr
o
n
g
=
β
M
o
d
er
at
e
(p
-v
al
u
e)

2.
4
2
(0
.1
3)

0
.8
1
(0
.3
7
)

3.
07

(0
.0
9)

0.
81

(0
.3
7)

0.
1
8
(0
.6
7)

1
.8
9
(0
.1
8)

0.
22

(0
.6
4)

1.
16

(0
.2
9)

χ
2
of

β
S
tr
o
n
g
=
β
W
ea
k
(p
-v
al
ue
)

2.
9
4
(0
.0
9)

3
.0
0
(0
.0
9
)

1.
09

(0
.3
0)

0.
59

(0
.4
5)

1.
7
2
(0
.2
0)

4
.5
2
(0
.0
4)

0.
35

(0
.5
6)

3.
08

(0
.0
9)

P
an
el
II
I:
S
tr
en
gt
h
of

sc
h
oo
l
d
is
tr
ic
t
po
li
cy

m
an
da
te

S
tr
o
n
g
d
is
tr
ic
t
p
ol
ic
y
m
an
da
te

−0
.0
74

(0
.1
25
)

−0
.1
38

(0
.1
04
)

−0
.1
00

(0
.1
2
4)

−0
.1
9
4*
*

(

0
.0
75

)

−0
.4
63

(0
.2
8
1)

−0
.0
5
5
(0
.0
59
)

−0
.4
51
*
(0
.2
4
0)

0.
12

9
(0
.2
26
)

M
od
er
at
e

di
st
ri
ct
po
li
cy

m
an
d
at
e

−0
.0
21

(0
.0
88
)

−0
.0
27

(0
.1
07
)

−0
.0
45

(0
.0
6
3)

−0
.1
0
3
(0
.1
00
)

−0
.0
12

(0
.0
5
1)

−0
.0
3
6

(0
.0
45
)

−0
.0
39

(0
.0
67
)

−0
.0
71
*
(0
.0
40
)

W
ea
k
d
is
tr
ic
t
p
ol
ic
y

−0
.0
02

(0
.0
93
)
0
.0
78

(0
.1
40
)

−0
.0
60

(0
.1
15
)

−0
.0
8
4

(0
.1
12
)

0.
0
39

(0
.0
61
)

0
.0
68
*
*
(0
.0
31
)

−0
.0
48

(0
.1
11
)

−0
.0
24

(0
.0
45
)
χ
2
of
β
S
tr
o
n
g
=
β
M
o
d
er
at
e
(p
-v
al
u
e)

0.
11

(0
.7
4)

0
.6
1
(0
.4
4
)

0.
12

(0
.7
3)

0.
37

(0
.5
5)

2.
4
3
(0
.1
3)

0
.0
7

(0
.7
9)

2.
69

(0
.1
1
)

0.
76

(0
.3
9)

χ
2
of
β
S
tr
o
n
g
=
β
W
ea
k
(p
-v
al
ue
)

0.
2
1
(0
.6
5)

2
.8
2
(0
.1
0
)

0.
05

(0
.8
3)

0.
58

(0
.4
5)

2.
8
0
(0
.1
0)

3
.5
4
(0
.0
7)

2.
16

(0
.1
5)

0.
42

(0
.5
2)

P
an
el
IV
:
S
tr
o
ng

di
st
ri
ct
po
li
cy

co
m
po
ne
nt
s

W
ri
tt
en

re
co
rd
s
&

re
p
o
rt
in
g

−0
.1
47

(0
.1
22
)

−0
.2
5
2*

(0
.1
38
)

−0
.1
68

(0
.1
12
)

−0
.2
6
8*
*
(0
.1
32
)

−0
.0
40

(0
.2
0
9)

0
.0
53

(0
.0
8
7)

−0
.0
59

(0
.2
00
)

−0
.0
01

(0
.1
24
)

In
ve
st
ig
at
io
n
s

0.
0
17

(0
.1
25
)

−0
.0
20

(0
.1
36
)

0.
07

7
(0
.1
20
)

0.
09
3
(0
.1
0
5)

−0
.0
79

(0
.2
7
6)

0
.1
43

(0
.1
0
7)

0.
00

2
(0
.2
60
)

0.
14
2
(0
.2
07
)

C
on
se
qu
en
ce
s

−0
.0
13

(0
.1
02
)

0
.1
05

(0
.1
31
)

0.
05
8
(0
.0
67
)

0.
13

0*

(0
.0
75
)

0.
1
51

(0
.1
29
)

−0
.1
0
5
(0
.0
98
)

0.
11
3
(0
.1
38
)

−0
.1
36

(0
.1
01
)

T
ra
in
in
g
&

tr
an
sp
ar
en
cy

0.
0
35
(0
.1
12
)

0
.0
05

(0
.1
45
)

−0
.0
20

(0
.0
8
6)

−0
.0
0
8
(0
.1
14
)

−0
.0
27

(0
.1
5
3)

0
.0
30

(0
.0
9
2)

−0
.1
08

(0
.1
53
)

−0
.1
95
*

(0
.1
16
)

L
eg
al
d
ef
in
it
io
ns

−0
.0
70

(0
.0
99
)

−0
.1
83

(0
.1
49
)

−0
.1
98
**

(0
.0
8
5)

−0
.3
72
**
*
(0
.1
19
)

−0
.3
2
9*
*
(0
.1
5
9)

−0
.0
4
3
(0
.1
04
)

−0
.3
01
*
(0
.1
6
3)

0.
06
9
(0
.1
13
)

494 J.J. Sabia, B. Bass

T
ab
le
7

(c
o
nt
in
ue
d)

S
ch
o
ol

sh
oo
ti
n
gs

A
rr
es
t
ra
te
s
D
ea
th
s

H
om

ic
id
e
P
ro
pe
rt
y

V
io
le
nt

D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
(1
)
(2
)
(3
)
(4
)
(5
)
(6
)
(7
)
(8
)

P
an
el
V
:
N
um

be
r
of

st
ro
n
g
d
is
tr
ic
t
po
li
cy

co
m
po
ne
nt
s

M
ul
ti
p
le
co
m
po
ne
nt
s

−0
.1
14

(0
.1
16
)

−0
.2
26

(0
.1
37
)

−0
.1
39

(0
.1
15
)

−0
.2
26
*
(0
.1
3
1)

−0
.1
3
1
(0
.1
19
)

−0
.1
04

(0
.0
68
)

−0
.1
71

(0
.1
1
2)

−0
.1
33
**

(0
.0
56
)

F
ew

co
m
po
n
en
ts

−0
.0
4
7
(0
.1
07
)

−0
.1
01

(0
.0
95
)

−0
.0
99

(0
.1
11
)

−0
.1
96
**

(0
.0
74
)

−0
.3
0
1
(0
.2
52
)

−0
.0
72

(0
.0
63
)

−0
.2
54

(0
.2
33
)

0
.0
75

(0
.1
95
)

Z
er
o
or

on
e
co
m
p
on
en
t

−0
.0
1
5
(0
.0
79
)

0.
01
7
(0
.1
16
)

−0
.0
36

(0
.0
60
)

−0
.0
76

(0
.0
89
)

−0
.0
0
5
(0
.0
45
)

0.
02
6
(0
.0
35
)

−0
.0
77

(0
.0
56
)
−0
.0
39

(0
.0
28
)

χ
2
of

β
M
u
lt
ip
le
=
β
F
ew
(p
-v
al
ue
)

0.
20

(0
.6
6)

0.
69

(0
.4
1)

0.
07
(0
.7
9)
0.
00

(0
.9
9
)

0.
31

(0
.5
8)

0.
13

(0
.7
2)

0.
0
8
(0
.7
8)

1
.0
4
(0
.3
1
)

χ
2
of

β
M
u
lt
ip
le
=
β
Z
er
o
o
r
O
n
e
(p
-v
al
ue
)

0.
36

(0
.5
5)

1.
48

(0
.2
3)

0.
47

(0
.5
0)

0.
64

(0
.4
3
)

1.
04

(0
.3
1)

3.
23

(0
.0
8)

0.
6
4
(0
.4
3)

2
.5
2
(0
.1
2
)

N

1,
02
0

2,
04
0

1,
02
0
2,
04
0

98
1

1,
96
2

97
9

1
,9
59

W

ei
gh
te
d
O
L
S
es
ti
m
at
es

in
co
lu
m
n
s
(1
)
th
ro
ug
h
(4
)
ar
e
ge
ne
ra
te
d
us
in
g
sc
h
oo
ls
ho
ot
in
g
d
at
a
be
tw
ee
n
19
9
3
an
d
2
01
2
.W

ei
gh
te
d
O
L
S
es
ti
m
at
es

in
co
lu
m
ns

(5
)
th
ro
ug
h
(8
)
ar
e
g
en
er
at
ed

us
in
g
da
ta
fr
o
m

th
e
19
9
3
an
d
20
1
2
U
ni
fo
rm

C
ri
m
e
R
ep
or
ts
.
S
ta
nd
ar
d
er
ro
rs
co
rr
ec
te
d
fo
r
cl
us
te
ri
ng

on
th
e
st
at
e
ar
e
in

pa
re
nt
h
es
es
.
A
ll
m
od
el
s
in
cl
ud
e
co
n
tr
o
ls
fo
r
st
at
e
fi
xe
d
ef
fe
ct
s,

ye
ar

fi
xe
d
ef
fe
ct
s,
st
at
e-
sp
ec
if
ic
li
ne
ar

ti
m
e
tr
en
d
s,
th
e
sh
ar
e
o
f
th
e
po
p
ul
at
io
n
th
at
is
m
al
e
an
d
no
n-
w
h
it
e,
av
er
ag
e
ag
e
of

th
e
p
op
u
la
ti
on
,
av
er
ag
e
te
ac
he
r
sa
la
ry
,
av
er
ag
e
p
up
il
/t
ea
ch
er

ra
ti
o,
N
at
io
na
l
sc
ho
ol

lu
nc
h
pa
rt
ic
ip
at
io
n
ra
te
s,
sh
ar
e
of

po
p
ul
at
io
n
w
it
h
B
ac
he
lo
r’
s
de
gr
ee
,
ze
ro

to
le
ra
n
ce

sc
h
oo
l
vi
ol
en
ce

la
w
s,
C
A
P
la
w
s,
sh
al
l
is
su
e
la
w
s,
be
er

ta
x
es
,
ze
ro

to
le
ra
nc
e

dr
un
k
dr
iv
in
g
la
w
s,
ci
ga
re
tt
e
ta
xe
s,
st
at
e
u
ne
m
pl
o
ym

en
t
ra
te
,
an
d
p
er

ca
pi
ta
in
co
m
e

**
*
S
ig
n
if
ic
an
t
at
1
%

le
v
el

*
*
at
5
%

le
v
el
*
at
10
%

le
v
el

Do anti-bullying laws work? 495

DD estimates on the control groups, which can be interpreted as falsification tests
(see Appendix Table 8), produce no evidence that strong ABLs or any strong
school district policy components are related to college school shootings or young
adult crime rates.

In the first four columns of Table 7, we estimate the effect of ABLs on the
probability of a minor teen school shooting death in a given state-year.19 The
results show that strong state ABLs are associated with a significant decline in
school shooting events involving a death, particularly shootings involving a
homicide (see Panels II through V). This finding, consistent with YRBS findings
above, suggests that ABLs generate potentially important social benefits.

In the final four columns of Table 7, we present findings on criminal arrest rates
for those ages 13–17 using the Uniform Crime Data. The dependent variable is the
natural log of the minor crime arrest rate. DDD estimates point to evidence that
state ABLs are associated with a reduction in minor teen arrest rates, on the order
of 9.3 to 10.8 %, relative to older young adults in their 20s (Panel I). In the
remaining panels, we continue to find evidence that state ABLs with strong
district policy components, and multiple strong district policy components are
associated with substantial declines in minor property and violent crime arrest
rates. Together, the findings in Table 7 suggest that the social benefits of ABLs
may extend to amelioration of violent behavior.20 21

6 Conclusions

This study presents the first comprehensive examination of the effect of state
ABLs on school safety, bullying, and youth violence. Difference-in-difference
estimates suggest that the enforcement of the typical ABL is associated with
small and statistically insignificant changes in student safety in school. How-
ever, when we explore heterogeneity in anti-bullying statutes, we find that
comprehensive, strong ABLs are associated with significant improvements in
student safety. Specifically, we find that strong ABLs are associated with a 7 to
13 % reduction in school violence and an 8 to 12 % reduction in bullying. We
also find that strong ABLs are associated with a reduction in teen school
shootings and a 9 to 11 % reduction in violent crime arrests of minor teens.

Despite the fact that ABLs appear to generate social benefits, a question
remains: are they cost-effective? Restricting our attention to the criminal benefits
of ABLs, the estimates presented in this study suggest an average reduction in
property and crime offenses of approximately 1100. While there is substantial

19 These models are estimated via linear probability model. Estimates using Poisson models that capture the
count of school shooting fatality events generate a qualitatively similar pattern of results.
20 The results for crime and safety persist when we expand the age group examined for the falsification tests to
those ages 19–29, matching the school shooting age group.
21 In unreported falsification tests, we draw data from the General Social Survey from 1993 to 2010 to
examine the effect of ABLs on (1) fear of walking around in one’s neighborhood and (2) presence of shotguns
in one’s home or garage, for 20–24-year-olds. Our findings suggest little evidence that strong school district
policies or multiple strong district policy components are associated with changes in young adult neighbor-
hood safety or shotgun ownership. In addition, we also experiment with an additional falsification test on
minor high school students’ helmet use and find no evidence that state ABLs are associated with this outcome.

496 J.J. Sabia, B. Bass

heterogeneity in per-victimization costs of crime, back-of-the-envelope calcula-
tions from Miller et al. (1996) suggest a per-offense cost of approximately
$20,000 (in 2014 dollars). This implies crime-reducing benefits of ABLs of
$22 million for the average state. However, if ABLs alter the career paths of
some juvenile criminals, the criminal cost savings could be much larger. For
instance, Cohen (2000) estimates an external cost per criminal career of $1.3 to
$1.5 million.

What about the costs of implementing a high intensity state ABL? Minne-
sota, which implemented a low intensity ABL in 2007, recently considered
amending their law to include (unfunded) mandates for comprehensive school
staff training and strict reporting requirements. The Minnesota Management and
Budget Office estimated this high intensity legislation as costing local school
districts approximately $20 million per year (Minnesota Management and
Budget 2013). While this cost estimate represents only one state’s ABL—and
costs could vary widely across states—it does suggest that the social benefits
will have to be quite large for a high intensity ABL to be cost-effective.

There are a number of limitations of this study worthy of note. Although our
study contributes to estimating the effect of ABLs on school safety, public
safety, and student well-being, our analysis could benefit from more compre-
hensive measures of bullying. Because ABLs are intended to deter aggressive
bullying behavior and harassment on school property, questions regarding bully
victimization and physical or verbal harassment would be useful. Additionally,
numerous psychology and sociology studies on bullying and victimization
suggest that victims of bullying and bullies themselves exhibit adverse health
and psychological effects later in life. Utilizing data over longer time periods
would be useful in order to explore whether anti-bullying policies alter indi-
viduals’ life trajectories. In addition, future work using better data to identify
lesbian/gay/bisexual/transgendered (LGBT) and disabled youth will be important
to explore whether the benefits of ABLs extend to these groups.

Acknowledgments The authors thank Sara Markowitz, Mark Duggan, and Rosa Minhyo Cho for
useful comments and suggestions on this paper. We thank Thanh Tam Nguyen for excellent
research assistance. Special thanks are owed to D. Mark Anderson, who graciously made school
shooting data available to these authors. We also thank conference participants at the Southern
Economic Association (SEA) and the Association of Public Policy Analysis and Management
(APPAM) as well as seminar participants at San Diego State University for useful comments and
suggestions on an earlier draft of this paper. The authors declare that we have no relevant or
material financial interests that relate to the research described in this paper.

Compliance with ethical standards

Conflict of interest During the three years prior to the acceptance of this article, Dr. Sabia has
been awarded grants from the Charles Koch Foundation (CKF) and the Employment Policies
Institute (EPI) totaling over $10,000. Travel support has also been received from EPI to participate
in a minimum wage panel in Washington, DC. Dr. Sabia’s research effort on the current project
was not funded by these foundations.

Funding This study did not receive grant funding.

Do anti-bullying laws work? 497

Appendix

Table 8 Falsification tests on violent behavior for older young adults

College shootings Arrest rates

Deaths Homicide Property Violent

(1) (2) (3) (4)

Panel I: Any ABL

ABL 0.004 0.061 −0.100 −0.139
(0.054) (0.037) (0.094) (0.123)

Panel II: Intensity rating

Strong ABL −0.027 0.001 −0.001 −0.004
(0.059) (0.049) (0.079) (0.105)

Moderate ABL 0.046 0.094 −0.178 −0.242
(0.066) (0.062) (0.182) (0.241)

Weak ABL −0.074 0.037* −0.001 −0.011
(0.101) (0.021) (0.070) (0.109)

χ2 of βStrong = βModerate (p value) 0.66 (0.42) 1.32 (0.26) 0.64 (0.43) 0.65 (0.43)

χ2 of βStrong = βWeak (p value) 0.24 (0.63) 0.47 (0.50) 0.00 (1.00) 0.00 (0.96)

Panel III: School district policy intensity rating

Strong district policy mandate 0.064 0.095 −0.415 −0.592
(0.110) (0.103) (0.335) (0.463)

Moderate district policy mandate 0.006 0.058 0.024 0.031

(0.051) (0.047) (0.079) (0.081)

Weak district policy −0.080 0.024 −0.035 −0.028
(0.097) (0.021) (0.061) (0.097)

χ2 of βStrong = βModerate (p value) 0.21 (0.65) 0.10 (0.76) 1.59 (0.21) 1.73 (0.19)

χ2 of βStrong = βWeak (p value) 1.17 (0.28) 0.43 (0.51) 1.15 (0.29) 1.32 (0.26)

Panel IV: Individual high intensity components

Written records and reporting 0.105 0.100 −0.067 −0.055
(0.071) (0.077) (0.226) (0.295)

Investigations 0.036 −0.017 −0.019 −0.141
(0.094) (0.091) (0.335) (0.455)

Consequences −0.118 −0.072 0.252 0.252
(0.064) (0.049) (0.163) (0.225)

Training and transparency 0.030 −0.012 0.006 0.079
(0.076) (0.079) (0.170) (0.227)

Legal definitions 0.113 0.174** −0.369* −0.380
(0.083) (0.081) (0.197) (0.259)

Panel V: Number of strong district policy components

Multiple components 0.112 0.086 −0.031 −0.039
(0.080) (0.078) (0.105) (0.101)

498 J.J. Sabia, B. Bass

References

American Psychological Association (2014) Bullying. Retrieved on 11 Aug 2014
Anderson DM, David E (2015) “The Meth Project and Teen Meth Use: New Estimates from the National and

State Youth Risk Behavior Surveys.” Health Economics
Anderson DM, Sabia JJ (2016) “Child Access Prevention Laws, Youth Gun Carrying, and School Shootings.”

IZA Discussion Paper No. 9830.
Anderson DM, Hansen B, Rees DI (2013) Medical marijuana laws, traffic fatalities, and alcohol consumption.

J Law Econ 56(2):333–69
Anderson DM, Hansen B, Rees DI (2015) Medical marijuana laws and teen marijuana use. Am L & Econ Rev

24(12):1644–50
Arseneault L, Walsh E, Trzesniewski K (2006) Bullying victimization uniquely contributes to adjustment

problems in young children: a nationally representative cohort study. Pediatrics 118(1):130–8
Becker G (1968) Crime and punishment: an economic approach. J Polit Econ 76:169–217
Bender D, Friedrich L (2011) Bullying at school as a predictor of delinquency, violence and other anti-social

behaviour in adulthood. Crim Behav Ment Health 21(2):99–106
Bertrand M, Duflo E, Mullainathan S (2004) How much should we trust difference and difference estimates. Q

J Econ 119(1):249–75
Blake JJ, Kim ES, Lund EM, and Zhou Q (2014) Predictors of bully victimization in students with disabilities

a longitudinal examination using a national data set. J Disabil Policy Stu. DOI: 0.1177/
1044207314539012.

Bond L, Carlin JB, Thomas L, Rubink K, Patton G (2001) Does bullying cause emotional problems? A
prospective study of young teenagers. Br Med J 323(7311):480–4

Broward County Public Schools. Anti-Bullying Policy 5.9, Procedural Manual 2010–2011. Retrieved 30
Jul 2012 from http://www.broward.k12.fl.us/sbbcpolicies/docs/P5.9A

Bushaw W, Lopez S (2012) “Public Education in the United States: A Nation Divided.” 2012 Gallop Poll Full
Report. Retrieved on 11 Aug. 2014.

California Education Code D2003] Section 48900-48927. Suspension or expulsion., Available at: http://www.
leginfo.ca.gov/cgi-bin/displaycode?section=edc&group=48001-49000&file=48900-48927

Cappadocia MC, Weiss JA, Pepler D (2012) Bullying experiences among children and youth with autism
spectrum disorders. J Autism Dev Disord 42(2):266–77

Carlyle KE, Steinman KJ (2007) Demographic differences in the prevalence, co-occurrence, and correlates of
adolescent bullying at school. J Sch Health 77(9):623–9

Table 8 (continued)

College shootings Arrest rates
Deaths Homicide Property Violent
(1) (2) (3) (4)

Few components 0.054 0.097 −0.233 −0.337
(0.093) (0.088) (0.307) (0.418)

Zero or one component −0.032 0.039 −0.034 −0.040
(0.064) (0.043) (0.056) (0.057)

χ2 of βMultiple = βFew (p value) 0.24 (0.63) 0.01 (0.93) 0.32 (0.57) 0.40 (0.53)

χ2 of βMultiple = βZero or One (p value) 1.99 (0.16) 0.30 (0.59) 0.00 (0.98) 0.00 (1.00)

N 1020 1020 981 980

Estimated marginal effects are obtained using weighted OLS regressions. All models include controls for state
fixed effects, year fixed effects, state-specific linear time trends, the share of the population that is male and
non-white, average age of the population, and all state-level controls listed in Table 1

***Significant at 1 % level, **at 5 % level, *at 10 % level

Do anti-bullying laws work? 499

http://www.broward.k12.fl.us/sbbcpolicies/docs/P5.9A

http://www.leginfo.ca.gov/cgi-bin/displaycode?section=edc&group=48001-49000&file=48900-48927

http://www.leginfo.ca.gov/cgi-bin/displaycode?section=edc&group=48001-49000&file=48900-48927

Carney JV (2000) Bullied to death perceptions of peer abuse and suicidal behaviour during adolescence. Sch
Psychol Int 21(2):213–23

Cohen, MA (2000) “Measuring the costs and benefits of crime and justice”, in volume 4: measurement and
analysis of crime and justice. 4:263–316. Criminal Justice 2000. Available at: https://www.ncjrs.
gov/criminal_justice2000/vol_4/04f

Columbine-angels.com (2015) “School Violence Around the World.” Available at: http://www.columbine-
angels.com/School_Violence.htm

Connecticut General Statutes (2010) Section 10-222d. Policy on bullying behavior. Available at: http://www.
cga.ct.gov/coc/PDFs/bullying/bullying_statutes_11-10

Craig WM (1998) The relationship among bullying, victimization, depression, anxiety, and aggression in
elementary school children. Personal Individ Differ 24(1):123–30

Daley A, Solomon S, Newman PA (2008) Traversing the margins: intersectionalities in the bullying of lesbian,
gay, bisexual and transgender youth. J Gay Lesbian Soc Serv 19(3–4):9–29

Doll J. Forthcoming. Ending school shootings: a guide to prevention and action
Due P, Holstein BE, Lynch J (2005) Bullying and symptoms among school-aged children: international

comparative cross sectional study in 28 countries. Eur J Public Health 15(2):128–32
Duncan N (1999) Sexual bullying: gender conflict and pupil culture in secondary schools. Routledge, London
Eisenberg ME, Neumark-Sztainer D, Perry CL (2003) Peer harassment, school connectedness, and academic

achievement. J Sch Health 73(8):311–6
Eriksen, Tine Louise M, Helena Skyt N, Marianne S (2014) Bullying in elementary school. J Hum Resour

49(4):839–71
Everytown.org (2015a) “Analysis of School Shootings.” Available at: http://everytown.org/documents/2014/10/analysis-

of-school-shootings
Faris R, Felmlee D (2011) Status struggles network centrality and gender segregation in same- and cross-

gender aggression. Am Sociol Rev 76(1):48–73
Fekkes M, Frans IM P, Pauline Verloove-Vanhorick S (2006) Effects of antibullying school program on

bullying and health complaints. Arch Pediatr Adolesc Med 160(6):638–44
Finland Ministry of Education and Culture (2014) “KiVa International.” Retrieved from: http://www.

kivaprogram.net. 30 June 2016.
Fox S, Stallworth LE (2005) Racial/ethnic bullying: exploring links between bullying and racism in the US

workplace. J Vocat Behav 66(3):438–56
Friedman MS, Koeske AJ, Silvestre AJ, Korr WS (2006) The impact of gender-role nonconforming behavior,

bullying, and social support on suicidality among gay male youth. J Adolesc Health 38(5):621–3
Gini G, Pozzoli T (2009) Association between bullying and psychosomatic problems: a meta-analysis.

Pediatrics 123(3):1059–65
Gladstone GL, Parker GB, Malhi GS (2006) Do bullied children become anxious and depressed adults?: a cross-

sectional investigation of the correlates of bullying and anxious depression. J Nerv Ment Dis 194(3):201–8
Glew GM, Ming-Yu F, Wayne K, Frederick P, Rivara FP, Kernic MA (2005) Bullying, psychosocial adjust-

ment, and academic performance in elementary school. Arch Pediatr Adolesc Med 159(11):1026–31
Green JC, Estlund A, Nishant J, Milan P (2014) “Bullying Prevention Policies and Child Behavior: Evidence Using

State Laws.” 142nd APHA Annual Meeting and Exposition (November 15-November 19,2014).APHA.
Hansen, Benjamin and Michael Lang (2014) “Bullying and Youth Suicide”. Working Paper, University of Oregon
Hansen B, Daniel IR, and Sabia JJ (2016) “Have cigarette taxes lost their bite? New estimates of the

relationship between cigarette taxes and youth smoking.” Forthcoming in American Journal of Health
Economics

Hatzenbuehler ML, Schwab-Reese L, Ranapurwala SI, Hertz MF, Ramirez MR (2015) Associations between
antibullying policies and bullying in 25 states. JAMA Pediatrics 169(10):e152411

Hepburn L, Azrael D, Molnar B, Miller M (2012) Bullying and suicidal behaviors among urban high school
youth. J Adolesc Health 51(1):93–5

Hu W (2011) “Bullying Law Puts New Jersey Schools on Spot.” The New York Times 30 August 2011: A1.
The New York Times

Jeong S, Lee BH (2013) A multilevel examination of peer victimization and bullying preventions in schools. J
Crim 2013:1. doi:10.1155/2013/735397

Juvonen J, Graham S, Schuster M (2003) Bullying among young adolescents: the strong, weak, and troubled.
Pediatrics 112:1231–7

Juvonen J, Wang Y, Espinoza G (2010) Bullying experiences and compromised academic performance across
middle school grades. J Early Adolesc 31(1):152–73

Kaltiala-Heino R, Rimpelä M, Marttunen M, Rimpelä A, Rantanen P (1999) Bullying, depression, and
suicidal ideation in Finnish adolescents: school survey. BMJ 319(7206):348–51

500 J.J. Sabia, B. Bass

https://www.ncjrs.gov/criminal_justice2000/vol_4/04f

https://www.ncjrs.gov/criminal_justice2000/vol_4/04f

http://www.columbine-angels.com/School_Violence.htm

http://www.columbine-angels.com/School_Violence.htm

http://www.cga.ct.gov/coc/PDFs/bullying/bullying_statutes_11-10

http://www.cga.ct.gov/coc/PDFs/bullying/bullying_statutes_11-10

http://everytown.org/documents/2014/10/analysis-of-school-shootings

http://everytown.org/documents/2014/10/analysis-of-school-shootings

http://www.kivaprogram.net/

http://www.kivaprogram.net/

http://dx.doi.org/10.1155/2013/735397

Kim Young S, Yun-Joo K, Bennett L (2005) School bullying and suicidal risk in Korean middle school
students. Pediatrics 115(2):357–63

Klein J (2012) The Bully Society: school shootings and the crisis of bullying in America’s schools. NYU
Press, New York

Konishi C, Hymel S, Zumbo BD, Li Z (2010) Do school bullying and student–teacher relationships matter for
academic achievement? A multilevel analysis. Can J Sch Psychol 25(1):19–39

Kosciw JG, Greytak EA, Diaz EM (2009) Who, what, where, when, and why: demographic and ecological
factors contributing to hostile school climate for lesbian, gay, bisexual, and transgender youth. J Youth
Adolesc 38(7):976–88

Kumpulainen K, Eila R, Irmeli H (1999) Children involved in bullying: psychological disturbance and the
persistence of the involvement. Child Abuse Negl 23(12):1253–62

Langdon SW, Preble W (2007) The relationship between levels of perceived respect and bullying in 5th
through 12th graders. Adolescence 43(171):485–503

Lieberman J (2008) School shootings: what every parent and educator needs to know to protect our children.
Kensington Publishing Corp, New York City

Markowitz S (2001) The role of alcohol and drug consumption in determining physical fights and weapon
carrying by teenagers. East Econ J 27(4):409–32

MassachusettsAdvocatesforChildren Legislation(2010)Section37O.Schoolbullyingprohibited; bullyingprevention
plans. Available at: https://malegislature.gov/Laws/GeneralLaws/PartI/TitleXII/Chapter71/Section37O/Print

Miller T, Cohen M, Wiersema B D1996] Victim costs and consequences: a new look., Final Summary Report
to the National Institute of Justice

Minnesota Management and Budget (2013) “Safe & Supportive Schools Act Explanation of the Bill.”
Available at: http://www.mn.gov/mmb/images/2013%2520SF783

Nansel TR, Overpeck M, Pilla RS, June Ruan W, Simons-Morton B, Scheidt P (2001) Relationships between
bullying and violence among US youth. Arch Pediatr Adolesc Med 157(4):348–53

Nansel TR, Craig W, Overpeck MD, Saluja G, June Ruan W (2004) Cross-national consistency in the
relationship between bullying behaviors and psychosocial adjustment. Arch Pediatr Adolesc Med
158(8):730–6

National Center for Education Statistics (2012) “Indicators of School Crime and Safety.” Available at:
http://nces.ed.gov/pubs2013/2013036

National School Safety and Security Services (2010) “School Associated Violent Deaths and School
Shootings.” Available at: http://www.schoolsecurity.org/trends/violent-deaths-and-school-shootings/

Ng Josephine WY, Sandra KM T (2008) School bullying and the mental health of junior secondary school
students in Hong Kong. J Sch Violence 7(2):3–20

O’Brennan LM, Bradshaw CP, Sawyer AL (2009) Examining developmental differences in the social-
emotional problems among frequent bullies, victims, and bully/victims. Psychol Sch 46(2):100–15

Oregon Revised Statutes (2009) Section 339. Harassment, Intimidation and Bullying. Available at:
http://www.oregonlaws.org/ors/chapter/339

PACER’s National Bullying Prevention Center (2014) About us. Available at: http://www.pacer.
org/bullying/about/

Peters, K (2013) “National Survey Reveals Parents’ Deep Concern About Protecting Kids from Violence.”
Center for American Progress. Retrieved on 11 Aug. 2014

Piehl AM, Kennedy DM, Braga AA (2000) Problem solving and youth violence: an evaluation of the Boston
Gun Project. Am L & Econ Rev 2(1):58–106

Radliff KM, Wheaton JE, Robinson K, Morris J (2012) Illuminating the relationship between bullying and
substance use among middle and high school youth. Addict Behav 37(4):569–72

Rothon C, Head J, Klineberg E, Stansfeld S (2011) Can social support protect bullied adolescents from
adverse outcomes? A prospective study on the effects of bullying on the educational achievement and
mental health of adolescents at secondary schools in East London. J Adolesc 34(3):579–88

Sabia J, Anderson DM D2016] Parental involvement laws, birth control, and mental health: new evidence from
the YRBS., Forthcoming in Journal of Health Economics

Sabia JJ, Pitts MM, Argys L (2016) Do minimum wages really increase youth drinking and drunk driving?
Working Paper, University of Colorado Denver

Salmivalli C, Antti K, Elisa P (2011) Counteracting bullying in Finland: the KiVa program and its effects on
different forms of being bullied. Int J Behav Dev 35(5):405–11

Saluja G, Iachan R, Scheidt PC, Overpeck MD, Sun W, Giedd JN (2004) Prevalence of and risk factors for
depressive symptoms among young adolescents. Arch Pediatr Adolesc Med 158(8):760–5

Seals D, Young J (2003) Bullying and victimization: prevalence and relationship to gender, grade level,
ethnicity, self-esteem, and depression. Adolescence 38(152):735–47

Do anti-bullying laws work? 501

https://malegislature.gov/Laws/GeneralLaws/PartI/TitleXII/Chapter71/Section37O/Print

http://www.mn.gov/mmb/images/2013%2520SF783

http://nces.ed.gov/pubs2013/2013036

School Associated Violent Deaths and School Shootings

http://www.oregonlaws.org/ors/chapter/339

http://www.pacer.org/bullying/about/

http://www.pacer.org/bullying/about/

Stevens V, De Bourdeaudhuij I, Van Oost P (2000) Bullying in Flemish schools: an evaluation of anti-bullying
intervention in primary and secondary schools. B J Ed Psychol 70:195–210

Stoptheshootings.org (2013) “List of School Shootings.” Available at: http://www.stoptheshootings.org
Strøm IF, Thoresen S, Wentzel-Larsen T, Dyb G (2013) Violence, bullying and academic achievement: a study

of 15-year-old adolescents and their school environment. Child Abuse Negl 37(4):243–51
Sweden Ministry of Education and Research (2016) “Education in Sweden.” Retrieved from https://sweden.

se/society/education-in-sweden 30 June 2016
Taylor-Tharp et al (2009) Victimization from mental and physical bullying and substance use in early

adolescence. Addict Behav 34:561–7
Turner HA, Vanderminden J, Finkelhor D, Hamby S, Shattuck A (2011) Disability and victimization in a

national sample of children and youth. Child Maltreat 16(4):275–86
U. K. Department for Education (2013) “Preventing and tackling bullying.” Retrieved from: https://www.gov.

uk/government/uploads/system/uploads/attachment_data/file/444862/Preventing_and_tackling_bullying_
advice . 30 June 2016

U.S. Department of Education (2011) Analysis of State Bullying Laws and Policies. Retrieved from
http://www2.ed.gov/rschstat/eval/bullying/state-bullying-laws/state-bullying-laws

U.S. Department of Health and Human Services (2014a) What Is Bullying? Retrieved on 11 Aug 2014
U.S. Department of Health and Human Services (2014b) About us. Retrieved on 11 Aug 2014
Undheim AM, Sund AM (2010) Prevalence of bullying and aggressive behavior and their relationship to

mental health problems among 12- to 15-year-old Norwegian adolescents. Eur Child Adolesc Psychiatry
19(11):803–11

Van Fleet DD, Van Fleet EW (2010) The violence volcano: reducing the threat of workplace violence.
Information Age Publishing, Inc, Charlotte

Washington Clover Park School District (2016) “Prohibition of Harassment, Intimidation and Bullying
(HIB)”. http://www.cloverpark.k12.wa.us/General/AntiBullyPolicy.aspx. Retrieved 30 June 2016

West K (2014) Public health approach in the evaluation of two states’ anti-bullying laws and local school
district policies. Working Paper, George Washington University

Wilkins‐Shurmer A, Mj O’C, Najman JM, William B, Williams GM, Anderson MJ (2003) Association of
bullying with adolescent health-related quality of life. J Pediatr Child Health 39(6):436–41

Wolke D, Copeland WE, Angold A, Jane Costello E (2013) Impact of bullying in childhood on adult health,
wealth, crime, and social outcomes. Psychol Sci 24(10):1958–70

502 J.J. Sabia, B. Bass

http://www.stoptheshootings.org/

https://sweden.se/society/education-in-sweden

https://sweden.se/society/education-in-sweden

https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/444862/Preventing_and_tackling_bullying_advice

https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/444862/Preventing_and_tackling_bullying_advice

https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/444862/Preventing_and_tackling_bullying_advice

http://www2.ed.gov/rschstat/eval/bullying/state-bullying-laws/state-bullying-laws

http://www.cloverpark.k12.wa.us/General/AntiBullyPolicy.aspx

Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.

• c.148_2016_Article_622

Do anti-bullying laws work? New evidence on school safety and youth violence
Abstract
Introduction
Background
Data and measures
YRBS data
School shooting and crime data
Anti-bullying laws
Empirical approach
Main results
School safety
Bullying
Minor teen school shootings and crime
Conclusions
Appendix
References