Career Counseling and Youth Crime. Evidence from …...Career Counseling and Youth Crime. Evidence...
Transcript of Career Counseling and Youth Crime. Evidence from …...Career Counseling and Youth Crime. Evidence...
Career Counseling and Youth Crime.Evidence from Career Compass of
Louisiana
Stephen Barnes, Louis-Philippe Beland and Swarup Joshi∗
October 2017
Abstract
We investigate the impact of career counseling services in high school on youthcrime. To do so, we study the impact of Career Compass of Louisiana, which providescounseling services to students regarding college admissions, enrollment, financial aid,and career exploration. We use a difference-in-differences framework to analyze ad-ministrative student-level data from the Louisiana Department of Education and indi-vidually matched administrative crime data from the Louisiana Department of PublicSafety and Corrections and the Office of Juvenile Justice. We find that the Career Com-pass program in Louisiana reduced youth crime by over 4 percent in treated districts.Moreover, the effect is more pronounced for students with low test scores, studentsreceiving free and reduced cost lunch, male and minority students. Our estimates arerobust to different specifications and placebo tests.
JEL Classification: I21, I26, K42
Keywords: Career Counseling, Youth Crime, Education.
∗Barnes: Louisiana State University, [email protected]; Beland: Louisiana State University, [email protected]; Joshi: Louisiana State University, [email protected].
1
1 Introduction
A wide range of policies and interventions aim to increase college attendance. Several papers
have documented that interventions providing career coaching and college admissions coun-
seling increase college enrollment, especially for marginal students. This literature shows
that simplifying information about college and financial aid and providing students access
to professional assistance can generate substantial improvements in students’ postsecondary
outcomes (e.g. Hoxby and Turner (2013, 2015), Castleman and Page (2015), Carrell and
Sacerdote (2017)).
In this paper, we investigate if career counseling also has an impact on crime. Youth
crime is a lasting concern for policymakers and scholars due to large associated social costs.
Research shows that juvenile delinquency has long-term consequences. For example, juvenile
delinquents are more likely to be unemployed, have lower wages and be incarcerated as an
adult (e.g Waldfogel (1994a,1994b), Hjalmarsson (2008), and Aizer and Doyle (2015)).
Our research studies the impact of Career Compass of Louisiana, which provides college
and career counseling services and coaching to high school seniors regarding college admis-
sions, enrollment, financial aid, and career exploration. Career Compass partners with local
school districts and charitable foundations to secure funding to support a district-wide con-
tract such that all public high schools within a school district receive services once a district
contracts with the organization. Career Compass started in a single district in 2006 and
expanded to other districts gradually over time. By 2012, Career Compass was operating in
23 districts in Louisiana.1 We use a difference-in-differences framework and administrative
student-level data from the Louisiana Department of Education matched individually with
administrative crime data from the Louisiana Department of Public Safety and Corrections
and Office of Juvenile Justice to investigate the impact of college and career counseling on
youth crime.2
1The expansion path of Career Compass across Louisiana is illustrated in Figure 1.2A unique state identification number allows us to match students in the education and crime databases.
2
We find that the Career Compass program decreases youth crime in Louisiana by over 4
percent in districts receiving services. We also observe a decrease in in-school misbehavior.
We investigate the heterogeneity of effects and find that results are more pronounced for
students with low standardized test scores in grade 8, students receiving free and reduced
cost lunch, male and minority students. Our results suggest that career counseling, in
addition to increasing college attendance as previously shown in the literature, decreases
youth crime. Our estimates are robust to different specifications and placebo tests.
The rest of the paper is organized as follows: Section 2 discusses the related literature;
Section 3 provides a description of Career Compass, the data and presents descriptive statis-
tics; Section 4 presents the empirical strategy; Section 5 is devoted to the main results,
heterogeneity of the results, robustness checks and potential mechanisms; and Section 6
concludes with a discussion of policy implications.
2 Literature
Our paper is related to the literature on school counseling and college-going interventions.
Several papers find a positive impact of school counseling and college-going interventions on
student outcomes. Hoxby and Turner (2013 and 2015) find that college counseling raises
students’ applications, admissions, enrollment, and progress at selective colleges. They
also show that interventions change students’ knowledge and decision-making. Stephan
and Rosenbaum (2013) use data on high school seniors in Chicago to find that coaches im-
prove the types of colleges students attend. Their results suggest that targeting resources
may improve high-school-to-college transitions for disadvantaged students. Carrell and Sac-
erdote (2017) present evidence from a series of field experiments employing college coaching
and mentoring, and find large impacts on college attendance and persistence. Castleman
and Page (2015) study the impact of two interventions: personalized text messaging and
near-aged peer mentors. Both cost-effective approaches substantially increased college en-
3
rollment for students with lower-quality college counseling or information. Castleman and
Goodman (2017) study the impact of intensive college counseling provided to college-seeking,
low-income students by a Massachusetts program that admits applicants partly on the basis
of a minimum GPA requirement. They find that counseling successfully shifts enrollment
toward four-year colleges encouraged by the program and appears to improve persistence
through the third year of college. Their results suggest that intensive college counseling
might improve degree completion rates for disadvantaged students.
The number of school counselors has also been documented as an important factor for
students. Carrell and Hoekstra (2014) find that additional school counselors increase student
achievement and decrease student misbehavior in high-school. Moreover, related papers show
that college attendance is positively affected by mandated college entrance tests, access to
tests and access to test centers (see Bulman (2015); Goodman (2016) and Pallais (2015)).
The literature also documents a positive impact of providing information about financial aid
on college attendance, especially for disadvantaged students (e.g. Bettinger et al. (2012)
and Dinkelman and Mart́ınez (2014)). In sum, the literature shows that coaching and other
targeted interventions increase postsecondary outcomes, especially for marginal students.
Another set of literature related to this study has investigated the determinants of youth
crimes. For example, Currie and Tekin (2012) document that childhood maltreatment greatly
increases the probability of engaging in crime later in life. Others have investigated the im-
pact of particular policies such as Sunday alcohol sale restrictions (Heaton, (2012)) and
juvenile curfews (Carr and Doleac, (2015)) on youth crime. Another line of research has
studied the impact of school calendar and hours spent in school on youth crime (eg. Jacob
and Lefgren (2003), Berthelon and Kruger (2011) and Akee, Halliday, and Kwak, (2014)). In
addition, previous research has shown that juvenile delinquency has long-term consequences,
underscoring the importance of understanding the determinants of youth crime and effec-
tiveness of potential interventions. For example, juvenile delinquents are more likely to be
unemployed, have lower wages and be incarcerated as adults (e.g. Waldfogel (1994a,1994b);
4
Hjalmarsson, (2008) and Aizer and Doyle (2015)).
Our paper is related to the growing body of literature on the link between education and
crime. Several studies document the positive impact of education on reducing juvenile crime.
For example, Anderson (2014) investigates the link between minimum high school drop out
age and juvenile arrests. He finds that minimum dropout age requirements significantly
decrease property and violent crime arrest rates for individuals 16 to 18 years old, which
is consistent with an incapacitation effect of schooling. Landerso et al. (2016) exploit
discontinuity with school starting age in Denmark to find that higher school starting age
lowers the propensity to commit crime at young ages and the number of crimes committed
for boys. Cook and Kang (2016) study six cohorts of school children in North Carolina and
find that those born soon after the cut date for enrolling in public kindergarten are more
likely to drop out of high school before graduation and to commit a crime by age 19 (see also
Machin et al. (2011), Brugard and Falch (2013), and Bell et al. (2016)). Our paper is related
to Doleac and Gibbs (2016), which studies the impact of the Kalamazoo Promise program in
Michigan in which graduates from local high schools are guaranteed full tuition to an in-state
public university for up to four years. They study the impact of the program announcement
on risky behaviors of teenagers. They find evidence that the program announcement lowered
arrests and teen birth rates. Recent studies have also documented a long-term impact of
education on crime. For example, Deming (2011), and Lochner and Moretti (2004) illustrate
that education is negatively related to adult crime.
Our contribution is to document an additional benefit of an intervention providing college
and career counseling: a decrease in youth crime. We use administrative data from Louisiana
to identify treated students and track criminal activity. This research also contributes to the
literature on the relationship between education and crime. We present evidence that career
counseling, in addition to increasing college attendance as shown in the literature, reduces
youth crime. Our results also point to benefits of increasing attendance at community
colleges and technical schools for marginal students.
5
3 Career Compass, Data and Descriptive Statistics
3.1 Career Compass and Louisiana
Career Compass of Louisiana is a non-profit organization with a mission to provide guidance
in career choice and college admission to high school students. The Career Compass model
focuses on partnerships at the school and district level in order to operate in all secondary
public schools within a school district with the core program service providing college and
career coaching to all students in grade 12 at those schools. Career Compass started in 2006
and expanded gradually over time. By 2012, Career Compass was operating in 23 districts.
The expansion path of Career Compass across Louisiana is illustrated in Figure 1.3
Career Compass aims to help students identify post-secondary options that match a stu-
dent’s interests and abilities and facilitate college enrollment through application assistance.
Students in treated districts have consistent access to a coach for queries regarding admis-
sions, enrollment, financial aid, and career exploration. Career Compass coaches provide
one-on-one assistance, which includes goal setting through a College Success Plan, career
aptitude assessments, career and technical education options, high school course selection,
selection of programs of study, financial aid awareness including Free Application for Federal
Student Aid (FAFSA), and financial assistance with college application and exam registra-
tion fees for low-income students. Career Compass has waived or paid more than $20,000
in college application fees. Similarly, they have also assisted in getting more than $100,000
of exam fees waived or paid. Career Compass focuses on maintaining a low-cost structure,
averaging $110 to $150 per student per year.
One important feature of this program is an emphasis on helping students consider the full
range of post-secondary options including community and technical colleges and four-year
schools. Coaches initiate meetings with each high school senior throughout the year to ensure
that the student is completing necessary steps to complete requirements and submit a post-
3Data on the timing and location of Career Compass expansions were provided by Career Compass ofLouisiana. See http://www.careercompassla.org/ for more details.
6
secondary application. Their mission is to increase the number of students who attend a post-
secondary institution (technical, community and four-year universities). Some of the most
frequent placements are at community colleges and technical schools. In 2016, the top two
placements were Baton Rouge Community College and Bossier Parish Community College.4
This differs from other interventions promoting post-secondary enrollment discussed in the
literature.
Career Compass has expanded over time using a combination of contracts from local
school districts and grants from private foundations and community foundations. This com-
bination of funding has helped Career Compass expand to a variety of school districts includ-
ing a mix of urban and rural sites as well as districts with above and below average resources
per student. While the expansion path and current set of schools served do not suggest any
systematic selection of districts into the program based on characteristics associated with
crime, we conduct a variety of robustness checks and find no indications of selection into the
program (see Table 7 and Figures A.1, A.2 and A.3).
Louisiana provides a particularly good backdrop for studying the impact of a program
like Career Compass on youth crime due to the relatively high crime rate in the state. In
2015, Louisiana’s juvenile arrest rate was above the national average for aggravated assault,
larceny, drug abuse and weapons charges, four of the five categories of crime reported by the
U.S. Department of Justice’s Office of Juvenile Justice and Delinquency Prevention (OJJDP
2017). In addition, Louisiana’s overall crime rate is well above the national average with a
violent crime rate of 539.7 compared to a national rate of 383.2 per 100,000 residents and a
property crime rate of 3,353.4 compared to a national rate of 2,487.0 per 100,000 residents
according to the Federal Bureau of Investigation’s 2015 Crime in the United States report.
4More details are available here: http://www.careercompassla.org/service-locations/
7
3.2 Data and Descriptive Statistics
In this paper, we use administrative data on students and crime to identify how Career
Compass intervention affects criminal activity. We use student-level administrative data
from the Louisiana Department of Education from 2005 to 2012 for all public high school
students in the state of Louisiana. Our data include approximately 250,000 students from
238 public schools in the state of Louisiana. Among them, Career Compass was implemented
in 99 schools. Our treatment group is composed of students who were enrolled in districts
with Career Compass services in place when the student began grade 12. Schools not yet
treated in a given year and the remaining 139 untreated schools are used as our control
group. Our data contains information on past student performance on a standardized test in
grade 8 (LEAP test) and prior schools attended, as well as a range of student characteristics
including gender, age, ethnicity, and eligibility for free or reduced cost school meals.5 The
data include records of in-school discipline incidents over time, which allow us to study
discipline events relative to a pre-treatment baseline. Each student is matched individually
to administrative state data on crime.
The crime data used in this paper come from the Louisiana Department of Public Safety
and Corrections, and the Office of Juvenile Justice including all case records from 2005 to
2012. We have information on the crime date and crime type (i.e. felony and misdemeanor).
Our main outcome of interest is whether or not the student committed a crime at any
point after the start of grade 12. We also investigate if Career Compass has a different
impact by type of crime, using two main categories of crime - felony and misdemeanor.
Felonies are serious criminal acts usually punishable by imprisonment of more than one
year. Misdemeanors and minor juvenile crimes, are grouped together in our analysis and
referred to collectively as misdemeanor crimes. Furthermore, the data set has a violent crime
5The Louisiana Department of Education administers a test given to eighth graders as part of theLouisiana Educational Assessment Program commonly referred to as the LEAP test. Since 1999, studentshave been tested in the subjects of English Language Arts (ELA), Mathematics, Science and Social Studies.Unfortunately, we do not have a standardized test score after treatment (after grade 8).
8
identifier, a specific subset of felony crimes. Therefore, we also consider the impact of Career
Compass exposure on incidence of violent crime. Over the study period, 13,461 crimes were
committed by youth in our data. Among those crimes, 5,465 are classified as misdemeanor,
7,996 as felony and 1,030 of those felonies are classified as violent crimes. Finally, these data
track criminal activity over time allowing us to investigate recidivism.
Table 1 presents descriptive statistics on students in our treatment group and students
in our control group. The data show that our treatment and control groups are similar in
terms of age, gender and test scores.6 However, treated students are more likely to be from
a minority group, and are more likely to receive free or reduced cost school lunch.7 Table 1
also presents descriptive statistics for the control group based on propensity score matching
(PSM).
4 Empirical Strategy
We use a difference-in-differences (DiD) strategy to analyze the effect of Career Compass on
youth crime. Our treatment group is composed of students who were enrolled in districts
with Career Compass services in place when the student began grade 12. The comparison
group consists of students in districts not covered by the Career Compass program during
the student’s 12th grade year. We estimate the following equation:
Yisdt = β0 + β1CareerCompassdt +Xisdt + µs + γt + τdt + εisdt (1)
6High and low test score students are distinguished based on their 8th grade LEAP test score. High-performing students are those who have LEAP test scores in the top 25 percent of statewide scores on theLEAP test, and low-performing students are those who have scores in the bottom 25 percent of statewidescores. All others are considered as middle-performing students. The average scores in each of the fourLEAP test subjects are used individually for the classification (i.e., a student in high-performing group isscoring top 25 percent in each of the four LEAP subjects.) . Moreover, students with missing test scoresin any one of the four subjects have been dropped from this part of the analysis. Results are robust toalternative classifications.
7We group hispanic and black students under minority students.
9
where Yisdt is the outcome of interest for student i in school s, district d and year t. Our
main outcome variable of interest is a dummy variable that takes a value of 1 if a student
i committed a crime at any point after the start of grade 12 and 0 otherwise. We also
present results for different types of crime: felony and misdemeanor. CareerCompassdt is
a binary variable that takes a value of 1 if the service is in place in district d in year t
when a student is in grade 12 and is equal to 0 otherwise. β1 is our parameter of interest
and represents the impact of Career Compass on the outcome variable. Xisdt is a vector
of student characteristics. We consider the following characteristics: gender, age, ethnicity,
free or reduced cost school meal eligibility, and past student performance in grade 8. In the
heterogeneity subsection, we test if results vary by student characteristics.
We include school fixed effects, µs to control for any time-invariant school-level factors
that may be correlated with outcome variables for students in school s. We also include
year fixed effects, γt, to control for any changes or trends from 2005 to 2012. In our anal-
ysis, observations are organized based on the year when the student enters grade 12. τdt
represents a district-specific time trend.8 In our subsection on heterogeneity and robustness,
we investigate whether results are robust to different control groups, conduct a number of
robustness checks including placebo tests, permutation tests and present event study graphs,
including investigation of compositional changes at treated schools and districts around the
implementation of the intervention.9
8We cluster standard errors at the school level to account for potential correlations among students inthe same school.
9We exclude the school districts of East Baton Rouge and Orleans (New Orleans) from our main analysisas major changes occurred in each of those districts during our sample period (creation of new city schooldistricts within the parish and Hurricane Katrina, respectively.) In appendix Table A.4, we show that ourresults are robust to the inclusion or exclusion of those districts.
10
5 Results
5.1 Main Results
Table 2 presents the impact of the college and career counseling intervention, Career Com-
pass, on youth crime. Column (1) displays the key result from estimating equation (1) for
the outcome variable all crime, and shows that Career Compass significantly decreases youth
crime by 4.55% in treated districts. Columns (2) and (3) display results based on two differ-
ent categories of crime; column (2) presents results for crimes classified as felony and column
(3) for crimes classified as misdemeanor. Columns (2) and (3) show that Career Compass
significantly decreases felony (-1.13%) and misdemeanor (-4.34%) offenses. Results in Table
2 show that exposure to Career Compass leads to a decrease in criminal activity among
treated students.
5.2 Robustness, Heterogeneity and Discussion of Mechanisms
5.2.1 Heterogeneity
Next, we investigate heterogeneity in impact by student characteristics. Table 3 presents
separate results for male students (Panel A), female students (Panel B), white students
(Panel C); minority students (Panel D); students with free or reduced cost lunch eligibility
(Panel E); and high, middle and low-performing students (Panels F, G and H).10 Table
3 shows that the Career Compass intervention has a significant impact on crime for male
students (-5.81% for all crimes); minority students (-7.54% for all crimes); students on free
and reduced cost lunch (-8.89% for all crimes); and low-performing students (-7.02% for all
crimes). Results for white students, female and high-performing students are not statistically
significant for all crimes. Results are similar for the more narrowly-defined crime categories
(felony and misdemeanor), as presented in columns (2) and (3) of Table 3.
10As defined above, high-performing students are those who have scored in the top 25 percent in their 8thGrade LEAP test and low-performing students have test scores at the bottom 25 percent.
11
Table 4 presents results for violent crime (a subset of felony) for all students and by
student characteristics (white; minority; free or reduced cost school lunch; and high, middle
and low-performing students). Results for this category of crime show that there is a small
and significant decrease in the propensity to commit a violent crime for all students (-0.56%).
Results for male students (-1.28%), minority students (-2.20%), middle-performing students
(-0.93%) and low-performing students (-0.20%) are statistically significant. Once again, we
find no statistically significant impact of Career Compass on violent crime among white
students, female students and high-performing students.
Table 5 investigates if the Career Compass intervention has an impact on the propensity
for recidivism, for all crime and by type of crime. Table 5 limits the sample to students
with juvenile or criminal records prior to grade 12. Table 6 shows that Career Compass
significantly decreases the propensity to commit any other crime (-1.15%), felony (-0.93%)
and misdemeanor (-0.78%).
We also investigate heterogeneity by district performance score and district crime rate.
The district performance score is a comprehensive measure designed to assess how the school
district as a unit has performed.11 Results by district type are summarized in the appendix in
Table A.1 and show that the impact of Career Compass is concentrated in low-performance
districts. In a similar way, we investigate how the impact of Career Compass differs between
districts with high and low crime rates.12 Table A.2 shows that the impact of the intervention
is concentrated in districts with high crime rates. Taken together, Tables A.1 and A.2 suggest
that this type of counseling intervention could be a valuable tool for reducing crime in schools
districts with low student performance and high rates of crime.
11It is calculated based on student scores on standardized tests as well as attendance, dropout rates, andgraduation outcomes for all students in a district. District Performance Score for 2012 was retrieved fromLouisiana Department of Education. We divide districts into two categories: high- and low-performancedistricts. High-performance districts are school districts with district performance score above the statewidemedian in 2012 and low-performance districts are all others.
12High crime rate districts are defined as school districts with a crime rate above the median of all districts.
12
5.2.2 Robustness
We next implement placebo tests, which are presented in Table 6. The placebo tests are
generated by turning on the Career Compass dummy in different years, specifically in years
(t-1) and (t-2), before it was actually implemented in each district. These placebo interven-
tions should have no significant impact on crime or any subcategory of crime. If there is a
significant relationship, then we would assume there is correlation between the crime trend
and Career Compass intervention. Table 6 shows that placebo treatments do not produce a
significant effect on youth crime or any subcategory. We take this as further evidence that
prior trends are not generating these results. We also present a series of event study graphs in
Figure 2 by category of crime. These graphs show a decrease in crime after the interventions
and no pre-treatment trend can be detected, providing further support to the main results in
Table 2. Figure 3 presents a similar exercice for all crime by students characteristics (male
students, female students, white students, minority students, students with free or reduced
cost lunch eligibility; and high and low-performing students). Results are similar to Table
3, with no prior trends.
Table 7 investigates if Career Compass program has an impact on characteristics of
students (fraction of minority, white, on free and reduced lunch, high performing and low
performing students in grade 8 tests) in the treated schools, characteristics of Parishes (ex-
penditures per pupil, unemployment rate, revenue per pupil, district performance score and
average earning of teachers), and characteristics of schools (full-time equivalent teachers per
100 enrolled, enrollment per guidance counselor, school performance score, enrollment count
and share of free and reduced lunch students). We find no evidence that Career Compass has
an impact on those characteristics thereby reducing the concern that the effects are driven
by compositional changes in student characteristics or other changes in treated schools and
districts that occur at the same time as the expansion of Career Compass. Similarly, Figures
A.1, A.2 and A.3 present a series of event study graphs investigating potential compositional
changes, using the same characteristics as Table 7. Once again, Figures A.1, A.2 and A.3
13
suggest that our results are not driven by compositional changes or other changes in schools
and districts, with no prior trends.
A number of other robustness checks are also included in the appendix. Table A.3
presents estimates using alternative estimation methods. Table A.3, panel A presents results
using a logit specification and Table A.3, panel B presents results using propensity score
matching.13 Tables A.3, panel A and B both present qualitatively similar results to those
found in our primary analysis summarized in Table 2. Table A.3, Panel C replicates the
main results of Table 2 but excludes the year 2012, which represents a large portion of
the interventions. Results, again, are qualitatively similar. Table A.3, Panels D and E
present results without controls and alternative clustering, respectively. Results are once
again qualitatively the same. In our primary analysis, we exclude the school districts of East
Baton Rouge and Orleans because major changes occurred in each district during the study
period (creation of new city-based school districts within the parish and Hurricane Katrina,
respectively). In appendix Table A.4, we shows that our results are robust to the inclusion
or exclusion of those districts. Finally, we conduct permutation tests. We randomize the
interventions and rerun baseline regressions for our main outcome variables: all crimes, felony
and misdemeanor. We do 1,000 replications and figures for coefficient estimate (along with
a vertical line representing our baseline estimate) presented in the appendix suggest it is
unlikely that the results are due to chance (see appendix Figure A.4).
Overall, results are robust to alternative specifications and robustness checks, supporting
our finding that Career Compass does significantly reduce youth crime.
13Propensity score matching, as shown in panel B of Tabel A.3, is done on student characteristics: gender;age; ethnicity; free or reduced cost school lunch eligibility; past student performance and characteristicsof districts and schools such as district performance score and number of full time equivalent teachers per100 students. We also replicated all other main tables using propensity score matching and results werequalitatively the same. In adition, we tested the robustness of propensity score matching results usinga wide variety of student-level, school-level, and parish-level characteristics, and found the results to bequalitatively similar.
14
5.2.3 Discussion: Potential Mechanisms
We next investigate potential mechanisms to explain why coaching intervention might affect
youth crime. We investigate the impact of the intervention on school misbehavior and post-
secondary enrollment. Table 8 presents results for the propensity to get disciplined at school
for all students and by student characteristics (gender; race; free or reduced cost school lunch;
and high-, middle- and low-performing students). Table 8 shows that the Career Compass
intervention significantly decreases the propensity to get disciplined (-4.40%) in school for all
students. It shows once again that the effect is more pronounced for male students (-5.05%),
minority students (-8.25%), students receiving free or reduced cost lunch (-8.86%) and low-
performing students (-7.28%). This suggests an increase in student effort in school following
the intervention. Table 9 investigates if Career Compass affects post-secondary enrollment,
using school level regressions.14 Table 9, column (1) presents results for all schools, columns
(2)-(5) presents results for different subsets of schools. As documented previously in the
literature, we find that a coaching intervention, Career Compass of Louisiana, increases
post-secondary enrollment. Tables 8 and 9 suggest that Career Compass leads to student
increasing their effort and post-secondary enrollment, which ultimately lead to lower youth
crime.
6 Conclusion
In this paper, we investigate the impact of Career Compass of Louisiana, a college and career
counseling service providing coaching to high school students regarding college admissions,
enrollment, financial aid, and career exploration. Using a difference-in-differences framework
and student-level data from the Louisiana Department of Education linked with individual
crime data from the Louisiana Department of Public Safety and Corrections, and the Office
of Juvenile Justice, we investigate the impact of a college and career counseling program on
14We conduct analysis for post-secondary enrollment outcome at school level because enrollment data isonly available at the school level.
15
youth crime. We find that implementation of Career Compass in schools decreases youth
crime. We investigate heterogeneity of the effects and find that results are more pronounced
for male students, students with low test scores, students receiving free or reduced cost lunch
and minority students. We also find that the impact is larger in high-crime school districts
and in low-performance districts. This suggests that coaching interventions should be pri-
oritized in those districts as a way to reduce crime. Our estimates are robust to different
specifications and placebo tests. Our results have important policy implications as juvenile
delinquency has long-term consequences and it suggests that a low-cost intervention, student
coaching and career counseling of high school seniors, decreased youth crime and misbehav-
ior in school. Our results also points to benefits to interventions leading to an increase in
community colleges and technical schools attendance for marginal students.
References
Aizer, Anna, and Joseph J. Doyle. “Juvenile incarceration, human capital, and future
crime: Evidence from randomly assigned judges.” Quarterly Journal of Economics,
(2015): qjv003.
Akee, Randall Q., Timothy J. Halliday, and Sally Kwak. “Investigating the effects of fur-
loughing public school teachers on juvenile crime in Hawaii.” Economics of Education
Review 42 (2014): 1-11.
Anderson, D. Mark. “In school and out of trouble? The minimum dropout age and juvenile
crime.” Review of Economics and Statistics 96.2 (2014): 318-331.
Anderson, D. M., B. Hansen, and M. B. Walker, 2013, “The Minimum Dropout Age and
Student Victimization.” Economics of Education Review, 35: 66-74
Beland, Louis-Philippe, and Dongwoo Kim. “The effect of high school shootings on schools
and student performance.” Educational Evaluation and Policy Analysis, 38.1 (2016):
16
113-126.
Beland, Louis-Philippe, and Richard Murphy. “Ill communication: technology, distraction
& student performance.” Labour Economics, 41 (2016): 61-76.
Bell, Brian, Rui Costa, and Stephen Machin. “Crime, compulsory schooling laws and
education.” Economics of Education Review 54 (2016): 214-226.
Berthelon, Matias E., and Diana I. Kruger. “Risky behavior among youth: Incapacitation
effects of school on adolescent motherhood and crime in Chile.” Journal of Public
Economics 95.1 (2011): 41-53.
Bulman, George. “The effect of access to college assessments on enrollment and attain-
ment.” American Economic Journal: Applied Economics 7.4 (2015): 1-36.
Brugard, Kaja Hiseth, and Torberg Falch. “Post-compulsory education and imprisonment.”
Labour Economics 23 (2013): 97-106.
Bettinger, E. P., Long, B. T., Oreopoulos, P., & Sanbonmatsu, L. (2012). “The role of
application assistance and information in college decisions: Results from the H&R
Block FAFSA experiment.” Quarterly Journal of Economics , 127(3), 1205-1242.
Bettinger, Eric P., and Rachel B. Baker. “The effects of student coaching: An evaluation
of a randomized experiment in student advising.” Educational Evaluation and Policy
Analysis 36.1 (2014): 3-19.
Carr, Jillian B., and Jennifer L. Doleac. “Keep the Kids Inside? Juvenile Curfews and
Urban Gun Violence.” (2015).
Carrell, Scott, and Bruce Sacerdote. “ Why Do College Going Interventions Work?.” Amer-
ican Economic Journal: Applied Economics , 2017, forthcoming
Carrell, Scott E., and Mark Hoekstra. “Are school counselors an effective education input?”
Economics Letters, 125.1 (2014): 66-69.
17
Castleman, Benjamin L., and Lindsay C. Page. “Summer nudging: Can personalized text
messages and peer mentor outreach increase college going among low-income high
school graduates?.” Journal of Economic Behavior & Organization 115 (2015): 144-
160.
Cook, Philip J., and Songman Kang. “Birthdays, schooling, and crime: regression-discontinuity
analysis of school performance, delinquency, dropout, and crime initiation.” American
Economic Journal: Applied Economics 8.1 (2016): 33-57.
Currie, Janet, and Erdal Tekin. “Understanding the cycle childhood maltreatment and
future crime.” Journal of Human Resources 47.2 (2012): 509-549.
Damm, Anna Piil, and Christian Dustmann. “Does growing up in a high crime neigh-
borhood affect youth criminal behavior?.” American Economic Review 104.6 (2014):
1806-1832.
Deming, D. J., 2011, “Better Schools, Less Crime?.” Quarterly Journal of Economics,
126(4): 2063-2115
Dinkelman, Taryn, and Claudia Mart́ınez A. “Investing in schooling in Chile: The role
of information about financial aid for higher education.” Review of Economics and
Statistics 96.2 (2014): 244-257.
Dills, A. K. and R. Hernandez-Julian, 2011, “More Choice, Less Crime.” Education Finance
and Policy, 6(2): 246-266
Doleac, Jennifer L., and Chloe R. Gibbs. “A Promising Alternative: How Making College
Free Affects Teens Risky Behaviors.” (2016).
Federal Bureau of Investigation. 2015. Crime in the United States 2015 Uniform Crime Re-
ports online. Baltimore. Accessed 7 July 2017. Available from https://ucr.fbi.gov/crime-
in-the-u.s/2015/crime-in-the-u.s.-2015. Internet.
18
Flanders, Will, and Corey A. DeAngelis. “More Graduates, Less Criminals? The Economic
Impacts of the Milwaukee Parental Choice Program.” (2017).
Goodman, Sarena. “Learning from the test: Raising selective college enrollment by provid-
ing information.” Review of Economics and Statistics 98.4 (2016): 671-684.
Grogger, J., 1997, “Local Violence and Educational Attainment.” Journal of Human Re-
sources, 32(4): 659-682
Hoxby, Caroline M., and Sarah Turner. “What high-achieving low-income students know
about college.” American Economic Review 105.5 (2015): 514-517.
Hoxby, Caroline, and Sarah Turner. “Expanding college opportunities for high-achieving,
low income students.” Stanford Institute for Economic Policy Research Discussion
Paper 12-014 (2013).
Heaton, Paul. “Sunday liquor laws and crime.” Journal of Public Economics 96.1 (2012):
42-52.
Hjalmarsson, Randi. 2008. “Criminal Justice Involvement and High School Completion.
Journal of Urban Economics 63: 613-630.
Hyman, Joshua. “ACT for all: The effect of mandatory college entrance exams on postsec-
ondary attainment and choice.” Education Finance and Policy (2016).
Imberman, S. A., 2011, “The Effect of Charter Schools on Achievement and Behavior of
Public School Students.” Journal of Public Economics, 95(7): 850-863
Jacob, Brian A., and Lars Lefgren. “Are idle hands the devil’s workshop? Incapacitation,
concentration, and juvenile crime.” American Economic Review 93.5 (2003): 1560-
1577.
Joshi, Swarup and Stephen Barnes. ”Impact of Low Cost post-secondary enrollment inter-
vention: Evidence from Louisiana”, (2017).
19
Landers, Rasmus, Helena Skyt Nielsen, and Marianne Simonsen. “School Starting Age and
the Crimeage Profile.” Economic Journal (2016).
Lochner, L. and E. Moretti, 2004, “The Effect of Education on Crime: Evidence from Prison
Inmates, Arrests, and Self-Reports.” American Economic Review, 94(1): 155-189
Machin, Stephen, Olivier Marie, and Sunica Vuji. “The crime reducing effect of education.”
Economic Journal 121.552 (2011): 463-484.
Marvell, T. B., 2001, “The Impact of Banning Juvenile Gun Possession.” Journal of Law
and Economics, 44(S2): 691-713
OJJDP Statistical Briefing Book. http://www.ojjdp.gov/ojstatbb/crime/qa05103.asp?qaDate=2015.
Released on March 27, 2017.
Sharkey, P., 2010, “The Acute Effect of Local Homicides on Children’s Cognitive Performance.”
Proceedings of the National Academy of Sciences, 107(26): 11733-11738
Stephan, Jennifer L. & Rosenbaum, James E. (2013). “Can High Schools Reduce College Enroll-
ment Gaps With a New Counseling Model?.” Educational Evaluation and Policy Analysis
June 2013 35: 200-219
Pallais, Amanda. “Small differences that matter: Mistakes in applying to college.” Journal of
Labor Economics 33.2 (2015): 493-520.
Waldfogel, Joel. 1994a. “The Effect of Criminal Conviction on Income and the Trust Reposed in
the Workmen.” Journal of Human Resources 29: 62-81.
Waldfogel, Joel. 1994b. “Does Conviction Have a Persistent Effect on Income and Employment?
International Review of Law and Economics 14: 103-119.
20
Figure 1: Career Compass locations, Career Compass is operational in shaded parishesExpansion by year: 2006 - East Baton Rouge; 2008 - Iberville, Pointe Coupee, and West BatonRouge; 2009 - Assumption; 2010 - Caddo, Webster; 2011 - Bossier, Claiborne, St. Mary, St. James,and St. John; 2012 - Allen, Avoyelles, Catahoula, Lasalle, Concordia, Grant, Nachitoches, Rapides,Sabine, Vernon, and Winn.Source: Career Compass of Louisiana.
21
(a) All Crimes (b) Felony Crimes
(c) Misdemeanor (d) Violent Crime
Figure 2: Event Study graphs for the impact of Career Compass of Louisiana.Sources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC).
22
(a) Male Students (b) Female Students
(c) Students on Free and Reduced Lunch (d) White Students
(e) Minority Students (f) High Performing Students
(g) Low Performing Students
Figure 3: Event Study graphs for the impact of Career Compass of Louisiana on all crimeby student characteristics.Sources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC). 23
Table 1: Summary Statistics
All Treatment Control Control(All Data) (PSM)
Age Exposed 17.58 17.57 17.58 17.69(0.756) (0.754) (0.756) (0.807)
Minority Race 0.389 0.462 0.358 0.479(0.488) (0.499) (0.48) (0.499)
Male 0.473 0.467 0.476 0.485(0.499) (0.499) (0.499) (0.499)
Free and Reduced Cost Lunch 0.403 0.438 0.388 0.462(0.491) (0.496) (0.487) (0.498)
High Performing Students 0.192 0.177 0.198 0.176(0.394) (0.381) (0.399) (0.381)
Middle Performing Students 0.323 0.324 0.323 0.305(0.468) (0.468) (0.468) (0.461)
Low Performing Students 0.194 0.198 0.192 0.210(0.395) (0.399) (0.394) (0.407)
Expenditure per Pupil (in Thousands) 10.81 10.44 10.96 10.51(2.459) (1.922) (2.636) (2.63)
Revenue per Pupil (in Thousands) 10.43 10.18 10.54 10.21(2.246) (1.675) (2.437) (2.430)
Full Time Faculty per 100 Students 9.65 6.721 10.87 6.611(14.26) (0.846) (16.81) (1.017)
Enrollment per Guidance Counselor 358.45 386.4 346.76 415.91(420.4) (481.5) (391.6) (537.66)
Unemployment rate 6.18 6.70 5.96 6.52(1.995) (1.85) (2.01) (2.37)
District Performance Score 105.8 100.6 107.9 100.6(10.44) (8.77) (10.33) (8.99)
School Performance Score 83.41 80.17 84.76 78.86(15.32) (17.05) (14.32) (13.35)
Note: Table 1 presents descriptive statistics (mean and standard errors) for key variables fortreatment and control groups. Standard deviations are in parentheses. The treatment groupconsists of 73,529 students and the control group consists of 175,672 students. Control (PSM)represents the control characteristics from group matched using a propensity score method.Sources: Administrative data from Louisiana Department of Education (DOE) and Depart-ment of Corrections (DOC).
24
Table 2: Main Results
(1) (2) (3)All Crime Felony Misdemeanor
CareerCompass -0.0455** -0.0113*** -0.0434**(0.0160) (0.0026) (0.0158)
Observations 249,201 249,201 249,201
Note: Table 2 presents difference-in-differences regression esti-mates for crime rates on all crimes, felony, and misdemeanor. Thecoefficient of interest is CareerCompass. Coefficients for school,year, year-district fixed effects, and individual characteristics arenot shown. Student characteristics are gender, age, ethnicity,and eligibility for free or reduced cost school meals and past stu-dent performance in grade 8. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Ed-ucation (DOE) and Department of Corrections (DOC).
25
Table 3: Heterogeneity in Main Results
(1) (2) (3)All Crime Felony Misdemeanor
Panel A: Impact on Male StudentsCareerCompass -0.0581** -0.0248* -0.0543**
(0.0208) (0.0105) (0.0195)Observations 117,922 117,922 117,922Panel B: Impact on Female StudentsCareerCompass -0.0184 0.0017 -0.0181
(0.0141) (0.0069) (0.0146)Observations 131,279 131,279 131,279Panel C: Impact on White StudentsCareerCompass -0.0134 -0.0050 -0.0096
(0.0296) (0.0035) (0.0278)Observations 152,240 152,240 152,240Panel D: Impact on Minority StudentsCareerCompass -0.0754*** -0.0237*** -0.0744***
(0.0209) (0.0023) (0.0212)Observations 90,494 90,494 90,494Panel E: Impact on Students on Free and Reduced Cost LunchCareerCompass -0.0889** -0.0020 -0.0883**
(0.0271) (0.0025) (0.0272)Observations 100,438 100,438 100,438Panel F: Impact on High Performing StudentsCareerCompass -0.0185 -0.0034 -0.0063
(0.0256) (0.0094) (0.0245)Observations 47,796 47,796 47,796Panel G: Impact on Middle Performing StudentsCareerCompass -0.0340 -0.0282** -0.0338
(0.0430) (0.0100) (0.0431)Observations 80,606 80,606 80,606Panel H: Impact on Low Performing StudentsCareerCompass -0.0702* -0.0151** -0.0723**
(0.0367) (0.0070) (0.0361)Observations 48,330 48,330 48,330
Note: Table 3 presents difference-in-differences regression estimates for crime rates onall crimes, felony, and misdemeanor for male students, female students, white students,minority students, students on free and reduced cost lunch, high, middle and low per-forming students. High performing students are those who have scored in the top 25thpercentile in their 8th Grade LEAP test and Low performing students are those who havescored in the bottom 25th percentiles. The coefficient of interest is CareerCompass.Coefficients for school, year, year-district fixed effects, and individual characteristics arenot shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and De-partment of Corrections (DOC).
26
Table 4: Propensity to Commit Violent Crime and Heterogeneity
Panel A: Propensity to commit Violent Crimes - all studentsCareerCompass -0.0056**
(0.0027)Observations 249,201Panel B: Impact on Male StudentsCareerCompass -0.0128**
(0.0055)Observations 117,922Panel C: Impact on Female StudentsCareerCompass 0.0013
(0.0006)Observations 131,279Panel D: Impact on White StudentsCareerCompass 0.0031
(0.0031)Observations 152,240Panel E: Impact on Minority StudentsCareerCompass -0.0220***
(0.0048)Observations 90,494Panel F: Impact on Students on Free and Reduced Cost LunchCareerCompass -0.0047
(0.0037)Observations 100,438Panel G: Impact on High Performing StudentsCareerCompass -0.0054
(0.0097)Observations 47,796Panel H: Impact on Middle Performing StudentsCareerCompass -0.0093**
(0.0032)Observations 80,606Panel I: Impact on Low Performing StudentsCareerCompass -0.0020*
(0.0010)Observations 48,330
Note: Table 4 presents difference-in-differences regression estimates for propensity to com-mit a violent crime (a subset of felony) for male students, female students, white students,minority students, students on free and reduced cost lunch, high, middle and low per-forming students. High performing students are those who have scored in the top 25thpercentile in their 8th Grade LEAP test and low performing students are those who havescored in the bottom 25th percentiles. The coefficient of interest is CareerCompass. Co-efficients for school, year, year-district fixed effects, and individual characteristics are notshown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and De-partment of Corrections (DOC).
27
Table 5: Recidivism
(1) (2) (3) (4)All Crime Violent Felony Misdemeanor
CareerCompass -0.0115*** -0.0045 -0.0093** -0.0078***(0.0025) (0.0027) (0.0036) (0.0019)
Observations 1,075 1,075 1,075 1,075
Note: Table 5 presents difference-in-differences regression estimates for re-cidivism rates on all crimes, violent crimes, felony, and misdemeanor. Thecoefficient of interest is CareerCompass. Coefficients for school, year, year-district fixed effects, and individual characteristics are not shown. ***p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education(DOE) and Department of Corrections (DOC).
28
Table 6: Placebo Tests
(1) (2) (3)All Crime Felony Misdemeanor
CareerCompass -0.0457** -0.0113*** -0.0436**(0.0158) (0.0025) (0.0157)
CareerCompass(t−1) 0.0075 0.0037 0.0002(0.0433) (0.0054) (0.0431)
CareerCompass(t−2) -0.0514 0.0084 -0.0583(0.0450) (0.0089) (0.0450)
Observations 249,201 249,201 249,201
Note: Table 6 presents estimates for placebo tests of crime rateson all crimes, felony, and misdemeanor. The coefficient of interestis CareerCompass, CareerCompass(t−1), and CareerCompass(t−2).Coefficients for school, year, year-district fixed effects, and individualcharacteristics are not shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Educa-tion (DOE) and Department of Corrections (DOC).
29
Table 7: The Effect of Carrer Compass on Characteristics of Students, Parishes and School:Selection investigation
(1) (2) (3) (4) (5)
Panel A: On Student level CharacteristicsMinority White FRL High Low
Performing PerformingCareerCompass 0.0395 -0.0160 -0.0476 0.0125 -0.0307
(0.0244) (0.0128) (0.0352) (0.0118) (0.0195)Observations 249,201 249,201 249,201 249,201 249,201
Panel B: On Parish level CharacteristicsExpen. Log Revenue District Average
Per Pupil) Unemp. Per Pupil Performance Earning(in ’000) Rate (in ’000) Score (in ’000)
CareerCompass 0.6817 0.0567 0.7131 -4.9004 1.4194(0.7255) (0.0547) (0.6819) (3.5209) (1.0957)
Observations 488 488 488 488 488
Panel C: On School level CharacteristicsFTE Enroll. per School Enroll. FRL
per 100 Guidance Performance Count ShareEnrolled Counselor Score
CareerCompass -1.3065 -22.57 -0.0511 40.50 0.0026(1.5316) (54.99) (0.9471) (21.15) (0.0092)
Observations 1,878 1,878 1,878 1,878 1,878
Note: Table 7 presents difference-in-differences regression estimates on Student (Individual),Parish and School level Characteristics. Student level characteristics shows the selection byMinority student, white students, students on free and reduced cost lunch, high performingstudents and low performing students. Parish level characteristics include selection of expen-diture per Pupil (in ’000), log unemployment rate, revenue per Pupil (in ’000), district Per-formance Score, and average earning (in ‘000). School level characterisitcs include selection ofFTE per 100 enrolled students, enrollment per guidance counselor, school performance score,enrollment count, and share of students on free and reduced cost lunch. The coefficient of in-terest is CareerCompass. Standard errors are shown in parentheses. Coefficients for year fixedeffects are not shown in panel A. Coefficients for school, and year fixed effects are not shown.*** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and Depart-ment of Corrections (DOC).
30
Table 8: Propensity to Commit In-School Misbehavior and Heterogeneity
Panel A: Propensity to get disciplined - all studentsCareerCompass -0.0440**
(0.0143)Observations 249,201Panel B: Impact on Male StudentsCareerCompass -0.0505**
(0.0175)Observations 117,922Panel C: Impact on Female StudentsCareerCompass -0.0230
(0.0149)Observations 131,279Panel D: Impact on White StudentsCareerCompass -0.0050
(0.0252)Observations 152,240Panel E: Impact on Minority StudentsCareerCompass -0.0825***
(0.0205)Observations 90,494Panel F: Impact on Students on Free and Reduced Cost LunchCareerCompass -0.0886***
(0.0284)Observations 100,438Panel G: Impact on High Performing StudentsCareerCompass -0.0016
(0.0246)Observations 47,796Panel H: Impact on Middle Performing StudentsCareerCompass -0.0269
(0.0387)Observations 80,606Panel I: Impact on Low Performing StudentsCareerCompass -0.0728**
(0.0358)Observations 48,330
Note: Table 8 presents difference-in-differences regression estimates for propensity to getdisciplined for all students, male students, female students, white students, minority stu-dents, students on free and reduced cost lunch, high, middle and low performing students.High performing students are those who have scored in the top 25th percentile in their 8thGrade LEAP test and Low performing students are those who have scored in the bottom25th percentiles. The coefficient of interest is CareerCompass. Coefficients for school,year, year-district fixed effects, and individual characteristics are not shown. *** p<0.01,** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and De-partment of Corrections (DOC).
31
Table 9: Results for Post Secondary Enrollment - school level regressions
(1) (2) (3) (4) (5)All Schools Low Income High Fraction minority High Income Majority white
CareerCompass 0.0628** 0.0706*** 0.1087*** 0.0578*** 0.0332(0.0028) (0.0163) (0.0087) (0.0059) (0.342)
Observations 1,904 978 1,080 926 823
Note: Table 9 presents difference-in-differences regression estimates for post-secondary enrollment on allschools, low income schools, and school with a High Fraction of students from minority groups. Low in-come schools are those with above median students on free and reduced lunch while High Fraction minorityare those with above median share of students as minority. The coefficient of interest is CareerCompass.Coefficients for school, year, year-district fixed effects, and individual school characteristics are not shown.*** p<0.01, ** p<0.05, * p<0.1.Sources: Enrollment data from Louisiana Department of Education (DOE).
32
Appendix
Table A.1: Results by District Performance Score
(1) (2) (3)All Crime Felony Misdemeanor
Panel A: Impact on High Performance DistrictsCareerCompass -0.0456 0.0099 -0.0462
(0.0613) (0.0064) (0.0613)
Observations 143,521 143,521 143,521Panel B: Impact on Low Performance DistrictsCareerCompass -0.0439** -0.0111*** -0.0418**
(0.0157) (0.0026) (0.0155)
Observations 105,680 105,680 105,680
Note: Table A.1 presents difference-in-differences regression estimates for crime rates on all crimes,felony, and misdemeanor for high performance districts and low performance districts. Districtperformance score is a comprehensive measure of assessing how the school district as a unit hasperformed. It includes grades 3-8 assessment index, dropout credit accumulation index, end-of-course exams assessment index, ACT assessment index, strength of diploma (graduation index),cohort graduation rate index, cohort graduation rate, and progress points. High performance dis-tricts are school districts with above median district performance score for 2012. The coefficient ofinterest is CareerCompass. Coefficients for school, year, year-district fixed effects, and individualcharacteristics are not shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and Departmentof Corrections (DOC). District Performance Score for 2012 was retrieved from Louisiana Depart-ment of Education.
33
Table A.2: Results by Overall Crime Rates
(1) (2) (3)All Crime Felony Misdemeanor
Panel A: Impact on Low Crime DistrictsCareerCompass -0.0287 0.0011 -0.0286
(0.0214) (0.0017) (0.0215)
Observations 118,364 118,364 118,364Panel B: Impact on High Crime DistrictsCareerCompass -0.0442** -0.0112*** -0.0421**
(0.0157) (0.0025) (0.0155)
Observations 121,019 121,019 121,019
Note: Table A.2 presents difference-in-differences regression estimates for crime rates onall crimes, felony, and misdemeanor for high crime districts and low crime districts. Lowcrime districts are school districts with below median for overall crime rates. The coeffi-cient of interest is CareerCompass. Coefficients for school, year, year-district fixed effects,and individual characteristics are not shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and Depart-ment of Corrections (DOC).
34
Table A.3: Robustness checks of Main Results
(1) (2) (3)All Crime Felony Misdemeanor
Panel A: Main Results using Logit RegressionCareerCompass -0.0358*** -0.0166*** -0.0337***
(0.0116) (0.0030) (0.0113)
Observations 249,201 249,201 249,201Panel B: Main Results using Propensity Score MatchingCareerCompass -0.0316*** -0.0083*** -0.0325***
(0.0026) (0.0025) (0.0026)
Observations 147,058 147,058 147,058Panel C: Main Results without the year 2012CareerCompass -0.0429** -0.0104*** -0.0418**
(0.0180) (0.0022) (0.0179)Observations 216,046 216,046 216,046Panel D: Main Results without controlsCareerCompass -0.0426** -0.0109*** -0.0406**
(0.0144) (0.0023) (0.0142)Observations 249,201 249,201 249,201Panel E: Main Results clustered at Parish levelCareerCompass -0.0455*** -0.0113*** -0.0434***
(0.0040) (0.0010) (0.0044)Observations 249,201 249,201 249,201
Note: Table A.3 presents difference-in-differences regression estimates forcrime rates on all crimes, felony, and misdemeanor. The coefficient of inter-est is CareerCompass. Coefficients for school, year, year-district fixed effects,and individual characteristics are not shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education(DOE) and Department of Corrections (DOC).
35
Table A.4: Estimates Including Schools in East Baton Rouge and Orleans Parishes
(1) (2) (3) (4)All parishes Excluding Orleans Excluding EBR Excluding Both
(Main results)
CareerCompass -0.0452** -0.0443** -0.0430** -0.0409**(0.0166) (0.0160) (0.0150) (0.0135)
Observations 266,438 263,682 251,957 249,201
Note: Table A.4 presents difference-in-differences regression estimates for crime rates on all crimesfor all parishes, all parishes excluding Orleans (New Orleans), and all parishes excluding EastBaton Rouge (EBR). Column (4) displays our main results for comparison. The coefficient of in-terest is CareerCompass. Coefficients for school, year, year-district fixed effects, and individualcharacteristics are not shown. *** p<0.01, ** p<0.05, * p<0.1.Sources: Administrative data from Louisiana Department of Education (DOE) and Departmentof Corrections (DOC).
36
(a) Minority Students (b) White Students
(c) Students on Free and Reduced Lunch (d) High Performing Students
(e) Low Performing Students
Figure A.1: Event Study graphs for characteristics of StudentsSources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC).
37
(a) FTE per 100 Enrolled Students (b) Enrollment Per Guidance Counselor
(c) School Performance Score (d) Enrollment Count
(e) Share on Free and Reduced Lunch
Figure A.2: Event Study graphs for School Level CharacteristicsSources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC).
38
(a) Expenditure per Pupil (in ‘000) (b) Unemployment Rate
(c) Revenue per Pupil (in ‘000) (d) District Performance Score
(e) Average Pay
Figure A.3: Event Study graphs for Parish Level Characteristics.Sources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC).
39
(a) Coefficients for All Crime (b) Coefficients for Felony Crime
(c) Coefficients for Minor Crime
Figure A.4: Permutation Test plotsInterventions are randomized with 1000 replications for our main outcome variables: all crimes,felony and misdemeanor. Vertical line represents our baseline estimate.Sources: Administrative data from Louisiana Department of Education (DOE) and Department ofCorrections (DOC).
40