Content uploaded by David P Kennedy
Author content
All content in this area was uploaded by David P Kennedy on Dec 18, 2014
Content may be subject to copyright.
DOI: 10.1542/peds.2013-3573
; originally published online July 21, 2014;Pediatrics
Tseng, Peter Bergman and Paul J. Chung
Richard Buddin, Martin F. Shapiro, Sheryl H. Kataoka, Arleen F. Brown, Chi-Hong
Mitchell D. Wong, Karen M. Coller, Rebecca N. Dudovitz, David P. Kennedy,
Successful Schools and Risky Behaviors Among Low-Income Adolescents
http://pediatrics.aappublications.org/content/early/2014/07/16/peds.2013-3573
located on the World Wide Web at:
The online version of this article, along with updated information and services, is
of Pediatrics. All rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.
Boulevard, Elk Grove Village, Illinois, 60007. Copyright © 2014 by the American Academy
published, and trademarked by the American Academy of Pediatrics, 141 Northwest Point
publication, it has been published continuously since 1948. PEDIATRICS is owned,
PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
Successful Schools and Risky Behaviors Among
Low-Income Adolescents
WHAT’S KNOWN ON THIS SUBJECT: Graduating from high school
is associated with better health and health behaviors. However,
no rigorous studies have tested whether exposure to a high-
performing school improves health or health behaviors, thus the
causal relationship is unknown.
WHAT THIS STUDY ADDS: Exposure to successful schools can
reduce very risky health behaviors among low-income
adolescents. The primary mechanism is mostly due to better
school retention and also due to better academic achievement.
abstract
OBJECTIVES: We examined whether exposure to high-performing
schools reduces the rates of risky health behaviors among low-income
minority adolescents and whether this is due to better academic
performance, peer influence, or other factors.
METHODS: By using a natural experimental study design, we used the
random admissions lottery into high-performing public charter high
schools in low-income Los Angeles neighborhoods to determine
whether exposure to successful school environments leads to fewer
risky (eg, alcohol, tobacco, drug use, unprotected sex) and very
risky health behaviors (eg, binge drinking, substance use at school,
risky sex, gang participation). We surveyed 521 ninth- through twelfth-
grade students who were offered admission through a random lottery
(intervention group) and 409 students who were not offered admission
(control group) about their health behaviors and obtained their state-
standardized test scores.
RESULTS: The intervention and control groups had similar demo-
graphic characteristics and eighth-grade test scores. Being offered
admission to a high-performing school (intervention effect) led
to improved math (P , .001) and English (P = .04) standard test
scores, greater school retention (91% vs 76%; P , .001), and lower
rates of engaging in $1 very risky behaviors (odds ratio = 0.73, P ,
.05) but no difference in risky behaviors, such as any recent use of
alcohol, tobacco, or drugs. School retention and test scores explained
58.0% and 16.2% of the intervention effect on engagement in very risky
behaviors, respectively.
CONCLUSIONS: Increasing performance of public schools in low-
income communities may be a powerful mechanism to decrease
very risky health behaviors among low-income adolescents and to
decrease health disparities across the life span. P ediatrics 2014;134:
e389–e396
AUTHORS: Mitchell D. Wong, MD, PhD,
a
Karen M. Coller,
PhD,
a
Rebecca N. Dudovitz, MD, MSHS,
a
David P. Kennedy,
PhD,
b
Richard Buddin, PhD,
c
Martin F. Shapiro, MD, PhD,
a
Sheryl H. Kataoka, MD, MSHS,
a
Arleen F. Brown, MD, PhD,
a
Chi-Hong Tseng, PhD,
a
Peter Bergman, PhD,
d
and Paul J.
Chung, MD, MS
a
a
University of California Los Angeles, David Geffen School of
Medicine, Los Angeles, California;
b
RAND Corporation, Santa
Monica, California;
c
Act, Inc, Iowa City, Iowa; and
d
Teachers
College, Columbia University, New York, New York
KEY WORDS
disparities, education, risk-taking behavior
ABBREVIATIONS
API—Academic Performance Index
CST—California Standards Tests
OR—odds ratio
Dr Wong conceptualized and designed the study, conducted all
analyses, and drafted the initial manuscript; Dr Coller assisted
with the study design, data collection, and data analysis and
reviewed and revised the manuscript; Dr Dudovitz assisted with
the conceptual framework for the study, survey content and
design, and data analysis and reviewed and revised the
manuscript; Dr Kennedy assisted with the study design,
particularly in social network data collection and analysis, and
reviewed and revised the manuscript; Dr Buddin assisted with
the study design in sampling and collection and analysis of
educational outcomes data and reviewed and revised the
manuscript; Drs Shapiro, Kataoka, and Brown assisted with the
study design and reviewed and revised the manuscript;
Drs Tseng and Bergman assisted with data analysis, statistical
methods, and interpretation and preparation of the data tables
of the manuscript and reviewed and revised the manuscript;
Dr Chung assisted with the study design, data collection, and
data analysis and reviewed and provided critical revisions of the
manuscript; and all authors approved the final manuscript as
submitted and agree to be accountable for all aspects of the
work.
www.pediatrics.org/cgi/doi/10.1542/peds.2013-3573
doi:10.1542/peds.2013-3573
Accepted for publication Apr 21, 2014
Address correspondence to Mitchell D. Wong, MD, PhD, UCLA
Division of General Internal Medicine and Health Services
Research, 911 Broxton Ave, Los Angeles, CA 90024. E-mail:
mitchellwong@mednet.ucla.edu
(Continued on last page)
PEDIATRICS Volume 134, Number 2, August 2014 e389
ARTICLE
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
Many aspects of the school environment
may influence adolescent behaviors and
health outcomes.
1,2
For example, school
climate factors such as greater school
connectedness and engagement, posi-
tive teacher relationships, and respect
in school are associated with better
academic outcomes, fewer behavioral
problems, and better mental health.
3–7
Other school-level factors, such as health
and disciplinary policies, are associated
with better health
3,8
but are less well
studied. Although these and other
studies have bee n very helpful in
unders tandin g h ow sc hool envi r on-
ments might be modified to improve
health, almost all have relied on obser-
vational study designs. Thus, selection
bias and other confounders, such as pa-
rental influence and neighborhood pov-
erty, severely limit the ability of these
studies to determine the causal relation-
ship between school envir onment and
adolescent outcomes. A few randomized
trials or natural experiments have shown
that lower barriers to education and
some early childhood programs have
economic and health benefits.
9–11
Despite
recent education reform efforts in the
United States, whether better health
can be achieved through public educa-
tion improvements is unproven.
The first charter school in the United
States opened in 1992, and since then
the charter school movement has spread
r apidly.
12
These schools receive public
funding and are designed to have more
autonomy over curriculum and oper-
ations to foster innovation and better
academic outcomes. Given this auton-
omy, their educational models can vary
substantially, but admission is free and
open to all. Although charters are not
a panacea for improving education,
12
the random admissions lottery of
successful char ter schools provides
an opportunity to observe a natural
experiment. The Reducing Health In-
equities Through Social and Educational
Change (RISE) Study surveyed appli-
cants to 3 high-performing public
charter high schools in l ow-income
neighborhoods of Los Angeles to de-
termine if exposure to these schools
redu ces risky behaviors.
METHODS
Study Design and Sampling
In 2010 we identified all 20 public charter
high schools run by charter management
organizations in low-income neighbor-
hoods of Los Angeles and open since
2007 or earlier. Among 7 of these with
a California 2009 Growth Academic Per-
formance Index ( A P I )
13,14
in the top tertile
of public high schools, 4 had enough
applicants to hold an admissions lottery ,
and 3 of these agr eed to participate in our
study .
Also in 2010 we sought to obtain a study
sample of current ninth- through twelfth-
gr ade students wh o had applied to a
participating charter school for the ninth
gr ade and enter ed the random admis-
sions lottery. We randomly selected po-
tential subjects from each school’sninth
gr ade applicant list from 2007 to 2010,
with the goal of r ecruiting into our study
equal numbers who were offered ad-
mission (intervention group) and who
were not offered admission (control
group). To minimize contamination of
our sample to other high-performing
schools, we excluded those who went
to another charter or private school for
ninth gr ade, but not those who tr ans-
ferred to another charter or private
school after ninth grade. We excluded
subjects who could not be contacted or
had moved out of the area and siblings
receiving admission outside the lottery.
After obtaining written informed pa-
rental consent, research assistants
conducted a 90-minute face-to-face
computer-assisted interview with each
subject. An audio-enhanced, computer-
assisted self-interview was used to
collect information about substance
use and sexual behaviors. With parental
consent, we also obtained student-level
math and English California Standards
Tests (CST) scores from the California
Department of Education for study par-
ticipants since eighth grade. The human
subjects research review board approved
all research activities.
Measures
Fr om surveys, we collected demogra ph-
ics, and self-r eported school information
(school tr ansfer and dropout), substance
use, sexual beha viors, and exposure to
violence.
15,16
We assessed depression,
17
school engagement,
18
parenting style,
19
and hopelessness.
20
To assess subjects’
personal social network, we asked them
to name 20 persons whom they know and
are most important to them. For each
person in their network (alter), we asked
subjects (ego) about their relationship
(eg, relative, friend), and the alter’schar-
acteristics (relative age, gender) and
behaviors (eg, smoking, alcohol, drug use,
sexual behaviors). We ca lculated the
propor tions of same-aged peers in a
subject’s network who used substances
or had sexual intercourse. CST scores
were categorized into below basic
(,300), basic (300–349), and proficient
and above ($350).
21
Our primary outcome measures were
risky behaviors, defined as any use of to-
bacco, alcohol, marijuana, and other drugs
in the past 30 days. W e also examined less
common but very risky behaviors, which
included binge drinking ($5drinkson1
occasion), alcohol use at school, any drug
use (excluding marijuana), carrying a
weapon to s chool in the past 30 da ys, gang
membership in the past 12 months, cur-
rent pregnancy , multiple sexual partners,
sex without condoms, sex without con-
traception, and alcohol or drug use with
sex in the past 90 days. We compared
those who engaged in none versus any of
these very risky behaviors.
Analytic Methods
We conducted an intent-to-treat analy-
sis comparing those offered admission
e390 WONG et al
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
to a study char ter school (intervention
group), regardless of whether they
accepted the offer, with those not of-
fered admission (control group). We
used generalized estimating equations
to account for nonindependence of
students clustered within schools,
adjusting for grade, gender, race/
ethnicity, language, parental educa-
tion, parental employment, parental
birthplace, and parenting style.
22
The
main analyses were limited to 10th
through 12th graders because we hy-
pothesized that the “intervention ef-
fect” would be absent among ninth
graders who had limited exposure to
the high-performing schools. We per-
formed a sensitivity analysis examining
ninth graders only , but the small sample
permitted controlling for gender only . We
formally tested for mediation effect by
using a method by Breen et al.
23
This test
does not take into account hierarchical
models, so we performed this mediation
test by using logistic regre ssion. We im-
puted missing data by using multiple
imputation based on methods devel-
oped by Rubin
24
and Schafer.
25
Stata
version 11 was used for all analyses
(StataCorp, College Station, TX).
26
RESULTS
Among all ninth-grade applicants to
the 3 study charter schools between
2007 and 2010, we randomly selected
1432 applicants who were not offered
admission (control group) and 952
who were offered admission (inter-
vention group) (Fig 1). We tried to lo-
cate applicants to screen them for
study eligibility and recruitment but
could not find 253 (17.7%) of the po-
tential control sample and 157 (16.5%)
of the potential intervention sample
using contact information from their
admissions application. Of those remain-
ing, 574 (48.7%) of the control and 162
(20.4%) of the intervention group were
ineligible, mainly because they attended
another charter school for ninth grade.
Of remaining e ligible subjects, 196
(32.4%) in the control group and 112
(17.7%) in the intervention group refused
participation (P , .001). The final sample
included 409 and 521 subjects in the
control and intervention groups, re-
spectively. Our original target sample was
950 subjects with 50% in each study arm.
Because of the higher ineligibility and
refusal r ate in the control arm, we were
unable to obtain equal numbers of sub-
jects, and we had to select from a much
larger sample of school applicants to
obtain the final control group.
Student demographic characteristics,
eighth grade math and English CST
scores, and parentaleducation, nativity,
and employment were not statistically
different between the 2 groups (Ta-
ble 1). Some differences were noted
that might indicate lower socioeco-
nomic status and potentially higher
risk for adverse education and health
outcomes among the control group.
Specifically, the control group was less
likely to own a home as reported by the
student (40.6% vs 45.0%; P = .02) and
parents were slightly less likely to have
graduated from high school (48.2%
vs 52.6%; P = .15). However, the in-
tervention group was less likely to in-
clude native English speakers (42.2%
vs 35.8%; P = .07), which might increase
their risk for worse academic out-
comes. Whether a student had come
from a high- or low-performing middle
school might influence the impact of
being exposed to a high-performing
high school. We found the Growth API
of the students’ previous middle schools
were similar between the 2 study arms,
and the baseline characteristics of the
2 study arms were similar after strati-
fying on the performance of their mid-
dle school.
Table 2 shows the characteristics of
the subjects’ school and school perfor-
mance. There was some “contamination”
between the control and intervention
groups as measured by the Growth API of
the actual school attended. Nine percent
of the intervention group ended up in
FIGURE 1
Subject recruitment, eligibility, and participation.
ARTICLE
PEDIATRICS Volume 134, Number 2, August 2014 e391
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
a school that performed worse than the
study charter schools compared with
82.9% of the control group, and 17.1% of
the control group ended up in a school
that was equal to or better than the
study charter schools compared with
90.8% of the intervention group. The
intervention group was less likely to have
transferr ed high schools or dropped out
(8.8%vs24.2%;P , .001) and to have
“cut” school in the past 12 months (P ,
.001). Also, the intervention group was
more likely to have attended a higher
performing high school (average Gr owth
API of 748 [68th percentile] vs 673 [45th
percentile]; P , .001). Student-level CS T
scores fr om eighth grade or the first
available score were similar in the 2
study arms (Table 1), but according to
the most recent CST scores, the In-
tervention group was more likely t o be
proficient or above in English (P =.04)
andmath(P , .001).
Our main outcomes were engagement in
risky and very risky behaviors. Because
exposure to a high-performing high
school would not likely have an imme-
diate impact in ninth grade, we limited
our analysis to 10th through 12th
graders. Although we observed no sta-
tistically significant differences between
the control and intervention groups in
any of the risky behaviors (tobacco, al-
cohol, and marijuana use in the past 30
days), the effect of the intervention on
cigarette smokingand alcohol wasin the
direction that we hypothesized and the
confidence intervals were relatively
wide (Table 3). We also examined 11 very
risky behaviors: the most common were
sex without contr aception (13.4%), in-
consistent condom use (11.6%), binge
drinking (8.3%), alcohol or drug use with
sex (7.1%), marijuana use at school
(6.4%), and drug use other than mari-
juana (6.3%). Prevalence was ,5% for
each remaining very risky behaviors (al-
cohol use at school, current pregnancy,
multiple sex partners, carrying a weapon
to school, and gang membership). The
proportion engaging in $1ofthese11
very risky behaviors was 41.9% in the
control group and 36.3% in the in-
tervention group, with an adjusted odds
ratio (OR) of 0.73 (P = .048). The in-
tervention effect on engaging in a very
riskybehaviorwasmorepronounced
among those who had attended a low-
performing middle school (OR = 0.67,
P = .032) than among those who had
attended a high-performing middle
school (OR = 0.85, P = .40). However, the
interaction effect between being offered
charter school admission and middle
school performance was not statistically
significant (P =.77).
We also conducted a sensitivity analysis
examining ninth graders only and, as
expected, found no statistically signifi-
cant differences in tobacco, alcohol or
marijuana use, or engagement in $1
very risky behaviors. Compared with
those in the intervention group, ninth
graders in the control group were
somewhat more likely to smoke ciga-
rettes (9.3% vs 5.3%; P = .22) and slightly
less likely to drink alcohol (18.3% vs
21.3%; P = .56) and use marijuana
(10.0% vs 11.5%; P =.72)inthepast30
days. Engagement in $1veryrisky
behaviors was almost identical between
groups (16.7 vs 16.5%; P =.98).
We examined whether various potential
mediators were associated with engag-
ing in $1 very risky beha viors, and
whether controlling for the potential
mediator diminished the intervention
effect. Very risky behaviors were more
likely if the student was not retained
in the same high school (ie, changed
schools or dropped out), was less en-
gaged in school, had below basic English
TABLE 1 Baseline Characteristics of the Control and Intervention Groups
Control Group Intervention Group P
N 409 521
Current grade, % .35
9th 26.4 25.5
10th 25.2 24.0
11th 26.9 25.0
12th 21.5 25.5
Race/ethnicity, % .60
Non-Latino white 0.5 0.4
Non-Latino African American 13.9 11.5
Latino 81.7 85.8
Other/mixed race 3.9 2.3
Male, % 47.2 42.8 .17
English as the native language, % 42.2 35.8 .07
Parent is a high school graduate, % 48.2 52.6 .15
Parent is US born, % 25.5 28.3 .35
Parent working full time, % 91.2 91.9 .65
Family owns home, % 40.6 45.0 .02
Parenting style, % .64
Average 60.9 64.6
Neglectful 15.7 15.8
Indulgent 6.1 5.8
Authoritarian 6.1 5.6
Authoritative 11.1 8.3
Middle school growth API, mean
a
681.0 679.5 .14
Eighth-grade CST math score, % .59
Below basic 48.8 48.6
Basic 28.8 31.6
Proficient or above 22.3 19.8
Eighth-grade CST English score, % .76
Below basic 25.1 23.0
Basic 34.2 34.1
Proficient or above 40.7 42.9
a
The API is based on the CST (score range: 200–1000).
e392 WONG et al
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
and math CST scores on their most re-
cent test, had a lower grade point av -
erage, had a gr eater proportion of peers
who use substance s or have had sex,
was more depressed, or felt more
hopeless about the futu r e (Table 4).
Havingateachernamedinthestudent’s
social network was the only potential
mediator not associated with engaging
in very risky behaviors. Only school re-
tention and math and English CST scores
were significant mediators (P , .001
and P = .06) and reduced the in-
tervention effect on engagement in very
riskybehaviorsafterbeingindividually
added into the regression model. These
2 mediators explained 58% and 16.2% of
the intervention effect, respectively. This
analysis suggests that exposure to a
high-performing school improves r e-
tention and standar dized test scores,
which then reduces very risky behaviors.
An alternative mechanism may be that
high-performing schools reduce very
risky behaviors, which then improve
test scores and school retention, but
this was not supported by additional
analyses. We conducted a mediation
analysis with retention as the outcome
and very risky behaviors as the medi-
ator and found that the intervention
effect without any mediators was
strong (OR = 5.2, P , .001), but very
risky behaviors mediated only 4% of
the intervention effect on retention
(P = .08). We also performed a similar
analysis with test scores as the out-
come. The intervention effect was as-
sociated with a 27-point increase on
standardized test scores (P , .001),
but engagement in very risky behaviors
TABLE 2 School Characteristics and Performance Measures by Intervention and Control Group
Control Group Intervention Group P
Current school, % ,.001
Charter public school 1.2 87.5
Other public school 85.6 10.8
Alternative school
a
8.8 1.7
Dropped out 4.4 0.0
Changed high schools or dropped out, % 24.2 8.8 ,.001
Cut school in the last 12 mo, % ,.001
Never 59.7 82.8
1 or 2 times 25.6 11.4
3 or 6 times 8.2 4.4
$7 times 6.5 1.4
Suspended from school in last 12 mo, % 14.3 17.5 .18
Expect to graduate, % 94.1 99.2 ,.001
Expect to go to college, % 95.6 96.0 .95
Mean engagement in school score 3.2 3.2 .65
Mean API score of current school
b
673.5 748.5 ,.001
API score of current school, % ,.001
Worse than study charter schools 82.9 9.2
Same as study charter schools 2.7 87.5
Better than study charter schools 14.4 3.3
Most recent CST math score, % ,.001
Below basic 67.9 55.3
Basic 23.3 27.1
Proficient or above 8.8 17.6
Most recent CST English score, % .04
Below basic 25.4 17.9
Basic 34.2 37.0
Proficient or above 40.4 45.1
a
Alternative school includes continuation school, adult school, home school, and independent study.
b
The API is based on the CST (score range: 200–1000).
TABLE 3 Proportions and Adjusted ORs of Engaging in Risky Health Behaviors Comparing Control With Intervention Groups
Smoked Cigarettes in
Last 30 Days
Drank Alcohol in
Last 30 Days
Used Marijuana in
Last 30 Days
Engaged in $1 Very Risky
Behaviors in Last 30 Days
a
All 10th–12th graders
Control group, % 11.4 39.2 23.8 41.9
Intervention group, % 9.4 37.6 24.7 36.3
Intervention versus control group, adjusted OR
b
(95% CI) 0.81 (0.49–1.33) 0.92 (0.67–1.26) 1.07 (0.74–1.52) 0.77 (0.56–1.05)
Intervention versus control group, adjusted OR
c
(95% CI) 0.79 (0.44–1.42) 0.84 (0.58–1.21) 1.07 (0.68–1.67) 0.73 (0.53–1.00)
10th–12th graders who went to a low-performing middle school
Control group, % 11.5 43.4 27.2 43.5
Intervention group, % 8.9 38.7 23.5 38.2
Intervention versus control group, adjusted OR
b
(95% CI) 0.72 (0.36–1.45) 0.82 (0.54–1.25) 0.86 (0.53–1.4) 0.79 (0.52 - 1.21)
Intervention versus control group, adjusted OR
c
(95% CI) 0.59 (0.33–1.06) 0.73 (0.46–1.14) 0.84 (0.54–1.31) 0.67 (0.46 - 0.97)
10th–12th graders who went to a high-performing middle school
Control group, % 11.1 33.5 19.8 40.5
Intervention group, % 10.0 37.2 27.4 35.8
Intervention versus control group, adjusted OR
b
(95% CI) 0.90 (0.42–1.96) 1.17 (0.71–1.93) 1.52 (0.86–2.69) 0.82 (0.5–1.34)
Intervention versus control group, adjusted OR
c
(95% CI) 1.14 (0.50–2.62) 1.01 (0.63–1.64) 1.79 (0.88–3.62) 0.85 (0.58–1.24)
Low- and high-performing middle schools were categorized on the basis of school-level API scores below and above the 50th percentile of all public middle schools in California between 2007
and 2010. CI, confidence interval.
a
Very risky behaviors included binge drinking, alcohol use at school, any drug use (excluding marijuana), carrying a weapon to school, membership in a gang in the last 30 days, currently
pregnant, multiple sexual partners, sex without condoms, sex without contraception, alcohol or drug use with sex in the last 90 days.
b
Adjusted for grade and gender only.
c
Adjusted for grade, gender, race/ethnicity, language, parental education, parental employment, parental birthplace, and parenting style.
ARTICLE
PEDIATRICS Volume 134, Number 2, August 2014 e393
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
mediated only 7% of the intervention
effect on test scores (P = .08).
DISCUSSION
A few experimental or quasi-experimental
studies have found that early child-
hood schooling and more primary
schooling have important economic and
health benefits.
9,10,27
However, whether
improving the school environment
causally results in fewer risky behav-
iors has not been fully explored with the
use of more rigorous study designs. We
used a natural experiment design and
found that, among minority adolescents
from low-income families, exposure to
high-performing public charter high
schools (intervention effect) led to bet-
ter standardized test scores, fewer
dropouts and school transfers, and
lower rates of engagement in very risky
behaviors, such as drinking at school,
gang participation, or use of alcohol or
drugs with sex. However, we found no
statistically significant differences in
less risky behaviors, such as any recent
use of tobacco, alcohol, or marijuana.
The use of tobacco and alcohol was
slightly lower in the intervention group,
but this finding did not reach statistical
significance perhaps due to lack of
statistical power.
A number of potential mechanisms
might explain why successful school
environments are associated with
fewer risky health behaviors:
1. Better cognitive skills may lead to
better health outcomes,
28,29
per-
haps through better health knowl-
edge, medical decision making, or
health literacy.
2. Factors that improve educational
achievement may also reduce
risky behaviors. For example, at-
tending a high-performing school
may lower exposure to “risky”
peers. Successful school environ-
ments may also improve persistence,
resiliency, and other noncognitive
skills, which may lead to better out-
comes.
29–32
3. Better academic achievement may
lead to a better future outlook and
less risk taking.
33
4. Being in school or doing more
homework may leave less time
and opportunity to engage in risky
behaviors.
We examined a number of factors re-
lated to these potential mechanisms
and tested them for mediation effects.
Although many were associated with
very risky behaviors, only math and
English CST scores and retention in the
same school were significant mediators.
This finding has important implications
for future interventions. For example,
school policy changes that reduce
suspensionsandexpulsions mightlower
rates of adolescent risky behaviors.
Several study limitations are worth
noting. Because the study was not a true
TABLE 4 Effect of Potential Mediators on the Intervention Effect on Engaging in $1 Very Risky Behaviors
Mediator (reference group) Engaging in $1 Very Risky Behaviors, OR (95% CI) Intervention Effect Explained
by Mediator, %
Mediator Effect Intervention Effect
Without Mediator
Intervention Effect
With Mediator
School retention 0.39 (0.25–0.62) 0.74 (0.53–1.03) 0.88 (0.63–1.24) 59.3
Standardized test score
a
0.75 (0.60–0.93) 0.73 (0.52–1.01) 0.78 (0.56–1.08) 21.6
Self-reported grade point average (3.6–4.0) 0.71 (0.51–1.00) 0.71 (0.51–1.00) 21.8
3.1–3.5 1.69 (0.97–2.94) 24.6
2.6–3.0 2.53 (1.45–4.41) 20.1
2.0–2.5 3.61 (2.08–6.26) 23.3
,2.0 7.37 (3.46–15.7) 210.7
None/refused 4.96 (1.90–12.9) 16.9
School engagement
a
0.71 (0.60–0.84) 0.72 (0.52–1.01) 0.71 (0.51–0.99) 25.3
Time spent on homework versus with friends (lowest tertile) 0.71 (0.51–0.99) 0.70 (0.51–0.98) 23.7
Middle tertile 0.61 (0.42–0.89) 24.0
Highest tertile 0.40 (0.26–0.62) 0.3
Proportion
b
of peers who used alcohol in last month 1.38 (1.28–1.48) 0.69 (0.48–0.98) 0.65 (0.45–0.92) 216.5
Proportion
b
of peers who used drugs in last month 1.47 (1.35–1.61) 0.7 (0.49–1) 0.6 (0.42–0.86) 242.7
Proportion
b
of peers who ever had sex 1.39 (1.29–1.49) 0.7 (0.49–0.99) 0.72 (0.5–1.02) 7.5
Has $1 teachers in their social network 0.99 (0.71–1.39) 0.72 (0.52–1) 0.72 (0.52–1) 20.2
Depression (none) 0.72 (0.52–1) 0.69 (0.50–0.97) 210.8
Mild 1..20 (0.80–1.79) 21.4
Severe 2.65 (1.40–5.02) 29.4
Hopeless about future (lowest tertile) 0.72 (0.52–1.00) 0.70 (0.51–0.98) 26.3
Middle tertile 1.57 (1.10–2.25) 24.9
Highest tertile 2.29 (1.34–3.92) 21.4
All models including the model without any mediators were adjusted for subject’s grade, gender, race/ethnicity, language, parental education, parental employment, parental birthplace, and
parenting style. CI, confidence interval.
a
Standardized score so that a 1-point change refers to a 1-SD change.
b
Proportion of peers is scaled so that a 1-point difference refers to a 10% change in proportion of peers.
e394 WONG et al
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
randomized trial, ensuring compara-
bility between the 2 study arms is dif-
ficult. The control sample was almost
twice as likely to refuse participation,
potentially introducing sampling bias.
Although limited information is avail-
able for those who refused, we suspect
that they would have more behavioral
and academic problems, which would
bias the results toward the null. We also
excluded many potential intervention
and control subjects because they
had chosen to attend another high-
performing charter or private school
in ninth grade. Had we included them,
a much greaterpropor tion ofthe control
group would have been “contaminated”
by exposure to another high-performing
school. As expected, exclusion for this
reason was more common among the
control group, which might have in-
troduced additional bias. For example,
applying to more schools might be
associated with differences in risky
behaviors, parenting, or academic
motivation. Despite these challenges,
among study participants, the 2 study
arms were similar in demographic
characteristics and early test score
performance, suggesting that group
assignment was random and equal.
Still, some unobserved differences that
could confound the results might exist
due to imperfect randomization or sam-
pling bias.
Additional limitations include the in-
ability to generalize to adolescents who
do not apply to charter schools or to
other cities, populations, or school
environments, including successful
noncharter public or private schools.
We relied on student self-reported
behaviors, and those with better
school outcomes might tend to provide
more socially acceptable responses.
We also did not observe intermediate-
or long-term outcomes beyond high
school. Finally, the intervention group
reported lower tobacco and alcohol use
that did not reach statistical signifi-
cance, but our study may have been
underpowered as suggested by the
large confidence intervals. Of note, for
our ex ante sample size calculation, we
used previously reported behavior
rates among high school dropouts and
graduates for smoking, the least
prevalent outcome in our study, and
fur ther inflated our sample because
the estimated effect size was based on
observational studies.
15,34
However,
the observed differences in our s tudy
were much smaller than anticipated.
The sampling bias issues mentioned
previously may have fur ther ham-
pered the study to achieve ad equate
power.
CONCLUSIONS
The current study encouragingly re-
veals that successful public charter
high schools in low-income neighbor-
hoods might have early beneficial
health effects. Future studies will need
to determine if the effects are long
lasting or can be observed in other
populations and school settings. The
academic achievement gap in the
United States appears to be growing
between the wealthy and poor,
35
which
is concerning not only because of the
economic implications but also be-
cause of the potential effects on pop-
ulation health and health disparities.
The present findings highlight the
impor tance of improving academic
achievement and attainment, and in
par ticular school retention, on ado-
lescent health behaviors and suggest
that health policy makers may need
to pay greater attention to educa-
tional policies and trends.
REFERENCES
1. Saab H, Klinger D. School differences in
adolescent health and wellbeing: findings
from the Canadian Health Behaviour in
School-aged Children Study. Soc Sci Med.
2010;70(6):850–858
2. Komro KA, Flay BR, Biglan A; Promise
Neighborhoods Research Consortium. Cre-
ating nurturing environments: a science-
based fr amework for promoting child health
and development within high-poverty neigh-
borhood s. Clin Child Fam Psychol Rev . 2011;14
(2):111–134
3. Jamal F, Fletcher A, Harden A, Wells H,
Thomas J, Bonell C. The school environment
and student health: a systematic review and
meta-ethnography of qualitative research.
BMC Public Health. 2013;13(1):798
4. LaRusso MD, Romer D, Selman RL. Teachers
as builders of respectful school climates:
implications for adolescent drug use
norms and depressive symptoms in high
school. J Youth Adolesc. 2008;37(4):386–
398
5. Frech A. Healthy behavior trajectories be-
tween adolescence and young adulthood.
Adv Life Course Res. 2012;17(2):59–68
6. Carter M, McGee R, Taylor B, Williams S.
Health outcomes in adolescence: associa-
tions with family, friends and school en-
gagement. J Adolesc. 2007;30(1):51–62
7. Bond L, Butler H, Thomas L, et al. Social and
school connectedness in early secondary
school as predictors of late teenage sub-
stance use, mental health, and academic
outcomes. J Adolesc Health. 2007;40(4):357.
e9–e18
8. Maes L, Lievens J. Can the school make
a difference? A multilevel analysis of ado-
lescent risk and health behaviour. Soc Sci
Med. 2003;56(3):517–529
9. Schweinhart LJ, Montie J, Xiang Z, Barnett
WS. Lifetime Effects: The High/Scope Perry
Preschool Study Through Age 40. Ypsilanti,
MI: HighScope Press; 2005
10. Lager ACJ, Torssander J. Causal effect of
education on mortality in a quasi-experiment
on 1.2 million Swedes. Proc Natl Acad Sci USA.
2012;109(22):8461–8466
11. Raver CC, Jones SM, Li-Grining C, Zhai F,
Metzger MW, Solomon B. Targeting children’s
behavior problems in preschool classrooms:
ARTICLE
PEDIATRICS Volume 134, Number 2, August 2014 e395
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
a cluster-randomized controlled trial. J
Consult Clin Psychol. 2009;77(2):302–316
12. Zimmer R et al. Charter Schools in Eight
States: Effects on achievement, attainment,
integration, and competition, Santa Mon-
ica, CA: Rand Corporation;1–133
13. Executive Summary Explaining the Academic
Perfo rmance Index (API). Analysis, Measure-
ment, and Accountability Report ing Division,
California Department of Education. Available
at: http://www .cde.ca. gov/ta/ac/ap/documents/
apiexecsummary .pdf . Accessed May 16, 2014
14. API Data Files - Academic Performance Index.
Analysis, Measurement, and Accountability
Reporting Division, California Department of
Education. Available at: http://www .cde.ca.gov/
ta/ac/ap/apidatafiles.asp. Accessed May 22,
2014
15. Johnston LD et al. Monitoring the Future
national results on adolescent drug use:
Overview of key findings, 2008 (NIH Publi-
cation No. 09-7401). Bethesda, MD: National
Institute on Drug Abuse; 2009
16. Centers for Disease Control and Preven tion,
2010. Y outh risk behavior surveillance - United
States, 2009. Survellance Summaries, 2010.
MMWR. 2010;59(No. SS-5): 1–149
17. Garrison CZ, Addy CL, Jackson KL, McKeown
RE, Waller JL. The CES-D as a screen for
depression and other psychiatric disorders
in adolescents. J Am Acad Child Adolesc
Psychiatry. 1991;30(4):636–641
18. Yazzie-Mintz E. Voices of students on en-
gagement: A report on the 2006 high school
survey of student engagement, Bloomington,
IN: Center for Evaluation & Education Policy,
Indiana University;2007. Available at: http://
eric.ed.gov/?id=ED495758. Accessed May 7,
2014
19. Lamborn SD, Mounts NS, Steinberg L,
Dornbusch SM. Patterns of competence
and adjustment among adolescents from
authoritative, authoritarian, indulgent, and
neglectful families. Child Dev. 1991;62(5):
1049–1065
20. Bolland JM. Hopelessness and risk behaviour
among adolescents living in high-po verty
inner-city neighbourhoods. JAdolesc. 2003;
26(2):145–158
21. California Standards Tests Technical Report
Spring 2011 Administration. Assessment
Development and Administration Division,
California Department of Education. 2012.
Available at: http://www.cde.ca.gov/ta/tg/sr/
documents/csttechrpt2011.pdf. Accessed May
18, 2014
22. Hardin JW, Hilbe JM. Generalized Estimating
Equations. 2nd ed. Boca Raton, FL: Chapman
& Hall; 2013
23. Breen R, Karlson KB, Holm A. Total, direct,
andindirecteffectsinlogitmodels.
Centre for Strategic Research in Educa-
tion Working Paper Series. 2010. Avail-
able at: http://cser.dk/fileadmin/www.cser.
dk/wp_005rbkkahfin_01.pdf. Acces sed March
13, 2013
24. Rubin DB. Statistical matching using file
concatenation with adjusted weights and
multiple imputations. J Bus Econ Stat. 1986;
4(1):87–
94
25. Schafer JL. Analysis of Incomplete Multi-
variate Data. Boca Raton, FL: Chapman &
Hall/CRC Press; 1997
26. StataCorp. Stata: Release 11 [statistical
software]. College Station, TX: StataCorp;
2009
27. Barnett WS. Benefi t-cost analysis of the
Perry preschool program and its policy
implications. Educ Eval Policy Anal. 1985;7
(4):333–342
28. Kaestner R, Callison K. Adolescent cognitive
and noncognitive correlates of adult health.
J Hum Capital. 2011;5(1):29–69
29. Heckman JJ, Stixrud J, Urzua S. The effects
of cognitive and noncognitive abilities on
labor market outcomes and social be-
havior (Working Paper No. 12006). National
Bureau of Economic Research; 2006: 1–34.
Available at: www.nber.org/papers/w12006.
Accessed March 14, 2013
30. Heckman JJ, Rubinstein Y. The importance
of noncognitive skills: lessons from the GED
testing program. Am Econ Rev. 2001;91(2):
145–149
31. Li Y, Lerner RM. Interrelations of behavioral,
emotional, and cognitive school engagement
in high school students. J Youth A dolesc.
2013;42(1):20–32
32. Resnick MD, Bearman PS, Blum RW, et al.
Findings from the National Longitudinal
Study on Adolescent Health: protecting ado-
lescents from harm. JAMA. 1997;278(10):
823–832
33. Kelly DB, Rollings AL, Harmon JG. Chronic
self-destructiveness, hopelessness, and risk-
taking in college students. Psychol Rep. 2005;
96(3 pt 1):620–624
34. Green MP, McCausland KL, Xiao H, Duke JC,
Vallone DM, Healton CG. A closer look at
smoking among young adults: where to-
bacco control should focus its attention.
Am J Public Health. 2007;97(8):1427–1433
35. Reardon SF. Whither opportunity? Rising
inequality, schools, and chi ldren’s life chances.
In: Duncan GJ, Murnane RJ , eds. Book Review .
2011 ed. New Y ork, NY: Russell Sage Founda-
tion;91–114
(Continued from first page)
PEDIATRICS (ISSN Numbers: Print, 0031-4005; Online, 1098-4275).
Copyright © 2014 by the American Academy of Pediatrics
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Support for this study was provided by a grant from National Institutes of Health (NIH)/National Institute on Minority Health and Health Disparities
(RC2MD004770) and by additional support from NIH/National Center for Advancing Translational Sciences (UL1TR000124). Funded by the National Institutes of
Health (NIH).
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
e396 WONG et al
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
DOI: 10.1542/peds.2013-3573
; originally published online July 21, 2014;Pediatrics
Tseng, Peter Bergman and Paul J. Chung
Richard Buddin, Martin F. Shapiro, Sheryl H. Kataoka, Arleen F. Brown, Chi-Hong
Mitchell D. Wong, Karen M. Coller, Rebecca N. Dudovitz, David P. Kennedy,
Successful Schools and Risky Behaviors Among Low-Income Adolescents
Services
Updated Information &
/peds.2013-3573
http://pediatrics.aappublications.org/content/early/2014/07/16
including high resolution figures, can be found at:
Permissions & Licensing
tml
http://pediatrics.aappublications.org/site/misc/Permissions.xh
tables) or in its entirety can be found online at:
Information about reproducing this article in parts (figures,
Reprints
http://pediatrics.aappublications.org/site/misc/reprints.xhtml
Information about ordering reprints can be found online:
rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.
Grove Village, Illinois, 60007. Copyright © 2014 by the American Academy of Pediatrics. All
and trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk
publication, it has been published continuously since 1948. PEDIATRICS is owned, published,
PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly
at Library Sm-Per-Ebs on October 7, 2014pediatrics.aappublications.orgDownloaded from
