1. Original article
Influence of Family and School-Level Factors on Age of Sexual
Initiation
Candace N. White, Ph.D., M.Ed. *, and Lynn A. Warner, Ph.D., M.S.W.
University at Albany, SUNY, School of Social Welfare, Albany, New York
Article history: Received June 17, 2014; Accepted September 27, 2014
Keywords: Adolescence; Initiation of sexual intercourse; School factors; Family factors; Multilevel models
A B S T R A C T
Purpose: This study examined the association of individual, family, and school-level characteristics
with age of sexual initiation (ASI) and focused specifically on school context as a moderator of
known predictors of ASI.
Methods: Data are from Waves I and IV of the National Longitudinal Study of Adolescent Health (N ¼
10,596). Predictors include grade point average, physical development, attitudes about sex, likeli-
hood of higher education, alcohol use, delinquency, family structure, parents’ education level,
childhood abuse, maternal approval of sex, parental monitoring, and parentechild relationship
quality. School-level predictors are averages of adolescents’ attitudes about sex and likelihood of
higher education and parents’ education. Hierarchical linear models run separately by sex were used
to predict ASI.
Results: When school-level attitudes about sex are more favorable, both boys and girls report
younger ASI, and school mean parental education attainment moderates the influence of
individual adolescents’ attitudes about sex on ASI. More of the predictors are significant for girls
than boys, whereas perception of maternal and peer approval of sexual activity are the most salient
predictors of younger ASI for boys.
Conclusions: Results highlight the importance of school context for understanding adolescents’
motivations for early ASI. Findings support the need for school-wide prevention interventions that
engage adolescents, peers, and parents in addressing attitudes about early sex.
Ó 2015 Society for Adolescent Health and Medicine. All rights reserved.
IMPLICATIONS AND
CONTRIBUTION
This nationally represen-
tative longitudinal study
contributes to a better
understanding of the
multiple contexts that in-
fluence adolescent sexual
activity. Findings support
the need for school-wide
prevention approaches
that incorporate parent
echild and peer-to-peer
communication about at-
titudes about early sex and
pregnancy for both boys
and girls.
Bioecological systems theory emphasizes the importance of
considering multiple systems to understand individual behavior
[1]. However, most studies of adolescents’ age of heterosexual
intercourse (hereafter age of sexual initiation [ASI]) have focused
exclusively on individual, peer, and family factors, despite
research that demonstrates the influence of more distal contexts
on other adolescent risk taking behaviors [2,3]. For example, low
neighborhood socioeconomic status (SES) has been found to be
associated with adolescents’ viewing sex and pregnancy in a
positive light [4] and with higher rates of early initiation [5].
A better understanding of the influence of school-level fac-
tors, as well as the interaction between school characteristics and
other known ASI predictors, is particularly important because
sexual risketaking behavior has been associated with adoles-
cents’ perceptions of their peers’ attitudes and behaviors [6e11],
and teen pregnancy and sexually transmitted disease prevention
efforts are often school based [12]. Current intervention evalua-
tions show successful reduction of adolescent sexual risketaking
behaviors using school-based interventions that directly address
the predictors of ASI (e.g., perception of peer norms, academic
Conflicts of Interest: There are no known conflicts of interest for either author.
The study was a secondary analysis using the National Longitudinal Study of
Adolescent Health (Add Health). Neither author has an affiliation with Add
Health study sponsors. No honorarium, grant, or other form of payment was
given to anyone to produce the article.
* Address correspondence to: Candace N. White, Ph.D., M.Ed., New York State
Office of Mental Health, 44 Holland Ave., Albany, NY 12229.
E-mail address: cwhite2@albany.edu (C.N. White).
www.jahonline.org
1054-139X/Ó 2015 Society for Adolescent Health and Medicine. All rights reserved.
http://dx.doi.org/10.1016/j.jadohealth.2014.09.017
Journal of Adolescent Health 56 (2015) 231e237

2. failure, parentechild relationship) [13]. However, findings have
been inconsistent, possibly because of variation in the risk factors
addressed, or unique characteristics of the target population.
Moreover, there has been little examination into the effects of
school-wide norms and values with regard to sexual activity, or
the social and economic status of students’ parents who serve as
role models for adolescents by setting expectations for adulthood
achievement such as college attendance. Given that neighbor-
hood SES has been found to moderate the influence on ASI of
parental involvement and decision making regarding youth’s
activities [10], it is also likely that aspects of the more proximal
school context interact with individual and family characteris-
tics. However, to the authors’ knowledge, only one study [14] has
examined the direct effect of “school” characteristics on sexual
initiation, with the finding that initiation occurred earlier in
private versus public schools and in schools with positive norms
about adolescents’ sexual activity. None have examined the
interaction of school characteristics with other known ASI
predictors.
The purpose of the present study is to examine multisystemic
influences on ASI, focusing on school-level characteristics as
possible moderators of previously identified individual and family
characteristics associated with ASI. The study examines the mul-
tiple levels of influence separately for boys and girls, given previous
studies that show sex differences for many ASI risk factors, and that
some prevention interventions have been effective for one but not
both sexes. For example, individual-level factors associated with
early initiation for both boys and girls include delinquency
[6,7,15e17], substance abuse [16,18e22], and childhood sexual and
physical abuse and neglect [23,24]. Low academic achievement
[9,15,18e21,25], low educational aspirations [26], and early physical
development [10,25] have been significantly associated with ASI for
both sexes and for girls only [17,27]. Family characteristics associ-
ated with early ASI for both boys and girls include single-parent
family structure [10,28,29], low parental income and education
[19,28,29], poor parentechild relationship quality [19,25,30,31],
low parental monitoring [30], and parents’ permissive attitudes
about sex [25,30e32]. However, some studies have found these
family factors to be significant for girls only [6,15,17,32e34]. The
adolescent’s sex therefore plays a critical role in which factors are
associated with age of initiation. Existing research on sexual initi-
ation is limited because of the lack of studies that simultaneously
examine the influence of individual, family, and school-level factors
while also noting any sex differences in predictors. The current
analysis of a large sample allows for further examination of these
sex differences and a comprehensive set of multisystemic pre-
dictors to inform evidence-based practice guidelines for social
workers practicing with families and in the schools.
Methods
Data source
This study uses data from the National Longitudinal Study of
Adolescent Health (Add Health), a nationally representative
survey that began in 1995, when students were in Grades 7e12
[35]. The Add Health study used a clustered and school-based
stratified random sampling design to ensure that the selected
schools were representative of schools in the United States.
Eighty communities with schools spanning Grades 7e12 were
randomly selected. The randomly selected school in 65 of the
communities did not span all grades, so feeder middle schools
were selected. After random selection of students stratified by
grade level and sex, and oversampling of particular sub-
populations, the baseline sample included 20,745 adolescents.
At Wave I, researchers conducted an extensive in-home
interview with the student and a half an hour interview with
one parent (88% of the sample had a participating parent).
Computer assisted self-interview protocols were used for
respondents to record answers to sensitive questions such as
those regarding sexual intercourse. Wave I respondents were
followed into young adulthood with four in-home interviews.
Wave IV interviews were conducted from 2007 to 2008 when
the sample was aged 24e32 years (N ¼ 15,701). The current
analysis excludes Wave IV respondents who are missing
sampling weights, those who had not had sexual intercourse as
of Wave IV, outliers on age of initiation, individuals who did not
report or inconsistently reported ever having sex or age of
initiation, those with missing family variable information, and
individuals living with foster parents. Because of these exclu-
sions, four strata were missing data and thus were also excluded
from analysis. Compared with the analytic sample (N ¼ 10,596),
the excluded group reported a significantly lower age of initia-
tion at Wave IV (mean, 16.11 years) and includes a significantly
higher proportion of adolescents with characteristics associated
with not completing high school (e.g., African-American,
nonresident fathers).
Measures
Outcome variable: age of sexual initiation
At Waves IeIV, adolescents were asked, “Have you ever had
sexual intercourse?” (“yes” or “no”). Sexual intercourse was
defined for respondents as “when a male inserts his penis into a
female’s vagina.” In Waves IIeIV, respondents who answered
“yes” were also asked “At what age did you have sexual inter-
course for the very first time?” The current analysis uses ASI as
reported at Wave IV.
Independent variables
Most of the independent variables are based on data collected
at Wave I when participants reported on behaviors and feelings
at or before that time. Two-parent family structure includes
having both a mother and father in the house, which could
include biological, step, adoptive, or other. Parentechild rela-
tionship quality is based on five questions concerning warmth,
satisfaction with mother/father relationships, and satisfaction
with communication style with each residential parent. Because
different Likert scale response categories were used for the five
items, “positive” responses (“strongly agree” or “agree,” “quite a
bit” or “very much”) were counted to obtain an overall score,
ranging from 0 to 5, with a higher score indicating more positive
relationship. For individuals in two-parent families, the highest
parent score is used. To address the negative skew, a dichoto-
mous variable was created to differentiate the 79% with a value of
five from the others. Childhood maltreatment, not assessed at
Wave I, was measured by two Wave IV questions about physical
and sexual abuse by parents or other adult caregivers by the time
respondents were in the sixth grade. Responses to both items
were summed to obtain an overall childhood abuse indicator,
ranging from 0 (“never happened”) to 10 (“more than 20 times”),
and then rescaled to a three-point scale to address the high
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237232

3. positive skew and kurtosis. Residential mother’s approval of
sexual initiation was based on the question, “How would mom
feel about you having sex at this time in your life,” with responses
on a scale of 1 (“strongly disapprove”) to 5 (“strongly approve”).
Informed by results from a previous study [10], parental moni-
toring is based on the summed responses to two items that
capture adolescents’ reports of parental decisions regarding their
“outside activities”: whether residential mother and/or father
decide whom the youth associates with and when the youth
must be home on weekends (0 ¼ no; 1 ¼ yes).
For the current analysis, school-level (Level 2) variables, mean
values were calculated from the responses provided by all ado-
lescents in each stratum. School-level parents’ education level is
based on the highest parental education level or the level of one
parent if the other parent’s information was unavailable, rated on
a scale of 1 (eighth grade or less) to 9 (professional training
beyond 4-year college). The score of 10 (never went to school)
was recoded to 0. For 88% of the sample, parent report is used; for
the remaining 12%, adolescent report is used. School (strata)-
level adolescents’ sex approval is assessed using the mean value
on nine items found to be associated with early sexual initiation
and lower rates of condom use during sex in prior Add Health
analyses [11]. Responses ranged from 1 to 5, indicating strong
agreement to strong disagreement. Items assess attitudes such as
whether sex and pregnancy would lead to gained respect,
embarrassment, guilt, quitting school, marrying the wrong per-
son, and growing up too fast. Alpha reliability was .76. An outlier
indicator was created for two strata that had an average age of
initiation greater than 19 and sex approval averages that were
more than three standard deviations below the average. School-
level higher education likelihood is based on the average of the
adolescent responses in each stratum to a question about how
likely it was she/he would go to college (range, 1e5; higher
values indicate higher likelihood).
Control variables include the variables used in stratification of
the schools (high-school metropolitan location, size, quartile
percentage white, school type, and region of the country) and
oversampling of particular groups; age and a dichotomous
indicator of high school versus middle school status at Wave I to
address the multicohort sampling design; a dichotomous indi-
cator of sex initiation before Wave I; individual-level variables
known to contribute to early initiation (race/ethnicity, sex,
alcohol use, delinquency, early physical development, and grade
point average); and the individual-level variables that would be
aggregated to the school level in the final model (higher educa-
tion likelihood, sex approval, and parents’ education level).
Dummy variables were created for all race categories except
Native American and other (62 individuals were Native American
and 79 individuals were in the other group). In addition to
African-American race and parents’ education level, an interac-
tion variable was included in the multivariate analyses to address
the oversampling of African-American adolescents with college
educated parents [36]. Academic achievement is assessed using
the mean, on a 4.0 scale, of self-reported grades from Wave I
because transcripts were not available for those in Grades 7 and 8
at Wave I. Scaled responses to the question “During the past
12 months, on how many days did you drink alcohol?” is used to
assess alcohol use. Responses ranged from 1 (every day/almost
every day) to 7 (never) and were reverse scored, so higher scores
represent higher alcohol use. Delinquency is assessed using the
average of the 15-item Wave I self-report of delinquent behaviors
occurring over the past 12 months, rated on a scale of never (0),
one or two times (1), three or four times (2), or five or more times
(3). Alpha reliability was .84. This is categorized according to
quartile distribution to address the high positive skew and
kurtosis. Physical development is assessed using Wave I adoles-
cent responses to the question, “How advanced is your physical
development compared to other girls/boys your age?” Response
choices ranged from 1 (look younger than most) to 5 (look older
than most), with a higher score indicating earlier maturation.
Analyses
Univariate and bivariate analyses were conducted to describe
the sample and examine sex differences on all variables. A series
of hierarchical linear models was estimated to determine the
influence of individual factors, family factors, school-level char-
acteristics, and the cross-level interaction of school level with
individual and family factors, on age of first sexual intercourse. A
multilevel approach is appropriate when using Add Health data
because participants are nested within schools, therefore
violating the assumption of independence across observations
[37]. Given previous sex difference findings, all steps were con-
ducted separately for the two sexes as follows: a null model with
no predictor variables; all control and individual-level variables;
all family variables; school-level variables; and the cross-level
interaction of school-level parental educational attainment
with family characteristics, and with the individual-level sex
approval variable. The intercept was random, and all predictors
were fixed in all models; except for the model with the cross-
level interaction, all models were fully nested. Level 1 variables
were group mean centered, and all models were estimated using
restricted maximum likelihood because the dependent variable
is normally distributed. The use of restricted maximum likeli-
hood and fixed predictors allows for examination of improve-
ment in model fit using Akaike Information Criterion [38]. For all
these measures, smaller values represent better fitting models.
SAS 9.4 (SAS Institute, Inc., Cary, NC) was used for all analyses.
All research was approved by the institutional review board at
[institution blinded for peer review] and an “Agreement for the
Use of Restricted-Use Data” and “Pledge of Confidentiality” were
provided to the Interuniversity Consortium for Political and
Social Research at the University of Michigan where Add Health
data are stored.
Results
Table 1 presents descriptive statistics for girls and boys and
significant differences between the sexes based on bivariate
analyses. The average age of the sample at Wave IV is 29 years,
with 24% in middle school and the remainder in high school at
Wave I. There is a slight over representation of female adoles-
cents (55%). Girls score significantly higher on more of the family
risk factors, and boys score significantly higher on more of the
individual-level risk factors. School-level descriptives for the
entire sample are reported in Table 2. There are no significant sex
differences at the school level or on age of initiation.
Hierarchical linear model results
The null models for both boys and girls show variation in the
average age of initiation across schools, and convergence criteria
are met for these and all subsequent models. Table 3 presents
results of the last three hierarchical linear model steps for males
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237 233

4. and females separately. Model 3 includes all control and family
variables. All the individual-level factors are associated with age
of initiation for both sexes. For girls, all family factors are
significant predictors of age of initiation. For boys, although other
family factors are significant, parentechild relationship quality
and parental monitoring are not significant predictors, even
when all other family variables are removed in a post hoc
analysis.
Model 4 shows results of the addition of school-level factors.
For girls, all school-level factors are significant. For boys, all
school-level factors are significant except school-level parental
educational attainment. It was significant however in a post hoc
analysis where it was the only school-level variable included,
suggesting that other school-level characteristics might mediate
the relationship of school-level parental education attainment
with ASI.
Cross-level interactions are reported in Model 5. The
interaction of school mean parents’ education level with
individual-level sex approval is the only significant interaction
for both girls (p < .01) and boys (p < .05). The coefficient for the
sex approval interaction is negative; as the school-level parents’
education level increases, the influence of individual-level atti-
tudes on age of initiation decreases. None of the interactions
with family variables are significant for either sex, nor are they
when the school-type indicator is removed. Removal of the
insignificant school parents’ education with family factors cross-
level interactions results in the best fitting model for both sexes.
Results for both sexes are shown in Table 4.
Discussion
Results confirm the influence of multisystemic factors and
their interactions on ASI. For both sexes, family- and school-level
factors explain the variation in ASI, even after controlling for
previously identified fixed (race and physical development)
and mutable (delinquency, alcohol use, grade point average)
individual-level factors. Consistent with prior research, single-
parent family structure, maternal approval of sex, and child-
hood abuse are associated with earlier age of initiation for both
boys and girls. Among girls, but not boys, the significance of
parentechild relationship quality and parental monitoring is also
consistent with previous sex difference findings [15,32e34].
The relationship between school-level higher education
likelihood and ASI, and attitudes about early sex/pregnancy and
ASI, is consistent with previous research showing the strong
influence of peer attitudes on sexual risk behaviors for both sexes
[6e11]. Teitler and Weiss [14] had also demonstrated that school
norms about acceptable timing of youths’ transitions mediated
the relationship between neighborhood SES and early initiation.
The current findings extend this research by showing that
attendance at a school where students disapprove of early sex
and pregnancy, and will likely pursue higher education, is a
protective factor against early sex. In addition, as demonstrated
by the significance of cross-level interactions for both sexes,
Table 2
Descriptive results for school-level variables (N ¼ 76)a
School-/community-level variables
(strata N ¼ 76)
Frequency (%)
or mean (SD)
Range of
strata means
School type
Proportion public school 68 (90%)
Proportion private school 6 (8%)
High school proportion white students
67% or more
41 (54%)
Metropolitan location
High school in urban location 23 (30%)
High school in rural location 14 (18%)
High school in suburban location 39 (51%)
High school has 351 or more students 38 (50%)
High school region West or Midwest 33 (44%)
Number of participating schools in a strata 1.68 (.47) 1e2
Number of participating students per strata 139 (113) 47e885
Proportion male students per strata .45 (.05) .35e.59
Parents’ education 6.07 (.99) 3.78e8.62
Approval of sexual activity 2.53 (.22) 1.73e3.15
Proportion pregnant .01 (.01) 0e.04
Higher education likelihood 4.18 (.30) 3.53e4.93
GPA 2.84 (.20) 2.36e3.48
Age of initiation 16.93 (.77) 15.36e19.74
GPA ¼ grade point average; SD ¼ standard deviation.
a
All Level 2 (“school level”) variables are calculated based on the 76 randomly
selected strata/communities, which contain one to two schools: one combined
high/middle school or one high school and one middle/junior high school.
Table 1
Sample description and sex differencesa
Variable Girls
(N ¼ 5,821)
Boys
(N ¼ 4,775)
Mean (SD) or
frequency (%)
Mean (SD) or
frequency (%)
Dependent variable
Initiation age 16.91 (2.76) 16.93 (2.81)
Independent family-level variables
Two-parent family structure 4,005 (69%)** 3,412 (71%)
Mother’s approval of sexual activity 1.59 (.84)*** 1.86 (.92)
Residential parentechild relationship
quality
4.47 (1.10)*** 4.68 (.82)
Residential parental monitoring 1.15 (.60)*** 1.25 (.64)
Childhood abuse .66 (1.61)** .58 (1.41)
Control variables
Initiation before Wave I 2,185 (37%) 1,834 (38%)
Age at Wave IV 28.98 (1.65)*** 29.18 (1.66)
Grade at Wave I
7th grade 673 (12%) 516 (11%)
8th grade 742 (13%) 637 (13%)
9th grade 1,078 (19%) 894 (19%)
10th grade 1,207 (21%) 979 (21%)
11th grade 1,160 (20%) 985 (21%)
12th grade 949 (16%) 755 (16%)
Ethnicity
White 3,263 (56%)** 2,812 (59%)
African-American 1,274 (22%)*** 858 (18%)
Latino 870 (15%) 726 (15%)
Asian American 340 (6%) 312 (7%)
Native American 37 (.64%) 25 (.52%)
Other 37 (.64%) 42 (.88%)
Parents’ education 6.01 (2.33)*** 6.19 (2.27)
Alcohol use 2.04 (1.34)*** 2.18 (1.52)
Delinquency .24 (.29)*** .33 (.38)
Physical development 3.26 (1.09)* 3.22 (1.11)
GPA 2.89 (.75)*** 2.69 (.77)
Self-reported approval of sexual
activity
2.53 (.71)*** 2.58 (.65)
Self-reported higher education
likelihood
4.34 (1.03)*** 4.08 (1.16)
*p < .05; **p < .01; ***p < .001.
GPA ¼ grade point average; SD ¼ standard deviation.
a
All independent variables are taken from Wave I report with the exception of
childhood abuse, which is taken from Wave IV. Age of initiation is taken from
Wave IV report. If initiation age was missing, it was taken from a previous
wave. Although the parentechild relationship quality, childhood abuse, and
delinquency variables were categorized for the multivariate analyses, descriptive
information for the original scale values is reported here.
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237234

5. school-level parents’ education appears to be an important
moderator of the association between attitudes about sex and
ASI. Attendance in a school where the parents’ education level is
high reduces the influence that approving attitudes about
sex/pregnancy have on age of initiation for girls. The lack of
moderation of the family characteristics by the school-level
variables is not consistent with results of a prior study in
which neighborhood SES moderated the influence of parental
monitoring and involvement [10], but that study includes only
middle school students and uses a neighborhood rather than
school-level indicator of SES as a moderator. It could be that
school-level parents’ education and related factors protect
against the risk of more liberal attitudes of individuals about sex,
but they do not impact the stronger influence of family factors.
The findings should be interpreted in light of study limita-
tions. Adolescents’ self-reports of ASI may not be recalled reli-
ably, and some adolescents may misrepresent parenting aspects
and approval, possibly to justify their behaviors. To minimize the
consequence of nonreliable ASI reports, only adolescents with
consistent reports across data collection waves were included in
the analyses. Future research using methods to estimate and
account for possible social desirability bias may be needed.
Because of missing data exclusions, youth with risks for a range
of negative outcomes were not included in this study. These
youth, however, would likely benefit from targeted and more
intensive interventions in addition to the universal prevention
programs suggested by the current analysis. Finally, in-
terpretations of causal relationships between the predictors and
ASI should be made cautiously, especially given possible lack of
temporal precedence between predictors and ASI for the 38% of
the sample who had sex before Wave I.
Despite these limitations, the present study has notable
strengths. Although previous studies identified contextual level
factors like neighborhood SES, none examined the influence of
Table 3
Hierarchical linear model results of Models 3e6: individual-, family-, and school-level predictors of age of initiation, Add Health, coefficient (SE)
Model 3 Model 4 Model 5
Girls
(N ¼ 5,821)
Boys
(N ¼ 4,775)
Girls
(N ¼ 5,821)
Boys
(N ¼ 4,775)
Girls
(N ¼ 5,821)
Boys
(N ¼ 4,775)
Intercept 17.25 (.47)*** 17.12 (.49)*** 17.66 (1.8)*** 18.77 (1.95)*** 17.67 (1.81)*** 18.77 (1.95)***
Region
West .35 (.24) .5 (.25) .32 (.18) .38 (.19) .33 (.18) .37 (.19)
Midwest À.09 (.23) .09 (.24) À.04 (.16) .12 (.17) À.04 (.16) .12 (.17)
South À.16 (.22) À.03 (.22) À.25 (.15) À.11 (.16) À.24 (.15) À.11 (.16)
Northeast 0 0 0 0 0 0
Metropolitan location
Urban .3 (.26) .31 (.27) .21 (.18) .22 (.19) .21 (.18) .22 (.19)
Suburban 0 (.22) .06 (.23) À.06 (.15) 0 (.16) À.05 (.15) À.01 (.16)
Rural 0 0 0 0 0 0
School type
Public À1.41 (.36)*** À1.71 (.37)*** À.15 (.31) À.32 (.34) À.15 (.31) À.32 (.34)
Catholic À.91 (.53) À1.05 (.54) .17 (.39) .09 (.41) .17 (.39) .1 (.41)
Private 0 0 0 0 0 0
Proportion white .11 (.1) .26 (.11)* .08 (.09) .19 (.09) .08 (.09) .18 (.09)
High school size .11 (.11) .13 (.11) .16 (.08)* .17 (.08)* .17 (.08)* .17 (.08)
African-American À.21 (.1)* À.41 (.12)*** À.16 (.09) À.34 (.12)** À.16 (.09) À.35 (.12)**
Latino .24 (.11)* À.11 (.13) .26 (.11)* À.1 (.13) .26 (.11)* À.09 (.13)
Asian .4 (.15)** .45 (.17)** .42 (.15)** .47 (.17)** .42 (.15)** .46 (.17)**
Age at interview .48 (.03)*** .39 (.03)*** .48 (.03)*** .39 (.03)*** .48 (.03)*** .39 (.03)***
In high school at Wave I À.1 (.1) .09 (.12) À.09 (.1) .08 (.12) À.1 (.1) .09 (.12)
Sex before Wave I À2.56 (.07)*** À2.39 (.08)*** À2.56 (.07)*** À2.4 (.08)*** À2.57 (.07)*** À2.4 (.08)***
Approval of sex À.45 (.05)*** À.33 (.06)*** À.45 (.05)*** À.33 (.06)*** .32 (.28) .44 (.37)
Parents’ Ed level .03 (.01)* .03 (.02) .03 (.01) .02 (.02) .03 (.01) .02 (.02)
African-American  parents’ Ed À.03 (.03) À.02 (.04) À.03 (.03) À.02 (.04) À.03 (.03) À.02 (.04)
Higher Ed likelihood À.01 (.03) .04 (.03) À.01 (.03) .04 (.03) 0 (.03) .04 (.03)
Physical development À.17 (.03)*** À.06 (.03)* À.17 (.03)*** À.06 (.03)* À.17 (.03)*** À.06 (.03)*
GPA .25 (.04)*** .25 (.05)*** .25 (.04)*** .25 (.05)*** .25 (.04)*** .25 (.05)***
Alcohol use À.18 (.02)*** À.16 (.03)*** À.18 (.02)*** À.16 (.03)*** À.18 (.02)*** À.16 (.03)***
Delinquency À.23 (.03)*** À.24 (.03)*** À.23 (.03)*** À.24 (.03)*** À.23 (.03)*** À.24 (.03)***
Two parent family .24 (.07)*** .24 (.08)** .25 (.07)*** .24 (.08)** .25 (.07)*** .24 (.08)**
Parentechild relationship quality .15 (.07)* À.08 (.09) .15 (.07)* À.07 (.09) À.1 (.43) À.04 (.59)
Childhood abuse À.25 (.04)*** À.15 (.05)** À.25 (.04)*** À.15 (.05)** À.25 (.04)*** À.15 (.05)**
Maternal approval of sex À.12 (.04)** À.12 (.04)** À.12 (.04)** À.12 (.04)** À.12 (.04)** À.12 (.04)**
Parental monitoring .14 (.05)** .05 (.05) .14 (.05)** .05 (.05) .22 (.3) À.45 (.34)
School mean parents’ Ed level À.35 (.11)** À.21 (.12) À.36 (.11)** À.21 (.12)
School outlier indicator À1.48 (.52)** À1.44 (.55)* À1.49 (.52)** À1.43 (.55)*
School mean sex approval À1.6 (.4)*** À1.7 (.44)*** À1.59 (.4)*** À1.71 (.44)***
School mean higher Ed likelihood 1.44 (.37)*** 1.02 (.4)* 1.45 (.37)*** 1.01 (.39)*
School mean parents’ Ed  parental monitoring À.01 (.05) .08 (.06)
School mean parents’ Ed  parentechild relationship .04 (.07) À.01 (.1)
School mean parents’ Ed  individual approval of sex À.13 (.05)** À.13 (.06)*
AIC 25,595 21,676.9 25,540.4 21,630.9 25,543.5 21,634.8
*p < .05; **p < .01; ***p < .001.
AIC ¼ Akaike Information Criterion; Ed ¼ education; GPA ¼ grade point average; SE ¼ standard error.
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237 235

6. students’ attitudes about sex and likelihood of higher education
aggregated to the school level. In addition, previous studies
did not account for the clustering of individuals within these
contexts, and relied heavily on cross-sectional data and logistic
regression models that used dichotomous outcome variables.
Their smaller sample sizes also precluded analysis of the
comprehensive list of predictors possible in this study.
Implications
The present study provides information about the ecological
systems involved in adolescents’ initiation of sex. The nationally
representative findings underscore the need for universal
prevention efforts that involve the adolescents’ families and
schools and interventions that address not only individual-level
risk factors but the adolescents’ and parents’ attitudes about
early sex. The present study found, for example, that boys report
more maternal approval of sex than girls, and this was one of only
a few significant predictors of initiation age for boys. Increased
efforts should be made to help parents understand the potential
protective effect of the expression of disapproval of early sexual
activity to girls and boys. Previous intervention research supports
the need for communication about these attitudes. For example, a
decrease in rates of early ASI was achieved with an intervention
aimed at increasing parentechild communication about the social
consequences of early sex in clinics [39], rather than just the health
and economic consequences on which parents commonly focus.
Parentechild and peer-to-peer communication regarding
these attitudes as well as education that addresses the social
pressures to engage in sexual behavior and provides opportunity
to practice refusal skills can be facilitated in the schools [12]. The
present study supports the need for interventions that address
not only the attitudes of immediate peers but also the school-
wide peer culture. Previous intervention has shown success in
changing school-wide attitudes about academic achievement
[40]. Similar efforts could be facilitated with regard to attitudes
about early sex and pregnancy, for example, with alumni groups
who can model resistance skills and have achieved higher edu-
cation and other positive milestones. Given the sex difference
findings, prevention programs might also consider implement-
ing programs that focus on strengthening family relationships
for girls and providing positive peer role models for boys.
Acknowledgments
The authors acknowledge Drs. Carolyn Smith, Kathryn Schiller,
and Glenn Deane for helpful comments and statistical assistance,
the Interuniversity Consortium for Political and Social Research at
the University of Michigan for access to the Add Health data, and
the University of North Carolina’s Carolina Population Center for
access to Add Health user guides and technical assistance. Find-
ings from the present study were presented at the Add Health
Users Conference in June 2014. C.N.W., PhD, wrote the first draft
of this article.
References
[1] Bronfenbrenner U. Ecological systems theory. In: Vasta R, ed. Six theories of
child development: Revised formulations and current issues. London,
England: Jessica Kingsley Publishers; 1992:187e249.
[2] Brookmeyer K, Fanti K, Henrich C. Schools, parents, and youth violence: A
multilevel, ecological analysis. J Clin Child Adolesc Psychol 2006;35:504e14.
[3] Denny SJ, Robinson EM, Utter J, et al. Do schools influence student
risk-taking behaviors and emotional health symptoms? J Adolesc Health
2011;48:259e67.
[4] Harding DJ. Cultural context, sexual behavior, and romantic relationships in
disadvantaged neighborhoods. Am Sociol Rev 2007;72:341e64.
[5] Browning CR, Leventhal T, Brooks-Gunn J. Neighborhood context and racial
differences in early adolescent sexual initiation. Demography 2004;41:
697e720.
[6] Dupéré V, Lacourse É, Leventhal T, et al. Neighborhood poverty and early
transition to sexual initiation in young adolescents: A developmental
ecological approach. Child Dev 2008;79:1463e76.
[7] Donenberg GR, Bryant FB, Emerson E, et al. Tracing the roots of early sexual
debut among adolescents in psychiatric care. J Am Acad Child Adolesc
Psychiatry 2003;42:594e608.
[8] Mitchell CM, Whitesell NR, Spicer P, et al. Cumulative risk for early sexual
initiation among American Indian youth: A discrete-time survival analysis.
J Res Adolesc 2007;17:387e412.
[9] Laflin MT, Wang J, Barry M. A longitudinal study of adolescent transition
from virgin to nonvirgin status. J Adolesc Health 2008;42:228e36.
[10] Roche KM, Mekos D, Alexander CS, et al. Parenting influences on early sex
initiation among adolescents: How neighborhood matters. J Fam Issues
2005;26:32e54.
[11] Henry DB, Schoeny ME, Deptula DP, et al. Peer selection and socialization
effects on adolescent intercourse without a condom and attitudes about
the costs of sex. Child Dev 2007;78:825e38.
Table 4
Hierarchical linear model results for the best fitting model, Add Health, coeffi-
cient (SE)
Girls
(N ¼ 5,821)
Boys
(N ¼ 4,775)
Intercept 17.68 (1.81)*** 18.72 (1.96)***
Region
West .33 (.18) .37 (.19)
Midwest À.04 (.16) .12 (.17)
South À.24 (.15) À.11 (.16)
Northeast 0 0
Metropolitan location
Urban .21 (.18) .22 (.19)
Suburban À.05 (.15) À.01 (.16)
Rural 0 0
School type
Public À.15 (.31) À.32 (.34)
Catholic .17 (.39) .1 (.41)
Private 0 0
Proportion white .08 (.09) .18 (.09)
High school size .17 (.08)* .17 (.08)*
African-American À.16 (.09) À.34 (.12)**
Latino .26 (.11)* À.09 (.13)
Asian .42 (.15)** .46 (.17)**
Age at interview .48 (.03)*** .39 (.03)***
In high school at Wave I À.1 (.1) .09 (.12)
Sex before Wave I À2.57 (.07)*** À2.4 (.08)***
Approval of sex .33 (.27) .42 (.37)
Parents’ Ed level .03 (.01) .02 (.02)
African-American  parents’ Ed À.03 (.03) À.02 (.04)
Higher Ed likelihood 0 (.03) .04 (.03)
Physical development À.17 (.03)*** À.06 (.03)*
GPA .25 (.04)*** .25 (.05)***
Alcohol use À.18 (.02)*** À.16 (.03)***
Delinquency À.23 (.03)*** À.24 (.03)***
Two parent family .25 (.07)*** .24 (.08)**
Parentechild relationship quality .15 (.07)* À.08 (.09)
Childhood abuse À.25 (.04)*** À.15 (.05)**
Maternal approval of sex À.12 (.04)** À.12 (.04)**
Parental monitoring .14 (.05)** .05 (.05)
School mean parents’ Ed level À.36 (.11)** À.21 (.12)
School outlier indicator À1.5 (.52)** À1.43 (.56)*
School mean sex approval À1.59 (.4)*** À1.7 (.44)***
School mean higher Ed likelihood 1.45 (.37)*** 1.02 (.4)*
School mean parents’ Ed  individual
sex approval
À.13 (.05)** À.13 (.06)*
AIC 22,536.3 21,630.3
*p < .05; **p < .01; ***p < .001.
AIC ¼ Akaike Information Criterion; Ed ¼ education; GPA ¼ grade point average;
SE ¼ standard error.
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237236

7. [12] Layzer C, Rosapep L, Barr S. A peer education program: Delivering highly
reliable sexual health promotion messages in schools. J Adolesc Health
2014;54:S70e7.
[13] Kirby D. Emerging answers 2007: Research findings on programs to reduce
teen pregnancy and sexually transmitted diseases. Washington, DC:
National Campaign to Prevent Teen and Unplanned Pregnancy; 2007.
[14] Teitler JO, Weiss CC. Effects of neighborhood and school environments on
transitions to first sexual intercourse. Sociol Educ 2000;73:112e32.
[15] Jordahl T, Lohman BJ. A bioecological analysis of risk and protective factors
associated with early sexual intercourse of young adolescents. Child Youth
Serv Rev 2009;31:1272e82.
[16] Lanctot N, Smith CA. Sexual initiation, pregnancy, and deviance in a
representative urban sample of African-American girls. J Youth Adolesc
2001;30:349e72.
[17] Price MN, Hyde JS. When two isn’t better than one: Predictors of early
sexual initiation in adolescence using a cumulative risk model. J Youth
Adolesc 2009;38:1059e71.
[18] Lohman BJ, Billings A. Protective and risk factors associated with adoles-
cent boys’ early sexual debut and risky sexual behaviors. J Youth Adolesc
2008;37:723e35.
[19] Lynch CO. Risk and protective factors associated with adolescent sexual
initiation. Adolesc Fam Health 2001;2:99e107.
[20] Perkins DF, Luster T, Villarruel FA, et al. An ecological, risk-factor exami-
nation of adolescents’ sexual initiation in three ethnic groups. J Marriage
Fam 1998;60:660e73.
[21] Santelli JS, Kaiser J, Hirsch L, et al. Initiation of sexual intercourse among
middle school adolescents: The influence of psychosocial factors. J Adolesc
Health 2004;34:200e8.
[22] Smith CA. Factors associated with early sexual initiation among urban
adolescents. Soc Work 1997;42:334e46.
[23] Hillis SD, Anda RF, Felitti VJ, et al. Adverse childhood experiences and
sexual risk behaviors in women: A retrospective cohort study. Fam Plann
Perspect 2007;33:206e11.
[24] Jones DJ, Runyan DK, Lewis T, et al. Trajectories of childhood sexual abuse
and early adolescent HIV/AIDS risk behaviors: The role of other maltreat-
ment, witnessed violence, and child gender. J Clin Child Adolesc Psychol
2010;39:667e80.
[25] Resnick MD, Bearman PS, Blum R, et al. Protecting adolescents from harm:
Findings from the National Longitudinal Study on Adolescent Health. JAMA
1997;10:823e32.
[26] Taylor-Seehafer M, Rew L. Risky sexual behavior among adolescent
women. J Soc Pediatr Nurs 2000;5:15e25.
[27] Zimmer-Gembeck MJ, Helfand M. Ten years of longitudinal research on U.S.
adolescent sexual behavior: Developmental correlates of sexual inter-
course, and the importance of age, sex and ethnic background. Dev Rev
2008;28:153e224.
[28] Bearman PS, Bruckner H. Promising the future: Virginity pledges and first
intercourse. Am J Sociol 2001;106:859e912.
[29] Santelli JS, Lindberg LD, Abma J, et al. Adolescent sexual behavior:
Estimates and trends from four nationally representative surveys. Fam
Plann Perspect 2000;32:156e65.
[30] Deptula DP, Henry DB, Schoeny ME. How can parents make a difference?
Longitudinal associations with adolescent sexual behavior. J Fam Psychol
2010;24:731e9.
[31] Dittus PJ, Jaccard J. Adolescents’ perceptions of maternal disapproval of
sex: Relationship to sexual outcomes. J Adolesc Health 2000;26:268e78.
[32] Rose A, Koo HP, Bhaskar B, et al. The influence of primary caregivers on the
sexual behavior of early adolescents. J Adolesc Health 2005;37:135e44.
[33] Pearson J, Muller C, Frisco ML. Parental involvement, family structure, and
adolescent sexual decision making. Sociol Perspect 2006;49:67e90.
[34] Browning CR, Leventhal T, Brooks-Gunn J. Sexual initiation during early
adolescence: The nexus of parental and community control. Am Sociol Rev
2005;70:758e78.
[35] Harris KM, Udry JR. National Longitudinal Study of Adolescent Health (Add
Health), 1994-2008. ICPSR21600-v12. Ann Arbor, MI: Inter-university
Consortium for Political and Social Research; 2013.
[36] Chantala K. Guidelines for analyzing Add Health data. National Longitudi-
nal Study of Adolescent Health, University of North Carolina at Chapel Hill;
Chapel Hill: 2006.
[37] Raudenbush SW, Bryk AS. Hierarchical linear models. 2nd edition.
Thousand Oaks: Sage Publications; 2002.
[38] Singer J. Using SAS PROC MIXED to fit multilevel models, hierarchical
models, and individual growth models. J Educ Behav Stat 1998;24:323e55.
[39] Guilamo-Ramos V, Bouris A, Jaccard J, et al. A parent-based intervention to
reduce sexual risk behavior in early adolescence: Building alliances
between physicians, social workers, and parents. J Adolesc Health 2011;48:
159e63.
[40] Hamm JV, Farmer TW, Lambert K, et al. Enhancing peer cultures of
academic effort and achievement in early adolescence: Promotive effects of
the SEALS intervention. Dev Psychol 2014;50:216e28.
C.N. White and L.A. Warner / Journal of Adolescent Health 56 (2015) 231e237 237