Buffering Effect of Religiosity for Adolescent Substance Use

Psychology of Addictive Behaviors
Copyright 2003 by the Educational Publishing Foundation
2003, Vol. 17, No. 1, 24 –31
0893-164X/03/$12.00
DOI: 10.1037/0893-164X.17.1.24
Buffering Effect of Religiosity for Adolescent Substance Use
Thomas Ashby Wills
Alison M. Yaeger
Albert Einstein College of Medicine of Yeshiva University
Ferkauf Graduate School of Psychology of Yeshiva University
James M. Sandy
Albert Einstein College of Medicine of Yeshiva University
This research examined the hypothesis that religiosity buffers the impact of life stress on adolescent
substance use. Data were from a sample of 1,182 participants surveyed on 4 occasions between 7th grade
(mean age
12.4 years) and 10th grade. Religiosity was indexed by Jessor’s Value on Religion Scale
(R. Jessor & S. L. Jessor, 1977). Zero-order correlations showed religiosity inversely related to alcohol,
tobacco, and marijuana use. Significant Life Events
Religiosity buffer interactions were found in
cross-sectional analyses for tobacco, alcohol, and marijuana use. A latent growth analysis showed that
religiosity reduced the impact of life stress on initial level of substance use and on rate of growth in
substance use over time. Implications for further research on religiosity and substance use are discussed.
Recent investigations have indicated that religiosity is a protec-
shown to reduce the impact of negative life events on adolescent
tive factor with regard to health status. Epidemiologic studies have
substance use (Wills, Vaccaro, & McNamara, 1992). There is
indicated that religiosity is inversely related to adult mortality
some evidence for religiosity as a protective factor, indicated by
rates, and lower rates of substance use among individuals with an
studies showing that measures of religiosity are inversely corre-
involvement in religion have been suggested as contributing to this
lated with indices of adolescent substance use (Wallace & Wil-
mortality differential (Levin, 1996; McCullough, Hoyt, Larson,
liams, 1997), and considerable evidence indicating that life stress
Koenig, & Thoresen, 2000). A relation of religious involvement to
is a risk factor for adolescent alcohol and other drug use (e.g.,
lower rates of alcohol use and problem behavior among adoles-
Chassin, Pillow, Curran, Molina, & Barrera, 1993; Wills, 1990).
cents has also been observed in several areas of the United States
Given the existence of buffering processes it is thus plausible to
(e.g., Bahr, Maughan, Marcos, & Li, 1998; Brody, Stoneman, &
predict that religiosity has a buffering effect for adolescent stres-
Flor, 1996), but at present there is limited understanding of the
sors, reducing the impact of life events on alcohol and other
nature of the relation between religiosity and substance use as it
substance use. In the following sections we outline the background
occurs during adolescence.
of the research and the basis for the hypothesis.
The present research was focused on buffering effects. Studies
of adolescents have shown several examples of factors that reduce
the impact of adverse experiences, that is, buffering effects (Wills,
Religiosity and Adolescent Substance Use
Blechman, & McNamara, 1996). For example, measures of family
support and of problem-solving skills have been shown to reduce
Studies of adolescent substance use have used various defini-
the effect of life stress on outcomes such as adjustment and
tions of religiosity (Wallace & Williams, 1997). Categorical indi-
academic achievement (Dubow & Tisak, 1989; Wolchik, Ruehl-
ces of religious affiliation (vs. none) have shown inconsistent
man, Braver, & Sandler, 1989), and parental support has been
results (e.g., Amey, Albrecht, & Miller, 1996), but consistent
inverse relations with alcohol and other substance use have been
found for measures that tap frequency of attendance at religious
services (Adlaf & Smart, 1985; Amey et al., 1996; Hadaway,
Thomas Ashby Wills and James M. Sandy, Department of Epidemiol-
Elifson, & Petersen, 1984). Effects for religiosity have been found
ogy and Social Medicine, Albert Einstein College of Medicine of Yeshiva
in later adolescence as well as among younger persons (Foshee &
University; Alison M. Yaeger, Ferkauf Graduate School of Psychology of
Yeshiva University.
Hollinger, 1996).
This research was supported by a Research Scientist Development
Studies that index the perceived importance of religion have
Award (K02 DA00252) from the National Institute on Drug Abuse and by
consistently shown inverse relations with substance use. Jessor and
Grant R01 DA05950 from the National Institute on Drug Abuse. We thank
colleagues (Jessor, Chase, & Donovan, 1980; Jessor & Jessor,
the superintendents of the school districts for their support; the parents and
1977) have found that a scale on the perceived importance of
participating students for their cooperation; and Sean Cleary, Elizabeth
religion was inversely related to measures of problem drinking in
Hirky, Grace McNamara, and Donato Vaccaro for assistance with the
regional and national samples of high school students. Inverse
research.
relations of similar measures to frequency of tobacco, alcohol, and
Correspondence concerning this article should be addressed to Thomas
marijuana use have also been found in other samples (Bahr et al.,
Ashby Wills, Department of Epidemiology and Social Medicine, Albert
Einstein College of Medicine of Yeshiva University, 1300 Morris Park
1998; Barnes, Farrell, & Banerjee, 1994; Resnick et al., 1997).
Avenue, Bronx, New York 10461. E-mail: wills@aecom.yu.edu
Although the attributes of the studies varied, results have generally
24

---

BUFFERING EFFECT OF RELIGIOSITY
25
been significant with control for demographic characteristics such
Method
as ethnicity and parental socioeconomic status.
Participants
Buffering Effects in Adult Samples
The data were derived from a school-based longitudinal study in which
a sample of adolescents was surveyed on four occasions from 7th grade
Several studies of adult samples, with psychological distress as
through 10th grade. The participants were students in public school dis-
the outcome, have investigated whether there is evidence for
tricts in the New York metropolitan area. The school districts are in mixed
urban–suburban communities that are indicated by census statistics as
buffering effects of religiosity. Williams, Larson, Buckler, Heck-
socioeconomically representative of the New York State population (U.S.
man, and Pyle (1991) investigated this question with longitudinal
Department of Commerce, 1993). The initial survey administration was
data from a community sample of 720 adults. Multiple regression
done in 7th grade (M age
12.4 years, SD
0.7), when students were in
analyses showed a significant interaction between life stress and
six junior high schools, and continued at yearly intervals through the 9th
religiosity: The positive relation between stress and anxiety/de-
grade and 10th grade, when students were in four high schools. The
pression symptomatology was reduced among persons with a
baseline sample was 29% African American, 23% Hispanic, 3% Asian
higher frequency of religious attendance, that is, a buffering effect.
American, 37% Caucasian, 5% other ethnicity, and 3% mixed ethnicity,
This effect was found both for a measure of general life stressors
and was 47% female and 53% male. Data on family structure indicated that
53% of the participants were living with both biological parents, 34% were
and a measure of physical health problems. Buffering effects were
in a single-parent structure, and 13% were in a blended family (one
found for religious attendance but not for a dichotomous index of
biological parent and one stepparent). Data on parental education indicated
religious affiliation.
that the mode was high school graduate and the mean on a 1-to-6 scale was
Other studies with adults have provided evidence for buffering
3.7 (SD
1.4), a level just above high school graduate.
effects with various indices of symptomatology. A study con-
ducted by Kendler, Gardner, and Prescott (1997) with a commu-
Procedure
nity sample found that the impact of life stress on adults’ depres-
sion was buffered by a measure of the importance of religion, and
A self-report questionnaire was administered to students in classrooms
buffering effects have been found in other studies with interactions
by trained research project staff using a standardized protocol. One class
involving gender (Siegel & Kuykendall, 1990) or denomination
period (approximately 40 min) was allotted for questionnaire administra-
(Park, Cohen, & Herb, 1990). A study of older adults conducted by
tion. The survey was administered under confidential conditions, and the
Strawbridge, Shema, Cohen, Roberts, and Kaplan (1998) found
responses were protected by a Certificate of Confidentiality from the U.S.
buffer interactions for a measure of organizational religiosity with
Public Health Service. Questionnaires were identified only with a code
number. Students were instructed not to write their name on the question-
respect to nonfamily stressors (e.g., financial problems) but not
naire and were assured that their answers were strictly confidential and
with respect to measures of family stressors (e.g., marital prob-
would not be known to their parents or teachers. Methodological research
lems, physical abuse). Finally, a prospective study of older adults
has shown that when participants are assured of confidentiality, self-reports
(Krause, 1997) found buffering effects with mortality as the out-
of substance use have good validity (Murray & Perry, 1987).
come: The effect of life stress on mortality was reduced among
Students participated under a consent procedure in which parents were
people with higher scores on religious coping. In this study, the
sent, by direct mail, a notice that informed them about the purpose of the
buffering effect was found primarily among participants with
research and the nature of the measures. A parent could have his or her
lower levels of education.
child excluded from the research, if he or she wished, by contacting the
investigator or a designated administrator at the school. Students also were
informed about the purpose and nature of the research at the time of
The Present Research
questionnaire administration and were told that they could refuse or dis-
continue participation.
Evidence from studies with adults suggests that religiosity may
The completion rates (number of usable questionnaires
total class
enrollment from school lists) were 92%, 88%, 85%, and 83% for 7th grade
operate as a buffering factor, reducing the impact of life stress on
through 10th grade, respectively. Case loss from parent and student refusal
adverse outcomes. Because buffering processes have been found
was approximately 1% in each wave; the majority of case loss occurred
among children and adolescents (Wills, Blechman, & McNamara,
because of student absenteeism. At the questionnaire administration in 8th
1996), we hypothesized that religiosity would have a buffering
grade, 9th grade, and 10th grade, students new to the schools were included
effect for the relation between life stress and adolescent substance
in the surveys. The size of the sample was 1,702 cases in 7th grade, 1,827
use. We investigated this prediction in a representative sample of
cases in 8th grade, 1,895 cases in 9th grade, and 1,699 cases in 10th grade.
adolescents studied over the period from 12 to 16 years of age.
For longitudinal data collection, the retention rate for the study variables
In this research, religiosity was indexed with a measure on the
was approximately 70%.
perceived importance of religion, a construct that prior research
has indicated to have consistent relations with adolescent sub-
Measures
stance use. We tested the relation of religiosity with measures of
overall alcohol use and heavy drinking, as well as with measures
The questionnaire began with demographic items. The participant was
of tobacco and marijuana use, and considered whether the relation
asked about his or her age, gender, and ethnicity (five options, multiple
responding allowed). An item on family structure asked the participant
was found in various demographic subgroups. We initially tested
what adult(s) he or she was currently living with (eight options, multiple
the hypothesized buffering effect in cross-sectional analyses with
responding allowed); this was recoded for analysis to three levels (single
multiple regression and then examined moderation effects in a
parent, blended family, or intact family). Items about parental education
longitudinal analysis based on latent growth modeling of four-
had the anchor points grade school and post-college (master’s or doctoral
wave data on substance use.
degree or other professional education).

---

26
WILLS, YAEGER, AND SANDY
Religiosity.
Religiosity was indexed with Jessor’s Value on Religion
higher compared with Caucasians (p
.0001). An effect of
Scale (Jessor & Jessor, 1977). A lead-in instruction stated, “Here are some
ethnicity indicated that both African Americans and Hispanics
questions on what you think about things. Read each one, and circle a
experienced more life events compared with Caucasians (p
number to show what you think.” Responses were made on scales that
.0001), and families with lower levels of education experienced
ranged from 1 to 4, with response points “not at all important,” “a little
more life events compared with families with more education (p
important,” “pretty important,” and “very important.” Internal consistency
.001). Participants from single-parent and blended families expe-
reliability (Cronbach’s alpha) was .78 –.81 over assessments. The items
rienced more adolescent-related events (p
.0001). For substance
were “To believe in God,” “To be able to rely on religious teachings when
you have a problem,” “To be able to turn to prayer when you’re facing a
use, the most consistent effects were for ethnicity, with Caucasians
personal problem,” and “To rely on your religious beliefs as a guide for
showing the highest rates of alcohol and other substance use,
day-to-day living.”
Hispanics showing intermediate rates, and African Americans
Negative life events.
A 20-item inventory based on previous measures
showing the lowest rates. Participants from both single-parent and
of adolescent stressors (Newcomb & Harlow, 1986; Wills et al., 1992) was
blended families had somewhat higher rates of substance use
administered with a dichotomous (no–yes) response scale to describe
compared with participants from intact families. Gender effects
events that occurred during the previous year. A subscale of 11 family
were not totally consistent but tended to show girls with higher
events, those that could have occurred to a family member and did not
rates of smoking and boys with higher rates of marijuana use.
directly involve the respondent (e.g., “Father/mother was unemployed”)
Zero-order correlations of religiosity with the substance use
had alphas of .58 –.62. A 9-item scale of adolescent events, those that could
indices and the composite score are presented in Table 1. A
occur directly to the respondent (e.g., “I had a serious illness”) had alphas
of .54 –.61.1
significant inverse correlation with overall alcohol use was found
Adolescent alcohol and other substance use.
Substance use by the
at the initial assessment and at all subsequent assessment points.
participant was measured with items that asked about the typical frequency
Relations of religiosity with the other indices were significant for
of his or her alcohol, tobacco, and marijuana use. Three items were
tobacco and marijuana use from 8th grade onward and for heavy
introduced to participants with the stem: “How often do you smoke
drinking from 9th grade onward. These results are consistent with
cigarettes/drink alcohol/smoke marijuana?” Responses were made on
previous research on religiosity in adolescence (Bahr et al., 1998;
scales that ranged from 0 to 5, with scale points “never used,” “tried
Wallace & Williams, 1997). The results also suggest that the effect
once–twice,” “used four–five times,” “usually use a few times a month,”
of religiosity on substance use becomes more generalized over the
“usually use a few times a week,” and “usually use every day.” An item on
period from early through middle adolescence.
heavy drinking asked the participant whether in the past month he or she
had had three or more drinks on one occasion; response points were “no,”
“happened once,” “happened twice,” and “happened more than twice.” The
Test for Buffering Effects
indices of cigarette, alcohol, and marijuana use were intercorrelated, con-
We had hypothesized that religiosity would have a buffering
sistent with prior methodological research (e.g., Needle, Su, & Lavee,
1989). Alpha for a 4-item composite score was .60 –.80 over assessments.
effect for adolescent substance use with respect to negative life
events. We initially tested this in multiple regression analyses with
substance use as the criterion, including terms for religiosity and
Results
life events and the interaction of Life Events
Religiosity. Be-
Prevalence data indicated that regular substance use had fairly
cause of the demographic effects, the regression models included
low levels at 7th grade and a steady increase over the study period.
binary indices for gender, ethnicity (African American vs. His-
For example, the proportions of participants who smoked weekly
panic or Caucasian and Hispanic vs. African American or Cauca-
or more often were 2%, 6%, 12%, and 16% for 7th through 10th
sian), and family structure (single vs. blended or intact and blended
grade, respectively, and the proportions of participants who had
vs. single or intact).2
engaged in heavy drinking more than once in the past month were
Following the procedure outlined by Aiken and West (1991), we
3%, 5%, 10%, and 15% for 7th through 10th grade, respectively.
first standardized the predictor variables and formed the cross-
A similar pattern, with initially low level and a steady increase
product by multiplying the standardized terms, and the criterion
over time, was found for the indices of overall alcohol use and
variable was also standardized. The two (standardized) main-effect
marijuana use. These rates are generally comparable to data from
terms and the cross-product term were then entered together in
other studies (Johnston, O’Malley, & Bachman, 1995).
multiple regression, with substance use score as the criterion, and
Descriptive statistics for the religiosity measure indicated that it
unstandardized regression coefficients from this analysis are re-
had negative skewness, as the majority of the sample reported
ported (Aiken & West, 1991, pp. 40 – 44). We performed analyses
placing a moderate value on religion. For 7th grade, the scale (with
for each of the four assessments to test for replicability of effects.
a possible range of 4 –16) had M
12.21 (SD
3.18) for 8th
To minimize the number of statistical tests, we based the primary
grade, M
11.97 (SD
3.20) for 9th grade, M
11.92 (SD
analysis on the composite substance use score, but we also per-
3.29), and M
11.72 (SD
3.57) for 10th grade. Skewness
formed analyses for each of the substance use indices. Because
values were – 0.60, – 0.57, – 0.52, and – 0.51 for 7th through 10th
Strawbridge et al. (1998) suggested that buffering interactions
grade, respectively. These values are consistent with data for
similar measures from other studies (Jessor et al., 1980; Jessor &
1 The study included other measures, such as parental support and
Jessor, 1977).
adolescent coping (see Wills, McNamara, Vaccaro, & Hirky, 1996), which
Relations of demographic characteristics to study variables were
are beyond the scope of the present article.
examined in analyses of variance. We focus on results that were
2 Parental education was inversely related to smoking but was not
consistent over assessments. For religiosity, significant effects
consistently related to value on religion or to other substance use indices,
were found for gender, with girls scoring higher (p
.001), and
so it was not included in the analyses. Results from analyses including
for ethnicity, with African Americans and Hispanics both scoring
parental education were quite similar to those reported here.

---

BUFFERING EFFECT OF RELIGIOSITY
27
might be found only for family-related life events, we performed
Table 2
subsidiary analyses separately for family events and adolescent
Coefficients From Multiple Regression Analysis for Religiosity
events.
and Life Events, With Substance Use as the Criterion, for Four
Results are presented in Table 2. Religiosity had an inverse
Grade Levels
relation to the composite substance use score at each assessment
point, and life events had a positive relation to the substance use
Grade
score at each point. Life Events
Religiosity interactions, con-
Variable
7th
8th
9th
10th
sistent in form with buffering effects, were observed for the
composite substance use score at 7th grade, 8th grade, and 9th
Religiosity
.07**
.09***
.09***
.08***
grade. At the 10th-grade assessment, significant buffer interactions
Life events
.33****
.32****
.36****
.32****
Events
Religiosity
.05*C,H
.07**C,H
.05*C,M
.04H,M
were found for heavy drinking and marijuana use, but the inter-
R2
.12
.13
.15
.15
action for the composite substance use score was nonsignificant.3
For the individual substance use indices, buffering effects were
Note.
Values are coefficients from a regression model with composite
found for heavy drinking, cigarette smoking, and marijuana use
substance use score as the criterion, including terms for gender, ethnicity,
but not for the index of overall alcohol use, possibly because this
and family structure (not shown in table). N for analysis is 1,440 –1,580
cases. Superscript letters indicate that a significant interaction was ob-
taps a more normative aspect of drinking. Effects for demographic
served for an individual substance use index; C
cigarette smoking, H
variables were found consistently only for the ethnicity indices,
heavy drinking, M
marijuana use.
which indicated that African Americans and Hispanics had lower
* p
.05.
** p
.01.
*** p
.001.
**** p
.0001.
levels of substance use, relative to Caucasians. Over assessments,
the regression models accounted for 12%–15% of the variance in
use at 7th grade, 8th grade, 9th grade, and 10th grade were
the substance use score.
specified as a growth model with constructs for intercept, the
We graphed the interactions using the method of J. Cohen and
initial level of substance use, and slope, the rate of change in
Cohen (1983), with estimated values of substance use plotted for
substance use over time (Windle, 1997). The intercept construct
cases at M
1 SD on the predictor variables. Graphs for the
was specified by setting the loadings of the four observed values of
composite substance use score are portrayed in Figure 1 for 7th
substance use on the intercept to 1, and the slope construct was
grade through 9th grade, respectively. The form of the interactions
specified by setting the loadings of the four observed values of
was a buffering process: The effect of life events on substance use
substance use on the slope to 0, 1, 2, and 3, representing the
was reduced among participants with higher religiosity.
assumption of linear growth with equal spacing of assessments
We addressed demographic differences by analyzing the inter-
over time. Exogenous variables were life events at Grade 7 to-
action of a demographic characteristic with religiosity predicting
gether with five demographic indices (for gender, ethnicity, and
the composite substance use score (e.g., Gender
Religiosity).
family structure). Covariances among the exogenous variables,
We then repeated these analyses across the four assessments. We
and a covariance of the intercept and slope constructs, were also
did not see consistent evidence that either the main effect or the
specified.
buffering effect of religiosity was observable in one demographic
Path effects were specified through regressing the intercept and
group but not in others. Although a difference could sometimes be
slope constructs on the exogenous variables. Analyses were per-
observed at one assessment, often it was not replicated at other
formed in Mplus Version 2.2 using the maximum likelihood
assessments. The most consistent difference was a larger main
method with the Expectation Maximization algorithm for missing
effect and buffering effect for girls compared with boys, but this
data (Muthe´n & Muthe´n, 1998). The model for the total sample
difference was found only at 9th and 10th grade and not at earlier
had reasonable fit to the data,
2(23, N
1,182)
49.14, and
assessments.
root-mean-square error of approximation (RMSEA) less than .050
(RMSEA
.031, confidence interval: .019 –.043).4 Significant
Longitudinal Analysis of Buffer Effects
paths for life stress indicated that negative life events were related
The buffering effect of religiosity was tested in a longitudinal
both to higher initial level of substance use (
.34), t(1180)
context using latent growth modeling. Observations of substance
3 Measures for family and adolescent events were correlated between .4
and .5, so subsidiary analyses used regression models with religiosity, one
Table 1
main-effect term for events, and the cross-product (e.g., Family Events
Correlation of Religiosity Measure With Substance Use Indices
Religiosity). For family events, significant interactions (p
.05) were
for Four Assessments
found at three time points. For adolescent events, significant interactions
(p
.05) were found at 7th grade and 8th grade; the interaction term was
Grade
nonsignificant at the two other time points. Thus the overall interaction was
Index
7th
8th
9th
10th
not attributable to just one type of event, and there was evidence for
buffering of family events.
Composite score
.07**
.11****
.10****
.13****
4 We performed analyses for participants with at least three values of
Alcohol use
.13****
.14****
.10****
.15****
substance use and a 7th-grade value for religiosity, an analytic sample of
Cigarette smoking
.02
.06**
.07**
.07**
1,182 cases. No buffering effects were found for overall alcohol use, so
Marijuana use
.01
.09****
.09****
.10****
the substance use measure for the growth analyses was based on the sum
Heavy drinking
.01
.01
.05*
.11****
of the scores for heavy drinking, cigarette smoking, and marijuana use at
Note.
N for correlations is 1,700 –1,800 cases.
a given time point. Slope loadings for the 9th and 10th grades were freely
* p
.05.
** p
.01.
**** p
.0001.
estimated.

---

28
WILLS, YAEGER, AND SANDY
11.07, p
.0001, and to greater rate of increase in substance use
over time (
.26), t(1180)
6.33, p
.0001. Significant paths
for demographic variables indicated a lower rate of growth in
substance use for African Americans (
–.20), t(1180)
5.07,
p
.0001, and for Hispanics (
–.14), t(1180)
3.67, p
.001. The predictors accounted for 12% of the variance in initial
level of use and 9% of the variance in rate of growth in use.
We tested the buffering effect of religiosity with a multiple-
group analysis. We divided the sample into subgroups based on
religiosity score from 7th grade, and we analyzed the growth
model for different subgroups as specified above (Rigdon, Schu-
macker, & Wothke, 1998).5 The base model, analyzed simulta-
neously in two subgroups with all parameters freely estimated, is
presented in Figure 2 with coefficients for the low- and high-
religiosity subgroups; unstandardized coefficients are presented
because standardized coefficients cannot be directly compared
across subgroups. The paths from life events to substance use
intercept and slope were both reduced in the high-religiosity sub-
group. We evaluated this effect with a multiple-group test, con-
straining these two coefficients to be equal across subgroups. This
test resulted in a difference chi-square (df
2) of 7.48, which is
significant (p
.025), and indicates that the impact of life events
on intercept and slope was significantly lower for the high-
religiosity subgroup. Tests for the individual paths indicated that
the difference chi-square (df
1) for the path from life events to
substance use intercept was 3.87 (p
.05), and the difference
chi-square (df
1) for the path from life events to substance use
slope was 4.04 (p
.05); hence each of the paths differed signif-
icantly according to level of religiosity. Thus, the buffering effect
of religiosity was demonstrated in a longitudinal context with
control for demographic characteristics.6 The ratio of the paths
from life events to substance use constructs in the high- versus
low-religiosity groups, an index of the moderation effect size, was
.71 for the path from life events to substance use intercept and .59
for the path from life events to substance use slope.
Discussion
The aim of this research was to test the prediction that religiosity
has a buffering effect for alcohol and other substance use over the
period from early to middle adolescence. The data we analyzed
were obtained from a representative sample of urban adolescents.
Inverse correlations of religiosity with adolescents’ alcohol, to-
bacco, and marijuana use were found. Results from multiple re-
gression analyses showed significant Life Events
Religiosity
5 We tested subgroupings based on median split, tertiles, and quartiles
(Rigdon et al., 1998), and found that the interaction effect was best
represented by the contrast of the lowest tertile on religiosity with the upper
two tertiles. Hence, we performed the multiple-group modeling as a two-
group analysis, which contrasted the lowest one third of the sample
(analytic n
357 cases) with the upper two thirds of the sample (analytic
n
825 cases).
Figure 1.
Interactions for Life Events
Religiosity. Panel A: 7th grade;
6 A separate multiple-group test with family events had a difference
Panel B: 8th grade; Panel C: 9th grade. Plotted are estimated values of
chi-square (df
2) of 5.72 (p
.05), and a separate test with adolescent
composite substance use score for cases at various combinations of pre-
events had a difference chi-square (df
2) of 5.66 (p
.05). In each case,
dictor variables. Note that the scale of the y-axis differs across grades
the two paths from events to intercept and slope were lower for the
because rates of substance use increase over time.
high-religiosity subgroup.

---

BUFFERING EFFECT OF RELIGIOSITY
29
Figure 2.
Multiple-group latent growth model for adolescent substance (Sub.) use from 7th grade to 10th grade,
with life events at 7th grade exogenous. Straight single-headed arrows are path effects; the curved double-headed
arrow is a covariance. Values are unstandardized coefficients (standard errors are in parentheses) except for
factor loadings, which are fixed. Coefficients above the line are for the low-religiosity group; coefficients below
the line are for the high-religiosity group.
buffering interactions, consistent with prediction. Buffering inter-
2000), and comprising domains of formal religious involvement
actions were observed for cigarette smoking, heavy drinking, and
and nonreligious spirituality that may not be highly correlated
marijuana use, with effects replicated over the study period, and
(John E. Fetzer Institute, 1999). These different aspects of religi-
these findings were obtained with control for demographic char-
osity could be compared in further research. Finally, we studied
acteristics. The buffering effect of religiosity was examined in a
religiosity and substance use during the early part of adolescence,
longitudinal context with latent growth modeling, and moderation
and studies conducted to investigate effects of religiosity in ado-
was observed for the impact of life events on initial level of
lescence or early adulthood would be desirable.
substance use and on rate of growth in use over time.
Consideration of effect sizes indicated that although the cross-
Analyses for Demographic and Events Subgroups
sectional correlations of religiosity with substance use (Table 1)
were modest in magnitude for individual indices, effects were
Specific analyses indicated that buffering effects were observed
observed for several indices and were consistent over time. The
at some point in the study for all demographic subgroups. There
effect size for the moderation for latent growth constructs (Figure
was a suggestion that effects of religiosity were more prominent
2) was such that the effect of life stress on rate of increase in
for females, but this difference was not totally consistent over the
substance use was about half as large in the high-religiosity group.
study period. Although previous research with adults has sug-
Thus, this research demonstrates that buffering effects of religios-
gested demographic differences, these were predominantly cross-
ity for adolescent substance use are a real phenomenon and are of
sectional studies, and there is little evidence of the replicability of
a magnitude that may have preventive significance.
demographic effects. In the present research, buffering effects
The present research had methodological advantages in that it
were observed in the total sample at several time points and for
was based on a sizable sample, a desirable attribute for demon-
several substance use indices, so given the number of demographic
strating interaction effects (Aiken & West, 1991; Chaplin, 1991).
subgroups in the sample it was no simple matter to conduct
In addition, the study tapped a range of life events and obtained
comprehensive tests for demographic differences while also min-
repeated assessments of the participants so that replicability of
imizing the number of statistical tests. Our approach relied on the
effects could be determined. We found that the basic effects for
observed replicability of effects for a composite substance use
religiosity were fairly replicable, although there was evidence that
score, and although this approach may be conservative, we con-
the main effects increased over the study period, whereas the
clude that effects of religiosity in adolescence are not limited to a
buffering effects seemed to be strongest during early adolescence.
particular demographic subgroup. In further research it would be
These observations may represent developmental trends but would
desirable to make efforts to determine the replicability as well as
need to be replicated in other studies in order to build a firmer base
the nature of demographic effects.
for developmental theory about the effects of religiosity.
We examined a suggestion from adult research that buffering
Some aspects of the present research could be noted as possible
effects occur only for family-related events (Strawbridge et al.,
limitations. This study assessed frequency of alcohol and other
1998). This was perhaps not a strong test (because of the intercor-
substance use, and the relation of religiosity to diagnostic indices
relation of event subtypes), but for adolescents we were unable to
of substance abuse or dependence may be considered in further
support the notion because buffering effects were observed with
research. Religiosity was indexed with one type of measure, but
respect to events that primarily affected the family as well as for
recent theoretical work suggests that the broader construct of
events that directly affected the adolescent him- or herself. We
religiousness is multidimensional, including facets of belief, par-
note that there are several differences between the studies; for
ticipation, and coping (George, Larson, Koenig, & McCullough,
example, the participants in Strawbridge et al.’s (1998) study were

---

30
WILLS, YAEGER, AND SANDY
50 years or older, the measure of religiosity was somewhat differ-
(1998). Drug use among African Americans: Ethnic identity as a pro-
ent, and the nonfamily events were of a largely different nature.
tective factor. Psychological Reports, 83, 1427–1446.
However, in future research it would seem justified to consider a
Chaplin, W. F. (1991). The next generation of moderator research in
variety of life events spanning the range of individual and family
personality psychology. Journal of Personality, 59, 143–178.
domains.
Chassin, L., Pillow, D. R., Curran, P. J., Molina, B. S. G., & Barrera, M.
(1993). Relation of parental alcoholism to early adolescent substance
use. Journal of Abnormal Psychology, 102, 3–19.
The Process of Buffering
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation
analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Buffering effects for adolescent substance use were observed
Cohen, S., & Wills, T. A. (1985). Stress, social support, and the buffering
consistently in this study. These effects could occur through one or
hypothesis. Psychological Bulletin, 98, 310 –357.
more different processes. In theory, a protective resource could act
Dubow, E. F., & Tisak, J. (1989). The relation between stressful life events
to directly counter the impact of a stressor; for example, financial
and adjustment in elementary school children: The role of social support
support would directly reduce the impact of financial problems
and problem-solving skills. Child Development, 60, 1412–1423.
(Peirce, Frone, Russell, & Cooper, 1996). However, we think it
Foshee, V. A., & Hollinger, B. R. (1996). Maternal religiosity, adolescent
unlikely that religiosity operates in such a manner, because it
social bonding, and adolescent alcohol use. Journal of Early Adoles-
probably affects multiple psychosocial domains. At the individual
cence, 16, 451– 468.
level, buffering could occur because religiosity affects attitudes
George, L. K., Larson, D. B., Koenig, H. G., & McCullough, M. E. (2000).
and values. For example, religiosity may be related to perceived
Spirituality and health: What we know, what we need to know. Journal
meaning and purpose in life (George et al., 2000; Newcomb &
of Social and Clinical Psychology, 19, 102–116.
Harlow, 1986) and could also be related to values and attitudes
Hadaway, C. K., Elifson, K. W., & Petersen, D. M. (1984). Religious
involvement and drug use among urban adolescents. Journal for the
about substance use (Brody et al., 1996). These factors could
Scientific Study of Religion, 23, 109 –128.
moderate the impact of negative life events through cognitive or
Jessor, R., Chase, J. A., & Donovan, J. E. (1980). Psychosocial correlates
attitudinal mechanisms.
of marijuana use and problem drinking in a national sample of adoles-
In theory, buffering might also occur because of relations to
cents. American Journal of Public Health, 70, 604 – 613.
coping processes, social networks, or both. Religiosity may influ-
Jessor, R., & Jessor, S. L. (1977). Problem behavior and psychosocial
ence the way people tend to cope with problems and their percep-
development. New York: Academic Press.
tions about the coping functions of substance use (Pargament,
John E. Fetzer Institute. (1999). Multidimensional measurement of reli-
1997; Wills & Hirky, 1996), hence an indirect mechanism through
giousness/spirituality for use in health research. Kalamazoo, MI:
coping processes could act to alter the effects of life stress on
Author.
various outcomes. In the domain of social processes, religiosity
Johnston, L. D., O’Malley, P. M., & Bachman, J. G. (1995). National
could be associated with the characteristics of an adolescent’s
survey results on drug use from the Monitoring the Future study,
1975–1994. Volume 1: Secondary school students. Rockville, MD: Na-
network of adults and peers and may be related to integration in the
tional Institute on Drug Abuse.
larger community through participation in social and service ac-
Kendler, K. S., Gardner, C. O., & Prescott, C. A. (1997). Religion,
tivities (Brook, Balka, Brook, Win, & Gursen, 1998; Umberson,
psychopathology, and substance use and abuse: A multimeasure,
1987; Wallace & Williams, 1997), a factor that also could work to
genetic– epidemiologic study. American Journal of Psychiatry, 154,
produce buffering effects (S. Cohen & Wills, 1985). Previous
322–329.
research and theory have emphasized that factors related to sub-
Krause, N. (1997). Stressors in highly valued roles, religious coping, and
stance use can be moderated in several different ways (e.g., Wills
mortality. Psychology and Aging, 13, 242–255.
& Cleary, 1996; Zucker, 1994), so for further research it seems
Levin, J. S. (1996). How religion influences morbidity and health: Reflec-
appropriate to consider multiple processes through which religios-
tions on natural history, salutogenesis and host resistance. Social Science
ity can affect functioning.
and Medicine, 43, 849 – 864.
McCullough, M. E., Hoyt, W. T., Larson, D. B., Koenig, H. G., &
Thoresen, C. (2000). Religious involvement and mortality: A meta-
References
analytic review. Health Psychology, 19, 211–222.
Adlaf, E. M., & Smart, R. G. (1985). Drug use, religious affiliation,
Murray, D. M., & Perry, C. L. (1987). The measurement of substance use
feelings, and behavior. British Journal of the Addictions, 80, 163–171.
among adolescents. Addictive Behaviors, 12, 225–233.
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and
Muthe´n, L. K., & Muthe´n, B. O. (1998). Mplus user’s guide. Los Angeles:
interpreting interactions. Newbury Park, CA: Sage.
Muthe´n & Muthe´n.
Amey, C. H., Albrecht, S. L., & Miller, M. K. (1996). Racial differences
Needle, R., Su, S., & Lavee, Y. (1989). A comparison of the empirical
in adolescent drug use: The impact of religion. Substance Use & Misuse,
utility of three composite measures of adolescent drug involvement.
31, 1311–1332.
Addictive Behaviors, 14, 429 – 441.
Bahr, S. J., Maughan, S. L., Marcos, A. C., & Li, B. (1998). Family,
Newcomb, M. D., & Harlow, L. L. (1986). Life events and substance use
religiosity, and the risk of adolescent drug use. Journal of Marriage and
among adolescents: The role of loss of control and perceived meaning-
the Family, 60, 979 –992.
lessness in life. Journal of Personality and Social Psychology, 51,
Barnes, G. M., Farrell, M. P., & Banerjee, S. (1994). Family influences on
564 –577.
alcohol abuse and other problem behaviors among Black and White
Pargament, K. I. (1997). The psychology of religion and coping: Theory,
adolescents in a general population sample. Journal of Research on
research, practice. New York: Guilford Press.
Adolescence, 4, 183–201.
Park, C., Cohen, L. H., & Herb, L. (1990). Intrinsic religiousness and
Brody, G. H., Stoneman, Z., & Flor, D. (1996). Parental religiosity, family
religious coping as life stress moderators for Catholics versus Protes-
processes, and youth competence in rural, two-parent African-American
tants. Journal of Personality and Social Psychology, 59, 562–574.
families. Developmental Psychology, 32, 696 –706.
Peirce, R. S., Frone, M. R., Russell, M., & Cooper, M. L. (1996). Financial
Brook, J. S., Balka, E. B., Brook, D. W., Win, P. E., & Gursen, M. D.
stress, social support, and alcohol involvement: A longitudinal test of the

---

BUFFERING EFFECT OF RELIGIOSITY
31
buffering hypothesis in a general-population survey. Health Psychology,
Wills, T. A., Blechman, E. A., & McNamara, G. (1996). Family support,
15, 38 – 47.
coping and competence. In E. M. Hetherington & E. A. Blechman
Resnick, M. D., Bearman, P. S., Blum, R. W., Bauman, K. E., Harris,
(Eds.), Stress, coping, and resiliency in children and the family (pp.
K. M., Jones, J., et al. (1997). Findings from the National Longitudinal
107–133). Hillsdale, NJ: Erlbaum.
Study on Adolescent Health. Journal of the American Medical Associ-
Wills, T. A., & Cleary, S. D. (1996). How are social support effects
ation, 278, 823– 832.
mediated: A test for parental support and adolescent substance use.
Rigdon, E. E., Schumacker, R. E., & Wothke, W. (1998). A comparative
Journal of Personality and Social Psychology, 71, 937–952.
review of interaction and nonlinear modeling. In R. E. Schumacker &
Wills, T. A., & Hirky, A. E. (1996). Coping and substance abuse. In M.
G. A. Marcoulides (Eds.), Interaction and nonlinear effects in structural
Zeidner & N. S. Endler (Eds.), Handbook of coping (pp. 279 –302). New
equation modeling (pp. 1–16). Mahwah, NJ: Erlbaum.
York: Wiley.
Siegel, J. M., & Kuykendall, D. H. (1990). Loss, widowhood, and psycho-
Wills, T. A., McNamara, G., Vaccaro, D., & Hirky, A. E. (1996). Escalated
logical distress among the elderly. Journal of Consulting and Clinical
substance use: A longitudinal grouping analysis from early to middle
Psychology, 58, 519 –524.
adolescence. Journal of Abnormal Psychology, 105, 166 –180.
Strawbridge, W. J., Shema, S. J., Cohen, R. D., Roberts, R. E., & Kaplan,
Wills, T. A., Vaccaro, D., & McNamara, G. (1992). The role of life events,
G. A. (1998). Religiosity buffers effects of some stressors on depression
family support, and competence in adolescent substance use. American
but exacerbates others. Journal of Gerontology: Social Sciences, 53B,
Journal of Community Psychology, 20, 349 –374.
S118-S126.
Windle, M. (1997). Latent variable approaches to modeling change in
Umberson, D. (1987). Family status and health behaviors: Social control as
adolescent alcohol involvement. In K. J. Bryant, M. Windle, & S. G.
a dimension of social integration. Journal of Health and Social Behav-
West (Eds.), The science of prevention: Methodological advances from
ior, 28, 306 –319.
alcohol and substance abuse research (pp. 43–78). Washington, DC:
U.S. Department of Commerce. (1993). 1990 Census of the Population:
American Psychological Association.
Social and economic characteristics (New York). Washington, DC: U.S.
Wolchik, S. A., Ruehlman, L. S., Braver, S. L., & Sandler, I. N. (1989).
Government Printing Office.
Wallace, J. M., Jr., & Williams, D. R. (1997). Religion and adolescent
Social support of children of divorce. American Journal of Community
health-compromising behavior. In J. Schulenberg, J. L. Maggs, & K.
Psychology, 17, 485–501.
Hurrelmann (Eds.), Health risks and developmental transitions during
Zucker, R. A. (1994). Pathways to alcohol problems. In R. A. Zucker,
adolescence (pp. 444 – 468). New York: Cambridge University Press.
G. M. Boyd, & J. Howard (Eds.), The development of alcohol problems
Williams, D. R., Larson, D. B., Buckler, R. E., Heckman, R. C., & Pyle,
(pp. 255–289). Rockville, MD: National Institute on Alcohol Abuse and
C. M. (1991). Religion and psychological distress in a community
Alcoholism.
sample. Social Science and Medicine, 32, 1257–1262.
Wills, T. A. (1990). Stress and coping factors in the epidemiology of
substance use. In L. T. Kozlowski, H. M. Annis, H. D. Cappell, F. B.
Received July 30, 2001
Glaser, M. S. Goodstadt, Y. Israel, et al. (Eds.), Research advances in
Revision received January 11, 2002
alcohol and drug problems (Vol. 10, pp. 215–250). New York: Plenum.
Accepted January 18, 2002

---