Prevention Science, Vol. 1, No. 4, 2000
Modeling Prevention Program Effects on Growth in
Substance Use: Analysis of Five Years of Data from the
Adolescent Alcohol Prevention Trial
Bonnie J. Taylor,
John W. Graham,
and William B. Hansen
The efﬁcacy of prevention programs is typically determined through analysis of covariance.
To date, a growth curve modeling approach is not used extensively in program evaluation.
However, for longitudinal data there are several advantages to using this approach as com-
pared to methods comparing means at two time points in a piecemeal fashion. In this study,
latent growth curve models were used to evaluate the effect of a program on the average
level of drug use, rate of change (growth) of drug use, and acceleration or deceleration in
the rate of change of drug use. The study relied on data from the Adolescent Alcohol
Prevention Trial, a randomized longitudinal drug use prevention program. The program
consists of drug use information, resistance skills training, and normative education compo-
nents. Data regarding cigarette and alcohol use were collected over a 5-year period, grade
7 to grade 11. Students receiving the normative education program had signiﬁcantly lower
average levels of reported cigarette and alcohol use, lower rates of growth for reported
cigarette and alcohol use, and less deceleration of reported levels of cigarette and alcohol
use as compared with the control group. Growth curve analysis is a powerful and effective
tool with which to model change and program efﬁcacy.
KEY WORDS: latent growth model; program evaluation; prevention; alcohol use; adolescents.
Evaluating the efﬁcacy of social psychology–
based prevention programs has been an integral part
of program development since the late 1970’s. For
evaluation purposes, researchers have typically col-
lected data on participants prior to program imple-
mentation and follow-up data on the same individuals
for up to several years after the program (Biglan et
al., 1987; Botvin & Eng, 1982; Ellickson & Bell, 1990;
Flay, d’Avernas, Best, Kersell, & Ryan, 1983; Han-
The Methodology Center, The Pennsylvania State University.
Department of Biobehavioral Health, The Pennsylvania State
Department of Human Development and Family Studies, The
Pennsylvania State University.
Tanglewood Research, Inc.
Correspondence should be directed to the Department of Biobe-
havioral Health, E315 Health & Human Development Building,
Penn State University, University Park, Pennsylvania 16802.
1389-4986/00/1200-0183$18.00/1 2000 Society for Prevention Research
sen, 1992; Hansen & Graham, 1991; Hansen, John-
son, Flay, Graham, & Sobel, 1988; Luepker, Johnson,
Murray, & Pechacek, 1983; Murray, Richards,
Luepker, & Johnson, 1987; Pentz et al., 1989; and
numerous others as cited in Flay, 1985). Despite the
collection of longitudinal data, the primary mode of
analysis for these studies has been some variant of
analysis of covariance (ANCOVA), examining group
differences at speciﬁc time points.
Some researchers have modeled the process of
the program’s effect on drug use outcomes by using
three waves of data (e.g., MacKinnon et al., 1991; also
see Donaldson, Graham, & Hansen, 1994; Graham,
Hofer, Donaldson, MacKinnon, & Schafer, 1997), but
even with these three-wave analyses, multiple years
of outcome data were analyzed in piecemeal fashion,
one outcome year at a time. Thus, although these
analyses are longitudinal in tracing the process of the
program’s effect through the mediating variable, they