V. SPECIAL TOPICS
A. Subject Attrition and Retention
- Biglan, A., Hood, D., Brozovsky, P., Ochs, L., Ary,
D., & Black, C. Subject attrition in prevention research. In:
Leukefeld, C.G., & Bukoski, W.J., eds. Drug Abuse Prevention
Intervention Research: Methodological Issues. National Institute on
Drug Abuse Research Monograph 107. DHHS Pub. No. (ADM)91-1761.
Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1991, pp.
213-234.
The chapter discusses the role of subject attrition
in substance abuse prevention research. Attrition may affect the validity
of experimental comparisons and may limit the extent to which findings
can be generalized to adolescents at highest risk. The authors examine
concerns about subject attrition, present methods for analyzing attrition
in evaluations of prevention programs, and make recommendations for
minimizing the extent and impact of attrition in such evaluations.
- Ellickson, P.L., Bianca, D., & Shoeff, D.C. Containing
attrition in school-based research: An innovative approach. Eval
Rev 12:331-351, 1988.
This article describes a successful approach for tracking
a highly mobile group of junior high school transferees and thereby
minimizing attrition in a longitudinal study of adolescent behavior.
Students were tracked through the home or the new school. When students
were located through the latter route, the authors sent the surveys
directly to the school itself instead of asking for a home mailing
address. This approach avoided asking school officials to give out
personal information and enhanced the likelihood of the survey being
delivered. Overall, the tracking effort cut nonresponse attributable
to between-school mobility by two-thirds and reduced the attrition
rate by one-half. The new-school strategy, which was a particularly
effective technique for finding student transferees, accounted for
a significant proportion of that improvement.
- Hansen, W., Tobler, N., & Graham, J. Attrition
in substance abuse prevention research: A meta-analysis of 85 longitudinally
followed cohorts. Eval Rev 14(6):677-685, 1990.
This meta-analysis of substance abuse prevention studies
reveals that the mean proportion of subjects retained dropped from
81.4 percent at 3-month followup to 67.5 percent at 3-year followup.
Time from pretest alone accounted for less than 5 percent of the variance.
Other available predictors of retention were not significant. Researchers
are encouraged to interpret their results in light of these normative
data and to adopt second-effort strategies to reduce attrition.
- Willett, J.B., & Singer, J.D. From whether to when:
New methods for studying student dropout and teacher attrition. Rev
Educ Res 61(4):407-450, 1991.
The authors shows how the methods of survival analysis
(also known as event history analysis) lend themselves naturally to
the study of the timing of educational events. Drawing examples from
the literature on teacher attrition and student dropout and graduation,
the authors introduce a panoply of survival methods useful for describing
the timing of educational transitions, e.g., student dropout and teacher
attrition, and for building statistical models of the risk of event
occurrence over time.
- Streissguth, A.P., & Giunta, C.T. Subject recruitment
and retention for longitudinal research: Practical considerations
for a nonintervention model. In: Kilbey, M.M., & Asghar, K., eds.
Methodological Issues in Epidemiological, Prevention, and Treatment
Research on Drug-Exposed Women and Their Children. National Institute
on Drug Abuse Research Monograph 117. DHHS Pub. No. (ADM)92-1881.
Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1992, pp.
137-154.
The chapter discusses strategies for subject recruitment
and retention for longitudinal research. It examines issues that need
to be considered in the planning of a study, factors that affect the
success of the recruitment effort, and special techniques for tracing
subjects and retaining subjects. It also discusses issues in staff
recruitment, training, and inspiration, as well as budgeting.
Back to Top
B. Statistical Analysis with
Missing Data
- Allison, P.D. Estimation of linear models with incomplete
data. In: Clogg, C., ed. Sociological Methodology 1987. San
Francisco: Jossey-Bass, 1987, pp. 71-103.
The chapter describes the uses of ML to handle missing
data problems in LISREL type of models. After describing the various
missing data mechanisms first described by Rubin (1976), the author
describes the assumptions underlying the method he proposed. The core
of the paper describes how to use LISREL to get ML estimates of linear
models with incomplete data. The method capitalizes on the ability
of LISREL to estimate simultaneously the same model for two or more
samples. For incomplete data problems, the sample is divided into
subsamples, each having a different set of variables present. The
model is then estimated simultaneously for all subsamples, constraining
corresponding parameters to be equal across subsamples. The author
further discusses the special techniques that are needed in such estimation.
There is also an appendix that contains proofs.
- Little, R.J.A., & Rubin, D.B. Statistical Analysis
With Missing Data. New York: Wiley, 1987.
The book surveys methodology for handling missing-data
problems and presents a likelihood-based theory for analysis with
missing data that systematizes these methods. Part I of the book discusses
ad hoc approaches to missing-value problems in three important areas
of statistics: analysis of variance of planned experiments, survey
sampling, and multivariate analysis. Part II presents a systematic
approach to the analysis of data with missing values, where inferences
are based on likelihoods derived from formal statistical models for
the data and the missing-data mechanisms. Applications of the approach
are presented in a variety of contexts, including regression, factor
analysis, contingency table analysis, time series, and sample survey
inference.
- Little, R.J.A., & Schenker, N. Missing Data. In:
Arminger, G., Clogg, C.C., & Sobel, M.E., eds. A Handbook for
Statistical Modeling in the Social and Behavioral Sciences. New
York: Plenum, 1992, pp. 39-75.
In this chapter, Little and Schenker discuss modern
methods for dealing with missing data. It covers important concepts
underlying missing data procedures, differences between naive approaches,
and more principled approaches. Then the chapter discusses various
missing data procedures, such as weighting adjustment for unit nonresponse,
maximum likelihood approaches, nonignorable nonresponse models, and
multiple imputation. It also covers other Bayesian simulation methods.
- Rindskopf, D. A general approach to categorical data
analysis with missing data, using generalized linear models with composite
links. Psychometrika 57:29-42, 1992.
Rindskopf describes a general approach for analyzing
categorical data when there are missing data. The method is based
on generalized linear models with composite links. The approach can
be used (among other applications) to fill in contingency tables with
supplementary margins. It is also used to fit log-linear models when
data are missing, fit latent class models (without or with missing
data on observed variables), fit models with fused cells, and fill
in tables or fit models to data when variables are more finely categorized
for some cases than others.
- Rubin, D.B. Inference and missing data. Biometrika
63:581-592, 1976.
This is the classical paper written by Donald Rubin
on the drawing of inferences about the parameter of the data when
missing values are present. It reviews the previous statistical literature
on missing data, and it provides a thorough theoretical discussion
of the problem. The paper outlines the different kinds of missing
data generating mechanisms and discusses the implication of each for
making inference about the true value of parameters. The paper also
discusses separately the use of direct-likelihood inferences and Bayesian
inferences in the context of missing data.
- Schafer, J.L. Analysis of Incomplete Multivariate
Data. London: Chapman and Hall, 1997.
The book presents a unified, Bayesian approach to the
analysis of incomplete multivariate data, covering data sets in which
the variables are continuous, categorical, or both. It is written
for applied statisticians, biostatisticians, practitioners of sample
surveys, graduate students, and other methodologically oriented researchers
in search of practical tools to handle missing data. The focus is
applied rather than theoretical, but technical details are also included.
- Clogg, C.C. Rubin, D.B., Schenker, D., Schultz, B.,
& Weidman, L. Multiple imputation of industry and occupation codes
from Census public-use samples using Bayesian logistic regression.
J Am Stat Assoc 86:68-78.
The authors describe methods used to create a new Census
database that can be used to study comparability of industry and occupation
classification systems. The project consists of extensive application
of multiple imputation to data and the fitting of hundreds of logistic
regression models. The paper summarizes the strategies used in the
project, and it shows how modifications of maximum likelihood methods
were made for the modeling and imputation phases of the project. These
methods include Bayesian methods that can deal with sparse data, which
usually present problems for traditional ML methods.
- Graham, J.W., & Schafer, J.L. On the performance
of multiple imputation for multivariate data with small sample size.
In: Hoyle, R., ed. Statistical Strategies for Small Sample Research.
Thousand Oaks, CA: Sage Publications, 1999.
The purpose of the chapter is to investigate the performance
of some of the latest missing-data technologiesin particular,
multiple imputation using NORMin realistic analyses where the
sample size, N, is relatively small. The authors review from
a users perspective the key ideas of multiple imputation and
the use of NORM software developed by Schafer (1997; NORM is available
at http://methcenter.psu.edu/).
They then describe a simulation study designed to test the limits
of NORM in a realistic setting where N is small and the data
do not conform to assumptions of normality.
- Schafer, J.L., & Olsen, M.K. Multiple imputation
for multivariate missing-data problems: A data analyst's perspective.
Multivariate Behav Res 33:545-571, 1998.
This article reviews the key ideas of multiple imputation,
discusses the software programs currently available, and demonstrates
their use on data from the Adolescent Alcohol Prevention Trial.
Back to Top
C. Selection of Covariates
in Prevention/Epidemiological Studies
- Baumrind, D. Familial antecedents of adolescent drug
use: A developmental perspective. In: Jones, C.L., & Battjes,
R.J., eds. Etiology of Drug Abuse: Implications for Prevention.
National Institute on Drug Abuse Research Monograph 56. DHHS Pub.
No. (ADM)87-1335. Washington, DC: Supt. of Docs., U.S. Govt. Print.
Off, 1985, pp. 13-44.
The author considers the impact of early childhood and
preadolescent socialization experiences on adolescent drug use from
a developmental perspective. She first review the processes defining
normal adolescent development and then present findings on the preadolescent
phase of the Family Socialization and Developmental Competence Project.
The findings do not support the presupposition that adolescent drug
use arises from pathological personal characteristics or pathogenic
socialization practices. The study also found that the use of illegal
drugs, such as marijuana, is not a deviant behavior for adolescents.
- Donovan, J.E., & Jessor, R. Structure of problem
behavior in adolescence and young adulthood. J Consult Clin Psychol
52(6):890-904, 1985.
The authors carried out a multivariate test of the existence
of a single behavioral syndrome that comprises problem drinking, illicit
drug use, delinquent-type behavior, and precocious sexual intercourse
among normal adolescents. Analyses were conducted using maximum likelihood
factor analyses based on self-report data from several samples of
adolescents and youth who participated in the 1978 National Study
of Adolescent Drinking (Rachal et al., 1980). Results of the study
consistently showed that one common factor accounts for the correlations
among the different problem behaviors. The authors argue that the
findings lend support to the notion of a syndrome of problem behavior
in both adolescence and young adulthood.
- Greenspan, S.I. Research strategies to identify developmental
vulnerabilities for drug abuse. In: Jones, C.L., & Battjes, R.J.,
eds. Etiology of Drug Abuse: Implications for Prevention. National
Institute on Drug Abuse Research Monograph 56. DHHS Pub. No. (ADM)
85-1335. Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1985,
pp. 136-154.
In this paper, Greenspan presents the developmental
structuralist approach to etiologic and intervention research. Such
an approach allows postulation of relationships between infancy and
subsequent behaviors associated with drug use. The author later uses
vignettes that illustrate how early intersensory integration, self-regulatory
mechanisms, and affective development relate to subsequent impairments
in affective relationships among individuals who develop antisocial
behavior and substance use.
- Hawkins, D.J., Catalano, R.F., & Miller, J.Y. Risk
and protective factors for alcohol and other drug problems in adolescence
and early adulthood. Psychol Bull 112(l):64-105, 1992.
The authors propose that a risk-focused approach to
prevention of adolescent alcohol and other drug problems is the most
promising route to effective intervention strategies. Such an approach
requires the identification of risk factors for drug abuse, identification
of methods by which risk factors have been effectively addressed,
and application of these methods to appropriate high-risk and general
population samples in controlled studies. The authors review risk
and protective factors to drug abuse, assess a number of approaches
for drug prevention potential with high-risk groups, and make recommendations
for research and practice.
- Hawkins, J.D., Lishner, D.M., & Catalano, R.F.,
Jr. Childhood predictors and the prevention of adolescent substance
abuse. In: Jones, C.L., and Battjes, R.J., eds. Etiology of Drug
Abuse: Implications for Prevention. National Institute on Drug Abuse
Research Monograph 56. DHHS Pub. No. (ADM)871335. Washington,
DC: Supt. of Docs., U.S. Govt. Print. Off., 1985. pp. 75-126.
The paper describes approaches to drug abuse prevention
with preadolescent children. The authors review these approaches in
light of the existing knowledge on the etiology of drug abuse among
children and adolescents. They also identify gaps in the prevention
intervention research and suggest etiological research that can aid
in the development and refinement of preventive intervention focused
on preadolescents.
- Jessor, R. Bridging etiology and prevention in drug
abuse research. In: Jones, C.L., & Battjes, R.J., eds. Etiology
of Drug Abuse: Implications for Prevention. National Institute on
Drug Abuse Research Monograph 56. DHHS Pub. No. (ADM) 85-1335.
Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1985.
Jessor comments on four papers in the monograph that
deal with the etiology of drug use during the transition from adolescents
to young adulthood. He concludes by highlighting two major issues
for drug abuse prevention research. The first is the appropriate goal
for prevention. Rather than targeting at complete abstinence, he presents
a more differentiated specification. The author also discusses the
implication of the findings on stages and sequences of drug initiation
and the role early onset plays in drug use initiation for drug abuse
prevention.
- Johnston, L.D. Contributions of drug epidemiology to
the field of drug abuse prevention. In: Leukefeld, C.G., & Brakoslu,
W.J., eds. Drug Abuse Prevention Research: Methodological Issues.
National Institute on Drug Abuse Research Monograph 107. DHHS
Pub. No. (ADM) 91-1761. Washington, DC: Supt. of Docs., Govt. Print
Off., 1991, pp. 57-80.
Johnston points out there are at least eight ways in
which epidemiological studies could inform the development of drug
prevention programs and the evaluations of such programs. He discusses
each of them in detail in this book chapter. The eight areas are (1)
drug use or drug-related problems that need to be prevented; (2) ages
at which such use is initiated or problems are occurring; (3) subgroups
in the population most "at risk" in terms of their demographic
and lifestyle characteristics; (4) changing backdrop against which
the effects of specific prevention efforts should be assessed; (5)
importance of certain key intervening variables such as attitudes
and beliefs; (6) behavioral and moral norms with regard to drug use
among young people and other groups having influence on them; (7)
extent to which major classes of prevention programming are reaching
targeted segments of the population and the subjective opinions of
those populations as to the helpfulness and effects of the interventions;
and (8) combined effectiveness of all forces in society that tend
to reduce drug use or abuse, including those that are planned programs,
more spontaneous efforts of groups or individuals, and other historical
events.
- Jones, C.L., & Battjes, R.J., eds. Etiology
of Drug Abuse: Implications for Prevention. National Institute on
Drug Abuse Research Monograph 56. DHHS Pub. No. (ADM)87-1335.
Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1985.
The monograph consists of a number of comprehensive
review papers written by distinguished scientists in the field of
etiological research on substance use. It focuses on results of etiologic
research on adolescent drug use could be utilized to improve efforts
to prevent drug abuse. The papers in the monograph are grouped into
two section, the first one deal with antecedents to drug use during
childhood and the impact of the transition to adolescence. The second
part deals with risk factors that become prominent during the transition
from adolescent to young adulthood.
- Kandel, D.B., & Yamaguchi, K. Developmental patterns
of the use of legal, illegal, and medically prescribed psychotropic
drugs from adolescence to young adulthood. In: Jones, C.L., &
Battjes, R.J., eds. Etiology of Drug Abuse: Implications for Prevention.
National Institute on Drug Abuse Research Monograph 56. DHHS Pub.
No. (ADM) 85-1335. Washington, DC: Supt. of Docs., U.S. Govt. Print.
Off., 1985, pp. 193- 235.
Kandel and Yamaguchi present their findings on the developmental
patterns of drug use among a cohort of young adults in a followup
study during 1980-1981. Using the methodology of Guttman Scaling,
they test specific sequential models of progression of drug use. Furthermore,
using event history analysis, they try to answer the question of whether
the use of certain drugs lower in the sequence influences the initiation
of higher drugs. The authors then discuss their findings on the initiation
of different types of illicit drugs.
- Robins, L.N., & Przybeck, T.R. Age of onset of
drug use as a factor in drug and other disorders. In: Jones, C.L.,
& Battjes, R.J., eds. Etiology of Drug Abuse: Implications
for Prevention. National Institute on Drug Abuse Research Monograph
56. DHHS Pub. No. (ADM)87-1335. Washington, DC: Supt. of Docs.,
U.S. Govt. Print. Off., 1985, pp. 178-192.
The paper presents findings from a large-scale epidemiological
study in three sites on drug disorder among adolescents. The high
rate of drug disorder is related to rise in drug use in this population
but not to increase in vulnerability to addiction or abuse among users.
Drug use disorders are associated with other psychopathology. Early
onset of use predict increased risk of drug disorders and higher severity
of the disorder. The findings suggest that a useful preventive strategy
is to try to postpone first drug use to age 18 or later.
- Shore, M.F. Correlates and concepts: Are we chasing
our tails? In: Jones, C.L., & Battjes, R.J., eds. Etiology
of Drug Abuse: Implications for Prevention. National Institute on
Drug Abuse Research Monograph 56. DHHS Pub. No. (ADM)87-1335.
Washington, DC: Supt. of Docs., U.S. Govt. Print. Off., 1985, pp.
127-135.
The paper is a response to three other papers in the
monograph. The author first lists seven major advances in drug abuse
research and then discusses the papers by Baumrind, Bush and Iannotti,
and Hawkins, all of which focus on the impact of early childhood factors
on later drug use. The author concludes by pointing out the difficulty
in developing theories in prevention research, the challenges in developing
useful research methodology for the field, and the need for longitudinal
research paradigm.
Back to Top
D. Reducing Selection Bias
and Adjusting for Initial Group Differences in Non-Randomized Treatment
Studies
- Berk, R.A. An introduction to sample selection bias,
American Sociological Review 48:386-398, 1983.
This paper is a brief review of some of the methods
in the diagnosis and correction for "sample selection bias."
It describes, with the aid of diagram, the nature and consequence
of sample selection bias. Drawing on Heckmans (1979) exposition
of the problem, the author further explains the phenomenon in statistical
terms. Later parts of the paper introduce the readers to several remedies
to the problem.
- Cochran, W.G. The effectiveness of adjustment by subclassification
in removing bias in observational studies. Biometrics 24:295-313,
1968.
In the adjustment by subclassification method, the distribution
of the covariate x is broken up into two, three, or more subclasses.
For each group of subjects, the mean value of the response variable
y, for the two groups under comparison, is calculated separately
within each subclass. Then a weighted mean of these subclass means
is calculated for each group, using the same weights for every group.
The effectiveness of such a method in removing bias in observational
studies was examined in this paper. The extent to which adjustment
reduces the sampling error of the estimated difference between the
y is also examined.
- Cochran, W.G., & Rubin, D.B. Controlling bias in
observational studies: A review. Snkhya, Series A 35(4):417-446,
1973.
This paper reviews work on the effectiveness of different
methods of matched sampling and statistical adjustment, alone and
in combination, in reducing bias due to confounding x-variables
when comparing two populations. The adjustment methods were linear
regression adjustment for x continuous and direct standardization
for x categorical. With x continuous, the range of situations
examined included linear relations between y and x,
parallel and nonparallel, monotonic nonlinear parallel relations,
equal and unequal variances of x, and the presence of errors
of measurement in x. The percent of initial bias that was removed
was used as the criterion. Overall, linear regression adjustment on
matched samples appeared superior to the matching methods, with linear
regression adjustment on matched samples the most robust method. Several
different approaches are suggested for the case of multivariate x.
- McKinlay, S.M. The effect of bias on estimators of
relative risk for pair-matched and stratified samples. J Am Stat
Assoc 70(352):859-864, 1977.
The author used Monte Carlo methods to compare the effectiveness
of pair-matched and independent stratified samples for estimating
relative risk in the presence of bias. Three approximations to the
maximum likelihood estimator for stratified samples suggested by Woolf,
Mantel and Haenszel and Birch, respectively, are also compared. The
results showed that the mean square error is always largest for the
matched-pairs estimator, while of the stratified estimators, Woolfs
consistently produces the smallest MSE, equaled by Birchs when
the samples are equal.
- Rosenbaum P.R., & Rubin, D.B. Reducing bias in
observational studies using subclassification on the propensity score.
J Am Stat Assoc 79(387):516-524, 1984.
The propensity score is the conditional probability
of assignment to a particular treatment given a vector of observed
covariates. It has been shown that subclassification on the propensity
score will balance all observed covariates. The paper shows how the
approach works using observational data on treatments for coronary
artery disease. Subclasses are first defined by the estimated propensity
score. These subclasses are applied within subpopulations, and model-based
adjustments are then used to provide estimates of treatment effects
within these subpopulations.
- Rosenbaum, P.R., & Rubin, D.B. The central role
of the propensity score in observational studies for causal effects.
Biometrika 70(1):41-55, 1983.
Both large and small sample theory show that adjustment
for the scalar propensity score is sufficient to remove bias due to
all observed covariates in observational studies. The authors show
that the approach can have the following applications: (1) matched
sampling on the univariate propensity score, which is a generalization
of discriminant matching, (2) multivariate adjustment by subclassification
on the propensity score where the same subclasses are used to estimate
treatment effects for all outcome variables and in all subpopulations,
and (3) visual representation of multivariate covariance adjustment
by a two-dimensional plot.
- Rubin, D.B. Using multivariate matched sampling and
regression adjustment to control bias in observational studies. J
Am Stat Assoc 74(366):318-328, 1979.
Rubin studied the efficacy of multivariate matched sampling
and regression adjustment for controlling bias due to specific matching
variables X when dependent variables are moderately nonlinear
in X, using Monte Carlo methods. The general conclusion he
had is that nearest available Mahalanobis metric matching in combination
with regression adjustment on matched pair differences is a highly
effective plan for controlling bias due to X.
- Rubin, D.B. Matching to remove bias in observational
studies. Biometrics 29:159-183, 1973.
Several matching methods that match all of one sample
from another larger sample on a continuous matching variable are compared
with respect to their ability to remove the bias of the matching variable.
One method is a simple mean-matching method, and three are nearest
available pair-matching methods. The methods abilities to remove
bias are also compared with the theoretical maximum given fixed distributions
and fixed sample sizes. A summary of advice to an investigator is
included.
- Rubin, D.B. The use of matched sampling and regression
adjustment to remove bias in observational studies. Biometrics
29:185-203, 1973.
The ability of matched sampling and linear regression
adjustment to reduce the bias of an estimate of the treatment effect
in two sample observational studies is investigated for a simple matching
method and five simple estimates. Monte Carlo results are given for
moderately linear exponential response surfaces, and analytic results
are presented for quadratic response surfaces. One of the conclusions
of the study is that the combination of regression adjustment in matched
samples generally produces the least biased estimate.
- Rubin, D.B. Multivariate matching methods that are
equal percent bias reducing, I: some examples. Biometrics 32:109-120,
1976.
Multivariate matching methods are commonly used in the
behavioral and medical sciences in an attempt to control bias when
randomization is not feasible. Some examples of multivariate matching
methods are discussed in Althauser and Rubin (1970) and Cochran and
Rubin (1973) but otherwise have received little attention in the literature.
The author presents examples of multivariate matching methods that
will yield the same percent reduction in bias for each matching variable
for a variety of underlying distribution. Eleven distributional cases
are considered and for each one, matching methods are described that
are equal percent bias reducing.
- Winship, C., & Mare, R.D. Models for sample selection
bias. Annu Rev Sociol 18:327-350, 1992.
This chapter reviews models that attempt to take account
of sample selection and their applications in research on labor markets,
schooling, legal processes, social mobility, and social networks.
Variants of these models apply to outcome variables that are censored
or truncatedwhether explicitly or incidentallyand include
the tobit model, the standard selection model, models for treatment
effects in quasi-experimental designs, and endogenous switching models.
It also discusses Heckmans two-stage estimator.
Back to Top
E. Meta-Analysis
- Bangert-Drowns, R.L. Review of developments in meta-analytic
methods. Psychol Bull 99:388-399, 1986.
The author provides a brief history of the development
of meta-analysis. The author also distinguishes five different approaches
to meta-analytic method and makes suggestions for the use.
- Bukoski, W.J., ed. Meta-Analysis of Drug Abuse Prevention
Programs. National Institute on Drug Abuse Research Monograph 170.
NTIS Pub. No. (ADM) 97-181598. Washington, DC: Supt. of Docs., U.S.
Govt. Print. Off., 1997.
The monograph provides firsthand guidance in the application
of research findings from meta-analysis and appropriate discussion
of key technical procedures that should be considered in conducting
future meta-analyses of drug abuse prevention research. It also helps
to delineate what prevention programs and policies appear to be the
most effective in combating drug abuse by adolescents and young adults
who may be entering the workplace. In the first section of the monograph,
Tobler presents a meta-analysis of adolescent drug abuse prevention
research findings; Schmidt and colleagues provide a meta-analysis
of integrity tests for predicting drug and alcohol abuse; and Becker
provides an approach for meta-analysis of drug-related risk and protective
factors research. In the second section of the monograph, several
chapters explore the appropriateness and special methodological considerations
that must be addressed when conducting a meta-analysis of the drug
abuse prevention research literature. Perry's chapter focuses upon
methods to calculate effect sizes; Devine's chapter discusses issues
in coding prevention intervention studies; Shadish and Heinsman assess
the differences in outcomes produced by experimental versus quasi-experimental
studies; Matt explores issues concerning generalized causal inferences
related to program effects; Hansen reviews approaches to classifying
independent variables and types of correlational relationships between
dependent and independent variables; in separate chapters, Lipsey
and Hedges discuss potential applications of meta-analysis for policy
development; and Bangert-Drowns presents general advantages and potential
limitations of conducting and utilizing meta-analysis in drug abuse
prevention research. Collectively these chapters provide a current
overview of the efficacy of drug abuse prevention programs and related
measurement systems and help define the techniques employed in meta-analysis
of drug abuse prevention programs.
- Hedges, L.V., & Olkin, L. Statistical Methods
for Meta-Analysis. Orlando, FL: Academic Press, 1985.
The main purpose of this book is to address the statistical
issues for integrating independent studies through the method of meta-analysis.
Chapter 3 of the book provides a review of omnibus procedures for
testing the statistical significance of combined results. Later chapters
discuss various methods to estimate and combine effect sizes. Chapters
7, 8, and 9 describe the different kinds of analysis that can be used
to analyze and compare effect sizes. Chapters 10 and 11 explore the
properties of correlated effect size estimates and their analysis.
Chapter 12 deals with outliers, and chapter 13 discusses clustering
procedures. Chapter 14 demonstrates the effects of censoring of effect
size estimates corresponding to nonsignificant mean differences.
- Hunter, J., & Schmidt, F. Methods of Meta-Analysis:
Correcting Error and Bias in Research Findings. Newbury Park,
CA: Sage Publications, 1990.
The authors present clear and detailed descriptions
of meta-analysis methods, with extensive examples that clarify the
applications of these methods. New methods for correcting statistical
artifacts not previously addressed are presented for both correlational
and experimental studies. Other features of the book include advances
in artifact distribution-based meta-analysis methods and applications
to examples, examination of the use of regression slopes and intercepts
in meta-analysis, detailed treatment of repeated measures of experimental
designs and meta-analysis methods for these designs, detailed treatment
of second-order sampling error and its associated problems, exploration
of the current and future role of meta-analysis in the social sciences,
and extensive analysis of criticisms of meta-analysis.
- Schmidt, F. What do data really mean? Research findings,
meta-analysis, and cumulative knowledge in psychology. Am Psychol
47(10):1173-1181, 1992.
In this article, the author presents an introduction
on the technology of meta-analysis. He first contrasts the approach
with traditional methods and then describes the application of meta-analysis
in industrial organizational psychology. Further, Schmidt outlines
the role of meta-analysis in theory development and the broader impact
it has on scientific research. The author argues that the broader
dissemination of meta-analysis will lead to major changes in the way
psychologists view the general research process.
- Rosenthal, R., & Rubin, D.B. Comparing effect sizes
of independent studies. Psychol Bull 92(2):500-504.
This article presents a general set of procedures for
comparing the effect sizes of two or more independent studies. The
procedures include a method for calculating the approximate significance
level for the heterogeneity of effect sizes of studies and a method
for calculating the approximate significance level of a contrast among
the effect sizes. Although the focus is on effect size as measured
by the standardized difference between the means (d) defined
as (M1 -M2)/S, the procedures
can be applied to any measure of effect size having an estimated variance.
This extension is illustrated with effect size measured by the difference
between proportions.
- Strube, M.J., & Miller, R.H. Comparison of power
rates for combined probability procedures: A simulation study. Psychol
Bull 99(3):407-415, 1986.
Rosenthal (1978) presented a thorough description of
six procedures for combining the significance levels from independent
tests of the same conceptual hypothesis. This simulation study compares
the power of these six methods. The results indicated that for large
numbers of studies to be combined, all procedures provide comparable
power. The techniques vary in their ease of computation, however,
making some procedures more preferable under certain conditions.
Back to Top
F. Sample Size Calculation
- Cohen, J. Statistical Power Analysis for the Behavioral
Sciences, 2nd ed. Hillsdale, NJ: Erlbaum, 1990.
The book provides a detailed description of the concept
of power analysis, and it is written as a handbook on the topics for
behavioral researchers. It covers the power calculation for various
statistical tests, t-test for means, significance of a product moment
r, differences between correlation coefficients, the test that a proportion
is .t and the sign test, differences between proportions, chi-square
test, ANOVA and ANCOVA, multiple regression and correlation analysis,
set correlation, and multivariate methods. The book concludes with
chapters on issues in power analysis and some computational procedures.
- Donner, A. Approaches to sample size estimation in
the design of clinical trials: A review. Stat Med 3:199-214,
1984.
The paper reviews various methods of sample size estimation
in the design of clinical trials. It is restricted to the discussion
of designs with the primary purpose of comparing two groups of patients
with respect to the occurrence of some specified event, such as death
or the recurrence of disease. It takes into account the special nature
of clinical trials that might preclude the direct applications of
formal sample size planning and simple and straightforward formulae.
- Kraemer, H.C., & Theimann, S. How Many Subjects?
Beverly Hills, CA: Sage Publications, 1988.
The book treats power in a unified fashion across hypothesis
testing techniques by calculating what the authors called critical
effect size. It offers a simple introduction for nonstatisticians
to power analysis and sample size determination as well as discussion
on other topics like the conditions under which a repeated measures
design will be more or less efficient than a cross-sectional design;
the considerations involved in deciding whether to match or stratify
subjects; the selection of variables for multiple regression analysis;
the value of equal (or near equal) N in analysis of variance designs;
how to insure, in a correlational study, that the study will be valid;
and the N required to make a reasonably rigorous test of one hypothesis
using the chi-square technique.
- Muller, K.E., LaVange, L.M., Ramey, S.L., & Ramey,
C.T. Power calculations for general linear multivariate models including
repeated measures applications. J Am Stat Assoc 87:1209-1226,
1992.
The paper reviews recently developed methods for power
analysis, in particular those applicable to general linear multivariate
models (GLMM). The paper first discusses the motivation for using
detailed power calculations, focusing on multivariate methods in particular.
Second, the authors survey available methods for the general linear
multivariate model (GLMM) with Gaussian errors and recommend those
based on F approximations. The paper covers the multivariate
and univariate approaches to repeated measures, MANOVA, ANOVA, multivariate
regression, and univariate regression. Third, the authors describe
the design of the power analysis for an example of a study that examines
the impact of mothers' verbal IQ on childrens intellectual development.
Then the authors describe the result of their power analysis and evaluate
the tradeoffs in using reduced designs and tests to simplify power
calculations. Lastly, the authors discuss the benefits and costs of
power analysis in the practice of statistics.
Back to Top
G. Human Subject Issues
- Hoagwood K., Jensen, P.S., & Fisher, C.B. Ethical
Issues in Mental Health Research with Children and Adolescents.
Mahwah, NJ: LEA, 1996.
The purpose of the book is to surface the key ethical
dilemmas that investigators who study child and adolescent emotional,
behavioral, developmental, and mental disorders are encountering,
and to offer practical suggestions for integrating ethical thinking
into such research. The book is organized into four parts. Part I
introduces the major scientific, regulatory, and community principles
that guide ethical practices in research on children's mental health.
Part II reviews major ethical issues across diverse research contexts.
Part II focuses attention on illustrative cases. Part IV discusses
the role of bioethicists.
- Porter, J.P. Regulatory considerations in research
involving children and adolescents with mental disorders. In: Hoagwood,
K., Jensen, P.S., & Fisher, C.B., eds. Ethical Issues in Mental
Health Research with Children and Adolescents. Mahwah, NJ: LEA,
1996, pp. 15-28.
The chapter highlights some of the regulatory considerations
in the Department of Health and Human Services (DHHS) regulations
for the protection of human subjects at Title 45 Code of Federal Regulations
Part 46. These regulations provide the minimum standards for investigators.
The author focuses on the regulatory aspects of research involving
particularly vulnerable children.
- Attkisson, C.C., Rosenblatt, A., & Hoagwood, K.
Research ethics and human subjects protection in child mental health
services research and community studies. In: Hoagwood, K., Jensen,
P.S., & Fisher, C.B., eds. Ethical Issues in Mental Health
Research with Children and Adolescents. Mahwah, NJ: LEA, 1996,
pp. 43-58.
The authors identify major ethical and human subject
challenges encountered in conducting research on services for children
and conducting research in communities more broadly, including epidemiological
surveys. They also identify strategies used by researchers to overcome
these obstacles.
- Hibbs, E.D., & Krener, P. Ethical issues in psychosocial
treatment research with children and adolescents. Hoagwood K., Jensen,
P.S., & Fisher, C.B., eds. Ethical Issues in Mental Health
Research with Children and Adolescents. Mahwah, NJ: LEA, 1996,
pp. 59-72.
In the chapter, the authors discuss five specific ethical
issues that confront investigators who study the efficacy or effectiveness
of psychosocial treatments for children. These issues are competence,
informed consent/assent, confidentiality, use of incentives, and selection
and involvement of control subjects.
- Putnam, F.W., Liss, M.B., & Landsverk, J. Ethical
issues in maltreatment research with children and adolescents. In:
Hoagwood K., Jensen, P.S., & Fisher, C.B., eds. Ethical Issues
in Mental Health Research with Children and Adolescents. Mahwah,
NJ: LEA, 1996, pp. 113-134.
This chapter addresses some of the complex ethical and
legal dilemmas raised in research with maltreated and victimized children.
In particular, the authors discuss the general legal and ethical principles
for informed consent/assent with minors and the role of parents in
consenting for their children to participate in research studies.
They also enumerate common concerns raised by researchers and institutional
review boards (IRBs) considering research protocols with maltreated
children. The authors also mention the mechanism available for clarification
of these problems.
Back to Top
|