Linda Collins
Linda Collins
Scroll
Professor of Social and Behavioral Sciences
-
Professional overview
-
Linda M. Collins is Professor of Global Public Health in the Department of Social and Behavioral Sciences, with a secondary appointment in the Department of Biostatistics. She earned her B.A. in Psychology at the University of Connecticut and her Ph.D. in Quantitative Psychology at the University of Southern California.
Collins’ research interests are focused on the development, dissemination, and application of the multiphase optimization strategy (MOST), a framework for the optimization of behavioral, biobehavioral, and social-structural interventions. The objective of MOST is to improve intervention effectiveness, efficiency, economy, and scalability. She is currently collaborating on research applying MOST in the areas of smoking cessation, the prevention of excessive drinking and risky sex in college students, and HIV services.
Collins’ research has been funded by the National Institute on Drug Abuse, the National Institute on Alcohol Abuse and Alcoholism, and the National Science Foundation, among others. She has given more than 150 presentations on MOST around the world, and her publications have appeared in journals in the fields of behavioral science, quantitative methodology, medicine, and engineering.
Collins has held tenured faculty positions at the University of Southern California and at Penn State University, where she was Distinguished Professor of Human Development and Family Studies and Director of The Methodology Center. She is a Fellow of the American Psychological Association, the Association for Psychological Science, the Society of Behavioral Medicine, and is a past president of the Society of Multivariate Experimental Psychology and the Society for Prevention Research.
-
Education
-
BA, Psychology, University of Connecticut, Storrs, CTPhD, Quantitative Psychology, University of Southern California, Los Angeles, CA
-
Honors and awards
-
Fulbright Specialist, National University of Ireland Galway (2018)Pauline Schmitt Russell Distinguished Career Award, Pennsylvania State University’s College of Health and Human Development (2017)Evan G. and Helen G. Pattishall Outstanding Research Achievement Award, Pennsylvania State University’s College of Health and Human Development (2011)President’s Award, Society for Prevention Research (2004)Faculty Scholar Medal for the Social and Behavioral Sciences, Pennsylvania State University (2000)Psychology Department Teacher of the Year, University of Southern California (1992)Psychology Department Mentorship Award, University of Southern California (1991)Society of Multivariate Experimental Psychology Award for Distinguished Early Career Contributions to Multivariate Behavioral Research (1991)
-
Areas of research and study
-
Behavioral ScienceCost EffectivenessCost-effective Health Programs and PoliciesDissemination and Implementation of Evidence-based Programs
-
Publications
Publications
Latent transition analysis for longitudinal data
AbstractVelicer, W. F., Martin, R. A., & Collins, L. M. (n.d.).Publication year
1996Journal title
AddictionVolume
91Page(s)
197-210AbstractAssessing outcome is a critical problem for the study of addictive behaviors. Traditional approaches often lack power and sensitivity. Latent Transition Analysis is an alternative procedure that is applicable to categorical latent variable models such as stage models. The method involves four different types of parameters, each of which may be relevant to different research questions. Two examples that employ the Stages of Change construct are used to illustrate the method. In the first example, three different models of longitudinal change are compared. In the second example, the effects of an expert system intervention for smoking is compared to a control condition. The method permits the investigation of a series of specific comparisons: (1) the effectiveness of the intervention for individuals in different stages can be assessed; (2) the effectiveness of the intervention can be evaluated for different time intervals; and (3) the effects of intervention on both progression through the stages and regression through the stages or relapse can be assessed. Other potential applications of the method are also discussed.The acquisition and maintenance of safer sexual behaviors among injection drug users
AbstractPosner, S. F., Collins, L. M., Longshore, D., & Anglin, M. D. (n.d.).Publication year
1996Journal title
Substance Use and MisuseVolume
31Issue
14Page(s)
1995-2015AbstractThe current research tests stage-sequential models of safer sexual behavior using a new method for data analysis called Latent Transition Analysis (LTA). Results are presented from data collected on 359 injection drug users participating in a seroconversion study conducted by the UCLA Drug Abuse Research Center. We identified a six-stage model that adequately represented the data. Results indicate that respondents moved back and forth among high, medium, and low risk stages. This finding highlights the need for continual, sustained interventions to help maintain safer sex and drug using behaviors in highrisk groups.Analysis of stage-sequential change in rehabilitation research
AbstractCollins, L. M., & Johnston, M. V. (n.d.).Publication year
1995Journal title
American Journal of Physical Medicine and RehabilitationVolume
74Issue
2Page(s)
163-170AbstractThe analysis of change is a critical topic for rehabilitation outcomes research, because the goal of rehabilitation is to improve patients' function. This article will sketch the issues and problems in statistical methods for analysis of change. Recovery after serious illness or injury often has been described in terms of transitions between a series of stages, but these theories of recovery have rarely been subjected to adequate statistical examination. This article presents two methods for the analysis of change, suitable for ordinal measures of function and for testing either simple cumulative or complex multipath stage-models of recovery. Progress in medical rehabilitation will be enhanced by explicitly specifying models of recovery, measuring recovery at multiple time points, and using the resulting data to test these models empirically.Comment on "How Many Causes Are There of Aging-Related Decrements in Cognitive Functioning?"
AbstractCollins, L. M. (n.d.).Publication year
1994Journal title
Developmental ReviewVolume
14Issue
4Page(s)
438-443AbstractSalthouse (1994) introduces a new type of correlation, termed the quasi-partial correlation (QPC). The QPC reflects the amount of age-related variance that is shared by two variables. In this commentary it is argued that the QPC is a model where age is a suitable proxy for time, change over time is linear, and causation is instantaneous or nearly instantaneous. Where these assumptions are reasonable, QPC′s are a highly valuable research tool. It is also argued that the "causes" of decline identified by Salthouse might be termed "dimensions" of decline instead, because the analysis did not include independently assessed causal variables. Finally, it is suggested that age is unsatisfying as a predictor of change across the lifespan.Crossvalidation of Latent Class Models of Early Substance Use Onset
AbstractCollins, L. M., Graham, J. W., Long, J. D., & Hansen, W. B. (n.d.).Publication year
1994Journal title
Multivariate Behavioral ResearchVolume
29Issue
2Page(s)
165-183AbstractCudeck and Browne (1983) were among the first to discuss the advantages of taking a crossvalidation approach to testing of covariance structure models. The purpose of this paper is to expand on Cudeck and Browne's work in two directions. The first direction of expansion is into testing of latent class models. The second direction of expansion involves using crossvalidation to examine differences between groups, where groups may be formed by gender, ethnicity, region, etc. In the present article crossvalidation is used to help select models of early substance use onset in a sample of young adolescents. The results suggest that the nature of the substance use onset process and the rate of movement through the process are the same for males and females at seventh and eighth grade. However, the present study did find evidence for gender differences in substance use experience at the beginning of seventh grade, with males somewhat more advanced in the onset process. The results also suggest that double crossvalidation is greatly to be preferred over single crossvalidation.Latent transition analysis and how it can address prevention research questions
AbstractCollins, L. M., Graham, J. W., Rousculp, S. S., Fidler, P. L., Pan, J., & Hansen, W. B. (n.d.).Publication year
1994Journal title
NIDA Research Monograph SeriesIssue
142Page(s)
81-111AbstractThe objective of this chapter is to introduce latent transition analysis (LTA) to the substance use prevention research community. LTA is a new methodological technique for testing stage-sequential models, such as models of substance use onset. LTA estimates several different sets of parameters. One of these sets is the transition probability matrix, which contains information about the probability of movement between stages in the model. LTA can be used to evaluate the effectiveness of prevention intervention programs by comparing the transition probability matrices of the program and control groups. If the prevention program is successful, the transition probability matrices will indicate that the probability of moving to a more advanced stage of drug use is lower for the program participants than for the control group. An advantage of taking a stage-sequential approach is that examining the transition probability matrix reveals how effective a program is for individuals entering the program with different levels and types of substance use experience.New statistical methods for substance use prevention research
Collins, L. M., & Seitz, L. A. (n.d.).Publication year
1994Journal title
NIDA Research Monograph SeriesIssue
142Page(s)
1-12Seven ways to increase power without increasing N
AbstractHansen, W. B., & Collins, L. M. (n.d.).Publication year
1994Journal title
NIDA Research Monograph SeriesIssue
142Page(s)
184-195AbstractMany readers of this monograph may wonder why a chapter on statistical power was included. After all, by now the issue of statistical power is in many respects mundane. Everyone knows that statistical power is a central research consideration, and certainly most National Institute on Drug Abuse grantees or prospective grantees understand the importance of including a power analysis in research proposals.Some design, measurement, and analysis pitfalls in drug abuse prevention research and how to avoid them: Let your model be your guide
Collins, L. M. (n.d.).Publication year
1994Journal title
NIDA Research Monograph SeriesIssue
139Page(s)
95-114Goodness-of-Fit Testing for Latent Class Models
AbstractCollins, L. M., Fidler, P. L., Wugalter, S. E., & Long, J. D. (n.d.).Publication year
1993Journal title
Multivariate Behavioral ResearchVolume
28Issue
3Page(s)
375-389AbstractLatent class models with sparse contingency tables can present problems for model comparison and selection, because under these conditions the distributions of goodness-of-fit indices are often unknown. This causes inaccuracies both in hypothesis testing and in model comparisons based on normed indices. In order to assess the extent of this problem, we carried out a simulation investigating the distributions of the likelihood ratio statistic G2, the Pearson statistic X2, and a new goodness-of-fit index suggested by Read and Cressie (1988). There were substantial deviations between the expectation of the chi-squared distribution and the means of the G2 and Read and Cressie distributions. In general, the mean of the distribution of a statistic was closer to the expectation of the chi-squared distribution when the average cell expectation was large, there were fewer indicator items, and the latent class measurement parameters were less extreme. It was found that the mean of the X2 distribution is generally closer to the expectation of the chi-squared distribution than are the means of the other two indices we examined, but the standard deviation of the X2 distribution is considerably larger than that of the other two indices and larger than the standard deviation of the chi-squared distribution. We argue that a possible solution is to forgo reliance on theoretical distributions for expectations and quantiles of goodness-of-fit statistics. Instead, Monte Carlo sampling (Noreen, 1989) can be used to arrive at an empirical central or noncentral distribution.Latent Class Models for Stage-Sequential Dynamic Latent Variables
AbstractCollins, L. M., & Wugalter, S. E. (n.d.).Publication year
1992Journal title
Multivariate Behavioral ResearchVolume
27Issue
1Page(s)
131-157AbstractStage-sequential dynamic latent variables are of interest in many longitudinal studies. Measurement theory for these latent variables, called Latent Transition Analysis (LTA), can be found in recent generalizations of latent class theory. LTA expands the latent Markov model to allow applications to more complex latent variables and the use of multiple indicators. Because complex latent class models result in sparse contingency tables, that may lead to poor parameter estimation, a simulation study was conducted in order to determine whether model parameters are recovered adequately by LTA, and whether additional indicators result in better measurement or in impossibly sparse tables. The results indicated that parameter recovery was satisfactory overall, although as expected the standard errors were large in some conditions with few subjects. The simulation also indicated that at least within the conditions examined here, the benefits of adding indicators outweigh the costs. Additional indicators improved standard errors, even in conditions producing extremely sparse tables. An example of LTA analysis of empirical data on math skill development is presented.Modeling Transitions in Latent Stage-Sequential Processes: A Substance Use Prevention Example
AbstractGraham, J. W., Collins, L. M., Wugalter, S. E., Chung, N. K., & Hansen, W. B. (n.d.).Publication year
1991Journal title
Journal of consulting and clinical psychologyVolume
59Issue
1Page(s)
48-57AbstractThis article illustrates the use of latent transition analysis (LTA), a methodology for testing stage-sequential models of individual growth. LTA is an outgrowth of latent class theory and is a particular type of latent Markov model emphasizing the use of multiple manifest indicators. LTA is used to compare the fit of two models of early adolescent substance use onset and to assess the effects of a school-based substance use prevention program on Ss measured in 7th grade and again in 8th grade. Several interesting findings emerged. First, a model of substance use onset including both alcohol and tobacco use as possible starting points fit better than a model that included alcohol use as the only starting point. Second, Ss who had tried tobacco but not alcohol in 7th grade seemed to be on an accelerated onset trajectory. Third, the normative education prevention program was generally successful, except for the students who had tried only tobacco in 7th grade.The measurement of dynamic latent variables in longitudinal aging research: Quantifying adult development
AbstractCollins, L. M. (n.d.).Publication year
1991Journal title
Experimental Aging ResearchVolume
17Issue
1Page(s)
13-20AbstractDynamic latent variables involve systematic intraindividual change over time. Although it seems natural to apply traditional measurement theory to dynamic latent variables, in fact this is often inappropriate. Traditional measurement theory is based on the idea of static latent variables and offers little guidance to the researcher who wishes to measure a dynamic latent variable with a high degree of accuracy and validity. It is the contention of this article that measurement of a dynamic latent variable must start from a clearly defined substantive theory about human development. Two approaches that take this perspective are presented: the longitudinal Guttman simplex (LGS), a measurement model for dynamic latent variables undergoing irreversible cumulative, unitary development; and latent transition analysis (LTA), a more general latent class measurement model.A note on the unbiased estimation of the intraclass correlation
AbstractDonoghue, J. R., & Collins, L. M. (n.d.).Publication year
1990Journal title
PsychometrikaVolume
55Issue
1Page(s)
159-164AbstractThe intraclass correlation, ρ, is a parameter featured in much psychological research. Two commonly used estimators of ρ, the maximum likelihood and least squares estimators, are known to be negatively biased. Olkin and Pratt (1958) derived the minimum variance unbiased estimator of the intraclass correlation, but use of this estimator has apparently been impeded by the lack of a closed form solution. This note briefly reviews the unbiased estimator and gives a FORTRAN 77 subroutine to calculate it.The Measurement of Dynamic Latent Variables in Longitudinal Aging Research: Quantifying Adult Development
AbstractCollins, L. M. (n.d.).Publication year
1990Journal title
GerodontologyVolume
9Issue
4Page(s)
127-134AbstractDynamic latent variables involve systematic intraindividual change over time. Although it seems natural to apply traditional measurement theory to dynamic latent variables, in fact this is often inappropriate. Traditional measurement theory is based on the idea of static latent variables and offers little guidance to the researcher who wishes to measure a dynamic latent variable with a high degree of accuracy and validity. It is the contention of this article that measurement of a dynamic latent variable must start from a clearly defined substantive theory about human development. Two approaches that take this perspective are presented; the longitudinal Guttman simplex (LGS), a measurement model for dynamic latent variables undergoing irreversible cumulative, unitary development; and latent transition analysis (LTA), a more general latent class measurement model.Using the longitudinal guttman simplex as a basis for measuring growth
AbstractCollins, L. M., & Cliff, N. (n.d.).Publication year
1990Journal title
Psychological bulletinVolume
108Issue
1Page(s)
128-134AbstractMany difficulties inherent in the measurement of growth stem from the use of traditional measurement methodologies. The longitudinal Guttman simplex (LGS), an alternative approach based on a model of growth, is discussed in this article. The LGS has several advantages over traditional methodology. First, interindividual differences in developmental rates are a part of the model. Second, the LGS procedure can easily handle any number of occasions of measurement. Third, the LGS is suited to nonlinear as well as linear monotonic growth. Fourth, a consistency index associated with the LGS methodology, CL, indicates the extent to which cumulative, unitary development characterizes a particular latent variable. Finally, and perhaps most important, because a model of the growth undergone by the latent variable being measured is incorporated in the LGS model the resulting instruments enjoy a high level of construct validity. The LGS is limited to cumulative, unitary development; additional measurement theories are needed for other kinds of development.An Ordinal I Scaling Method for Questionnaire and Other Ordinal I Data
AbstractCliff, N., Collins, L. M., Zatkin, J., Gallipeau, D., & McCormick, D. J. (n.d.).Publication year
1988Journal title
Applied Psychological MeasurementVolume
12Issue
1Page(s)
83-97AbstractThis paper reports the development and application of a method for ordering persons and items (or stim uli) when responses are ordinal. The method applies most directly to data where responses are dichoto mous, indicating agreement or acceptableness or simi larity, and can be assumed to reflect proximity rather than dominance. It orders rows and columns of the re sponse matrix into “parallelogram” form, using pair- wise interchange procedures, followed by other steps. The method was applied to several sets of question naire data and one set of archeological data, with rea sonable success. Other applications and extensions are suggested. Index terms: Dichotomous responses, Interchange methods, Ordinal scaling, Parallelogram scaling, Proximity data, Questionnaire responses.Omega: A General Formulation of the Rand Index of Cluster Recovery Suitable for Non-disjoint Solutions
AbstractCollins, L. M., & Dent, C. W. (n.d.).Publication year
1988Journal title
Multivariate Behavioral ResearchVolume
23Issue
2Page(s)
231-242AbstractCluster recovery indices are more important than ever, because of the necessity for comparing the large number of clustering procedures available today. Of the cluster recovery indices prominent in contemporary literature, the Hubert and Arabie (1985) adjustment to the Rand index (1971) has been demonstrated to have the most desirable properties (Milligan & Cooper, 1986). However, use of the Hubert and Arabie adjustment to the Rand index is limited to cluster solutions involving non-overlapping, or disjoint, clusters. The present paper introduces a generalization of the Hubert and Arabie adjusted Rand index. This generalization, called the Omega index, can be applied to situations where both, one, or neither of the solutions being compared is non-disjoint. In the special case where both solutions are disjoint, the Omega index is equivalent to the Hubert and Arabie adjusted Rand index.The Longitudinal Guttman Simplex: A New Methodology for Measurement of Dynamic Constructs in Longitudinal Panel Studies
AbstractCollins, L. M., Cliff, N., & Dent, C. W. (n.d.).Publication year
1988Journal title
Applied Psychological MeasurementVolume
12Issue
3Page(s)
217-230AbstractTraditional psychometric procedures can be inade quate for the measurement of dynamic constructs in longitudinal panel studies. This paper introduces an al ternative based on the longitudinal Guttman simplex (LGS) model, a measurement model developed espe cially for dynamic constructs measured longitudinally. The LGS is a model of cumulative, unitary develop ment. It is cumulative in the sense that as persons ac quire new skills (or abilities, or opinions), earlier ob tained skills are retained; it is unitary in the sense that all persons progress through a sequence of skills in the same skill order. CL, a consistency index that gives the researcher a measure of the extent to which the LGS model axioms are obeyed in a given dataset, is introduced. By making use of this consistency index, the researcher can develop scales uniquely sensitive to cumulative, unitary development. LGSCLUS, an explor atory procedure to find longitudinal Guttman scales in empirical datasets, is described. An artificial data study is reported, the purpose of which was to test the performance of LGSCLUS under controlled conditions. The artificial data study showed that, in general, LGSCLUS recovers longitudinal Guttman scales with a high degree of accuracy. There remains a need for measurement procedures for dynamic constructs exhib iting types of development other than cumulative and unitary. Index terms: Dynamic constructs, Gutt man simplex, Longitudinal panel studies, Mathemati cal models, Measurement theory, Scaling, Three-set data.Frequency and adequacy of breast cancer screening among elderly hispanic women
AbstractRichardson, J. L., Marks, G., Solis, J. M., Collins, L. M., Birba, L., & Hisserich, J. C. (n.d.).Publication year
1987Journal title
Preventive MedicineVolume
16Issue
6Page(s)
761-774AbstractStudies have demonstrated that Hispanic (relative to Anglo) women are at greater risk for late-stage breast cancer diagnosis. Screening irregularity may be a factor contributing to late-stage diagnosis, yet virtually nothing is known about the breast cancer-screening behavior of Hispanic women. We interviewed 600 elderly Hispanic women residing in Los Angeles to collect information on frequency of physician breast examinations and mammography and on regularity and competence of breast self-examination. Predictors of screening were also examined. Fifty percent of our sample indicated that they had had a breast exam within the past year; 12.5% had had a mammogram within the past year (74% never had had a mammogram); and 47% reported that they had performed breast self-examination within the past month. Few of the women were able to demonstrate adequate breast self-examination technique on a foam breast model, and only 1% found all five lumps present. Thus, although the observed frequency of screening and self-examination is comparable to national norms, it is unlikely that our subjects' attempts at self-examination would lead to early breast cancer detection. Age, educational level, emotional reactions to screening, and media cues predicted screening behavior. Physician instruction in breast self-examination increased the frequency and adequacy of self-examination. Perceived susceptiblity to cancer, perceived benefits of early detection, and level of acculturation were not strong predictors. The extent to which our results generalize to other subpopulations of Hispanic women is discussed.Health behavior of elderly Hispanic women: Does cultural assimilation make a difference?
AbstractMarks, G., Solis, J., Richardson, J. L., Collins, L. M., Birba, L., & Hisserich, J. C. (n.d.).Publication year
1987Journal title
American journal of public healthVolume
77Issue
10Page(s)
1315-1319AbstractThe role of cultural assimilation in Hispanic health behavior has received little empirical examination. Prior studies have operationalized assimilation primarily in terms of language preference and have obtained weak or no effects. We interviewed 603 elderly Hispanic women residing in Los Angeles to evaluate the usefulness of cultural factors as predictors of preventive health behavior (e.g., physical examination, screening for breast cancer) more rigorously. Factor analysis of responses yielded four dimensions of cultural assimilation: 'language preference', 'country of birth', 'contact with homeland', and 'attitudes about children's friends'. After controlling for education and age, no dimension of assimilation associated strongly or consistently with health behavior. Of the four dimensions, use of English language associated most closely with increased screening, although most of the effects for language were small in magnitude. These findings, coupled with those of other studies, strongly suggest that cultural factors may have little impact on the health behavior of Hispanics. Access to and availability of services, affective reactions toward screening, and sociodemographic factors are stronger determinants of Hispanic health practices.Implementation effectiveness trial of a social influences smoking prevention program using schools and television
AbstractFlay, B. R., Hansen, W. B., Johnson, C. A., Collins, L. M., Dent, C. W., Dwyer, K. M., Grossman, L., Hockstein, G., Rauch, J., Sobel, J. L., Sobol, D. F., Sussman, S., & Ulene, A. (n.d.).Publication year
1987Journal title
Health Education ResearchVolume
2Issue
4Page(s)
385-400AbstractResults are reported from a quasi-experimental implementation effectiveness trial of a television and school-based smoking prevention program. The program used the social influences approach, focusing on peer, family and media influences on adolescents to become smokers, and providing grade 7 students (age 12-13) with the knowledge and skills to resist them. The evaluation allowed for assessments of: the value of coordinated television programming in increasing school, student and parent availability and acceptance of the program; the effects of program context (whether all or half of a grade cohort received the program in school) on participation and subsequent smoking behavior; and the effects of parental participation in prevention activities on subsequent student smoking. Data on availability, acceptance, participation and implementation are reported and immediate post-test and 1- and 2-year follow-up results. Strong effects of television programming and context on availability, acceptance and participation were found. Significant associations were also found between each television viewing and parental involvement and subsequent student smoking, but a lack of overall program effects on smoking outcomes limit their interpretation. Constraints on programming and research design suggest: (i) possible limitations to program effects we might reasonably expect in real-world applications of the social influences approach and (ii) the need for future true experimental efficacy trials to determine exactly what level of programming will be needed to achieve significant real-world effects.Psychosocial Predictors of Young Adolescent Cigarette Smoking: A Sixteen‐Month, Three‐Wave Longitudinal Study
AbstractCollins, L. M., Sussman, S., Rauch, J. M., Dent, C. W., Johnson, C. A., Hansen, W. B., & Flay, B. R. (n.d.).Publication year
1987Journal title
Journal of Applied Social PsychologyVolume
17Issue
6Page(s)
554-573AbstractUnderstanding the psychosocial factors that predict cigarette smoking onset in young people is of crucial importance for prevention efforts. The present study examined prospective psychosocial predictors of smoking in a three‐wave longitudinal data set. Similar in design to an earlier study by Chassin, Presson, Sherman, Corty, and Olshavsky (1984), the present study replicated their work, and extended it by (a) using composite predictors derived from exploratory factor analysis, (b) including prior behavior as a predictor, (c) using a design extended over three waves of data collection, and (d) using a sample composed primarily of urban teenagers. Subjects were 3295 7th‐grade students at the beginning of the study. The subjects completed a questionnaire containing items tapping cigarette smoking behavior and psychosocial items that have previously been shown to predict smoking behavior. Forty‐one psychosocial items on the Wave 1 (initial) questionnaire were factor analyzed, and five factors were retained. Subscale scores were constructed based on these factors and were used as predictors. Regression analyses were performed using the subscales and pretest smoking frequency to predict a continuous measure of smoking, and discriminant analyses were performed to predict transitions between qualitative levels of smoking. Prior smoking behavior was the most important predictor of future smoking. Four of the subscales, Social Disapproval, Risk Taking/Rebelliousness, Perceived Smoking Prevalence, and Motivation to Comply, were significant predictors. One subscale, Physical Consequences from Smoking, was not predictive of smoking in any of the analyses. The effect sizes cross‐validated well. It is suggested that an integrative model of smoking initiation developed by Flay, d'Avernas, Best, Kersell, and Ryan (1983) best summarizes the results of the present study.BINCLUS: Nonhierarchical Clustering of Binary Data
AbstractCliff, N., McCormick, D. J., Zatkin, J. L., Cudeck, R. A., & Collins, L. M. (n.d.).Publication year
1986Journal title
Multivariate Behavioral ResearchVolume
21Issue
2Page(s)
201-227AbstractBINCLUS is a clustering procedure designed for aggregating binary variables into relatively homogenous clusters. It uses any of several indices of binary association and operates by a variation on the “average linkage principle. It was tried out on a number of sets of artificial data and found to be extremely successful. With real data, where clusters are typically less clearly defined, two modifications were useful in clarifying the results. Results of using BINCLUS with two sets of real data are given.Factor Recovery in Binary Data Sets: A Simulation
AbstractCollins, L. M., Cliff, N., McCormick, D. J., & Zatkin, J. L. (n.d.).Publication year
1986Journal title
Multivariate Behavioral ResearchVolume
21Issue
3Page(s)
377-391AbstractThe present study compares the performance of phi coefficients and tetrachorics ailong two dimensions of factor recovery in binary data. These dimensions are (a) accuracy of nontrivial factor identification, and (b) factor structure recovery given a priori knowledge of the correct number of factors to rotate. Nontrivial factor identification was poor for both indices, with phi's performing slightly better than tetrachorics. In contrast, factor structure recovery was quite good when the correct number of factors was rotated. Phi coefficients generally yielded better factor structure recovery than tetrachorics and were better at preventing items from intruding onto factors where they did not belong, while tetrachorics were better than phi's at preventing items from being omitted from factors where they should have been included. The solutions based on tetrachorics contained many Hey wood cases. It is suggested that for most applications it is preferable to base factor analysis on phi coefficients.