Abstract: Extraction of Latent Class Progressions in Behavioral Data

Extraction of Latent Class Progressions in Behavioral Data Jennifer Lord-Bessen and Ying Liu Department of Psychology, Fordham University This study developed and tested a new exploratory latent class methodology for repeated measures data, latent class progression analysis (LCPA), which identifies different subgroups’ developmental trajectories over three or more time points where the trajectories are in terms of a progression through levels of a discrete latent variable. The method employs a three-step procedure involving multigroup latent class analysis to establish the stages of behavior, individuals’ assignment to the stages, and then use of repeated measures latent class analysis to extract subgroups’ developmental trajectories (progressions) through the stages. To demonstrate the usefulness of LCPA in psychological research, it was applied to a behavioral dataset investigating adolescents’ change in sexual habits and risky behaviors. Using public-use data from the National Longitudinal Survey of Youth 1997 (Bureau of Labor Statistics, 2012). This survey collected longitudinal data from a nationally representative sample of households, where 8,984 respondents aged 12–16 in 1997 were interviewed annually. The subset used in this study (N = 2,937) consisted of youths aged 17–18 at Round 2 who did not report being married at Round 2, 3, or 4, and the data collected was measured during those three time points. The final sample comprised 51% boys and 49% girls, with 56% White, 29% African American, 2% Asian or Pacific Islander, and 13% other. Seven items were chosen covering the topics of dating, sexual, and risky behaviors. LCPA was used to analyze the data, and six stages of behavior were identified We would like to thank our SMEP sponsor, David Budescu, for his advice and support. Correspondence concerning this abstract should be addressed to Jennifer Lord-Bessen, Fordham University, Department of Psychology, Bronx, NY 10458. E-mail: jlord2@fordham.edu and named (safe multi-partners, risky multi-partners, safe monogamous, risky monogamous, complete abstainers, and sex abstainers). The names were indicative of the responses to the items. For example, subjects in the safe multi-partners stage were likely to identify as having two or more dating and sexual partners but not likely to engage in any risky behaviors (exposure to STD, smoking cigarettes, drunkenness, use of marijuana) in the past year. Membership in the stages was calculated at each time point and ranged from 4–37% of the sample at any time. In addition, five behavioral progressions were identified (serial sexual abstainers, safe eventual monogamists, serial safe multi-partners, serial complete abstainers, serial risky multi-partners). Here, the names were representative of subjects’ progression through the identified stages. For example, safe eventual monogamists are individuals who did not engage in the risky behaviors throughout all three times. They were equally likely at Time 1 to be in the safe multi-partners, safe monogamous, or complete abstainers stages, but were most likely to be found in the safe monogamous group by Time 3. The progressions ranged in size from 10–28% of the population. With the development of LCPA, researchers now have a tool to extract and identify the different ways that behavior can change over time, where change is represented by categorical variables across three or more time points.