LATENT CLASS ANALYSIS WITH INFORMATIVE PRIORS

Open Access
Author:
Root, Tammy Lynn
Graduate Program:
Human Development and Family Studies
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
August 31, 2007
Committee Members:
  • Linda Marie Collins, Committee Chair
  • Leann L Birch, Committee Member
  • Runze Li, Committee Member
  • Rhonda Belue, Committee Member
Keywords:
  • latent class analysis
  • small sample
  • identification
  • Bayesian methodology
  • informative priors
Abstract:
Latent class analysis (LCA; Goodman, 1974; Lazarsfeld & Henry, 1986) is frequently used in the social and behavioral sciences to identify and model latent classes composed of similar cases in multivariate categorical data. When samples are small in relation to the size of the contingency table and model complexity two challenges arising are biased parameter estimates and identification problems. Bayesian methods, an exciting perspective that is gradually beginning to make its way into the social and behavioral sciences (Howard, Maxwell, & Fleming, 2000; Raftery, 1995; Western, 1999) allow the incorporation of additional information gathered from other sources into an analysis thereby providing flexibility with sample size requirements. This prior information is based on what (we believe) we already know based on theory and prior empirical evidence. This additional information may help establish identification and reduce parameter bias in LCA, particularly when sample sizes are relatively small. LCA with informative priors is a practical approach for assisting with parameter recovery and model identification particularly with samples that are small relative to the size of the contingency table and model complexity. This dissertation is divided into two manuscripts with the purpose of describing LCA with informative priors and demonstrating its usefulness using data on disordered eating and substance use. The first manuscript is a methodological paper discussing the procedure, selected results from the simulation study, and an empirical example which examines patterns of substance use among a disordered eating sample. The second manuscript, a technical report, provides more extensive simulation results not presented in the methodological paper.