A Novel Approach to Latent Class Modelling: Identifying the Various Types of Body Mass Index Individuals. (25th February 2020)

• Record Type: Journal Article
• Authors: Brown, Sarah; Greene, William; Harris, Mark
• Is Part Of: Journal of the Royal Statistical Society. Volume 183:Number 3(2020)
• Page Start: 983
• Page End: 1004
• Publication Date: 2020-02-25
• Subjects: Body mass index -- Expected values -- Latent class models -- Obesity -- Ordered probability models; Social sciences -- Statistical methods -- Periodicals; Statistics -- Periodicals; 300.15195
• DOI: 10.1111/rssa.12552
• Languages: English
• ISSNs: 0964-1998

Abstract

Summary: Given the increasing prevalence of adult obesity, furthering understanding of the determinants of measures such as the body mass index ( BMI ) remains high on the policy agenda. We contribute to existing literature on modelling the BMI by proposing an extension to latent class modelling, which serves to unveil a more detailed picture of the determinants of BMI. Interest here lies in latent class analysis with a regression model and predictor variables explaining class membership, a regression model and predictor variables explaining the outcome variable within BMI classes and instances where the BMI classes are naturally ordered and labelled by expected values within class. A simple and generic way of parameterizing both the class probabilities and the statistical representation of behaviours within each class is proposed, that simultaneously preserves the ranking according to class-specific expected values and yields a parsimonious representation of the class probabilities. Based on a wide range of metrics, the newly proposed approach is found to dominate the prevailing approach and, moreover, results are often quite different across the two.