Latent class analysis applied to health behaviours

Abstract

The research literature points to health behaviour being multidimensional but is not consistent regarding the number or nature of dimensions. In this study, health behaviour data was subjected to latent class analysis, which is appropriate when underlying types rather than continua of health behaviour are hypothesized. Questionnaire data on various eating behaviours and on exercising, drinking alcohol, and smoking were obtained from 109 adult subjects. Each behaviour was dichotomized, the split being at the point which best corresponded to contemporary public health advice. Latent class analysis of the eating behaviours was consistent with a two class model of eating: healthy vs more mixed eating. Thereby, a single eating behaviour variable was created. Latent class analysis of this eating behaviour along with the exercising, alcohol, and smoking behaviours was modestly consistent with a two-type model of health behaviour: healthy behaviour vs more mixed behaviour. Latent class analysis may be an appropriate technique to apply to health behaviour data if there is reason to believe that the latent variables are categorical rather than continuous.