A parsimonious, stacked latent-class methodology for predicting behavioral heterogeneity in terms of life-constraint heterogeneity

Our conjecture is that for many recreational activities a significant amount of the variation in the sites visited can be explained, and predicted, by variation in life constraints such as kids, BMI (body-mass index) fitness, skill, and health. The objective is to develop a parsimonious method for identifying behavioral heterogeneity caused by life-constraint heterogeneity and separating it from that caused by preference heterogeneity. We estimate, for two different recreational activities, with two independent data sets, how much behavioral heterogeneity can be attributed to life-constraint heterogeneity. We develop and estimate a stacked latent-class approach to life constraints, assuming individuals have many correlated life constraints. First, at the bottom of the stack, a latent-class life-constraint model is specified and estimated; then life-constraint class becomes a covariate in a behavioral latent-class model of participation and site selection. We find, with both simple statistics and behavioral models, that life-constraint classes explain a significant amount of the observed behavioral heterogeneity. Prediction is a critical reason to distinguish the influence of current constraints from the influence of current preferences: it is easy to directly observe life-constraint levels. Stacked latent-class models have many potential applications, besides ours.