ANALYSIS OF RADIO LISTENING TIME USING A HIERARCHICAL BAYESIAN TWO-SIDED CENSORED TOBIT MODEL

Abstract

This study seeks to elucidate the generative mechanisms governing radio listeners’ listening durations. The proposed model formalizes the intrinsic periodicity of individual listening behaviors within the framework of harmonic regression, integrating explanatory variables such as the day of the week and program genre. To account for inter-individual heterogeneity, a hierarchical Bayesian model is employed, with parameter estimation conducted via the Markov Chain Monte Carlo (MCMC) method. Empirical validation is performed using data from the radio streaming service “radiko.” The results indicate that listening durations exhibit diverse periodic structures specific to individual listeners and that preferences for different program genres are highly heterogeneous. Furthermore, cluster analysis based on listener-specific heterogeneous response coefficients identifies four distinct periodic listening patterns.