Zero-inflated ordinal data models with application to sport (in)activity

Traditional models for ordinal data (as CUB models or cumulative models with logit/probit link, among others) present limits in explaining the surplus of zero observations, especially when the zeros may relate to two distinct situations of non-participation/inactivity and infrequent participation, for instance. We propose an extension of standard models: zero-inflated CUB models and zero inflated ordered cumulative (ZIOC) probit/logit models handling the GECUB models and using a double-hurdle combination of a split (logit/probit) model and an ordered probit/logit models, respectively. Both extensions, potentially, relate to different sets of covariates. Finally, models are applied to Sport surveys. Specifically the paper explores the determinants of sport (in)activity: the frequency and the probability of sports participation. It distinguishes between genuine ‘non-participants’ and the ones who do not participate at a time but might under different circumstances.