Objective To examine the appropriateness of different statistical models in analysing falls count data. Methods Six count models (Poisson, negative binomial (NB), zero-inflated Poisson (ZIP), zero-inflated NB (ZINB), hurdle Poisson (HP) and hurdle NB (HNB)) were used to analyse falls count data. Empirical evaluation of the competing models was performed using model selection criteria and goodness-of-fit through simulation. Data used were from a prospective cohort study of women aged 40-80 years. Results Of the 465 women analysed, 330 (71%) did not fall at all. The analyses identified strong evidence of overdispersion in the falls data. The NB-based regression models (HNB, ZINB, NB) were better performed than the Poisson-based regression models (Poisson, ZIP, HP). Vuong tests favoured the HNB model over the NB and ZINB models and the NB model over the ZINB model. Model accuracy measures and Monte Carlo simulation of goodness-of-fit confirmed the lack of fit of the Poissonbased regression models and demonstrated the best fit for the HNB model with comparable good fit for the ZINB and NB models. Conclusions Falls count data consisting of a considerable number of zeros can be appropriately modelled by the NBbased regression models, with the HNB model offering he best fit. The evaluation procedure presented in this aper provides a defensible guideline to appropriately odel falls or similar count data with excess zeros.
ASJC Scopus subject areas
- Public Health, Environmental and Occupational Health