Statistical modelling for falls count data.

摘要: Falls and their injury outcomes have count distributions that are highly skewed toward the right with clumping at zero, posing analytical challenges. Different modelling approaches been used in published literature to describe falls distributions, often without consideration of underlying statistical assumptions. This paper compares use modified Poisson negative binomial (NB) models as alternatives (P) regression, for analysis fall outcome counts. Four different count-based regression (P, NB, zero-inflated (ZIP), (ZINB)) were each individually fitted four separate datasets from Australia, New Zealand United States. The finite mixtures P NB also compared standard model. Both (F, Vuong bootstrap tests) graphical select compare models. Simulation studies assessed size power model fit. study confirms over-dispersed, but not dispersed due excess zero counts or heterogeneous population. Accordingly, generally provided poorest fit all datasets. improved significantly both was model, Although there little difference between ZINB models, interests parsimony it is recommended future involving data routinely preference mixture distribution. fact these conclusions apply across samples older people participating methodology, adds strength this general guiding principle.