Regression models for the analysis of psychiatric data

摘要

Three regression models were fitted to a set of psychiatric contacts: two parametric (based on the negative binomial distribution and on the generalised Waring distribution (GWR)) and one semiparametric (based on the cumulative mean function of the number of contacts). They were all able to account for the large amount of overdispersion and gave similar estimates of the regression coefficients and of their SEs. However, the GWR model could give additional information to the clinician owing to the possibility to quantificate the proneness and the liability of different sets of patients in contacting psychiatric services.