Analysis of mixed longitudinal (k,l)-Inflated power series, ordinal and continuous responses with sensitivity analysis to non-ignorable missing mechanism

Abstract

Authors propose a joint random effect model for analyzing longitudinal mixed count, ordinal and continuous responses, where the count response is inflated in two points (k and l) and there is the possibility of non-ignorable missing values for all responses. The random effect approach is used to investigate both of the correlation between mixed responses and the correlation of longitudinal nature. The likelihood-based methods are used to inference about the parameters of the model. However, the interpretation of the fitted model highly depends on the assumptions imposed on the missing mechanism, so the authors extend a general index of sensitivity to non-ignorability (ISNI) methodology to assess the impact of the parameters of non-ignorability in the missing mechanisms on key inferences. A simulation study is performed in which for count response (k,l)-inflated Poisson and (k,l)-inflated negative binomial distributions are considered. Also, an application using a clinical data set is discussed.