Modeling Longitudinal Count Data with Missing Values: A Comparative Study

Longitudinal data differs from other types of data as we take more than one observation from every subject at different occasion or under different conditions. The response variable may be continuous, categorical or count. In this article the focus is on count response. The Poisson distribution is the most suitable discrete distribution for count data. Missing values are not uncommon in longitudinal data setting. Possibility of having missing data makes all traditional methods give biased and inconsistent estimates. The missing data mechanism is missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). This article compares different methods of analysis for longitudinal count data in the presence of missing values. The aim is to compare the efficiency of these methods. The relative bias and relative efficiency is used as criteria of comparison. Simulation studies are used to compare different methods. This is done under different settings such as different sample sizes and different rates of missingness. Also, the methods are applied to a real data.