Handling missing data in the association-marginal model through longitudinal data analysis : a simulation study

Abstract EN

Missing data can frequently occur in a longitudinal data analysis, where repeated measurements are taken over time.

Unfortunately, missing data can lead to large standard errors in parameter estimates because nonresponse is compounded across times of data collection to produce small longitudinal sample sizes.

Also, the problems of survey nonresponse(i.

e.

, reduction in statistical power and threat of parameter bias) are a particularly salient challenge for longitudinal researchers.

Thus, the main goal of this paper is to introduce a new idea that describes simultaneously the association structure (A) with the marginal distributions (M) of the responses for longitudinal data in the presence of missing data (MS), through a composite link.

This new idea (AM-MS) is of great importance where it is applicable for large and sparse tables.

In addition, it can also be used for fitting log linear models to contingency tables with missing data (MS) and fitting models with various assumptions about the missing data mechanisms either MCAR, MAR or NMAR.

A simulation study will be performed to apply this new idea, under various situations including (missing mechanisms, missing rates and five methods for handling missing data).

The goodness-of-fit test statistics and the number of adjusted residuals greater than 2 are used as evaluation criteria.