In a longitudinal study the subjects are observed for a specified period of time and information about them is gathered relating to an event of concern at different time points. Researchers are often interested in analyzing data that arise from longitudinal or clustered design where there exists correlation among observations of a given subject. In analyzing longitudinal data, this must be taken into account to avoid misleading inference. But if the outcomes are binary or counts, general maximum likelihood based approaches are less tractable. Though logistic regression may be use to analyze and identify the risk factors and prognostic factors for the disease status, it ignores the possible correlation among the repeated observations. To overcome the difficulty, Generalized Estimating Equations (GEE) was suggested by Liang and Zeger (1986). The generalized estimating equation is an important and widely used approach in such analysis that does not require the complete specification of the joint distribution of repeated measurements. It is based on two moments of the outcome variables under the assumption that variance is known function of mean. Since the true correlation among the repeated responses is unknown, GEE offers to take a working correlation for analysis. In our study we consider four correlation structures, namely- independence, exchangeable, pairwise. The autoregressive is not applicable for unequal follow-up.

In this study we introduce bootstrap resampling of data in GEE and check the correlation structure, which structure gives the better estimate of coefficient considering standard error. We employ the several bootstrap samples of the maternal morbidity data in GEE. And to select best correlation structure which is our main interest in the study.