Analysis of clustered data: A combined estimating equations approach

Clustered data often arise in biomedical research. Examples include data from longitudinal studies and data sampled within clusters. This dissertation proposes a new regression analysis method for clustered data. The method results in increased estimation efficiency, over generalized estimating equations in particular settings, by optimally weighting and combining sources of information from the data through combinations of estimating equations. The proposed method also avoids modeling decisions regarding the true correlation structure of the data.;In this dissertation, important sources of information from clustered data settings are introduced, and by the proposed method, are optimally combined. Coefficient and variances estimates are proposed and the properties of these estimates are theoretically justified. An algorithm is developed to solve the estimating equations in practice. Increased estimation efficiency of the proposed method, relative to a generalized estimating equations approach, is demonstrated through derivations and simulations. Several applications of the methodology are also discussed.