Abstract:
|
Generalized estimating equations (GEE) are often used for the marginal analysis of longitudinal data. GEE utilizes a working correlation structure for regression parameter estimation. This working structure can be very important in terms of statistical power because the more accurately this structure is modeled, the more efficiently regression parameters are estimated. Numerous criteria have been proposed to select a working structure. Unfortunately, no selection criterion has been proven to always work better than all other criteria. Furthermore, some popular criteria can actually be detrimental in some settings. In practice, the data analyst will not know which criterion will work best. Therefore, we propose how to utilize information from multiple criteria. We demonstrate the benefits of our proposed approach via a simulation study in a variety of settings and then in an application example.
|