Abstract:
|
Many analyses involve fitting regression models to sample survey data and these surveys use complex design features such as unequal probabilities of selection, geographical clustering and stratification based on key variables. Typically, these design features are related to the outcome or covariates in the regression model. Ignoring the design features may lead to biased estimates, especially if the population model is not perfectly specified. Since all models are approximations, it is important to tease out these design features before fitting the regression models. Furthermore, many data sets may be subject to both unit and item nonresponse. The basic idea is to "uncomplex" the complex survey design, handle missing data and fit the population level regression model, all simultaneously. A Bayesian framework is used to implement this idea to derive an iterative approach for estimating the parameters of regression models under a variety of scenarios. A simulation study investigates the repeated sampling properties of the estimates. The method is illustrated using data from National Health and Nutrition Examination Survey.
|