Log Linear Modeling with Complex Survey Data
Chris Skinner
University of Southampton, United Kingdom

Different approaches have been proposed for fitting log linear models to contingency tables based on complex survey data. One approach seeks to handle differential sampling via an offset term in the model. A second pseudo maximum likelihood approach fits models to the weighted table. This paper will investigate and compare the properties of inference procedures for these methods. Point estimation, variance estimation and testing will be considered. The case when survey weights are constant within the cells of the table will be contrasted with the case when they are variable. Some empirical illustration will be provided using data from a survey in France collecting data on daughter's and father's social class.