Using marginal mean models with data from a longitudinal survey having a complex design: some advances in methods

In recent years, longitudinal surveys, where sample subjects are observed over two or more time points, are being undertaken by government agencies in order to provide longitudinal data for analytic studies and to aid in the development of public policy. At Statistics Canada, for example, the National Population Health Survey (NPHS), the National Longitudinal Survey of Children and Youth (NLSCY) and the Survey of Labour and Income Dynamics (SLID) were all launched in the mid 1990’s for this purpose and now have available several cycles of data on the same samples of individuals. Data from such surveys can be used for a variety of purposes including (a) gross flows estimation, (b) event history modeling, (c) conditional modeling of the response at a given time point as a function of past responses and present and past co-variables, and (d) modeling of marginal means of responses as functions of co-variables. Binder (1998) gives an excellent account of the possible uses of longitudinal survey data.