Data from household surveys are from clustered samples of persons who are often selected at differential rates. These aspects of the sampling result in nonindependence and unequal weighting of the observations that should be considered during the analysis stage. Health survey data are used extensively in cohort studies through long term followup of the sample, case-control studies by providing population controls, and cross-sectional studies. Research has been conducted into extending a variety of statistical methods for application to survey data. Some of the major areas of research are as follows: - Proportional hazard regression analysis was developed for complex health survey data with time dependent covariates and left truncated time to event. These methods were applied to the followup of the First National Health and Nutrition Examination Survey. - Graphical methods were generalized to survey data which address the sample weighting and the large number of observations found in many national health surveys. - Methods were developed for estimating super-population variances from survey data. Corrections to standard variance estimators provide consistent estimates for super-population variances. - Methods for computing confidence intervals for small samples from surveys are being developed. - Research was conducted on the estimation of adjusted treatment means for generalized linear and proportional hazard regression models. For each of these areas, manuscripts are either under submission or being written.
