Bias In Variance Estimation Using Re-Sampling Of Longitudinal And Nested Administrative Health Data.
*Bassam Dahman, Virginia Commonwealth University 

Keywords: resampling, bias, longitudinal, hierarchical

Re-sampling methods are widely used in estimating the variance and standard errors of parameter estimates and predicted values, and in determining the statistical significance in hypothesis testing using administrative health data. In longitudinal and nested models, bias might be introduced to these estimates if the samples generated by the re-sampling routine differ in their distribution and longitudinal or nesting structure from the original data. This bias might be particularly large in the presence of missing data. In this paper we study and demonstrate the properties of different methods of re-sampling longitudinal and nested data. Using a simulation study based on longitudinal hospital level data and nested discharge data we compare between the different re-sampling routines, and evaluate the bias corrections required for each method.