Introduction. This simulation study compares 3 methods for estimating b , in light of longitudinal data. The 3 methods will be compared in terms of: accuracy in their estimate of s.e.( b&d4; ), and relative efficiencies in their estimate of b Purpose of this study. In recent years, several methods for dealing with correlated residuals in regression analysis have been incorporated into standard statistical software, such as SAS and Stata. However, there are gaps in our knowledge as to how these methods compare with one another. The purpose of this simulation study is to compare several methods that have recently become available to applied statisticians. The methods to be compared are: (1) Generalized estimating equation (GEE) methodology, (2) Multilevel modeling (that is, hierarchical random effects models), (3) Resampling methodology (the bootstrap). Specifically, this study compares the above methods in terms of both efficiency of the estimators, as well as potential bias in the estimate of variance Methods. This study is based on the following regression model: Yij=b0+noj +b1+n1j Xij+3ij,where beta0 = the 'average' (across all individuals) part of the intercept, nuoj = the individual-specific deviation in the intercept, beta1 = the 'average' (across all individuals) part of the slope, nu1j = the individual-specific deviation in the slope, and &egr;ij = a measurement and individual-specific error term, or residual Through systematic Monte Carlo simulations, each of the three comparison methods (GEEs, Multilevel Modeling, and Bootstrapping) estimate the parameters in the above model. The specific parameter of interest is b1 , which is the fixed part of the slope. The results of the three methods are compared to determine relative efficiencies in the estimate of b1 and accuracy in the estimate of var( b&d4;1 )