Unweighted versus Weighted Regression Methods May Be Sufficient to Analyze Complex Survey Data in the Canadian Longitudinal Study on Aging

IntroductionComplex surveys use stratified or cluster sampling to recruit participants. Researchers analyzing these surveys often wish to make inferences about the source populations from which the participants are drawn. In such cases, methodologists recommend employing sample weights in regression analyses; however, the utilization of weights in studies of associations are not without dispute. Materials and methodsTo help guide analyses of complex surveys, we utilized baseline data from the Comprehensive Cohort of the Canadian Longitudinal Study on Aging (CLSA) (n = 30,097) and compared unweighted and weighted regression analyses of the association between social support availability (SSA) and cognitive function. We also conducted simulation studies to validate our findings in the CLSA. ResultsThe regression coefficients [Formula] for SSA were similar across the unweighted and weighted regression models; the standard errors of the [Formula] were lower in the unweighted models. The simulation studies mimicking CLSA confirmed these findings. Overall, our findings demonstrated a small advantage for the unweighted analysis of CLSA data due to the smaller standard errors. DiscussionAlthough we cannot guarantee that this would be the case for all association analyses with CLSA data, the current study showed the use of analytical weights was not necessary for our associations of interest.