Causal Inference with Complex Survey Designs

Public opinion surveys are a highly valuable resource for social scientists, as they allow researchers to learn about the determinants of attitudes toward diverse issues and to test hypotheses about political behavior. Using survey data to arrive at estimates of causal effects that are generalizable to the target population of interest (i.e., making population-level inferences) can be challenging, however, because it requires taking into account complex sampling designs and data collection issues such as unit nonresponse. Researchers who disregard important elements of the survey design run the risk of obtaining measures of causal effects that do not apply to the target population.