Classic Linear Mediation Analysis of Complex Survey Data Using Balanced Repeated Replication

Mediation analysis revealing whether an exposure affects a response via a mediator is scientifically and practically important. Applications to complex survey data from stratified multistage sampling have been technical challenges. Balanced repeated replication (BRR) is one common variance estimation method suggested for national/international complex surveys such as Tobacco Use Supplement to the Current Population Survey (TUS-CPS) and Program for International Student Assessment (PISA). Researchers working on complex survey data are commonly SAS users. To facilitate the application of mediation analysis to complex surveys, this study develops the SAS macro %MediationBRR implementing a recently proposed method that adjusts to complex sampling designs in classic linear mediation analysis and illustrates the applications using data from TUS-CPS and PISA.