A tutorial for using propensity score weighting for moderation analysis: an application to smoking disparities among LGB adults
share
Objective. To provide step-by-step guidance and STATA and R code for using propensity score (PS) weighted analyses when estimating moderation effects. Data Sources and Study Setting. Using data on 41,832 adults from the 2019 National Survey on Drug Use and Health (NSDUH), we examine if gender moderates the association between sexual minority status and adult smoking prevalence and illustrate the application of the key steps outlined in this tutorial Study Design. Tutorial illustrating the key steps for estimating and testing moderation. Steps include (1) examining covariate overlap across treatment/exposure groups within levels of the moderator, (2) estimating the PS weights, (3) assessing the quality of the PS weights, (4) estimating the moderated effects, and (5) assessing sensitivity of findings to unobserved confounding. Principal Findings. All balance criteria suggesting successful balance has been achieved within each level of the moderator when estimating the needed PS weights. In our outcome models, the estimated risk difference of current smoking for LGB vs heterosexual males is 0.05 (95 contrast, the risk difference for female LGB individuals is larger, equaling 0.15 (95 case study. Notably, the risk difference for females was highly robust to unobserved confounders while the risk difference for males was highly sensitive and easy to wipe away. Conclusions. Estimating robust moderation effects in observational studies requires thoughtful consideration of imbalances between the key groups being compared. Propensity score weighting within each level of the moderator can improve the estimated moderation effects by minimizing bias from imbalance within the moderator subgroups.
READ FULL TEXT