ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

Methods for integrating probability and nonprobability samples have grown in popularity in recent years. This session will present the work of three researchers in this area: Professors Michael Elliott and Brady West from the University of Michigan, and Professor Rebecca Andridge from The Ohio State University. Two of the three proposed sessions will focus on the issue of generalized randomized controlled trials (RCTs). While RCTs remain the gold standard for causal inference because of their ability to control for both observed and unobserved confounders, the inability to generalize their findings from their often non-representative samples is a major limitation. Prof. Elliott will consider generalizing settings where RCTs suffer from non-compliance by extending Bayesian principal stratification methods to accommodate information from probability benchmark samples in settings with and without outcomes measured in the probability sample. Prof. Andridge will consider extensions of methods to account for selection bias in RCTs when benchmark data is available but there may be remaining unmeasured selection bias by the use of proxy pattern mixture models (PPMM) that rely on sensitivity parameters that capture the deviation from random sample selection and can be varied systematically to determine how robust trial results are to a departure from ignorable sample selection. Prof. West will consider the use of PPMM to assess possible unmeasured selection bias in regression models in more general settings, with a focus on use of mass imputation. All speakers will discuss detailed simulation study results, as well as applications: Elliott considers generalizing non-compliance from the UK REFLUX trial that studied the effect of laparoscopic surgery versus standard medical management on gastrooesophageal reflux disease; Andridge considers an RCT to determine whether yoga could improve inflammation and fatigue among breast cancer survivors; West considers generalizing depression measures from the nonprobability US Census Household Pulse Survey.