The estimation of population/subpopulation average treatment effects has been a subject of extensive research. Randomized experiments are the ideal way to estimate the effects of interventions. However, in many applications, randomized experiments ARE impractical because of financial, logistical, or ethical considerations. In cases where an ideal randomized experiment cannot be performed, our knowledge of causal effects must come from non-randomized (i.e. observational) studies. This talk will focus on the proposal of an outcome-free three-stage procedure to estimate causal effects from non-randomized studies, which we call MITSS. First, we create subclasses that include observations from each group based on the covariates. Next, we independently estimate the response surface in each group using a flexible spline model. Lastly, multiple imputations of the missing potential outcomes are performed. A simulation analysis that resembles real life situations is conducted to compare MITSS to other commonly-used methods is carried out. In  many of the conditions examined, MITSS produced a valid statistical procedure while providing a relatively precise point estimate and a relatively short interval estimate. We will demonstrate extensions of MITSS to estimate causal effects in multiple health services research studies. The extended procedures address common limitations encountered in many observational studies: CLEAR definition of causal effect estimands, non-collapsibility, confounded assignment mechanisms, treatment heterogeneity and dealing with forms of missing data other that due to the assignment mechanisms.

<span font-size:medium;font-variant-numeric:normal;font-variant-east-asian:normal;line-height:25.6px;white-space:normal;widows:1;" style="color:#31353B">Roee Gutman, Department of Biostatistics, Brown University, Providence, RI

Zoom link: