2016 — Nonlinearity and Endogeneity in Health Services Research
Garrido MM, James J. Peters VAMC REAP/GRECC; Mount Sinai School of Medicine; Deb P, City University of New York; Burgess, Jr. JF, COLMR - VA Boston Healthcare System; Boston University School of Public Health; Penrod JD, James J. Peters VAMC REAP/GRECC; Mount Sinai School of Medicine;
Common outcomes in health services research, including costs and utilization, require a thorough understanding of how to approach issues of skewness, non-negative outcomes, censoring, and endogeneity. In this workshop, we will tackle some of these issues with the following two objectives: 1) To demonstrate several models for analyzing endogenous treatment effect models for nonlinear outcomes: a) propensity scores, b) control functions, and c) full information maximum simulated likelihood (FIMSL). 2) To distinguish between types of treatment effects commonly reported from these methods, including average treatment effect on the treated (ATET), average treatment effect (ATE), and local average treatment effect (LATE).
Using VA cost and utilization data, we will demonstrate the construction of models with propensity scores, control functions (including the special case of two-stage residual inclusion), and FIMSL. We will describe the advantages and disadvantages of each set of methods and will demonstrate the use of Stata code for control function models. In addition, we will use VA data from a single cost study to demonstrate the differences among types of treatment effects. Participants will be encouraged to send methodological questions to the lead author at least the day prior to the workshop that can be worked through as a group. There will be dedicated time for participant questions and discussion.
This workshop is targeted to health services researchers, statisticians, and methodologists.
Assumed Audience Familiarity with Topic:
The audience should have a basic understanding of regression techniques.