Causal Regression Discontinuity and Its Application to a Biomedical Example

• Activity Number: 621
• Type: Contributed
• Date/Time: Wednesday, August 3, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Marie-Abele Bind* and Fan Li and Farbizia Meali
• Companies: and Duke University and University of Florence
• Keywords: Regression discontinuity ; Epidemiology ; Biomedicine ; Causal inference

Abstract

The causal relationship between body mass index (BMI) and triglycerides is unknown. Considering the Normative Aging Study and a regression discontinuity approach (with a cutpoint at 30kg/m2), we estimate the causal effect of BMI on triglyceride levels. We assume that participants with BMI greater than 30kg/m2 are advised by their doctor to change their lifestyle (e.g., diet, physical activity). The optimal bandwidth in regression discontinuity is unclear. We choose one bandwidth for which participants on either side of the discontinuity have balanced background covariates and compare our results to the one found by the Imbens-Kalyanaraman Optimal Bandwidth. Using the Imbens-Kalyanaraman Optimal Bandwidth, the regression discontinuity of BMI on triglycerides is estimated to be equal to -48mg/dL (p-value=0.12), suggesting that participants with BMI lower than 30kg/m2 do not benefit from our hypothesized medical advice.