Restricted Estimation for More Efficient Causal Inference in Longitudinal Studies

• Activity Number: 416
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Edward H. Kennedy*+ and Marshall M. Joffe
• Companies: University of Pennsylvania and University of Pennsylvania
• Keywords: causal inference ; semiparametric inference ; longitudinal data

Abstract

Structural nested models (SNMs) are a powerful way to represent causal effects in longitudinal studies with treatments that change over time. Parameters of SNMs are typically estimated via semiparametric approaches, which yield robust results but can also be very imprecise. In particular, optimally (i.e., locally) efficient estimators for SNMs are very challenging to implement in practice, and more straightforward estimators can yield problematic losses in efficiency. To remedy these problems, we propose an approach for increasing efficiency of SNM estimators based on finding optimally efficient estimators within restricted classes. Our restricted estimation approach is straightforward to implement, does not require any extra modeling assumptions (nor any extra model fitting), and comes with guarantees of increased asymptotic efficiency. We explore finite sample properties using simulated data, and illustrate the approach in a real example to estimate the effect of erythropoietin on hematocrit among patients with kidney disease.