The Importance of Underlying Ordering Assumptions in Causal Inference

• Activity Number: 444
• Type: Invited
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Jessica Gerald Young*+ and Miguel A. Hernán and James M. Robins
• Companies: Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health
• Keywords: causal inference ; assumptions ; ordering ; parametric g-formula ; inverse probability weighting ; marginal structural models

Abstract

When estimating the causal effect of a time-varying treatment using longitudinal data, most applications assume that the ordering of confounders with respect to treatment is known. In interval cohorts, however, the ordering of covariates measured simultaneously is generally unknown. While much attention has been paid to the importance of assumptions in making causal inferences, the consequences of assumptions concerning ordering are often ignored. Effect estimates may be sensitive to the choice of ordering and investigators must be careful in defining their estimator, as well as selecting an appropriate estimation algorithm, depending on this choice. Here we consider various underlying ordering assumptions and how they impact effect estimators and associated estimation algorithms under interventions on single and multiple treatments. Focus is on estimators based on the parametric g-formula and inverse probability weighting of marginal structural models. Examples are given from the coronary heart disease and HIV literature.