West Coast Ballroom
Evaluating A Key Instrumental Variable Assumption Using Randomization Tests (306660)*Luke John Keele, University of Pennsylvania
Keywords: instrumental variables, natural experiments
Instrumental variable (IV) analyses are becoming common in health services research and epidemiology. Most of these analyses use instruments that are naturally occurring, such as distance to a capable hospital. With IVs of this type, the investigator must assume the instrument is as-if randomly assigned. While this assumption cannot be tested directly with data, it can be falsified. Currently a number of different falsification tests have been developed. Typically, these tests compare relative prevalence or bias in observed covariates between the instrument and the exposure. Under these tests, investigators are required to make a covariate by covariate judgement about the validity of the IV design. We propose an alternative falsification test that compares the balance on the IV or exposure to the balance that would have been produced under as-if randomization. Our method allows for graphical comparisons as well as a global balance test to understand the overall level of imbalance produces by the IV or the treatment. We demonstrate these methods on a recent IV application using bed availability in the intensive care unit (ICU) as an instrument for admission to the ICU.