Using stochastic frontier models to estimate treatment heterogeneity and inefficiency
Luo, Z., Bray, J. W., Cowell, A., & Gardiner, J. C. (2008, August). Using stochastic frontier models to estimate treatment heterogeneity and inefficiency. Presented at 2008 Joint Statistical Meetings (JSM), Denver, CO.
Researchers of clinical trials frequently encounter outcomes that are bounded between zero and a fixed upper limit. We present strategies for estimating heterogeneous treatment effects for Binomial outcomes using maximum simulated likelihood. We show that the effects of unobserved heterogeneity can also be interpreted as inefficiency in the underlying health production process. Our model is motivated and illustrated by the COMBINE trial for alcohol dependence. Our results indicate that naïve models do not adequately account for the variation in the outcomes. Unobserved patient heterogeneity needs to be appropriately modeled in clinical trials even if randomization balances both observed and unobserved characteristics of all patients.