The data structure and inferential goals for Health Services Research (HSR) and evaluation require use of a hierarchical model (HM) that accounts for the structure and specifies both population values and random effects for units such as clinics, physicians and health service regions. HMs properly account for the sample design and structure scientific and policy-relevant statistical inferences. Applications require valid and efficient estimation of population parameters (such as the average death rate), estimation of between unit variation (variance components) and inferences on unit-specific random effects. These include unit-specific ranks (to be used in profiling/league tables) estimates of the histogram of random unit-specific effects, estimation of how many and which unit-specific effects exceed a threshold, and identification of extremely poor and good performers. No single set of estimates can effectively address these multiple goals, and we will develop and evaluate a """"""""panel"""""""" of goal-specific summaries and inferences. The panel will combine efficiency (making good use of the available information) with robustness (high efficiency over a broad range of underlying assumptions).
Our specific aims are: structuring inferences via the Bayesian formalism; development, implementation, and application of robust priors; evaluation of the robustness, efficiency and operating characteristics of the panel of summaries; development and implementation of computational approaches, with focus on massive data sets; preparation of case studies based on Rand projects; hosting a summer intern at Rand. The proposed research will produce innovative statistical methodology and computer implementation, provide guidance on addressing multiple HSR goals, communicate interesting and informative case studies, educate and acculturate pre-doctoral students.
Showing the most recent 10 out of 17 publications
