How reliable are hospital efficiency estimates? Exploiting the dual to homothetic production

For scientific use, stochastic frontier estimates of hospital efficiency must be robust to plausible departures from the assumptions made by the investigator. Comparisons of alternative study designs, each well within the 'accepted' range according to current practice, generate similar mean inefficiencies but substantially different hospital rankings. The three alternative study contrasts feature (1) pooling vs partitioned estimates, (2) a cost function dual to a homothetic production process vs the translog, and (3) two conceptually valid but empirically different cost-of-capital measures. The results suggest caution regarding the use of frontier methods to rank individual hospitals, a use that seems to be required for reimbursement incentives, but they are robust when generating comparisons of hospital group mean inefficiencies, such as testing models that compare non-profits and for-profits by economic inefficiency. Demonstrations find little or no efficiency differences between these paired groups: non-profit vs for-profit; teaching vs non-teaching; urban vs rural; high percent of Medicare reliant vs low percent; and chain vs independent hospitals.