The Measurement and Production of Quality in Long-Term Care Facilities in the US: 2012-2017

The proposed research seeks to analyze the production of quality in long-term care. First, we present new evidence on quality variation across nursing homes for long-term care patients. We estimate substantial, persistent variation in quality, with 90-day stay in a standard deviation better nursing home increasing the subsequent 90-day survival rate of the average patient by 1 percentage point. While there are strong correlations between true quality and readily-observed proxies for quality, such as facility ownership, staffing, and Nursing Home Compare star ratings, there is substantial residual variation in quality, both within and across markets. To understand the determinants of this variation, our second exercise borrows tools from the production function literature to estimate a model oflong-term care production as a function oflabor, capital, and a residual productivity shifter. Preliminary estimates suggest a substantial quality-quantity tradeoff: all else equal, improving quality by 1 percentage point (ppt) would require a nursing home to decrease its patient load by 12.6%. We also show that quality variation is determined more by heterogeneity in preferences than by nursing homes' productivity or inputs. Our third exercise uses the model estimates to explore the role of reallocation in improving patient outcomes. A decomposition exercise suggests that allocative forces have had little impact on quality; secular improvement in patient outcomes over time is driven almost entirely by within-facility improvements. A counterfactual simulation of mortality gains under mechanical reallocation of patients to higher quality facilities similarly has little effect. Re-sorting patients to the best facilities in their markets, subject to capacity constraints, would increase the per-quarter change in survival rates by 0.35 ppt, about a third of a standard deviation. Next, we will dig deeper into these findings. First, we will explicitly model how capacity constraints discipline reallocative forces in nursing home markets, and how policies such as certificate of need laws and construction moratoria have contributed. Second, motivated by our finding that facility for-profit status and chain ownership impact quality, productivity, and preferences, and by sparse and error-prone nursing home ownership data, we will use new data and machine learning to create a high-quality panel of nursing home ownership and estimate the effects of ownership changes on our model fundamentals. Lastly, we will expand our model to allow for facilities to produce two different kinds of care with different production functions and potential profitability: long-term care for (typically) Medicaid patients and post-acute care for (typically) Medicare patients.