Market Structure as a Determinant of Patient Care Quality: The Importance of Controlling for Specialization

The marked lack of consensus about important issues like the relationship between competition and patient-care may seem surprising. After all, various databases track patient-utilization and outcomes, making it straightforward to construct and compare raw measures of quality. However, given the reasonable expectation that patients select the treatments most likely to benefit them, it is difficult for analysts to be sure that simple comparisons of facility averages capture the causal impact of different characteristics. For example, if higher quality treatments are associated with certain market characteristics, and unobservably sicker patients systematically prefer higher quality, then simple analyses of the impact of market structure on average quality will be biased.

This type of endogeneity is a significant concern in health-care settings. However, one might also worry about the possibility of expected differences in treatment across individuals. This is especially the case given that the medical literature has documented considerable heterogeneity – sometimes attributed to demographic variation – in responsiveness to treatment for many conditions. If present, such random treatment effects would mean that empirical methods that only account for unobserved variation in the level of sickness also produce biased estimates of the true average effect of treatment.

To address the possible importance of heterogeneous responses to treatment (i.e., random coefficients), I develop a control function (CF) approach to estimating the average quality of many different treatments. The CF model addresses selection problems by controlling directly for the impact of unobserved, as well as observed, patient characteristics. Thus, the likelihood of simultaneity bias is markedly reduced since the quality estimate takes into account the types of patient being treated. Once different treatments’ average qualities have been consistently estimated via the CF approach, they can be decomposed on facility- and market-level factors as in the structural productivity literature.

I apply my empirical approach to comprehensive data on hemodialysis patients in Atlanta between 2004 and 2008. For over thirty years, commentators have complained that the average quality of hemodialysis treatment in America lags that of other developed nations. However, little is accepted about whether or not patients' outcomes vary with factors like the degree of local competition. Addressing at most the possibility of selection based on unobserved severity, different papers have produced different results.

The Atlanta data strongly suggest that the lack of clarity in the previous literature may stem in part from the importance of selection due to heterogeneous treatment effects. In particular, I find a robust negative correlation between concentration and quality when using the CF estimates. In contrast, I find that the results about the relationship between competition and quality are much less strong -- and even sometimes inverted -- when the estimates of treatment quality come from models that ignore selection on both intercept and slope unobservables. Though contrary to some of the prior literature, these results are consistent with overall economic theory.