Despite convergence during the 20th century, there remain large Black-White disparities in health outcomes in the United States. As of 2019, Black life expectancy was approximately three years lower than White life expectancy (Dwyer-Lindgren et al., 2022).1 At the same time, Black Americans are dramatically underrepresented in medicine: only four percent of physicians in the U.S. are Black.2 Both phenomena are complex with numerous potential explanations, but policymakers and researchers have suggested a link for decades. For instance, in 1985 the U.S. Department of Health and Human Services published The Report of the Secretary's Task Force on Black and Minority Health, which argued that racially driven disparities in health outcomes should be a national priority while noting that “most minorities receive health care from providers who do not share their own ethnic/cultural background”. The report went on to assert that efforts should be made to improve minority representation in the health profession, implicitly with the goal of closing the Black-White gap in health outcomes.

A substantial literature, in medicine and other fields, has explored the effect of race-match between health care providers and patients on inputs to health. For instance, Alsan et al. (2019) measure the effect of Black men being assigned to Black outpatient providers on the uptake of preventative care. Cooper-Patrick et al. (1999) measure the effect of race-match on “participatory decision-making”, reporting positive effects. Ye and Yi (2023) find that patient-physician race concordance reduces revisit rates. Yet, to the best of our knowledge, only two studies have found a causal effect of physician-patient race-match on any health outcome: Greenwood et al. (2020) document that physician-patient race-match reduces Black-White disparities in mortality of newborns at birth, and Frakes and Gruber (2022) report reduced mortality from race-matches in the US military. Our study, along with the above papers, contributes to filling this gap.

We provide causal evidence of an effect of physician-patient race-match on a health outcome. Specifically, we find an effect on mortality. Focusing on non-Hispanic Black and White patients and physicians3, we find that when Black patients are matched with a same-race physician, they are less likely to die while in the hospital. The magnitude of the within-hospital mortality reduction is similar to the effect of gaining insurance (Card et al., 2009).

To execute this study, we collect unique records from the Florida Department of Health that identify the race of the universe of physicians in Florida. We then gather administrative hospital data from across the state of Florida over multiple years and pair these with the physician race data.4 To the best of our knowledge, this is the first time these data have been linked.

We ensure plausibly exogenous assignment of patients to attending physicians by employing two main sample restrictions and a rich set of fixed effects. First, we focus on uninsured patients. Uninsured patients typically do not have an established provider to coordinate (or provide) their care in a hospital, so their hospital physician is typically random. Second, we restrict our sample to uninsured patients admitted through the emergency department. Hospitalizations that originate in the emergency department are almost certainly unscheduled, instigating a “shock” in the hospital's inpatient caseload. Consequently, whoever is on call in the hospital is likely to become the attending physician. Our main specifications include a rich set of physician, hospital, and patient ZIP code fixed effects to rule out lingering selection concerns. We find that race-matched pairings between Black patients and physicians improve those patients’ outcomes: Black patients are 0.28 percentage points less likely to die while in the hospital when paired with a Black physician. Mortality is relatively rare, so this represents a nearly 27% decline relative to the mean. To further confirm that our results can be interpreted as causal and to enhance generalizability, we estimate an alternative specification using an instrumental variable to generate exogenous variation in physician-patient race-match and add all payers back into the sample; we indeed find similar results.