Editorial ColumnsHow Does Race Matter, Anyway?In two articles published since 2001, Balsa and McGuire (2001, 2003) haveused the language and tools of economic theory to examine several phenom-ena that can influence doctors’ decision making during clinical encounters thatpatients and result in racial disparities in health care. These articles, whichadapted and extended earlier theoretical developments in labor economics,were published in an economics journal and are unlikely to have been read bymany health services researchers. Therefore, a brief summary of the articles isin order.Balsa and McGuire (2003) identified prejudice, clinical uncertainty, andstereotyping as distinct mechanisms that can operate within the clinical en-counter and lead to racial disparities in care. Drawing on the literature in socialpsychology, they defined prejudice as the holding of a negative attitude oraffect against members of another racial group; e.g., white physicians may beprejudiced against black patients. By modeling prejudice as a psychologicalcost experienced by white physicians when they treat black patients, Balsa andMcGuire showed that prejudiced white physicians would provide less care toblacks than to whites.Balsa and McGuire (2003) also considered the effects of two types ofclinical uncertainty. In the first type, physicians were assumed to have onlynoisy indicators of patients’ clinical condition——perhaps as a result of imper-fect diagnostic tests——so they were necessarily uncertain about patients’ pre-cise diagnosis or severity. The investigators found that under suchcircumstances physicians would be forced to rely on prior probabilities——e.g., the prevalence of disease——in making diagnostic and treatment decisions.Thus, for instance, if the prevalence of disease were lower in blacks than inwhites, even unprejudiced physicians would be less likely to recommendtreatment to black patients than to white patients. As a result, more blackpatients than white patients who would benefit from treatment would remainuntreated.In the second type of uncertainty, Balsa and McGuire (2003) assumedthat physicians have no trouble assessing the diagnosis and severity of patientsin their own racial group, but the indicators of patients’ clinical condition arenoisy for patients from a different racial group, possibly due to miscommu-nication and misunderstanding resulting from cultural or language differences.They showed that even unprejudiced physicians would be forced to rely on1prior probabilities to a greater degree when treating black patients than whitepatients. Thus, diagnoses and treatment recommendations of white physicianswould be less well ‘‘matched’’ to individuals’ needs for black patients than forwhites. Balsa and McGuire also pointed out that because the recommendedcare would, on average, be less beneficial for blacks than for whites, blackscould rationally react by going to the physician less often or complying lesswith treatment. Balsa and McGuire (2001, 2003) referred to the racially con-tingent diagnostic and treatment decisions that can arise from either type ofuncertainty as statistical discrimination.Balsa and McGuire (2003) also considered the role of stereotypes. Fol-lowing contemporary social psychology, they defined stereotyping as theprocess by which people use social categories (e.g., race or gender) in acquir-ing, processing, and recalling information about others (Dovidio 1999). Theyemphasized that stereotypes are a cognitive mechanism for simplifying andorganizing social information in a complex world, that they tend to be neg-ative and exaggerated (Ashmore and Del Boca 1981), and that they are notnecessarily accompanied by a negative affect. Drawing on models of stere-otyping in labor economics, they showed that certain negative stereotypes——e.g., blacks are less likely to comply with treatment——can result in less care forblacks, especially if the stereotypes are self-fulfilling.Finally, in a brief analysis of policy implications, Balsa and McGuire(2003) observed that corrective actions for disparities must derive from anunderstanding of the underlying mechanisms. Specifically, efforts to improveinformation and reduce noise in the clinical encounter would reduce dispar-ities that arise from clinical uncertainty. However, disparities that arise fromprejudice and stereotypes are likely to be harder to deal with. Efforts to combatprejudice and eliminate stereotypes could help, and rule-based policies re-garding criteria for treatment and treatment rates in different racial groupscould be effective in some cases.In this issue of Health Services Research, Balsa, McGuire, and Meredith(2005) attempt to conduct empirical tests of the role of statistical discriminationin health care. Using data from the Medical Outcomes Study (Tarlov et al.1989), the investigators assess the factors that influence white doctors’ deci-sions to diagnose black and white patients with hypertension, diabetes,or depression after an ambulatory visit. For each disease, they estimate aThis work was supported in part by Grant No. P01-HS10770 from the Agency for Healthcare andResearch and Quality.2 HSR: Health Services Research 40:1 (February 2005)‘‘traditional disparities regression’’ in which they model a doctor’s decision todiagnose a patient as a function of patient and physician characteristics as wellas a ‘‘signal’’ emitted by the patient. For hypertension, a positive signal is thepatient’s affirmative response to a previsit question asking whether he has everbeen told he has hypertension. For diabetes, a positive signal is the patient’saffirmative response to an analogous previsit question or his report that hetakes insulin. For depression, a positive signal is constructed from the re-sponses to a previsit mental health screener and symptom questionnaire. Theinvestigators then compare the results of the traditional disparities regressionand a ‘‘statistical discrimination regression’’ in which they add disease prev-alence (to identify physicians’ use of prior probabilities) and an interactionbetween black race and a positive signal (to identify communication problemsbetween white doctors and black patients) to the model’s explanatory vari-ables.Balsa, McGuire, and Meredith (2005) find that sex, age, and a positivesignal are significantly associated with the doctor’s diagnosis in the traditionaldisparities regressions for