Elsevier

Journal of Health Economics

Volume 20, Issue 6, November 2001, Pages 881-907
Journal of Health Economics

Statistical discrimination in health care

https://doi.org/10.1016/S0167-6296(01)00101-1Get rights and content

Abstract

This paper considers the role of statistical discrimination as a potential explanation for racial and ethnic disparities in health care. The underlying problem is that a physician may have a harder time understanding a symptom report from minority patients. If so, even if there are no objective differences between Whites and minorities, and even if the physician has no discriminatory motives, minority patients will benefit less from treatment, and may rationally demand less care. After comparing these and other predictions to the published literature, we conclude that statistical discrimination is a potential source of racial/ethnic disparities, and worthy of research.

Introduction

“Statistical discrimination” refers to how an agent (an employer, a doctor), without intending to discriminate, might apply an otherwise reasonable decision making rule (pay according to productivity, treat according to need), that in practice leads to unequal treatment of members of two ethnic groups. Employers do not observe true productivity, and doctors do not observe true health need. Employers and doctors only observe signals or indicators of these concepts. The process of inferring the true from the signal can cause identical members of the two ethnic groups emitting the same signal to be treated differently. Aigner and Cain (1977), drawing on earlier work of Arrow (1973), Phelps (1972) and others, proposed that (white) employers can more readily interpret signals about productivity from white workers than from black workers. Employers in their model are willing to pay expected productivity to all workers, but the Bayesian employer downweights a noisy signal in relation to population mean values. The upshot is that Blacks earn less reward for higher productivity because the employer has a harder time (in comparison to white workers) distinguishing the more from the less productive. In a health care context, a white male doctor might have an easier time interpreting the signal, “doc, it really hurts” from a white male patient than from a black woman patient, or from a Latino woman patient (for whom English may be a second language). This paper begins with this simple observation and then works out the implications of the noisy signal hypothesis for explaining the unequal treatment of minorities in health care.

The literature describing differences in the use of health care resources between Whites and Blacks/Latinos is huge. Not only do minorities appear to receive less and inferior health care than Whites (Williams, 1994, Ford and Cooper, 1995), but they are also less likely to receive higher technology services/procedures (Escarce et al., 1993, Goldberg et al., 1992, Wennecker and Epstein, 1989, Yergan et al., 1987, Williams et al., 1979, Shi, 1999). While some of these differences may be explained by innate health risk differentials or differences in socio-economic status (SES), disparities seem to remain even after controlling for these factors. For example, the fact that Latinos underutilize mental health services, even at higher SES levels has been widely recognized (Ruiz, 1993, Ginzburg, 1991, Vega et al., 1999, Gallo et al., 1995, Sue et al., 1991, Scheffler and Miller, 1989, Wells et al., 1989, Hough et al., 1987). Also, data with good clinical information documents that even in the presence of extensive controls for “need”, ethnic minorities get less or inferior care (Peterson et al., 1997, Bach et al., 1999, Shapiro et al., 1999).1

There appears to be considerable agreement with the view that “it is time to stop documenting disparities, and to start doing something about them”.2 The theme of the American Public Health Association, November 2000 annual meeting, for example, was “Eliminating Health Disparities”. Nonetheless, the impetus to redress inequities and correct the inefficiencies associated with disparities can be better translated into effective policy if the mechanisms by which disparities arise are understood. Morgenstern (1997) points out that the empirical work necessary to quantify the magnitude of disparities is distinct from the work to identify the causes of disparities, a distinction not always kept clear in the literature. Reduced form empirical models can help identify the mediating role of readily measurable system factors, such as geographic access, provider practice patterns, and insurance plans (Gomes and McGuire, 2000). However, these models can neither explain the source of the differences in observable factors nor the disparities that remain once these factors have been controlled for. Identifying the potential role of discrimination in explaining disparities is especially important, but analysis of secondary data can only approach this issue indirectly.

Research employing actor-patients is better suited to ferreting out discrimination. This accords with Altonji and Blank’s (1999) view of how to study discrimination in labor markets. Discrimination exists insofar as objectively similar patients of different ethnicities or genders get different recommendations from doctors (Schulman et al., 1999).3 This empirical manifestation of discrimination can appear from different root causes. “Taste discrimination” (Becker, 1971) cannot be ruled out as an explanation for the results of these studies, but, as we will argue, “statistical discrimination” might be at work as well. As Schulman et al. (1999) put it, “bias may represent overt prejudice on the part of physicians or, more likely, could be the result of subconscious perceptions rather than deliberate actions or thoughts”.

When a doctor hears a symptom report from a patient the doctor must make an inference about the likely cause of the problem and what actions should be taken. This recommendation may depend in general on inferences about unobservable variables the doctor makes based on what he/she can see, including gender and race. Matching in a study like Schulman et al.’s can only encompass the “observable” variables. Suppose the doctor believes, based on experience, that men are less likely than women to renew prescriptions for anti-hypertensive drugs because men have more difficulty with the side effects of the drugs. When a treatment choice exists, the doctor is then less likely to recommend the drugs to a man than to a woman, even in an experiment in which a researcher constructs patients who are “objectively identical” except for gender. The researcher cannot control the doctor’s inferences about unobserved variables such as likelihood of compliance. These inferences may be made “rationally” and in the best interest of the patient as the doctor sees it; and they may lead to disparities in treatments.

Other work in health services points toward a role for miscommunication as being more of a problem when minority patients talk to their doctors. The importance of accurate communication between physician and patient has been widely recognized in the medical profession.4 Several studies have found that the quality of communication both in the history-taking segment of the visit and during discussion of the management plan influences patient health outcomes (Stewart, 1995). Communicational problems may be particularly exacerbated when doctor and patient belong to different ethnic groups, either because of language, cultural differences or both. Waitzkin (1985) found that SES was one of a set of factors associated with doctors’ willingness to talk and listen to patients. Einbinder and Schulman (2000) concluded, on the basis of a review, that race discordant physician–patient relationships affected symptom communication and recognition. There is some evidence that physician–patient matches in ethnicity and language are beneficial to patients in terms of satisfaction with treatment, compliance and quantity of care received (Cooper-Patrick et al., 1999, Sue et al., 1991, Saha et al., 1999, Takeuchi et al., 1995).5 However, the literature is not unanimous on this point. Chen et al. (2001) compared cardiac catheterization rates among Blacks and Whites (controlling for clinical status), and found lower rates for Blacks among black doctors than among white doctors.

Culture and ethnicity have also been shown to affect the interpretation of health conditions and other aspects of clinical care. Patients’ trust in hospitals and physicians, their perceptions of illness and suffering, their interpretation of lack of improvement and their proclivity to disclose information, among other things, appear all to have ethno-cultural correlation (Berger, 1998, Torres, 1986, Uba, 1992, Meredith and Siu, 1995). Effective communication depends on a willingness to communicate, and national surveys have repeatedly shown that Blacks mistrust their doctors more frequently than Whites (The Commonwealth Fund, 1995, The Henry J. Kaiser Family Foundation, 1999). Other research highlights the incidence of language in the interpretation of symptoms and in the outcomes of the medical encounter. Marcos et al. (1973) found that bilingual Hispanic schizophrenic patients demonstrated more content indicative of psychopathology when interviewed in English rather than Spanish. For members of some groups, no or limited proficiency in English inhibits communication with doctors and other health care workers. In 1990, 14% of Americans 5 years and older did not speak English at home. More than half of these spoke Spanish, and the percent will surely be higher in the 2000 Census numbers.6

Motivated by the importance of communication, culture and language in the physician–patient relationship, in this paper we consider the relatively poor communication between members of minority groups and (predominantly) white physicians as a potential fundamental cause of disparities in health outcomes and health services.7 Before proceeding we want to make three disclaimers. First, by embracing this hypothesis, we do not contend that it is the only form of discrimination that may be present in health care markets. On the contrary, we believe that health care markets, with attenuated competitive forces, could sustain stereotypes and taste discrimination in the long run more readily than labor markets. However, the importance of communication in health care and the particular beliefs, language and expression patterns associated with each ethnic group in the US justifies our concentration on “statistical” discrimination in this paper. Second, we do not study the determinants of bad communication. While it can be argued that bad communication in the diagnosis stage may result from an unwillingness of the provider to spend enough time with a minority patient (perhaps a prejudice issue), here we set up the model as if imperfect information about minorities where the only reason behind miscommunication. Third, we treat statistical discrimination as a hypothesis. Our tasks in this paper are to work out the implications of the hypothesis, and to lay the groundwork for empirical assessment.

Section 2 introduces the underlying problem that physicians may observe patients’ needs with less accuracy if the patient is a minority group member. In the simple context in which a benevolent doctor is called upon to match treatment to need, minorities will experience a poorer match and, anticipating this, may demand less care. The analysis in Section 2 shows how statistical discrimination can account for some salient problems in the health care of minorities, including, (1) minorities use health care less frequently, (2) minorities are more likely to drop out of treatment and (3) existing treatments are less effective for minorities. Statistical discrimination also generates unique predictions about how providers assign treatment to different ethnic groups. While other theories of discrimination expect doctors to provide less health care to minorities always, statistical discrimination predicts that minorities will receive less resources in some cases, but should receive more in some others. In Section 3, we call attention to additional implications of the basic model once we allow for multiple treatments, financial incentives in the physician’s objective function and learning. Assessing the quantitative importance of the mechanism of statistical discrimination in accounting for health services disparities is a matter for empirical investigation. Section 4 summarizes some of the empirical implications of statistical discrimination and checks them against published papers dealing with disparities. The conclusion is in Section 5.

Section snippets

Ethnicity and a poor match of treatment to need

Before turning to health care, it is worthwhile to briefly review the origin of the idea of statistical discrimination in the labor literature and to compare labor and health care markets in terms of the likelihood of the importance of communication problems in explaining differential economic outcomes for minority workers/patients.

Multiple treatments, financial incentives and learning

In this section, we modify some of the assumptions we have worked with so far and derive new implications of the statistical discrimination hypothesis. Allowing for more than one treatment generates predictions about the distributions of the “treatment portfolios” recommended by the physician to Whites and Blacks. Considering self-interest on the part of the physician can explain issues as why minorities suffer a bigger impact when the payment system changes from fee-for-service (FFS) to

Empirical relevance of the statistical discrimination hypothesis

In the previous sections, we proposed a model in which miscommunication in the clinical encounter resulted in disparities in the use of health resources between minorities and Whites. We reiterate at this stage that we do not claim that statistical discrimination is the only relevant force behind disparities. On the contrary, we believe that disparities stem from a complex interaction of phenomena that need to be studied carefully before addressing any particular policy. The quantitative

Conclusions

This paper provides an explanation of discrimination in health care (discussed under the title of “statistical discrimination”) in which differences in treatment rates arise from relatively poor communication between a minority patient and a health care provider. The model implies that minorities fare worse than Whites from health care services because they are more likely to be mismatched to treatments. Unlike other hypotheses of discrimination, the model predicts that minorities may or may

Acknowledgements

We acknowledge the support of Grant PO1_MH59876 from the National Institute of Mental Health. We are grateful to Victor Aguirregabiria, Margarita Alegrı́a, Naihua Duan, Randy Ellis, Peter Guarnaccia, Kevin Lang, Glenn Loury, Michael Manove, Lisa Meredith, Charles Phelps and two referees for helpful comments.

References (78)

  • Bach, P., Cramer, L., Warren, J., Begg, C., 1999. Racial differences in the treatment of early-stage lung cancer. The...
  • Balsa, A., McGuire, T., 2001. Prejudice, Clinical Uncertainty and Stereotyping as Sources of Health Disparities. Boston...
  • Becker, G.S., 1971. The Economics of Discrimination, 2nd Edition. The University of Chicago Press,...
  • J. Berger

    Culture and ethnicity in clinical care

    Archives of Internal Medicine

    (1998)
  • Cain, G., 1986. The economic analysis of labor market discrimination: a survey. In: Ashenfelter, O.A., Layard, R....
  • J. Chen et al.

    Racial differences in the use of cardiac catheterization after acute myocardial infarction

    New England Journal of Medicine

    (2001)
  • S. Coate et al.

    Will affirmative-action policies eliminate negative stereotypes?

    American Economic Review

    (1993)
  • Cooper-Patrick, L., Gallo, J., Gonzales, J., et al., 1999. Race, gender and partnership in the patient–physician...
  • L.C. Einbinder et al.

    The effect of race on the referral process for invasive cardiac procedures

    Medical Care Research and Review

    (2000)
  • Eisenberg, J.M., 1986. Doctors’ Decisions and the Cost of Medical Care. Health Administration Press, Ann Arbor,...
  • Eley, J., Hill, H., Chen V., et al., 1994. Racial differences in survival from breast cancer: results of the National...
  • R. Ellis et al.

    Provider behavior under prospective reimbursement: cost sharing and supply

    Journal of Health Economics

    (1986)
  • Escarce J.J., et al., 1993. Racial differences in the elderly’s use of medical procedures and diagnostic tests....
  • E.S. Ford et al.

    Racial/ethnic differences in health care utilization of cardiovascular procedures: a review of the evidence

    Health Services Research

    (1995)
  • J.J. Gallo et al.

    Filters on the pathway to mental health care II: socio-demographic factors

    Psychological Medicine

    (1995)
  • E. Ginzburg

    Access to health care for Hispanics

    Journal of the American Medical Association

    (1991)
  • Goldberg K.C., et al., 1992. Racial and community factors influencing coronary artery bypass surgery rates for all...
  • Gomes, C., McGuire, T., 2000. Identifying the Sources of Racial and Ethnic Differences in Health Care Use. Boston...
  • Hannan, E., Ryn, M., Burke, J., et al., 1999. Access to coronary artery bypass surgery by race/ethnicity and gender...
  • J.J. Heckman

    Detecting discrimination

    Journal of Economic Perspectives

    (1998)
  • R.L. Hough et al.

    Utilization of health and mental health services by Los Angeles Mexican-Americans and non-Hispanic Whites

    Archives of General Psychiatry

    (1987)
  • M. Kogan et al.

    Relation of content of prenatal care to the risk of low birth weight

    Journal of the American Medical Association

    (1994)
  • K. Lang

    A language theory of discrimination

    Quarterly Journal of Economics

    (1986)
  • K. Lang

    Language and economists’ theories of discrimination

    International Journal of the Sociology of Language

    (1993)
  • S. Lundberg

    The enforcement of equal opportunity laws under imperfect information: affirmative action and alternatives

    Quarterly Journal of Economics

    (1991)
  • S. Lundberg et al.

    Private discrimination and social intervention in competitive labor markets

    American Economic Review

    (1983)
  • L.R. Marcos et al.

    The effect of interview language on the evaluation of psychopathology in Spanish American Schizophrenic patients

    American Journal of Psychiatry

    (1973)
  • McGuire, T., 2000. Physician agency. In: Newhouse, J., Culyer, A. (Eds.), Handbook of Health Economics, Vol. 1....
  • R. Mayberry et al.

    Racial and ethnic differences in access to medical care

    Medical Care Research and Review

    (2000)
  • Cited by (0)

    View full text