Racial-Ethnic Biases,Time Pressure,and Medical Decisions

Abstract

This study examined two types of potential sources of racial-ethnic disparities in medical care: implicit biases and time pressure. Eighty-one family physicians and general internists responded to a case vignette describing a patient with chest pain. Time pressure was manipulated experimentally. Under high time pressure, but not under low time pressure, implicit biases regarding blacks and Hispanics led to a less serious diagnosis. In addition, implicit biases regarding blacks led to a lower likelihood of a referral to specialist when physicians were under high time pressure. The results suggest that when physicians face stress, their implicit biases may shape medical decisions in ways that disadvantage minority patients.

Keywords

disparities ethnicity quality of health care race time pressure

One of the most striking examples of persistent racial-ethnic inequality in the United States is the fact that minority patients who are otherwise comparable to their white counterparts receive significantly poorer health care (Institute of Medicine [IOM] 2003). Even after accounting for access barriers, differences in health status and behaviors, socioeconomic background, insurance, and health care preferences, a substantial proportion of the racial-ethnic gaps remains unexplained. That it occurs in medical contexts is especially puzzling, given the forces in the health care system that ought to discourage discrimination: ethical commitments, high education of health care professionals, legal and other regulations on medical practice, and the threat of malpractice litigation that is constantly present in physicians’ lives.

This puzzling observation has led scholars to look at health care providers’ racial-ethnic biases as a potential source of differences in health care. Theorists propose that racial-ethnic biases may in some instances lead to inferior care for minority patients (Burgess, Fu, and van Ryn 2004; Burgess et al. 2006, 2007; Dovidio et al. 2008; Kposowa and Tsunokai 2002; van Ryn 2002; van Ryn and Fu 2003). They argue that racism is embedded in American culture and society and so physicians have typically been socialized in a culture infused with racial prejudices. Although physicians genuinely may intend to provide the best possible care to all patients, they are not immune to racial-ethnic biases.

Yet, direct evidence for a link between providers’ racial-ethnic biases and health care outcomes is very limited. Methodological challenges contribute to the difficulty of demonstrating this link empirically. Social desirability bias is a formidable challenge in studies of racial-ethnic attitudes and discrimination (Pager and Shepherd 2008; Quillian 2006). The social unacceptability of expressing racist beliefs may lead research participants to refrain from giving responses that could be interpreted as biased. This tendency would produce an inaccurate picture of racial attitudes and behaviors in interracial interactions. In medicine, the social desirability bias is further exacerbated by providers’ fear of professional and legal repercussions should their racial-ethnic biases be uncovered (Dovidio et al. 2008). Nevertheless, these difficulties should not stop the search for empirical evidence, since such evidence is essential for formulating effective solutions to the problem and has potential long-term benefits to the medical profession and certainly for minority Americans’ health.

Background

Subliminal Priming and Implicit Biases

Given the concerns about social desirability bias in studies of racial attitudes, this study was specifically designed to minimize this problem. It uses subliminal priming, a method novel in sociology, to overcome the challenges posed by the social desirability biases. Subliminal priming limits the ability of respondents to consciously alter their responses in the effort to appear less biased. It presents race-ethnicity relevant information subliminally, namely, in a way that most participants do not register consciously. When racial-ethnic information is presented in this way, it is unlikely that conscious self-presentation strategies would play a role.

Subliminal priming is commonly used in psychology to activate information processing along a categorical dimension. This method, first developed by Bargh and Pietromonaco (1982) and later adapted by Devine (1989) and Graham and Lowery (2004) specifically for the study of race, relies on the assumption that the subliminally presented information about a social category passively (without an intervening act of will) affects perception, evaluation, and behavior (Bargh and Chartland 2000). This method works for subjects who have in their minds cognitive structures (e.g., biases or stereotypes) associated with the presented information. Since racial-ethnic biases and stereotypes are widespread in the American population (Greenwald and Banaji 1995; Nosek et al. 2007), the method typically works for a majority of subjects. To achieve the activation of racial biases, researchers expose subjects to subliminal primes in the form of race-related words, as in this study (Abreu 1999; Devine 1989; Graham and Lowery 2004; Kawakami, Dion, and Dovidio 1998; Lepore and Brown 1997; Wittenbrink, Judd, and Park 1997), or images of black versus white faces (Bargh, Chen, and Burrows 1996; Brown et al. 2003; Chen and Bargh 1997; Dovidio et al. 1997; Livingston and Brewer 2002).

Importantly, subjects who undergo racial priming are more likely to generate stereotypical responses, which suggest the predictive validity (for a review, see Fazio and Olson 2003). Devine (1989), for instance, subliminally primed participants with race-related versus race-neutral words. Subsequently, participants rated ambiguous behavior of a target described in a vignette. When the majority of words were related to the category “black” (as opposed to race-neutral), ratings of hostility of the target increased by about 10 percent. In other studies, subliminal priming by words representing the category “black” was associated with increased hostility and decreased friendliness in interracial interaction, more negative ratings of the target (e.g., lower likeability and friendliness), expectations of greater recidivism, and endorsement of harsher criminal punishment (Abreu 1999; Graham and Lowery 2004; Lepore and Brown 1997). Bargh and colleagues (1996), Chen and Bargh (1997), and Payne (2001) observed more stereotypical responses after subliminal priming by black faces. Subliminal priming was also used for the activation of stereotypes of non-black minorities, such as Asians Americans (Abreu et al. 2003). Subliminal priming methods show moderate test-retest reliability (Kawakami and Dovidio 2001).

It is believed that subliminal priming produces more stereotypical responses because it activates implicit racial-ethnic biases. These biases are culturally embedded and develop through socialization from an early age. They differ from more commonly studied explicit biases in that they operate automatically, outside of conscious awareness. Despite the fact that explicit racist beliefs are rare these days, most Americans show anti-minority preferences (Nosek et al. 2007). Whites’ implicit cognitions regarding blacks tend to be negative, in contrast to the positive cognitions they hold regarding other whites (Blair 2001). Importantly, implicit biases (activated by priming or measured by other methods) influence behaviors in interracial interactions, including nonverbal expressions of hostility, anxiety, and attention (Chen and Bargh 1997; Dovidio et al. 1997; Dovidio, Kawakami, and Gaertner 2002; Fazio et al. 1995), as well as more deliberate decisions, such as generosity with monetary resources (Stepanikova, Triplet, and Simpson 2011) or budget cuts to minority organizations (Rudman and Ashmore 2007).

Implicit biases have received abundant attention in psychology but represent a relatively new area in sociology. Only recently have sociologists started to study them in the domains of race (Kim 2003; Stepanikova et al. 2011) and gender (Tinkler, Li, and Mollborn 2007). The attention to implicit racial-ethnic biases in sociology is important, since the high prevalence of these biases, amply documented by psychologists, provides one plausible answer to a deeply sociological question: Why do racial-ethnic inequalities persist even though whites’ explicit attitudes over past decades have shown a marked change toward the endorsement of racial-ethnic equality (Pager and Shepherd 2008)? In this study, I examine the relevance of implicit racial-ethnic biases in one area of persistent racial-ethnic inequality: medical decisions.

Another important contribution of this study is the attention to the influence of time pressure on racially-ethnically biased medical decisions. I argue that physicians’ implicit racial-ethnic biases influence medical decisions and may lead to poorer quality of care for black and Hispanic patients, especially when physicians are under time pressure. Given that the subjective experience of time pressure on the job is common among American physicians (Linzer et al. 2000), the examination of clinical decision making under this type of stress is an important area of study. I support my argument with empirical observations from an experiment with primary care physicians as participants. This study is among the first to provide evidence that implicit processes, despite their nonconscious nature, play an important role in medical decisions that are made under time pressure.

Racial Biases in Medical Settings

Despite the long-standing reluctance in the medical community to acknowledge the possibility of racially-ethnically biased medicine (Byrd 1990), there are two suggestive lines of evidence. The first links patient race to medical decisions and clinicians’ perceptions of patients. Van Ryn and Burke (2000), for instance, reported that cardiologists perceived their black patients to be less intelligent, less educated, less likeable, less friendly, and less likely to adhere to medical advice, even after controlling for confounding factors. An earlier study by Schulman and coauthors (1999) found that physicians were less likely to recommend cardiac catheterization for black females than for white females with the same symptoms. These findings suggest that biased perceptions may lead to clinical decisions that disadvantage minorities. The second line of evidence comes from patient surveys. These reveal that perceptions of racial-ethnic bias and discrimination in health care are common among minority patients (Johnson et al. 2004). Such perceptions are reported by as many as 23 percent of blacks and 15 percent of Hispanics nationwide (LaVeist, Rolley, and Diala 2003). The problem is compounded for Hispanics, who often indicate that health care providers judge them unfairly because of their poor English language skills.

Even so, the argument that racial-ethnic biases affect medical decisions has been challenged on various grounds. It is certainly the case that if queried, most physicians would express explicit antiracist attitudes. Yet, it is becoming increasingly clear that physicians are not impervious to implicit racial biases. In a large national study, a majority of doctors showed implicit anti-black preferences (Sabin, Nosek, and Greenwald 2009). These preferences were particularly strong among white, Asian, and Hispanic doctors.

The evidence demonstrating the links between implicit biases and behavior reviewed earlier suggests that if doctors have implicit racial biases, then there are predictable negative consequences for medical care delivered to minority patients. This argument was first formulated by van Ryn (2002) and elaborated by Kposowa and Tsunokai (2002). Van Ryn (2002) theorizes that unconsciously activated stereotypical beliefs about patients affect how providers interpret symptoms and what treatments they recommend. Two studies examined the link between implicit biases and medical decisions empirically. The first study with cardiologists showed that higher levels of implicit anti-black biases measured by the Implicit Association Test (IAT) were related to a lower likelihood of recommending thrombolysis for a black patient with cardiovascular disease described in a vignette (Green et al. 2007). The second study with pediatricians found no relationship between IAT scores and treatment recommendations (Sabin et al. 2008).

While these two studies represent notable progress, their methodologies leave open the possibility that social desirability bias influenced the results. In both studies, physicians received explicit information about patient race, either in the form of racial category labels or as ostensive images of patients. In addition, both studies used the IAT, which asks participants to pair images of black versus white faces with value-laden words, such as good or bad, thus making the focus on racial attitudes transparent. The awareness of this focus may have led to a desire to present a nonprejudiced self-image and motivated socially desirable responses to black patients.

As explained earlier, the key strength of the present study is that it avoids the social desirability challenge by using subliminal priming to activate implicit racial-ethnic biases. It is important to reiterate that racial-ethnic biases are activated by this method only if these biases are present in an individual’s mind. Doctors do not differ from the general population in regard to their implicit racial biases (Sabin et al. 2009). Thus, it is reasonable to expect that the subliminal priming method will activate implicit racial-ethnic biases for most doctors.

Time Pressure

Time pressure is a common stressor among physicians (Linzer et al. 2000). Physicians working in resource-poor, underfunded, and understaffed health care facilities that disproportionately serve minority clientele are especially affected. These physicians face high volumes of patients but lack the time and resources to provide high-quality care. Organizational policies can exacerbate stress. Physicians working under managed care, for instance, face productivity pressures, excessive administrative loads, and time limits for patient visits. They often feel that they do not have adequate time to spend with their patients (Linzer et al. 2000). Time pressure is linked to low job satisfaction, fatigue, burnout, and errors (Bovier and Perneger 2003; Spickard, Gabbe, and Christensen 2002).

Building on van Ryn and Fu (2003) and Kposowa and Tsunokai (2002), I argue that time pressure can drive racial-ethnic inequalities by making it more likely that racial-ethnic biases will affect medical decision making. This is because time pressure is one notable example of a situation that depletes the supply of free cognitive resources. Under time pressure and in other stressful situations, individuals tend to resort to stereotypes that serve as heuristics and free up the much needed cognitive energy (Allport 1954; Macrae, Milne, and Bodenhausen 1994). In some cases, conscious effort helps to modify biases, especially among individuals who are highly motivated to act in an unbiased manner (Blair 2002), but such efforts may meet with limited success among persons who are facing a depletion of cognitive resources (Bartholow, Dickter, and Sestir 2006).

Medicine in general is a cognitively demanding profession but some medical situations, like those characterized by time pressure, are especially stressful. Such situations can increase physicians’ reliance on categorization, stereotyping, and other biases, possibly to save cognitive energy and increase cognitive efficiency. In low-stress situations, on the other hand, free cognitive resources can help physicians to combat biases. Under low stress, physicians can more easily attend to characteristics that are unique about the patient rather than resorting to categorization and stereotyping that may compromise the quality of care. Therefore, I hypothesize that implicit racial-ethnic biases will affect medical decision making more strongly under high time pressure compared to under low time pressure.

Data and Methods

Sample

Data were collected from 82 physicians. One participant was excluded because of a failed awareness check (described later), reducing the final sample to 81.¹ The sampling frame was provided by the American Medical Association (AMA). AMA maintains a master list of all physicians practicing in the United States who belong to the AMA, organized by specialty. To make the sample relatively homogeneous in terms of training, knowledge, and experience with consultations for chest pain, I selected only family physicians and general internists. Probability sampling was used, but the response rate of 2 percent makes it prudent to treat this as a convenience sample.

Low response rate is common in studies asking physicians to devote time out of their busy schedules to research participation. The response rates can be improved with substantial financial incentives, as with surveys fielded by pharmaceutical companies, but such competitive incentives were not available in this study. Convenience samples are acceptable in studies similar to this one that do not seek to describe a population but to elucidate relationships between variables. Prior experiments and vignette studies in health care research have relied on convenience samples and still led us to important conclusions. Nevertheless, it is of interest that this sample is close to the population of family physicians and general internists practicing nationwide in terms of its gender composition (males: 69 percent vs. 70 percent). It overrepresents whites (79 percent vs. 74 percent) and practices with 2 to 10 physicians (57 percent vs. 26 percent). It underrepresents solo practices (20 percent vs. 36 percent). It is roughly similar to national rates for HMOs and faculty practices.

Procedures

Subliminal priming

Physicians were contacted by e-mail and invited to participate in an Internet-based study of medical decision making. After accessing the study Web site and indicating their informed consent, they were told that they would participate in a “concentration exercise.” Its true purpose was to perform subliminal priming to activate implicit biases. Physicians were randomly assigned to one of four conditions (i.e., black, Hispanic, white, or control) and exposed to one of four possible sets of primes (stimuli words). Primes consisted of BLACK, AFRO, AFRICAN, and RAP for the black category; HISPANIC, LATINA, SPANISH, CHICANA, and MEXICAN for the Hispanic category; and WHITE, EUROPEAN, ANGLO, and CAUCASIAN for the white category.² The control group was exposed to racially-ethnically neutral words (BLOCK, TEXT, MAP, and PERCENT). Each participant received a priming sequence consisting of 57 randomly ordered stimulus words. Because there were only 4 to 5 words in each category, each word appeared on average 14 times in a single sequence but the exact number of repetitions varied because of random sampling of words. For the sake of consistency with prior studies, each sequence also included 8 race-ethnicity-unrelated words (SHOULD, PAPER, FORMAT, LOOK, TEACHER, HIGHWAY, DIFFERENCE, and LOOK). Each of these words was placed in a randomly selected position in the priming sequence, bringing the length of the whole sequence to 65 (57 race-ethnicity-related plus 8 race-ethnicity-unrelated words). Each word was flashed very briefly (80 ms) in one random corner of the computer screen. Before every new word, the respondent’s glance was directed back to an “X” in the middle of the screen. The purpose was to achieve parafoveal presentation, namely, presentation at the outskirts of the visual field. Parafoveal presentation is important because it minimizes the chances of directly glancing at the stimulus (Bargh and Chartrand 2000). Such accidental direct glances increase the probability of conscious processing.

Each priming word was immediately replaced by a race-ethnicity-neutral word or by randomly ordered letters, such as thnlju. This technique, known as masking, is needed to overwrite the previous image that is retained in iconic memory for some time after the image had disappeared from the display (Bargh and Chartrand 2000). Masking is an additional safeguard that decreases the chances of conscious processing. Masks remained on the screen for 600 ms, long enough for participants to read them comfortably. As a cover story, participants were asked to indicate by pressing a designated key whether the mask they saw was a word or a non-word.

Medical vignette

After subliminal priming, physicians read a vignette describing a patient with chest pain (see the appendix) and gave their medical judgments. Vignettes are a cost-efficient and valid method of studying decision making in clinical practice (Peabody et al. 2000). The vignette used in this study was extensively vetted with a family physician to ensure that (a) it realistically portrayed a patient whom a family physician or general internist might see and (b) the patient’s symptoms were ambiguous, namely, they possibly, but not conclusively, indicated cardiovascular disease. This was important, since clinical uncertainty increases the need for discretionary decisions. According to Mort et al. (1996), the influence of biases and other nonmedical factors is most pronounced in situations that call for discretionary medical decisions. This feature must be kept in mind for the interpretation of results. Unlike in earlier studies (i.e., Schulman et al. 1999), there was no right or wrong response to the vignette. This study simply explored whether in face of clinical uncertainty, implicit racial-ethnic biases would influence clinical judgment.

Chest pain was chosen as the chief complaint for two reasons. First, racial disparities in cardiovascular care are large and well documented. Blacks and Hispanics with cardiovascular disease are less likely than their white counterparts to get appropriate care, including diagnostic procedures, cardiac catheterization, revascularization, beta blockers, and thrombolytic therapy (IOM 2003). The second reason is the higher risk of cardiovascular disease among blacks and Hispanics compared to whites (Winkleby et al. 1998). Given this difference, it is important to understand the mechanisms that may affect decisions about cardiovascular care for these minorities.

The patient was described as a 62-year-old female. Gender and age were held constant across vignettes since it was important to distinguish racial-ethnic biases from sexism and ageism. The insurance status was also held constant to rule out the potential effects of stereotypes of the uninsured. To avoid the activation of explicit biases, the vignette included no explicit information about race-ethnicity.

To present medical information, I modified a method used by Schulman et al. (1999). Whereas they used blood pressure and cholesterol levels corresponding to the 20th to 30th percentiles or to the 70th to 80th percentiles for the risk of coronary heart disease (CHD), my study used values corresponding to mean risk to increase the need for discretionary decision making. The vignette also described the type of pain, hereditary and lifestyle factors, and results of diagnostic tests (electrocardiography and exercise EKG stress test). Again, indications of high risk were combined with some mitigating characteristics. For instance, the patient has a long smoking history, which is a well-known risk factor for cardiovascular disease, but she quit smoking a year ago, which may have lowered her current risk.

After reading the vignette, physicians answered questions about how they would diagnose the patient. Similar to Schulman et al.’s (1999) study, they characterized the type of chest pain and estimated the probability that the patient had cardiovascular disease: “In your opinion, how likely is the patient’s chest pain to be angina?” and “How likely is it that this patient has clinically significant CHD (i.e., at least 70 percent narrowing of one or more coronary arteries)?” Responses were recorded on 10-point scales (1 = not at all likely, 10 = extremely likely). Physicians were also asked whether or not they would refer the patient to a specialist.

Physicians received a time limit of three minutes to read the vignette and to respond to questions. To manipulate time pressure, a random half of the physicians received a second vignette describing a patient with back pain to respond to during the same time limit. Manipulation of time pressure represented a meaningful innovation to vignette-based research. The manipulation check performed at the end of the study revealed that this manipulation was successful. Subjective perceptions of stress were measured by the question, “How stressful did you find it to work under the time limit when answering questions about the patient(s)?” with responses on a 7-point scale (not at all stressful to extremely stressful). Stress was significantly higher among physicians who read descriptions of two patients (mean = 4.13) compared to physicians who read a description of one patient (mean = 3.28, p < .01).

In the next part of the study, physicians reported their professional and sociodemographic characteristics. The questions were based on national surveys of physician practice patterns (Center for Studying Health System Change 2006; Gold et al. 1995). Physicians reported their gender (male or female), ethnicity (“Do you consider yourself Hispanic or Latino/a?”), and race (American Indian/Alaska Native, Asian, black or African American, Native Hawaiian/Pacific Islander, and white/Caucasian). Professional characteristics included specialty (family medicine or general internal medicine), type of practice (solo office practice, an office with 2 to 10 physicians, an office with 11 or more physicians, a group-model HMO, a staff-model HMO, government practice, faculty practice— academic or residence based, or other), and years since graduation from the medical school. These questions no longer involved a time limit.

The study concluded with a debriefing and awareness check. The purpose of the awareness check was to identify respondents who consciously processed the racial-ethnic stimuli, either because of unusually high perceptual speed or because of an accidental direct glance at the stimulus, despite its extremely brief, parafoveal presentation. The awareness check was introduced with the following statements: “For some participants, we activated automatic cognitions about race and ethnicity by very briefly flashing words associated with different racial and ethnic categories during the concentration exercise. Did you notice any words related to race or ethnicity during the concentration exercise?” The participants who answered yes were asked to write the words they saw and later excluded if they specified at least one word correctly. The awareness check determined that a single respondent correctly specified some stimulus words. This suggests that for a large majority of respondents, the presentation of stimuli was truly subliminal, as intended, and that the racial-ethnic information did not enter conscious awareness. This bolsters confidence that the observed differences in medical decisions were caused by automatic cognitions and that conscious strategies, including those that involve self-presentation, did not play a role.

Analytic Plan

Statistical analysis was performed using Stata software. After calculating descriptive statistics for all variables, I conducted bivariate tests comparing the means for the three dependent variables (evaluations of the likelihood of CHD, evaluations of the likelihood of angina, and referral to specialist) by categories of racial-ethnic stimuli. These tests were conducted separately for low and high time pressure. t-tests were used for continuous variables (i.e., CHD and angina), and chi-square tests were used for the dichotomous indicator of referral. Comparisons were made between the black or Hispanic category versus the category containing white and control conditions (referred to henceforth as white/control). White and control conditions were combined into a single category to reduce the number of comparisons and to simplify results. Importantly, there were no significant differences between white and control conditions on any of the three dependent variables.

Finally, I estimated multivariate regression models to formally evaluate my hypothesis. Linear regression was used for evaluations of angina and CHD. Logistic regression was used for referral. Because of deviations from normality and to minimize the potential undue influence of outliers in a small sample, all models used robust estimators. Two models were estimated for each dependent variable. Model 1 contained main effects of time pressure and racial-ethnic category (black and Hispanic vs. white/control), along with control variables measuring physician gender, race, Hispanic ethnicity, years since graduation from medical school, specialty, and type of practice. Model 2 included interactions between racial-ethnic categories and time pressure, along with other variables used in model 1.

To better understand these interactions, I used estimates from model 2 to calculate predicted values for diagnostic decisions (CHD and angina) and predicted probabilities for referral by racial-ethnic stimuli and time pressure. I set control variables at their means (if continuous) or modes (if categorical). These calculations illustrate the relationships between implicit biases and medical decisions under high versus low time pressure for a typical respondent, namely, non-Hispanic, white, male, family practitioner with close to 19 years of experience who works in an office staffed with 2 to 10 physicians.

Results

Sample Characteristics

Table 1 shows descriptive statistics for medical decisions, sociodemographic indicators, and professional characteristics. In general, physicians evaluated the likelihood that the patient had CHD and angina as moderately high (7.06 for CHD and 6.93 for angina on a 1 to 10 scale). A majority (70 percent) indicated that they would refer the patient to a specialist. These results suggest that the physicians picked up on the heart disease risk factors and weighed them more heavily compared to the mitigating factors in formulating their judgments.

Table 1.

Characteristics of the Sample: Means and Standard Deviations (in Parentheses)

Medical Decisions
How likely is it that this patient has clinically significant CHD (1–10)	7.06 (1.81)
How likely is the patient’s chest pain to be angina (1–10)	6.93 (1.97)
Would you refer this patient to a specialist? (yes = 1; no = 0)	.70
Physicians’ sociodemographic characteristics
Male	.69
Hispanic ethnicity	.08
Race
White	.79
Black	.05
Other	.16
Physicians’ professional characteristics
Family medicine	.51
General internal medicine	.49
Type of practice
Solo office practice	.20
An office with 2 to 10 physicians	.57
An office with 11 or more physicians	.13
A group-model or staff-model HMO	.05
Government (VA, military)	.06
Faculty practice (academic or residence based)	.12
Years since graduation from medical school (4–49)	18.66 (9.94)

Note: N = 81. Standard deviations are given for continuous variables only. CHD = coronary heart disease; HMO = health maintenance organization; VA = Veterans Administration.

Bivariate tests revealed no significant differences between racial-ethnic categories under low time pressure. Under high time pressure, however, physicians who were exposed to black stimuli evaluated the likelihood of CHD and angina as significantly lower compared to physicians who were exposed to white/control stimuli. The magnitudes of these differences were 1.6 for CHD (p = .011) and 2.0 for angina (p = .003). A significant difference was also observed for referral. Under high time pressure, the probability of referral in the white/control category was 76 percent but only 39 percent in the black category (p = .008).

Multivariate Analysis

In multivariate models that include main effects of racial-ethnic categories (model 1 for each dependent variable), medical decisions in black and Hispanic conditions did not differ from the white/control condition (see Table 2). Model 2, however, reveals important differences in the relationships between racial-ethnic stimuli and medical decisions by time pressure, as indicated by significant coefficients for interaction effects. A significant interaction between the black category and time pressure was found in all models. A significant interaction between the Hispanic category and time pressure was observed for diagnostic decisions but not for referral.

Table 2.

Unstandardized Coefficients from Multivariate Regression Models of Medical Decisions

	CHD^a		Angina^a		Referral^b
	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2
High time pressure	.05	1.01	−.01	1.23	.56	2.35
	(.899)	(.081)	(.988)	(.035)	(.329)	(.002)
Racial-ethnic stimuli^c
Black	−.88	−.01	−.89	.14	−.57	.92
	(.149)	(.988)	(.175)	(.838)	(.385)	(.381)
Hispanic	−.14	.58	.27	1.35	−.33	−.26
	(.767)	(.251)	(.610)	(.003)	(.647)	(.760)
Interactions
Black × High time pressure		−2.32		−2.77		−4.38
		(.018)		(.007)		(.003)
Hispanic × High time pressure		−1.80		−2.60		−1.25
		(.036)		(.004)		(.371)

Note: N = 81. All models control for physician gender, race, Hispanic ethnicity, years since graduation, specialty, and type of practice. None of the control variables yielded significant effects. p values for two-tailed tests appear in parentheses.CHD = coronary heart disease.

Linear regression models using robust standard errors.

Logistic regression models using robust standard errors.

Reference category is white/control.

Predicted values for diagnostic decisions (CHD and angina) and predicted probabilities for referral help understand these interaction effects (see Figure 1). Significant differences suggesting racial-ethnic bias were not evident under low time pressure but emerged when the physicians were placed under high time pressure. After controlling for confounding factors, physicians exposed to black stimuli evaluated the likelihood of CHD as 1.3 points lower on a 1 to 10 scale compared to physicians exposed to white/control stimuli (p = .011). In the model of angina, this difference was 1.5 points (p = .002). A similar pattern was evident in the model of referral. The predicted probability of referral for physicians exposed to high time pressure and black stimuli was only 48 percent, while for physicians exposed to high time pressure and white/control stimuli, it was 94 percent (p < .001).

Figure 1.

Predicted Values for Diagnostic Decisions (A and B) and Predicted Probabilities for Referral to Specialist (C) by Time Pressure and by Categories of Racial-Ethnic Stimuli

The comparisons of medical decisions under high versus low time pressure for each racial-ethnic category yielded additional interesting insights. The most striking was the difference in the referral rates for blacks. Under high time pressure, the chances of referral fell by nearly 50 percent compared to low time pressure (p = .003). Considerable decreases were evident also for the likelihood of diagnosing angina (p = .007) and CHD (p = .018). A similar pattern was evident in the Hispanic condition for angina (p = .004) and CHD (p = .036). In contrast, the chances of referral in the white/control condition improved by nearly 30 percent under high time pressure (p = .002). Taken together, these results were consistent with my hypothesis.

Discussion

To investigate the sources of racial-ethnic inequality in medicine, the present study adapted a time-tested method from psychological research that focuses on implicit biases. It is among the first to provide evidence that these implicit processes, despite their mostly nonconscious nature, play an important role in physicians’ clinical judgment and decision making. The findings revealed that biases activated by implicit information about patient race-ethnicity influenced medical judgment, but only under high time pressure. When doctors were under the type of stress related to time pressure, they systematically leaned toward more aggressive pursuit of diagnostic leads for whites than for blacks. The magnitudes of these differences were considerable. For instance, the probability of referral to a specialist under high time pressure was 94 percent for whites but only 48 percent for blacks. While an aggressive pursuit of diagnostic leads is not necessarily beneficial in all medical situations, it is likely to be a better option in this particular case, since we know that non-whites have significantly poorer cardiovascular health. Thus, the results imply that under high time pressure, minority patients may receive poorer care compared to their white counterparts.

The findings have important implications for the theoretical understanding of race-ethnicity as a stratifying variable in medicine. The evidence suggesting a causal role of physicians’ implicit biases in disparities resonates with a fundamentally sociological view that racial-ethnic inequalities have macro-level, historical, and socio-structural origins but at the same time they are maintained on the micro level through interpersonal interaction (Malat 2006). Implicit biases are one mechanism through which this happens. These biases develop because we live in a racially-ethnically divided society and they affect many areas of everyday life, including professional decision making by doctors.

The interesting findings regarding the role of time pressure add another layer. Time pressure, a contextual factor rooted in the organizational structure of medicine, modified the influence of racial-ethnic biases on clinical judgment. This finding highlights the need for considering contextual factors along with individual factors (e.g., racial-ethnic biases) in studies of racial inequalities, since they jointly influence outcomes. For instance, the chances of referral under high time pressure fell by nearly a half for blacks. In contrast, these chances actually improved by nearly a third for whites. These findings show white privilege in action and are particularly interesting vis-à-vis prior literature that debates whether disparities in care are a result of too little care for minorities or too much care for whites (Epstein et al. 2000). The current study suggests that when physicians are under time pressure, they step up their pursuit of diagnostic clues for whites but step it down for blacks. Thus, majority privilege is evident under time pressure, and it works jointly with minority underprivilege to constitute disparities in care.

The results have additional implications for several specific perspectives on racial-ethnic inequalities. Statistical discrimination perspective, for instance, argues that doctors know about the differences in the prevalence of diseases in various population groups and use this information when formulating a diagnosis, especially under clinical uncertainty (Balsa and McGuire 2001). Yet, recent evidence suggests that the actual disease rates are less important than their socially constructed aspects. For instance, physicians are less likely to diagnose cardiovascular disease among women compared to men because they believe that women face a lower risk, even though epidemiological evidence contradicts such beliefs (Lutfey et al. 2010; Maserejian, Lutfey, and McKinlay 2009). My study yielded a similar pattern of less serious diagnoses for groups that are objectively more vulnerable to disease. These findings challenge the classic formulation of the statistical discrimination hypothesis and support instead the importance of stereotypical social representations in clinical judgment.

Another perspective on racial-ethnic disparities stresses patient-level factors beyond race-ethnicity, such as cultural expectations, treatment preferences, trust, and tendency to comply with medical advice, which may differ among racial-ethnic groups. These factors may act as cultural barriers and negatively influence clinical encounters between white doctors and minority patients. The danger of patient-level explanations is that they shift the focus away from the uncomfortable issue of racism and important structural factors. Therefore, the evidence indicating that patient-level factors cannot fully explain racial-ethnic differences in treatment is important. My results add to this body of evidence, since patient-level factors including socioeconomic status, insurance, and work and family situation (which may contain clues about the social position and health-relevant attitudes) were held constant and could not cause the observed differences. Yet, one possibility remains: Widely held stereotypes regarding individual characteristics of minority versus white patients may have played a role. If negative stereotypes about minority patients as distrustful and noncompliant (Bogart et al. 2000; Sabin et al. 2008; van Ryn and Burke 2000) were activated by racial-ethnic stimuli, they could have influenced physician responses even though the vignette did not contain any information consistent with such stereotypes. This possibility resonates with Lutfey et al.’s (2010) argument about socially constructed input into medical decisions and underscores the need for further attention to this area.

Strengths and Limitations of the Study

One key strength of this study is that it facilitated causal inference since it minimized the influence of confounding factors by the use of random assignment, experimental controls, and purposive manipulations. This is an important step forward since previous evidence regarding racial-ethnic biases in medicine, including the two studies on implicit biases in medical decisions reviewed earlier (Green et al. 2007; Sabin et al. 2008), is almost exclusively correlational. To design effective policies to eliminate racial-ethnic inequalities, we need to understand causes of these inequalities, not just their correlates.

Another strength is that the study used the Internet-based design to reach physicians practicing at diverse settings nationwide. This was particularly important to alleviate concerns about ecological validity that are common in experiments, especially in those conducted in laboratories with college students. Yet, some concerns regarding ecological validity remain as the necessary trade-off with the advantage of causal inference. It is not feasible to manipulate the characteristics of real patients, which is why vignettes were used. Despite the evidence that the ecological validity of vignettes favorably compares to other methods, such as chart abstraction or standardized patients (Peabody et al. 2000), the possibility that real-life medical decisions differ from responses to vignettes clearly remains. Vignettes represent simplified medical situations. In real life, physicians have opportunities to ask patients follow-up questions and to perform additional examinations and tests. Such opportunities are lacking when patient information is presented in vignettes, but such information may not be as routinely collected when physician face serious time constraints.³

Despite this limitation, vignettes provide several advantages that strengthen the study. First, they standardize the information delivered to physicians. This bolsters confidence that the observed differences in medical decisions do not result from clinically relevant differences in patient presentation or factors in the physician’s environment—a possibility that is always present in studies with real patients. Another advantage is the fact that vignetes can be presented without any racial-ethnic identifiers. This helps to isolate the effects of implicit biases from any potential explicit biases that physicians may have. When racial-ethnic information is explicit, for instance, in studies that use ostensive patients’ images, implicit biases are activated alongside any explicit biases that may exist. Even though explicit racism among physicians is uncommon (Sabin et al. 2008, 2009), it is entirely possible that some physicians harbor explicit prejudice (Kposowa and Tsunokai 2002). Given the nature of this study, the ability to isolate implicit biases far outweighs the disadvantage of lowered ecological validity of vignette design.

Implications for Efforts to Reduce Health Care Disparities

Much of the work on racial-ethnic bias and discrimination is motivated by the effort to reduce inequalities. This study provided some initial evidence about causal mechanisms linking provider bias, time pressure, and medical decisions. Based on its findings, it seems useful to consider a two-pronged approach, simultaneously targeting individual and contextual sources of inequalities.

Individual-level strategies directly target physicians. These interventions are based on the assumption that racial-ethnic biases, including their implicit components, are malleable, especially if individuals are motivated to preserve their self-image or to appropriately respond to social demands (Blair 2002). Educational interventions should consequently aim to raise physicians’ awareness of implicit racial-ethnic biases, educate them about the cognitive mechanisms responsible for biased decision making, motivate them to provide equitable care, and provide a nonthreatening environment to practice new skills (Burgess et al. 2007). Physicians may benefit from knowing that when they are busy, stressed, or multitasking, they are at an increased risk for biased decision making. Such knowledge would arm physicians in their efforts to resist biases during times of stress.

The second approach targets organizations. One could encourage organizations to create work environments that minimize contextual factors that contribute to disparities. This study pinpointed time pressure as one such factor, suggesting that taking steps to alleviate time pressure may help medical personnel to reduce the impact of clinically irrelevant information on their decisions. With adequate time, physicians can attend to individual patient information, instead of (perhaps unconsciously) turning to heuristic devices such as social categorization. They can also use free cognitive energy to make conscious effort to counter stereotypes.

Footnotes

Appendix

Acknowledgements

I would like to thank Karen Cook, Cecilia Ridgeway, Buzz Zelditch, Brian Lowery, Claude Steele, Brent Simpson, Barry Markovsky, and anonymous reviewers for their helpful comments at various stages of this project.

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The author received a Graduate Research Opportunity Grant from Stanford University and funding from Office of the Dean, College of Arts and Sciences, University of South Carolina.

Notes

Bio

Irena Stepanikova is an assistant professor at the Department of Sociology at The University of Alabama at Birmingham. She studies racial-ethnic disparities in health care and in health and well-being. Her interests also include physician-patient communication and trust, social psychology, and research methods.

References

Abreu

José M.

1999. “Conscious and Non-Conscious Black Stereotypes: Impact on First Impression and Diagnostic Ratings by Therapists.” Journal of Consulting and Clinical Psychology 67:387–93.

Abreu

José M.

Ramirez

Estrella

Kim

Bryan S. K.

Haddy

Chris

. 2003. “Automatic Activation of Yellow Peril Asian American Stereotypes: Effects on Social Impression Formation.” The Journal of Social Psychology 143:691–706.

Allport

Gordon

. 1954. The Nature of Prejudice. Cambridge, MA: Addison-Wesley.

Balsa

Ana I.

McGuire

Thomas G.

2001. “Statistical Discrimination in Health Care.” Journal of Health Economics 20:881–907.

Bargh

John A.

Chartland

Tanya L.

2000. “Mind in the Middle: A Practical Guide to Priming and Automaticity Research.” Pp. 253–85 in Handbook of Research Methods in Social and Personality Psychology, edited by Reis

H. T.

Judd

C. M.

New York, NY: Cambridge University Press.

Bargh

John A.

Chen

Mark

Burrows

Lara

. 1996. “Automaticity of Social Behavior: Direct Effects of Trait Construct and Stereotype Activation on Action.” Journal of Personality and Social Psychology 71:230–44.

Bargh

John A.

Pietromonaco

Paula R.

1982. “Automatic Information Processing and Social Perception: The Influence of Trait Information Presented Outside of Conscious Awareness on Impression Formation.” Journal of Personality and Social Psychology 43:437–49.

Bartholow

Bruce D.

Dickter

Cheryl L.

Sestir

Marc A.

2006. “Stereotype Activation and Control of Race Bias: Cognitive Control of Inhibition and Its Impairment by Alcohol.” Journal of Personality and Social Psychology 90:272–87.

Blair

Irene V.

2001. “Implicit Stereotypes and Prejudice.” Pp. 359–74 in Cognitive Social Psychology: The Princeton Symposium on the Legacy and Future of Social Cognition, edited by Moskowitz

G. B.

Mahwah, NJ: Erlbaum.

10.

Blair

Irene V.

2002. “The Malleability of Automatic Stereotypes and Prejudice.” Personality and Social Psychology Review 6:242–61.

11.

Bogart

Laura M.

Kelly

Jeffrey A.

Catz

Sheryl L.

Sosman

James M.

2000. “Impact of Medical and Non-Medical Factors on Physician Decision Making for HIV/AIDS Antiretroviral Treatment.” Journal of Acquired Immune Deficiency Syndrome 23:396–404.

12.

Bovier

Patrick A.

Perneger

Thomas V.

2003. “Predictors of Work Satisfaction among Physicians.” European Journal of Public Health 13:299–305.

13.

Brown

Rupert

Croizet

Jean-Claude

Bohner

Gerd

Fournet

Marion

Payne

Andrew

. 2003. “Automatic Category Activation and Social Behavior: The Moderating Role of Prejudiced Beliefs.” Social Cognition 21:167–93.

14.

Burgess

Diana J.

Steven S.

van Ryn

Michelle

. 2004. “Why Do Providers Contribute to Disparities and What Can Be Done about It?” Journal of General Internal Medicine 19:1154–59.

15.

Burgess

Diana J.

van Ryn

Michelle

Crowley-Matoka

Megan

Malat

Jennifer

. 2006. “Understanding the Provider Contribution to Race/Ethnicity Disparities in Pain Treatment: Insights from Dual Process Models of Stereotyping.” Pain Medicine 7:119–34.

16.

Burgess

Diana J.

van Ryn

Michelle

Dovidio

John

Saha

Somnath

. 2007. “Reducing Racial Bias among Health Care Providers: Lessons from Social-Cognitive Psychology.” Society of General Internal Medicine 22:882–87.

17.

Byrd

Michael W.

1990. “Race, Biology, and Health Care: Reassessing a Relationship.” Journal of Health Care for the Poor and Underserved 1:278–96.

18.

Center for Studying Health System Change. 2006. Community Tracking Study Physician Survey, 2004-2005. [Computer file]. ICPSR version. Washington, DC: Center for Studying Health System Change [producer]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].

19.

Chen

Mark

Bargh

John A.

1997. “Non-Conscious Behavioral Confirmation Process.” Journal of Experimental Social Psychology 33:541–60.

20.

Devine

Patricia G.

1989. “Stereotypes and Prejudice: Their Automaticity and Controlled Components.” Journal of Personality and Social Psychology 56:5–18.

21.

Dovidio

John F.

Kawakami

Kerry

Gaertner

Samuel L.

2002. “Implicit and Explicit Prejudice and Interracial Interaction.” Journal of Personality and Social Psychology 82:62–68.

22.

Dovidio

John F.

Kawakami

Kerry

Johnson

Craig

Johnson

Brenda

Howard

Adaiah

. 1997. “On the Nature of Prejudice: Automatic and Controlled Processes.” Journal of Experimental Social Psychology 33:510–40.

23.

Dovidio

John F.

Penner

Louis A.

Albrecht

Terrance L.

Norton

Wynne E.

Gaertner

Samuel L.

Shelton

J. Nicole

. 2008. “Disparities and Distrust: The Implications of Psychological Processes for Understanding Racial Disparities in Health and Health Care.” Social Science & Medicine 67:478–86.

24.

Epstein

Arnold M.

Ayanian

John Z.

Keogh

Joseph H.

Noonan

Susan J.

Armistead

Nancy

Cleary

Paul D.

Weissman

Joel S.

David-Kasdan

Jo Ann

Carlson

Diane

Fuller

Jerry

Marsh

Douglas

Conti

Rena M.

2000. “Racial Disparities in Access to Renal Transplantation: Clinically Appropriate or Due to Underuse or Overuse?” New England Journal of Medicine 343:1537–44.

25.

Fazio

Russell H.

Jackson

Joni R.

Dunton

Bridget C.

Williams

Carol J.

1995. “Variability in Automatic Activation as an Unobtrusive Measure of Racial Attitudes: A Bona Fide Pipeline?” Journal of Personality and Social Psychology 69:1013–27.

26.

Fazio

Russell H.

Olson

Michael A.

2003. “Implicit Measures in Social Cognition Research: Their Meaning and Use.” Annual Review of Psychology 54:297–327.

27.

Gold

Marsha R.

Hurley

Robert

Lake

Timothy

Endor

Todd

Berenson

Robert

. 1995. “A National Survey of the Arrangements Managed-Care Plans Make with Physicians.” New England Journal of Medicine 333:1678–83.

28.

Graham

Sandra

Lowery

Brian S.

2004. “Priming Unconscious Racial Stereotypes about Adolescent Offenders.” Law and Human Behavior 28:483–504.

29.

Green

Alexander R.

Carney

Dana R.

Pallin

Daniel J.

Ngo

Long H.

Raymond

Kristal L.

Iezzoni

Lisa I.

Banaji

Mahzarin R.

2007. “Implicit Bias among Physicians and Its Prediction of Thrombolysis Decisions for Black and White Patients.” Journal of General Internal Medicine 22:1231–38.

30.

Greenwald

Anthony G.

Banaji

Mahzarin

. 1995. “Implicit Social Cognition: Attitudes, Self-Eseem, and Stereotypes.” Psychological Review 102:4–27.

31.

Institute of Medicine. 2003. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC: National Academy Press.

32.

Johnson

Rachel L.

Saha

Somnath

Arbelaez

Jose J.

Beach

Mary Catherine

Cooper

Lisa A.

2004. “Racial and Ethnic Differences in Patient Perceptions of Bias and Cultural Competence in Health Care.” Journal of General Internal Medicine 19:101–10.

33.

Kawakami

Kerry

Dion

Kenneth

Dovidio

John F.

1998. “Racial Prejudice and Stereotype Activation.” Personality and Social Psychology Bulletin 24:407–16.

34.

Kawakami

Kerry

Dovidio

John F.

2001. “The Reliability of Implicit Stereotyping.” Personality and Social Psychology Bulletin 27:212–25.

35.

Kim

Do-Yeong

. 2003. “Voluntary Controllability of the Implicit Association Test (IAT).” Social Psychology Quarterly 66:83–96.

36.

Kposowa

Augustine J.

Tsunokai

Glenn T.

2002. “Searching for Relief: Racial Differences in Treatment of Patients with Back Pain.” Race and Society 5:193–223.

37.

LaVeist

Thomas A.

Rolley

Nicole C.

Diala

Chamberlain

. 2003. “Prevalence and Patterns of Discrimination among US Healthcare Consumers.” International Journal of Health Services 33:331–44.

38.

Lepore

Lorella

Brown

Rupert

. 1997. “Category and Stereotype Activation: Is Prejudice Inevitable?” Journal of Personality and Social Psychology 72:275–87.

39.

Linzer

Mark

Konrad

Thomas

Douglas

Jeffrey

McMurray

Julia E.

Pathman

Donald E.

Williams

Eric S.

Schwartz

Mark D.

Gerrity

Martha

Scheckler

William

Bigby

Elnora

. 2000. “Managed Care, Time Pressure, and Physician Job Satisfaction: Results from the Physician Worklife Study.” Journal or General and Internal Medicine 15:441–50.

40.

Livingston

Robert W.

Brewer

Marilynn B.

2002. “What Are We Really Priming? Cue-Based Versus Category-Based Processing of Facial Stimuli.” Journal of Personality and Social Psychology 82:2–18.

41.

Lutfey

Karen E.

Eva

Kevin W.

Gerstenberger

Eric

Link

Carol L.

McKinlay

John B.

2010. “Physician Cognitive Processing as a Source of Diagnostic and Treatment Disparities in Coronary Heart Disease: Results of a Factorial Priming Experiment.” Journal of Health and Social Behavior 51:16–29.

42.

Lutfey

Karen E.

Link

Carol L.

Grant

Richard W.

Marceau

Lisa D.

McKinlay

John B.

2009. “Is Certainty More Important Than Diagnosis for Understanding Race and Gender Disparities? An Experiment Using Coronary Heart Disease and Depression Case Vignettes.” Health Policy 89:279–87.

43.

Macrae

C. Neil

Milne

Alan B.

Bodenhausen

Galen V.

1994. “Stereotypes as Energy-Saving Devices: A Peek inside the Cognitive Toolbox.” Journal of Personality and Social Psychology 66:37–47.

44.

Malat

Jennifer

. 2006. “Expanding Research on the Racial Disparity in Medical Treatment with Ideas from Sociology.” Health 10:259–82.

45.

Maserejian

Nancy N.

Lutfey

Karen E.

McKinlay

John B.

2009. “Do Physicians Attend to Base Rates? Prevalence Data and Statistical Discrimination in the Diagnosis of Coronary Heart Disease.” Health Services Research 44:1933–49.

46.

Mort

Elizabeth A.

Edwards

Jennifer N.

Emmons

David W.

Convery

Karen

Blumenthal

David

. 1996. “Physician Response to Patient Insurance Status in Ambulatory Care Clinical Decision-Making: Implications for Quality of Care.” Medical Care 34:783–97.

47.

Nosek

Brian A.

Smyth

Frederick L.

Hansen

Jeffrey J.

Devos

Thierry

Lindner

Nicole M.

Ranganath

Kate A.

Smith

Colin Tucker

Olson

Kristina R.

Chugh

Dolly

Greenwald

Anthony G.

Banaji

Mahzarin R.

2007. “Pervasiveness and Correlates of Implicit Attitudes and Stereotypes.” European Review of Social Psychology 18:36–88.

48.

Pager

Devah

Shepherd

Hana

. 2008. “The Sociology of Discrimination: Racial Discrimination in Employment, Housing, Credit, and Consumer Markets.” Annual Review of Sociology 34:181–209.

49.

Payne

B. Keith

. 2001. “Prejudice and Perception: The Role of Automatic and Controlled Processes in Misperceiving a Weapon.” Journal of Personality and Social Psychology 181:185–86.

50.

Peabody

John W.

Luck

Jeff

Glassman

Peter

Dresselhaus

Timothy R.

Lee

Martin

. 2000. “Comparison of Vignettes, Standardized Patients, and Chart Abstraction. A Prospective Validation Study of 3 Methods for Measuring Quality.” JAMA 83:1715–22.

51.

Quillian

Lincoln

. 2006. “New Approaches to Understanding Racial Prejudice and Discrimination.” Annual Review of Sociology 32:299–328.

52.

Rudman

Laurie A.

Ashmore

Richard D.

2007. “Discrimination and the Implicit Association Test.” Group Processes and Intergroup Relations 10:359–72.

53.

Sabin

Janice A.

Nosek

Brian A.

Greenwald

Anthony G.

2009. “Physician Implicit and Explicit Attitudes about Race and MD Gender, Race and Ethnicity.” Journal of Health Care for the Poor and Underserved 20:896–913.

54.

Sabin

Janice A.

Rivara

Frederick P.

Greenwald

Anthony G.

2008. “Physician Implicit Attitudes and Stereotypes about Race and Quality of Medical Care.” Medical Care 46:678–85.

55.

Schulman

Kevin A.

Berlin

Jesse A.

Harless

William

Kerner

Jon F.

Sistrunk

Shyrl

Gersh

Bernard J.

Dube

Ross

Taleghani

Christopher K.

Burke

Jennifer E.

Williams

Sankey

Eisenberg

John M.

Escarce

Jose J.

Ayers

William

. 1999. “The Effect of Race and Sex on Physicians’ Recommendations for Cardiac Catheterization.” New England Journal of Medicine 340:618–26.

56.

Spickard

Anderson

Gabbe

Steven G.

Christensen

John F.

2002. “Mid-Career Burnout in Generalist and Specialist Physicians.” JAMA 288:1447–50.

57.

Stepanikova

Irena

Triplett

Jennifer

Simpson

Brent

. 2011. “Implicit Racial Bias and Prosocial Behavior.” Social Science Research 40:1186–95.

58.

Tinkler

Justine Eatenson

Yan E.

Mollborn

Stefanie

. 2007. “Can Legal Interventions Change Beliefs? The Effect of Exposure to Sexual Harassment Policy on Men’s Gender Beliefs.” Social Psychology Quarterly 70:480–94.

59.

van Ryn

Michelle

. 2002. “Research on the Provider Contribution to Race/Ethnicity Disparities in Medical Care.” Medical Care 40(Suppl.):I-140–51.

60.

van Ryn

Michelle

Burke

Jane

. 2000. “The Effect of Patient Race and Socio-Economic Status on Physicians’ Perceptions of Patients.” Social Science and Medicine 50:813–28.

61.

van Ryn

Michelle

Steven S.

2003. “Paved with Good Intentions: Do Public Health and Human Service Providers Contribute to Racial-Ethnic Disparities in Health?” The American Journal of Public Health 93:248–55.

62.

Winkleby

Marilyn A.

Kraemer

Helena C.

Ahn

David K.

Varady

Ann N.

1998. “Ethnic and SocioEconomic Differences in Cardiovascular Disease Risk Factors.” JAMA 280:356–62.

63.

Wittenbrink

Bernd

Judd

Charles M.

Park

Bernadette

. 1997. “Evidence for Racial Prejudice at the Implicit Level and Its Relationship with Questionnaire Measures.” Journal of Personality and Social Psychology 72:262–74.