Do Physicians' Implicit Views of African Americans Affect Clinical Decision Making?

M. Norman Oliver, MD, MA; Kristen M. Wells, MPH, PhD; Jennifer A. Joy-Gaba, PhD; Carlee Beth Hawkins, MA; Brian A. Nosek, PhD


J Am Board Fam Med. 2014;27(2):177-188. 

In This Article



Clinical Vignette. A clinical vignette (Appendix 1) indicated diagnostic criteria for OA for which TKR would be an appropriate treatment recommendation. The validity of this vignette was vetted a priori by a panel of family physicians. The race of the patient (black or white) was indicated by a small photograph of a man in his 50s or 60s set immediately above the vignette. We avoided drawing attention to race in the vignette to minimize suspicion of the purpose of the study. Otherwise, participants may have invoked deliberate strategies to avoid bias that they would not in normal contexts. To elicit variation in the likelihood of recommending TKR, the vignette was written such that other recommendations were possible. Overall, 44% of doctors recommended TKR.

A family physician or general internist would refer a patient to a surgeon for TKR, but primary care physicians often discuss with their patients the pros and cons of procedures, operations, and treatments that will be performed by specialists. In the clinical vignette, participants were guided through such a discussion.

Implicit Association Test. The Implicit Association Test (IAT)[32] is a widely used tool to measure implicit biases.[33] An IAT measuring implicit attitudes toward racial groups, for example, compares the average time it takes respondents to categorize images of black and white Americans and words with a good and bad meaning (eg, wonderful, terrible) in 2 critical conditions. In one condition, participants categorize black faces and good words with one response key and white faces and bad words with a second response key. In the other condition, participants categorize white faces and good words with one key and black faces and bad words with the other key. In one study, >70% of the >700,000 participants were faster in the second condition than the first, indicating stronger associations of good words with white faces compared with black faces.[14] This occurs despite whether the respondent is aware of possessing the associations in memory and whether they actively agree or disagree with them. The IAT has been conducted more than 15 million times since its inception in 1998 (eg, Nosek, Banaji, & Greenwald,[33] Nosek et al,[14]), and a large literature examines its psychometric properties and validity (see Nosek, Hawkins, & Frazier[34,35] for recent reviews). There is substantial evidence that physicians can complete the measure effectively with valid results (Sabin et al[15,36]). We used the previously recommended IAT procedure[33] and analysis protocol.[37]

The Race Preference IAT used the racial category labels black American and white American (stimuli: faces of black and white men, respectively) and evaluative category labels good (word stimuli: friendly, pleasure, happy, smiling) and bad (word stimuli: angry, noxious, brutal, terrible). The race Medical Cooperativeness IAT utilized the same racial category labels and images and the categories medically cooperative (word stimuli: takes physical therapy, accepts steroid injections, accepts NSAIDs, welcomes surgery) and medically uncooperative (word stimuli: declines physical therapy, refuses steroid injections, refuses NSAIDs, opposes surgery).

Survey Measures. Explicit bias was assessed by asking respondents whether they prefer black or white people (5-point Likert scale and 10-point thermometer scale for feelings). Participants also were asked about their beliefs about patients' willingness to undergo surgery (5-point Likert scale for both black and white patients), their opinions regarding the effectiveness of TKR, and opinions on unconscious bias and IATs before and after the test (see for a complete list of study materials).


Participants read the same vignette but were randomly assigned a picture of a black or a white patient. Participants were randomly assigned to complete the IATs before (intervention condition) or after reading the vignette (control condition). Then, participants reported whether the patient's knee pain was the result of severe OA (using a 5-point Likert scale), whether they would recommend TKR, the strength of their recommendation (using a 5-point Likert scale), and their explicit attitudes and beliefs about racial groups and the existence of implicit bias. To bolster the exposure of and education about implicit bias, the IAT ended with a debriefing that explained the differences between implicit and explicit biases and the possible role of implicit bias in decision making. All study methods were approved by the institutional review board at the University of Virginia, and all participants provided informed consent.


Two sampling methods were used for this study, with one change in the procedure between them. The samples are combined for primary reporting, and observed results did not differ meaningfully between the 2 samples when examined separately.

Direct Recruiting. We recruited 79 family medicine and internal medicine physicians from the University of Virginia Health System via an E-mail with a link to the study. Additional family medicine physicians were recruited at departmental "Grand Rounds" at the University of Virginia and at the Virginia Academy of Family Physicians annual meeting. All clinicians recruited through direct methods have a self-reported education level of MD and are likely to refer patients for TKR.

Indirect Recruiting. A second sample of clinicians was recruited through the volunteer participant pool at the Project Implicit website ( Hundreds of thousands of volunteers visit Project Implicit each year and are educated about implicit biases and given an opportunity to participate in research. After registering, participants complete a short demographic questionnaire that includes questions about educational degree(s) attained and current occupation. Participants then are randomly assigned to a study from a pool of studies—these are sometimes unobtrusively selected based on their demographic responses.

For this study, respondents who reported their education level as MD and listed their current occupation as "health care–diagnosing and treating practitioners" during registration were eligible to be assigned to this study. Once assigned, the clinicians indicated whether they felt qualified to recommend TKR. Those who agreed continued to the study (n = 587); those who disagreed were assigned to another study. Participants were subsequently excluded if they indicated that they were not currently or had never been employed as medical professionals (n = 43) or reported their age as ≤22 years old (n = 27), making it unlikely that they had completed medical school and residency. Another 53 participants dropped out of the study before viewing the patient vignette, leaving 464 participants. Other than these recruitment procedures, the study procedure was identical to the previous sample except, because of time constraints (Nosek, Sriram, & Umansky[38]), the online participants completed only the Race Medical Cooperativeness IAT and not the Race Preference IAT. We prioritized using the implicit measure that Green et al.[30] demonstrated had a stronger effect: the Race Medical Cooperativeness IAT.