implicit bias and its impact on diversityucd-advance.ucdavis.edu/sites/main/files/file...implicit...

Implicit Bias and Its Impact on Diversity

•  Implicit bias –  Experience –  Definition –  Measurement

•  Evidence of impact on behavior

•  Evidence of impact in bias in career-related evaluation

•  Is there any good news? What can be done?

•  The way we perceive, judge, remember is often full of errors Feeling confident ≠ Being accurate

•  What we already know affects what we perceive; preconceived expectations influence current judgments

•  Errors in perception (mindbugs) are ordinary byproducts of normal mental processes – Memory – Perception – Learned associations

•  Ordinary: – all humans are prone to these errors

–  they are unintentional, occurring without our awareness or control

Implicit bias

Implicit bias literature reviews

•  Some concepts automatically go together in our mind because we’ve learned these associations simply by being immersed in society

Implicit bias driven by learned associations

Name the color of each set of letters

SLB

SPRND

SLB

SPRND

HLMG

CFLTK

HLMG

SPRND

HLMG

SPRND

CFLTK

CFLTK

SLB

CFLTK

CFLTK

Name the color of each set of letters

RED

BLUE

RED

BROWN

YELLOW

GREEN

GREEN

YELLOW

BROWN

GREEN

YELLOW

BROWN

BLUE

BLUE

RED

•  Implicit Association Test (IAT)

Implicit bias measurement

•  Do mere associations show up in behavior?

–  Implicit bias predicts: •  the rate of callback for interviews (Rooth 2007)

•  awkward body language and feelings of discomfort (McConnell & Leibold 2001)

•  how we read the friendliness of facial expressions (Hugenberg & Bodenhausen 2003)

•  more negative evaluations of ambiguous actions by African Americans (Rudman & Lee 2003)

•  More negative evaluations of agentic (i.e., confident, aggressive, ambitious) women in hiring conditions (Rudman & Glick 2001)

Implicit bias impact on behavior

•  Bertrand & Mullainathan (2003) American Economic Review

To examine the effect of race on receiving job callbacks, the researchers responded with fictitious resumes to help-wanted ads in Boston and Chicago newspapers. The resumes were altered from actual ones found on job search Web sites. The researchers categorized the new resumes as high or low quality and assigned them an equal number of traditionally black names (e.g., Lakisha) or traditionally white names (e.g., Greg).

– Resumes with white names had a 50 percent greater chance of receiving a callback than did resumes with black names (10.08% vs. 6.70%, respectively).

– Higher-quality resumes elicited 30 percent more callbacks for whites, whereas they only elicited 9 percent more callbacks for blacks.

– Employers who listed "Equal Opportunity Employer" in their ad discriminated just as much as other employers.

Evidence of impact of bias in career-related evaluation

•  Trix & Psenka (2003) Discourse & Society

The researchers analyzed 312 real letters of recommendation that helped medical school faculty attain their clinical and research positions. The letters were received by a large U.S. medical school from 1992 to 1995.

– Compared with letters of recommendation for males, letters for females were

•  shorter

•  more likely to be "letters of minimal assurance" (e.g., lacking in specificity)

•  more likely to contain gender terms (e.g., "she is an intelligent young lady")

•  more likely to include "doubt raisers" (e.g., criticisms, hedges, faint praise)


•  Wennerås & Wold (1997) Nature

The researchers evaluated the peer- review system of postdoctoral fellowships at the Swedish Medical Research Council. They obtained the evaluator reviews through Freedom of the Press Act. Applicants included 62 men, 52 women; awardees included 16 men and 4 women.

– Women were graded below men in all 3 categories of scientific achievement

•  10% lower in scientific competence

•  7% lower for proposed methodology

•  5% lower for proposal relevance

Controlling for:

–  Number of publications (total, first-authored)

–  Summed journal impact factors (total, first-authored)

–  Number of citations (total, first-authored)

–  Other factors included in regression model: gender, nationality, discipline, post-doc abroad, affiliation with member of the evaluation committee


•  Wennerås & Wold (1997)

– “… a female applicant had to be 2.5 times more productive than the average male applicant to receive the same competence score as he…”

–  the positive impacts of being male and affiliated with a member of the review committee exceeded the influence of measures of scientific impact and productivity by 52 – 220%


!"##$%&'()

•  Ginther et al. (2011) Science

Analyzed the association between a U.S. National Institutes of Health (NIH) R01 applicant’s self-identified race or ethnicity and the probability of receiving an award using data from the NIH IMPAC II grant database, the Thomson Reuters Web of Science, and other sources.

– After controlling for the applicant’s educational background, country of origin, training, previous research awards, publication record, and employer characteristics, African-American applicants are 10 percentage points less likely than whites to be awarded NIH research funding.


•  Moss-Racusin, et al. (2012) PNAS

A broad, nationwide sample of biology, chemistry, and physics professors (n = 127) evaluated the application materials of an undergraduate science student who had ostensibly applied for a science laboratory manager position. All participants received the same materials, which were randomly assigned either the name of a male (n = 63) or a female (n = 64) student; student gender was thus the only variable that differed between conditions.

– The applications identified with female names were rated significantly lower than those male names on competence and hireability, as well as the amount of salary and amount of mentoring they would offer the student


bias because they have been rigorously trained to be objective.On the other hand, research demonstrates that people who valuetheir objectivity and fairness are paradoxically particularly likelyto fall prey to biases, in part because they are not on guardagainst subtle bias (24, 25). Thus, by investigating whether sci-ence faculty exhibit a bias that could contribute to the genderdisparity within the fields of science, technology, engineering,and mathematics (in which objectivity is emphasized), the cur-rent study addressed critical theoretical and practical gaps in thatit provided an experimental test of faculty discrimination againstfemale students within academic science.A number of lines of research suggest that such discrimination

is likely. Science is robustly male gender-typed (26, 27), resour-ces are inequitably distributed among men and women in manyacademic science settings (28), some undergraduate womenperceive unequal treatment of the genders within science fields(29), and nonexperimental evidence suggests that gender bias ispresent in other fields (19). Some experimental evidence sug-gests that even though evaluators report liking women more thanmen (15), they judge women as less competent than men evenwhen they have identical backgrounds (20). However, thesestudies used undergraduate students as participants (rather thanexperienced faculty members), and focused on performancedomains outside of academic science, such as completing per-ceptual tasks (21), writing nonscience articles (22), and beingevaluated for a corporate managerial position (23).Thus, whether aspiring women scientists encounter discrimi-

nation from faculty members remains unknown. The formativepredoctoral years are a critical window, because students’ expe-riences at this juncture shape both their beliefs about their ownabilities and subsequent persistence in science (30, 31). There-fore, we selected this career stage as the focus of the presentstudy because it represents an opportunity to address issues thatmanifest immediately and also resurface much later, potentiallycontributing to the persistent faculty gender disparity (32, 33).

Current StudyIn addition to determining whether faculty expressed a biasagainst female students, we also sought to identify the processescontributing to this bias. To do so, we investigated whetherfaculty members’ perceptions of student competence would helpto explain why they would be less likely to hire a female (relativeto an identical male) student for a laboratory manager position.Additionally, we examined the role of faculty members’ preex-isting subtle bias against women. We reasoned that pervasivecultural messages regarding women’s lack of competence in sci-ence could lead faculty members to hold gender-biased attitudesthat might subtly affect their support for female (but not male)science students. These generalized, subtly biased attitudes to-ward women could impel faculty to judge equivalent studentsdifferently as a function of their gender.The present study sought to test for differences in faculty

perceptions and treatment of equally qualified men and womenpursuing careers in science and, if such a bias were discovered,reveal its mechanisms and consequences within academic sci-ence. We focused on hiring for a laboratory manager position asthe primary dependent variable of interest because it functions asa professional launching pad for subsequent opportunities. Assecondary measures, which are related to hiring, we assessed: (i)perceived student competence; (ii) salary offers, which reflectthe extent to which a student is valued for these competitivepositions; and (iii) the extent to which the student was viewed asdeserving of faculty mentoring.Our hypotheses were that: Science faculty’s perceptions and

treatment of students would reveal a gender bias favoring malestudents in perceptions of competence and hireability, salaryconferral, and willingness to mentor (hypothesis A); Faculty gen-der would not influence this gender bias (hypothesis B); Hiring

discrimination against the female student would be mediated (i.e.,explained) by faculty perceptions that a female student is lesscompetent than an identical male student (hypothesis C); andParticipants’ preexisting subtle bias against women would mod-erate (i.e., impact) results, such that subtle bias against womenwould be negatively related to evaluations of the female student,but unrelated to evaluations of the male student (hypothesis D).

ResultsA broad, nationwide sample of biology, chemistry, and physicsprofessors (n = 127) evaluated the application materials of anundergraduate science student who had ostensibly applied fora science laboratory manager position. All participants receivedthe same materials, which were randomly assigned either thename of a male (n = 63) or a female (n = 64) student; studentgender was thus the only variable that differed between con-ditions. Using previously validated scales, participants rated thestudent’s competence and hireability, as well as the amount ofsalary and amount of mentoring they would offer the student.Faculty participants believed that their feedback would beshared with the student they had rated (see Materials andMethods for details).

Student Gender Differences. The competence, hireability, salary con-ferral, and mentoring scales were each submitted to a two (studentgender; male, female) ! two (faculty gender; male, female) be-tween-subjects ANOVA. In each case, the effect of student genderwas significant (all P < 0.01), whereas the effect of faculty partici-pant gender and their interaction was not (all P > 0.19). Tests ofsimple effects (all d > 0.60) indicated that faculty participantsviewed the female student as less competent [t(125) = 3.89, P <0.001] and less hireable [t(125) = 4.22, P < 0.001] than the identicalmale student (Fig. 1 and Table 1). Faculty participants also offeredless careermentoring to the female student than to themale student[t(125) = 3.77, P < 0.001]. The mean starting salary offered thefemale student, $26,507.94, was significantly lower than that of$30,238.10 to the male student [t(124) = 3.42, P < 0.01] (Fig. 2).These results support hypothesis A.In support of hypothesis B, faculty gender did not affect bias

(Table 1). Tests of simple effects (all d < 0.33) indicated thatfemale faculty participants did not rate the female student asmore competent [t(62) = 0.06, P = 0.95] or hireable [t(62) = 0.41,P = 0.69] than did male faculty. Female faculty also did notoffer more mentoring [t(62) = 0.29, P = 0.77] or a higher salary[t(61) = 1.14, P = 0.26] to the female student than did their male

Fig. 1. Competence, hireability, and mentoring by student gender condition(collapsed across faculty gender). All student gender differences are significant(P < 0.001). Scales range from 1 to 7, with higher numbers reflecting a greaterextent of each variable. Error bars represent SEs. nmale student condition = 63,nfemale student condition = 64.

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1211286109 Moss-Racusin et al.

colleagues. In addition, faculty participants’ scientific field, age,and tenure status had no effect (all P > 0.53). Thus, the biasappears pervasive among faculty and is not limited to a certaindemographic subgroup.

Mediation and Moderation Analyses. Thus far, we have consideredthe results for competence, hireability, salary conferral, andmentoring separately to demonstrate the converging resultsacross these individual measures. However, composite indices ofmeasures that converge on an underlying construct are morestatistically reliable, stable, and resistant to error than are each ofthe individual items (e.g., refs. 34 and 35). Consistent with thislogic, the established approach to measuring the broad conceptof target competence typically used in this type of gender biasresearch is to standardize and average the competence scaleitems and the salary conferral variable to create one compositecompetence index, and to use this stable convergent measure forall analyses (e.g., refs. 36 and 37). Because this approachobscures mean salary differences between targets, we chose topresent salary as a distinct dependent variable up to this point, toenable a direct test of the potential discrepancy in salary offeredto the male and female student targets. However, to rigorouslyexamine the processes underscoring faculty gender bias, wereverted to standard practices at this point by averaging thestandardized salary variable with the competence scale items tocreate a robust composite competence variable (! = 0.86). Thiscomposite competence variable was used in all subsequent me-diation and moderation analyses.

Evidence emerged for hypothesis C, the predicted mediation(i.e., causal path; see SI Materials and Methods: AdditionalAnalyses for more information on mediation and the results ofadditional mediation analyses). The initially significant impact ofstudent gender on hireability (" = !0.35, P < 0.001) was reducedin magnitude and dropped to nonsignificance (" = !0.10, P =0.13) after accounting for the impact of student compositecompetence (which was a strong predictor, " = 0.69, P < 0.001),Sobel’s Z = 3.94, P < 0.001 (Fig. 3). This pattern of resultsprovides evidence for full mediation, indicating that the femalestudent was less likely to be hired than the identical male be-cause she was viewed as less competent overall.We also conducted moderation analysis (i.e., testing for fac-

tors that could amplify or attenuate the demonstrated effect) todetermine the impact of faculty participants’ preexisting subtlebias against women on faculty participants’ perceptions andtreatment of male and female science students (see SI Materialsand Methods: Additional Analyses for more information on andthe results of additional moderation analyses). For this purpose,we administered the Modern Sexism Scale (38), a well-validatedinstrument frequently used for this purpose (SI Materials andMethods). Consistent with our intentions, this scale measuresunintentional negativity toward women, as contrasted witha more blatant form of conscious hostility toward women.Results of multiple regression analyses indicated that partic-

ipants’ preexisting subtle bias against women significantly inter-acted with student gender to predict perceptions of studentcomposite competence (" = !0.39, P < 0.01), hireability (" =!0.31, P < 0.05), and mentoring (" = !0.55, P < 0.001). To in-terpret these significant interactions, we examined the simpleeffects separately by student gender. Results revealed that themore preexisting subtle bias participants exhibited againstwomen, the less composite competence (" = !0.36, P < 0.01)and hireability (" = !0.39, P < 0.01) they perceived in the fe-male student, and the less mentoring (" = !0.53, P < 0.001) theywere willing to offer her. In contrast, faculty participants’ levelsof preexisting subtle bias against women were unrelated to theperceptions of the male student’s composite competence (" =0.16, P = 0.22) and hireability (" = 0.07, P = 0.59), and theamount of mentoring (" = 0.22, P = 0.09) they were willing tooffer him. [Although this effect is marginally significant, its di-rection suggests that faculty participants’ preexisting subtle biasagainst women may actually have made them more inclined tomentor the male student relative to the female student (al-though this effect should be interpreted with caution because ofits marginal significance).] Thus, it appears that faculty partic-ipants’ preexisting subtle gender bias undermined support forthe female student but was unrelated to perceptions and treat-ment of the male student. These findings support hypothesis D.

Table 1. Means for student competence, hireability, mentoring and salary conferral by student gender conditionand faculty gender

Male target student Female target student

Male faculty Female faculty Male faculty Female faculty

Variable Mean SD Mean SD Mean SD Mean SD d

Competence 4.01a (0.92) 4.1a (1.19) 3.33b (1.07) 3.32b (1.10) 0.71Hireability 3.74a (1.24) 3.92a (1.27) 2.96b (1.13) 2.84b (0.84) 0.75Mentoring 4.74a (1.11) 4.73a (1.31) 4.00b (1.21) 3.91b (0.91) 0.67Salary 30,520.83a (5,764.86) 29,333.33a (4,952.15) 27,111,11b (6,948.58) 25,000.00b (7,965.56) 0.60

Scales for competence, hireability, and mentoring range from 1 to 7, with higher numbers reflecting a greater extent of eachvariable. The scale for salary conferral ranges from $15,000 to $50,000. Means with different subscripts within each row differsignificantly (P < 0.05). Effect sizes (Cohen’s d) represent target student gender differences (no faculty gender differences weresignificant, all P > 0.14). Positive effect sizes favor male students. Conventional small, medium, and large effect sizes for d are 0.20,0.50, and 0.80, respectively (51). nmale student condition = 63, nfemale student condition = 64. ***P < 0.001.

Fig. 2. Salary conferral by student gender condition (collapsed across facultygender). The student gender difference is significant (P < 0.01). The scaleranges from $15,000 to $50,000. Error bars represent SEs. nmale student condition=63, nfemale student condition = 64.

Moss-Racusin et al. PNAS Early Edition | 3 of 6

PSYC

HOLO

GICALAND

COGNITIVESC

IENCE

S

•  Does knowledge of the illusion have any corrective effect on diminishing the illusion?

Is there any good news? What can be done?

•  An individual’s motivation to be fair does matter, but we must first believe that there is a potential problem before we try to fix it. –  Raise awareness of implicit bias

•  Create more objective, structured evaluation and interview processes –  Develop and prioritize evaluation criteria prior to evaluating candidates and apply them

consistently to all applicants. –  Administer training to all involved in the search and interview process on how to conduct

consistent, equitable review

•  Whenever possible engage in blind review by removing indicators of gender, race/ethnicity, etc. from application materials.

•  Spend sufficient time evaluating each candidate

•  Evaluate entire applications; don’t depend too heavily on only one element (such as letters of recommendation, or the prestige of the degree-granting institution)

•  Be able to defend every decision for eliminating or advancing a candidate.

Is there any good news? What can be done?

implicit bias and its impact on diversityucd-advance.ucdavis.edu/sites/main/files/file...implicit...

Documents