CoChat
Sign Up
Cited by 1,259Social and Intergroup PsychologyRacial and Ethnic Identity ResearchNames, Identity, and Discrimination Research

The DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index; none of the 45 sources its claims cite is checked for support yet.

Long-term reduction in implicit race bias: A prejudice habit-breaking intervention

Open access
Devine, Patricia G., Forscher, Patrick S., Austin, Anthony J., Cox, William T.L.·2012·Journal of Experimental Social Psychology
DOI 10.1016/j.jesp.2012.06.003PMC PMC3603687
Open this paper in CoChat — free account

Abstract

We developed a multi-faceted prejudice habit-breaking intervention to produce long-term reductions in implicit race bias. The intervention is based on the premise that implicit bias is like a habit that can be reduced through a combination of awareness of implicit bias, concern about the effects of that bias, and the application of strategies to reduce bias. In a 12-week longitudinal study, people who received the intervention showed dramatic reductions in implicit race bias. People who were concerned about discrimination or who reported using the strategies showed the greatest reductions. The intervention also led to increases in concern about discrimination and personal awareness of bias over the duration of the study. People in the control group showed none of the above effects. Our results raise the hope of reducing persistent and unintentional forms of discrimination that arise from implicit bias.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

A multifaceted prejudice habit-breaking intervention produced reductions in implicit race bias that persisted 8 weeks after the manipulation in a randomized controlled trial. Intervention participants showed lower IAT scores than controls, and greater week-2 concern about discrimination and strategy use predicted larger reductions.

Key findings

  • Intervention participants had lower implicit race bias than control participants after the manipulation, \(B = -.19\), \(t(88) = -2.82\), \(p = .006\), \(\Delta R^2 = .081\); the reductions at 4 and 8 weeks did not differ systematically, \(B = .091\), \(t(88) = .82\), \(p = .42\).
  • The intervention increased concern about discrimination, \(B = .43\), \(t(88) = 2.24\), \(p = .028\), \(\Delta R^2 = .049\), and prejudice-relevant discrepancies, \(B = .38\), \(t(88) = 2.31\), \(p = .024\), \(\Delta R^2 = .047\), but did not change racial attitudes or internal/external motivations to respond without prejudice (all \(ps \ge .53\)).
  • Week-2 concern about discrimination moderated the intervention effect on implicit bias, \(B = -.15\), \(t(86) = -2.46\), \(p = .016\), \(\Delta R^2 = .058\); high-concern intervention participants had IAT scores .38 units lower at week 4 and .31 units lower at week 8 than high-concern control participants.
  • Self-reported likelihood to use the strategies predicted lower implicit race bias, \(B = -.14\), \(t(50) = -3.18\), \(p = .003\), \(\Delta R^2 = .17\); implementation-related word stems at week 2 predicted lower week-4 IAT bias, \(B = -.068\), \(t(50) = -2.47\), \(p = .017\), \(\Delta R^2 = .11\), and likelihood plus implementation word use jointly accounted for 23% of the variance in week-4 implicit bias.

Methods

  • Ninety-one non-Black introductory psychology students (67% female, 85% White) completed a 12-week longitudinal study; participants were randomly assigned to intervention (\(n = 53\)) or control (\(n = 38\)).
  • Implicit race bias was measured with the Black–White IAT at baseline and 4 and 8 weeks after the manipulation; explicit measures were administered 4 weeks before and 2 and 6 weeks after the manipulation. The intervention was a 45-minute narrated interactive slideshow with education and training in five strategies: stereotype replacement, counter-stereotypic imaging, individuation, perspective taking, and increasing opportunities for contact.
  • Analyses used general linear models with the baseline dependent variable as a covariate, multiple imputation for missing data, moderation analyses, and text mining of free-response strategy-use reports.

Limitations

  • The paper states that the multifaceted intervention creates ambiguity about which components are necessary or sufficient, that the study lacked precise quantitative indicators of use of the five strategies, and that future work must show whether reductions in implicit bias lead to reductions in discriminatory outcomes.
  • Evident from the text: the sample was non-Black introductory psychology students, mostly White and female, limiting generalizability; the study used only one implicit measure, the Black–White IAT.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.jesp.2012.06.003

Cites

72 of 72 references verified0 citation errors · 0 not checkable

Statistics

25 tests recomputed · 1 with the reported and recomputed p on opposite sides of .05

Verification

Claims the paper makes

0 of 45 cited sources checked · 0 supported · 0 flagged

Show all 51 claims

  • The quote only partly states this claim. 3
  • It may state two separate claims as one. 24
  • Probably not a substantial claim. 10
34 of 34

  • The authors state this as an assumption.

    In the Black-White IAT, to the extent that negative valence is associated with Black people and positive valence with White people, compatible trials are easier than incompatible trials.

    “Underlying the race IAT is the assumption that, to the extent that negative valence is associated with Black people (and positive valence with White people), when Black faces and unpleasant words (and White faces and pleasant words) are paired together on the same response key (compatible trials), the task should be easier than with the reverse pairings (incompatible trials).”

  • The paper relies on this without support.

    Extant research has not yet examined whether interventions can produce long-term implicit bias reductions, nor clearly specified the type of effort required to yield such reductions.

    “Although this evidence is consistent with the prejudice habit-breaking framework, extant research has not yet examined whether interventions can produce long-term implicit bias reductions, nor has it clearly specified the type of effort required to yield such reductions.”

  • The paper relies on this without support.

    The IAT D-score is a valid measure of implicit race bias.

    “The intervention had a lasting effect on one measure of implicit race bias.”
    “indicating that people high in concern about discrimination at week 2 retained the reductions in IAT bias 8 weeks after the intervention.”

  • The paper relies on this without support.

    Participants’ free-response descriptions of strategy use accurately reflect their actual use of strategies.

    “Participants’ descriptions of their strategy use were rich and complex.”
    “we used a text mining approach to calculate frequencies of theoretically important word stems in the descriptions.”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Statistics that don’t add up

Each reported p-value recomputed from its test statistic and degrees of freedom.

  • Inconsistent p-value across alpha = .05: t(48) = -2.03, p = .051

    Major

    The paper reports p = .051 here; recomputing from the test statistic and its degrees of freedom gives p = .048 (two-tailed). The two fall on opposite sides of alpha = .05. The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .051 · recomputed p = .048

  • Inconsistent p-value: t(86) = -2.90, p = .009

    Minor

    The paper reports p = .009 here; recomputing from the test statistic and its degrees of freedom gives p = .005 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .009 · recomputed p = .005

  • Inconsistent p-value: t(89) = 2.10, p = .042

    Minor

    The paper reports p = .042 here; recomputing from the test statistic and its degrees of freedom gives p = .039 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .042 · recomputed p = .039

  • Inconsistent p-value: t(50) = -2.40, p = .021

    Minor

    The paper reports p = .021 here; recomputing from the test statistic and its degrees of freedom gives p = .020 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .021 · recomputed p = .020

  • Inconsistent p-value: t(49) = -2.66, p = .012

    Minor

    The paper reports p = .012 here; recomputing from the test statistic and its degrees of freedom gives p = .011 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .012 · recomputed p = .011

  • Inconsistent p-value: t(49) = -2.40, p = .028

    Minor

    The paper reports p = .028 here; recomputing from the test statistic and its degrees of freedom gives p = .020 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.

    Reported p = .028 · recomputed p = .020

6 of 6

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Doesn’t add upt(48) = -2.03reported p = .051 · recomputed p = .048The two fall on opposite sides of .05

    “t(48) = -2.03, p = .051”

  • Doesn’t add upt(89) = 2.1reported p = .042 · recomputed p = .039

    “t(89) = 2.10, p = .042”

  • Doesn’t add upt(86) = -2.9reported p = .009 · recomputed p = .005

    “t(86) = -2.90, p = .009”

  • Doesn’t add upt(50) = -2.4reported p = .021 · recomputed p = .020

    “t(50) = -2.40, p = .021”

  • Doesn’t add upt(49) = -2.4reported p = .028 · recomputed p = .020

    “t(49) = -2.40, p = .028”

  • Doesn’t add upt(49) = -2.66reported p = .012 · recomputed p = .011

    “t(49) = -2.66, p = .012”

  • Consistentt(88) = -2.82reported p = .006 · recomputed p = .006

    “t(88) = -2.82, p = .006”

  • Consistentt(88) = 0.82reported p = .42 · recomputed p = .414

    “t(88) = .82, p = .42”

  • Consistentt(88) = 2.24reported p = .028 · recomputed p = .028

    “t(88) = 2.24, p = .028”

  • Consistentt(89) = -0.22reported p = .83 · recomputed p = .826

    “t(89) = -.22, p = .83”

  • Consistentt(88) = 2.35reported p = .021 · recomputed p = .021

    “t(88) = 2.35, p = .021”

  • Consistentt(88) = -0.6reported p = .55 · recomputed p = .550

    “t(88) = -.60, p = .55”

  • Consistentt(88) = 2.31reported p = .024 · recomputed p = .023

    “t(88) = 2.31, p = .024”

  • Consistentt(88) = -0.93reported p = .36 · recomputed p = .355

    “t(88) = -.93, p = .36”

  • Consistentt(88) = 1.93reported p = .057 · recomputed p = .057

    “t(88) = 1.93, p = .057”

  • Consistentt(88) = -0.67reported p = .5 · recomputed p = .505

    “t(88) = -.67, p = .50”

  • Consistentt(88) = -0.37reported p = .72 · recomputed p = .712

    “t(88) = -.37, p = .72”

  • Consistentt(86) = -2.46reported p = .016 · recomputed p = .016

    “t(86) = -2.46, p = .016”

  • Consistentt(86) = -0.082reported p = .94 · recomputed p = .935

    “t(86) = -.082, p = .94”

  • Consistentt(86) = 0.55reported p = .59 · recomputed p = .584

    “t(86) = .55, p = .59”

  • Consistentt(82) = -2.31reported p = .024 · recomputed p = .023

    “t(82) = -2.31, p = .024”

  • Consistentt(50) = -3.18reported p = .003 · recomputed p = .003

    “t(50) = -3.18, p = .003”

  • Consistentt(50) = -0.69reported p = .5 · recomputed p = .493

    “t(50) = -.69, p = .50”

  • Consistentt(50) = -1.86reported p = .069 · recomputed p = .069

    “t(50) = -1.86, p = .069”

  • Consistentt(50) = -2.47reported p = .017 · recomputed p = .017

    “t(50) = -2.47, p = .017”

25 of 25

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The number of strategies changes from five to six

    Minor

    The training section says participants were taught five strategies, but the strategy-measures section says responses were averaged across six techniques.

  • Discrepancies are attributed to the wrong assessment weeks

    Minor

    The methods place the post-intervention explicit assessments, which include prejudice-relevant discrepancies, at weeks 2 and 6. The results instead describe the discrepancy effect at weeks 4 and 8, when the IAT was assessed.

  • Baseline racial-attitudes mean exceeds the scale maximum

    Minor

    The Attitudes Towards Blacks items are scored from 1 to 7 and averaged, so the reported baseline mean of 7.57 is impossible on the stated scale.

3 of 3

Citations

Cites

Flagged first, then by citations

Show all 69 references

Cited by

Sorted by citations

Items 701–800 of 1,000

Related

See all

Metadata

Authors
Devine, P. G., Forscher, P. S., Austin, A. J., & Cox, W. T. L.
Affiliations
University of Wisconsin–Madison (US)
Corresponding author
Devine, Patricia G.
ORCID
Venue
Journal of Experimental Social Psychology
Publisher
Elsevier BV
Published
2012-11
Volume
48
Issue
6
Pages
1267-1278
ISSN
0022-1031 (ISSN-L)
DOI
10.1016/j.jesp.2012.06.003
PMC
PMC3603687
OpenAlex
W2025640120
Semantic Scholar
6d2d8282627bfcaf933e3d68d3e9c170b75f2f1f
Cited by
1,259
References
69
Citations by year
2026: 82 · 2025: 140 · 2024: 111 · 2023: 104 · 2022: 136 · 2021: 127 · 2020: 155 · 2019: 112 · 2018: 90 · 2017: 70 · 2016: 53 · 2015: 37 · 2014: 26 · 2013: 13 · 2012: 1
FWCI
81.23
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Experimental Social Psychology (publishedVersion)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (acceptedVersion)
  • PubMed Central (acceptedVersion)
  • Journal of the Arkansas Academy of Science (submittedVersion, cc-by-nd)
Funders
National Institutes of Health (R01 GM088477)
Topics
  • Social Sciences › Sociology and Political Science › Social and Intergroup Psychology
  • Social Sciences › Sociology and Political Science › Racial and Ethnic Identity Research
  • Social Sciences › Sociology and Political Science › Names, Identity, and Discrimination Research
Keywords
prejudice reduction, unintentional discrimination, long-term intervention effects

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review