Neural signals for the
detection of unintended race bias
David
M. Amodio, Eddie Harmon-Jones, Patricia G. Devine, John J. Curtin, Sigan L.
Hartley, and Alison E. Covert
University
of Wisconsin – Madison
Interaction: F(1, 33) = 14.39, p < .001
Race of prime
Interaction: F(1, 33) = 10.84, p < .005
Race of prime
Time
(ms)
Error trials: t(33) = 2.94, p < .01
Time
(ms)
Error trials: t(33) = .97, p = .33
ØEvaluation system
•Continuously monitors ongoing neural activity for conflict between behavioral tendencies
ØAssociated with activity in
anterior cingulate cortex (ACC)
•When conflict is detected, second system is signaled
ØRegulatory system
•Organizes behavior to resolve conflict
ØAssociated with activity in
prefrontal cortex
Pattern Mask
Black or White Face Prime
Gun or Tool Target
Pattern Mask
Time
1s
200 ms
200 ms
response
From W. Gehring: www.personal.umich.edu/
~wgehring/lab/Learn.html
*Adapted
from Payne, 2001
*
Error-related negativity (ERN)
•Occurs concurrent with
error response
•Fronto-central scalp distribution
•High temporal
resolution
The dual-system model suggests two explanations for why prejudice control
fails:
1)Conflict
detection system not activated sufficiently
•Conflict between automatic race bias and intention to
respond without prejudice not detected
2)Regulatory
system not activated sufficiently
•Conflict is detected, but second system fails to regulate behavior
lPresent Study:
lIs the conflict
detection system sensitive to the potential for
a race-biased response?
ØExamined activity of conflict-detection process
associated with participants’ race-biased
responses
ØConflict detection was measured using error-related
negativity (ERN) component of the event-related
potential
Participant
instructions:
•“Task designed to measure racial prejudice”
•“Errors on certain trials attributed to race bias”
•Responding “gun” instead of “tool” after a Black face suggested influence of stereotype
•«Errors on Black-tool trials critical«
•Participants
categorized each target by pressing a computer keyboard button labeled “gun” or “tool”
•Responses were to be made within 500 ms of target
ØIncreased error rate
to facilitate examination of unintended race-biased responses
•Black face primes were designed to activate “violent”
stereotype
ØBlack face should facilitate “gun” responses and cause
conflict for “tool” responses
•A) Gun-tool task created race-biased response
conflict
•Black face primes
facilitated “gun” responses and inhibited “tool” responses
•After seeing a Black
face, participants were most likely to make stereotype-consistent errors (e.g., press “gun” when
target was “tool”)
•B) Greater conflict detection for race-biased
responses
•ERNs were largest for errors attributable to race bias (Black-tool errors), compared to ERNs for all other error types
•C. Race-biased ERNs predicted greater control
lDiscussion
ØUnintentional race bias not due to lack of
detection
•Errors attributable to race bias were associated with
larger ERNs than other errors
•Unintentional race bias most likely associated with
failure of PFC-related system to regulate
behavior
ØRecruitment of race-bias control begins very early in response stream and does not require awareness
•Suggests revision of predominant models of mental
correction, e.g., Wegener & Petty
(1997), Wilson & Brekke (1994)
ØGreater sensitivity to the potential for race-bias
predicted more controlled behavior throughout
task
•Suggests individuals more sensitive to race-biased
response conflict are more adept at
regulating race-biased behaviors
•Increasing one’s regulatory ability may require
enhancing one’s implicit sensitivity to
race-biased conflict detection
Method
•Participants
•34 White American students
•Procedure
•Completed 288 trials of gun-tool task
•EEG: 27 scalp sites, average earlobe reference
•ERN derivation
•1-15 Hz signal at frontocentral midline (Fcz)
•Averaged across error responses within each trial
Dual system model of control
(Botvinick, Braver,
Carter, Barch, & Cohen, 2001)
Stereotypes of Blacks
are so deeply imbedded in American culture that
they may be activated automatically (Devine, 1989). Once activated, racial
stereotypes can lead to unintentional discriminatory
behaviors (Dovidio, Kawakami, & Gaertner, 2002). Indeed,
many self-avowed egalitarians report that prejudices often slip through in their behavior,
despite their non-prejudiced intentions
(Devine, Monteith, Zuwerink, & Elliot, 1991; Monteith, 1993).
Although the conditions precluding control have been studied, previous research has not
examined the process underlying failures to control expressions of
prejudice.
Results
ØWhy does prejudice control sometimes fail?
•Has the mind not detected that race bias is present?
•Is the mind aware of the bias, but unable to inhibit prejudiced behavior?
To address these questions, we applied a neural model of cognitive control to the context of race bias:
r(32) = -.44, p = .01
r(32) = -.48, p < .005
•Larger race-biased ERNs (negatively valenced) predicted greater slowing of responses following errors
ØTo examine the behavioral effects of ERN
amplitude associated with race-bias–detection,
a “race-bias ERN” was computed, representing the ERN to Black-tool
errors with White-tool errors covaried
•Larger race-biased ERNs (negatively valenced) predicted greater post-error accuracy on “tool” trials but not “gun” trials
