Implicit Bias and the Illusion of Conscious Ill Will

Article
Social Psychological and
Personality Science
2014, Vol. 5(4) 500-507
ª The Author(s) 2013
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/1948550613506123
spps.sagepub.com
Implicit Bias and the Illusion
of Conscious Ill Will
Erin Cooley1, B. Keith Payne1, and K. Jean Phillips1
Abstract
Implicit bias is defined, in part, by a lack of intent. Yet the implicit attitudes literature has made little contact with research on the
experience of conscious will, which suggests that the feeling of conscious intent is an inference rather than a direct report of how
actions are caused. We tested the hypothesis that inferences about one’s intentions shape whether an automatically activated
attitude is endorsed explicitly. In a first study, individuals who perceived their attitudes toward gay men to be intended showed
stronger implicit–explicit correspondence. In a second study, we manipulated perceptions of intent. Inferences that implicit bias
was intended caused participants to express those biases on an explicit measure. A third study replicated the experimental effects
and found that metacognitions of intent were especially influential among individuals who were motivated to be unprejudiced.
Results suggest that metacognitive inferences about intent can shape whether automatically activated bias becomes explicitly
endorsed prejudice.
Keywords
implicit bias, implicit attitudes, homophobia, illusion of conscious will, automatic, controlled processing
Implicit bias is by definition unintended. Although explicit attitudes are deliberately expressed, implicit attitudes may come to
mind and affect behavior even despite intentions to the contrary. This is one reason that many people find research on
implicit bias to be surprising, even disturbing. It suggests that
people who intend to be fair might nonetheless discriminate.
We argue that it is important to distinguish between the causal influence of intentions and the subjective experience of
intentions. The causal influence of intentions can be studied
by experimentally controlling or manipulating intentions. Most
implicit measures of attitudes pit intentional response strategies
against unintended influences of attitudes. In priming tasks, for
example, participants are instructed to make judgments about
target items, but evaluations of prime stimuli may unintentionally influence the speed or content of responses. In the affect
misattribution procedure (AMP; Payne, Cheng, Govorun, &
Stewart, 2005) used here, participants are asked to rate the pleasantness of ambiguous target symbols, and they are explicitly
warned not to let their feelings toward the prime photos influence their ratings. Effects of the primes that persist despite
intentions to the contrary are taken as evidence of automatic
(i.e., unintentional) evaluations.
When completing an implicit test, respondents are sometimes surprised at their own performance, suggesting that their
behaviors may have diverged from their intentions. Yet, subjective experiences of intention do not necessarily track the real
causal influences of intent. Several theoretical perspectives
have argued that the feeling of consciously intending an action
or thought can be a construction (e.g., Gazzaniga, 1983, 1988;
Libet, 1985; Libet, Gleason, Wright, & Pearl, 1983). Wegner’s
theory of apparent mental causation provides an account of
how this may happen (Wegner, 2002; Wegner & Wheatley,
1999). The theory argues that unconscious brain processes
cause both action and conscious thoughts about the action. People observe this correlation and infer that their conscious
thoughts caused the action. In this view, inferring causality in
our own actions operates according to the same principles by
which people infer causality in objects such as one billiard ball
striking another. Thoughts are most likely to be interpreted as
causes if they occur prior to the act, are consistent with the act,
and are the only apparent cause.
Intentions are sometimes constructed after the action has
been completed. For example, Kühn and Brass (2009) reported
experimental evidence for post hoc confabulations of intent.
They found that after participants failed to inhibit a habitual
response in a stop-signal task, they often claimed that they
intended to respond as they did (even though response time
data suggested that the response was an impulsive error).
According to constructivist theories, the conscious experience
1
University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
Corresponding Author:
B. Keith Payne, University of North Carolina at Chapel Hill, Campus Box #
3270, Chapel Hill, NC 27599, USA.
Email: [email protected]
Cooley et al.
of intention is a matter of inferences that sometimes are confabulations generated after the fact (Kelley & Jacoby, 1990).
501
Study 1
Method
Metacognitions About Implicit Attitudes
Metacognitive models of attitudes locate the difference
between implicit and explicit cognition in the thoughts that
people have about their automatic evaluations. The Associative
Propositional Model of Evaluation (APE; Gawronski & Bodenhausen, 2006) claims that automatically activated associations
provide input into a propositional reasoning process in which
people accept the activated information as valid or reject it as
invalid. Similarly, the Metacognitive Model (MCM; Petty, Briñol, & DeMarree, 2007) distinguishes between evaluative associations that are activated from memory and validity tags that
are applied to them. The APE and MCM focus on metacognitive inferences specifically about whether the attitude is valid
(i.e., true, accurate, or justified). We suggest that the basic principle can be extended to encompass a variety of other distinct
inferences, including inferences about whether an evaluation
is intentional.
A useful framework for integrating these different kinds
of inferences is the Situated Inference Model of Priming
(Loersch & Payne, 2011). This model posits that priming
simply increases the accessibility of information that is
semantically or affectively related to the prime. Primed
information is then used to guide responses when people
misattribute the primed information as their own spontaneous thoughts or feelings. The Situated Inference Model
highlights that the kinds of metacognitive inferences people
make are flexibly determined by both the associations automatically activated by primes and the contextual demands of
the situation. As applied to implicit evaluations, the Situated
Inference Model suggests that people might make a wide
variety of inferences about what an automatic evaluation
means, including whether it was intentional.
According to constructivist theories, if participants infer
that they intended their automatic thoughts then they should
become—from the participants’ perspective—intentional
thoughts. This idea has striking implications for implicit
prejudice because it means that post hoc inferences of intent
may turn implicit bias into explicit prejudice. Because past
research has found moderate implicit–explicit correlations
and substantial variability on both implicit and explicit measures in attitudes toward gay men, we decided to examine
attitudes toward homosexuality across three experiments
(Banse, Seise, & Zerbes, 2001; Jellison, McConnel, &
Gabriel, 2004).
First, we assessed automatic evaluations of gay male couples and then measured (Study 1) and manipulated (Studies 2
and 3) inferences about whether those evaluations were intentional. We then measured explicit attitudes toward gay men to
test whether participants were more likely to explicitly endorse
antigay bias when they perceived their (implicit) bias to be
intentional.
Undergraduates (N ¼ 122, 66% female) completed the study
for course credit. They first completed an AMP (Payne et al.,
2005) to measure implicit attitudes toward gay male couples.
The AMP measures attitudes by presenting on each trial a
prime photo (75 ms) followed by a blank interval (125 ms) and
then a Chinese pictograph (100 ms) and finally a pattern mask
that remains until a response is made. Participants are
instructed to evaluate the pictographs as pleasant or unpleasant
without influence from the primes. Attitudes toward the primes
bias pictograph ratings and are measured by the proportion of
pleasant judgments following each prime type. Primes included
photos of male–male couples and male–female couples (as
controls) in affectionate poses. Following the AMP, participants were asked to reflect on the immediate feelings they
experienced toward the pictures of same-sex couples. They
were asked to rate on a 7-point scale whether their feelings during the implicit task were intentional on 3 items (‘‘My feelings
toward the photos of homosexuals were intentional,’’ ‘‘I did not
intend my feelings toward homosexuals,’’ and, ‘‘I intended my
immediate feelings toward homosexuals’’). Finally, they completed the 22-item Modern Homophobia scale (Raja & Stokes,
1998) to measure explicit bias against gay men (e.g., ‘‘I am
tired of hearing about gay men’s problems; I would remove
my child from class if I found out the teacher was gay.’’) and
reported their sex and sexual orientation along with other
demographics.
Results and Discussion
Preliminary Analyses
Seven participants were omitted from the analyses using the
AMP because four could read Chinese characters and three
pressed the same key on all trials, resulting in a final sample
of 115. Overall, participants indicated implicit bias toward gay
men. The proportion of pleasant responses to male–male
primes (M ¼ .44, standard deviation [SD] ¼ .27) was significantly less than the proportion of pleasant responses to male–
female primes (M ¼ .80, SD ¼ .19), F(1, 114) ¼ 115.80, p <
.001. Implicit bias scores were computed as the difference
between pleasant ratings on male–female trials when compared
to male–male trials. These scores were highly reliable (Cronbach’s a ¼ .93). Explicit bias scores were computed by averaging scores across the Modern Homophobia scale (a ¼ .95).
Perceived intent for implicit bias was measured with the sum
of all three intent items (a ¼ .83). Both the sex (male ¼ 1;
female ¼ 2) and the sexual orientation of the participant
(straight ¼ 1; nonstraight ¼ 2) were correlated with implicit
attitudes (r ¼ .21, p ¼ .03; r ¼ .31, p ¼ .001)
and explicit attitudes (r ¼ .16, p ¼ .10; r ¼ .27, p ¼
.004). As a result, we controlled for these demographic variables in the following analyses.
502
Figure 1. Adjusted means for explicit bias plotted against implicit bias,
by measured levels of perceived intent. All variables are standardized,
and implicit bias is plotted at 1 standard deviation above and below the
mean, Study 1.
Social Psychological and Personality Science 5(4)
measure of intent was actually capturing thoughts about validity. Other metacognitive models suggest that the primary factor
distinguishing implicit and explicit evaluations is perceived
validity (Gawronski & Bodenhausen, 2006; Petty et al.,
2007). Gawronski, Peters, and Lebel (2008) suggested that
attributions of validity may be an important basis for making
other metacognitive inferences such as ownership or authorship. From this perspective, one would expect perceptions of
intent and validity to be associated. It is therefore important
to determine whether perceptions of intent have effects independent of validity.
Study 2 addresses these issues by manipulating perceived
intentionality of implicit evaluations and subsequently measuring both perceived intent and perceived validity of implicit evaluations. By manipulating perceived intent following the
measurement of implicit bias, we can isolate the effects of post
hoc constructions. By measuring both perceived validity and
perceived intent we can examine whether our manipulation of
intent is affecting metacognitions about the intent behind implicit responses independent of metacognitions about validity.
Study 2
Main Analyses
To examine the effects of perceived intent on the association
between implicit and explicit attitudes toward gay men, we conducted a regression analysis predicting explicit bias scores from
perceived intent, implicit bias scores, and their interaction. All
variables were standardized using z-scores prior to analysis. The
main effect of implicit bias was significant, b ¼ .57, t ¼ 6.88, p <
.001. Importantly, the predicted interaction was also significant,
indicating that perceived intentionality of implicit responses
moderated the relationship between implicit and explicit attitudes
toward gay men, b ¼ .17, t ¼ 2.09, p ¼ .04.1
Figure 1 displays the simple slopes associating implicit bias
scores with explicit homophobia at high and low levels of perceived intention. Analyses of simple slopes indicated that the
association between implicit and explicit attitudes was significant at high levels of perceived intent (1 SD above the mean),
b ¼ .71, t ¼ 6.75, p < .001, and at low levels of perceived intent
(1 SD below the mean), b ¼ .39, t ¼ 3.29, p ¼ .001, although
the size of the association was much smaller for participants
with low ratings of intent.
We next examined the effects of perceived intent separately for individuals with high and low levels of implicit
bias (+1 SD from the mean). Among participants with high
levels of implicit bias, greater perceived intent was significantly associated with greater levels of explicit homophobia,
b ¼ .24, t ¼ 2.08, p ¼ .04. Among those with low levels of
implicit bias perceived intent was not associated with levels
of explicit homophobia, b ¼ .10, t ¼ .84, p ¼ .40.
These results suggest that perceived intent moderates
whether implicit bias is endorsed explicitly or denied. However, because intent was measured and not manipulated, ratings
of intent may have reflected either post hoc constructions or
preexisting intentions. Furthermore, it is possible that our
Method
Undergraduates (N ¼ 156, 70% female) completed the study
for course credit. They first completed the same AMP as in
Study 1 and then were randomly assigned to the intentional
or unintentional groups. Participants in the intentional group
were asked to think about two to three reasons why they may
have intended to have the feelings they experienced during the
AMP. Those in the unintentional group generated reasons why
the feelings may have been unintended. This manipulation
draws on the confirmation bias, in which searching for evidence in favor of a hypothesis increases belief in it (Nickerson,
1998). Next, participants rated whether their feelings during the
implicit task were intentional as in Study 1.
Importantly, in Study 2, participants also rated the perceived
validity of their feelings during the implicit task. To do this, they
reported on a 7-point scale whether their feelings during the implicit task were valid (‘‘My immediate feelings toward homosexuals
are valid,’’ ‘‘My feelings reflect accurate beliefs about homosexuals,’’ and ‘‘My feelings do NOT necessarily reflect true beliefs
about homosexuals’’). Finally, they completed the Modern
Homophobia scale and reported demographic variables.
Results and Discussion
Preliminary Analyses
Nine participants were removed from analyses because seven
could read Chinese characters and two pressed the same key
across all trials, resulting in a final sample of 147. Implicit
prejudice (a ¼ .91) and explicit prejudice (a ¼ .95) were
calculated in the same way as in Study 1. Perceived intent
(a ¼ .79) and perceived validity (a ¼ .77) were calculated
Cooley et al.
503
by summing across the 3 items that composed each measure. As in
Study 1, participants displayed implicit prejudice with the proportion of pleasant responses to male–male primes (M ¼ 0.49, SD ¼
0.26) significantly lower than the proportion of pleasant responses
to male–female primes (M ¼ 0.71, SD ¼ 0.20), F(1, 146) ¼ 59.38,
p < .001. Also replicating Study 1, both the sex and the sexual
orientation of the participant were correlated with both implicit
attitudes (r ¼ .17, p ¼ .04; r ¼ .25, p ¼ .003) and explicit attitudes (r ¼ .19, p ¼ .02; r ¼ .22, p ¼ .009). We controlled for
these demographic variables in the following analyses.
Manipulation Check
As expected, participants in the intentional condition reported
greater levels of intent than those in the unintentional condition,
b ¼ .24, t ¼ 3.02, p ¼ .003. Also, as expected on the basis of prior
theory, the manipulation affected perceptions of the validity of
implicit attitudes, b ¼ .24, t ¼ 2.93, p ¼ .004. We therefore examined the effects of both perceived intent and validity in the main
analyses reported below.
Main Analyses
A regression analysis using standardized variables tested the
effects of implicit bias (as a difference score), the intent manipulation, and their interaction, on explicit prejudice. There was a
main effect for both implicit bias and the manipulation of
intent. Higher levels of implicit bias were associated with
increased levels of explicit homophobia, b ¼ .63, t ¼ 10.03,
p < .001, and those in the intentional condition reported greater
levels of explicit bias as compared to those in the unintentional
condition, b ¼ .18, t ¼ 2.93, p ¼ .004.
Importantly, the predicted interaction was also significant, b ¼
.13, t ¼ 2.15, p ¼ .033 (see Figure 2).2 Simple slopes analyses
indicated that implicit bias was more strongly associated with
explicit bias in the intentional group, b ¼ .80, t ¼ 8.05, p <
.001, than in the unintentional group, b ¼ .52, t ¼ 6.68, p <
.001. The intentional group reported greater explicit prejudice
than the unintentional group when implicit bias was high (1 SD
above the mean), b ¼ .31, t ¼ 3.40, p < .001, but not when implicit
bias was low (1 SD below the mean), b ¼ .05, t ¼ .54, p ¼ .60. The
greatest explicit prejudice was thus expressed among participants
with high implicit bias who were led to infer that they intended
those thoughts and feelings.
Because the manipulation influenced not only perceived
intent but also perceived validity, we examined whether our
effects were mediated by perceived intent independent of validity. We tested simultaneous moderated mediation by perceived
intent and validity using Hayes’ (2013) PROCESS macro using
1,000 bootstrapped samples (Figure 3).
We found a significant indirect effect of condition on explicit attitudes via perceived validity among participants with
high implicit bias, b ¼ .06 (95% confidence interval [0.01,
0.15]) but not low in implicit bias, b ¼ .04 (95% confidence
interval [0.10, 0.001]). Critical to our hypotheses, there was
also a significant indirect effect of condition on explicit
Figure 2. Adjusted means for explicit bias plotted against implicit bias,
by intentionality inference condition. All variables are standardized,
and implicit bias is plotted at 1 standard deviation above and below the
mean, Study 2.
attitudes via perceived intent among participants high in implicit bias, b ¼ .06 (95% confidence interval [0.01, 0.13]), but not
low in implicit bias, b ¼ .01 (95% confidence interval [0.03,
0.04]). Thus, although the manipulation of perceived intent
also affected perceptions of validity, perceived intent had its
predicted effects independent of these effects on validity.
Study 3
In our first two studies, we showed that when participants perceived that automatic evaluations were intentional, they were
more likely to endorse similar evaluations explicitly. In our
final study, we replicated the manipulation of perceived intent,
while further clarifying the nature of these effects and ruling
out an alternative explanation based on changes in attitude
strength rather than perceived intent.
In this study, we used a different measure of explicit attitudes. Although the Modern Homophobia scale used in our previous studies is a reliable and validated scale, it measures a
mixture of behavioral intentions, beliefs, and evaluations. This
is a strength insofar as the complex items may provide participants with justifications for endorsing the items if they are
inclined to do so. Such complex statements have also been
argued to be ecologically valid, because prejudice in daily life
is often couched in justifications (Sears & Henry, 2005). However, the complex items are a limitation for theoretical precision because it is not clear which components of the attitude
(or justifying cognitions) may be driving the observed effects.
Therefore, in Study 3 we replicated our experimental effects
using simpler evaluations, such as directly asking participants
the extent to which they dislike gay men.
504
Social Psychological and Personality Science 5(4)
Perceived Intent
Indirect effect at +1 SD implicit bias
Indirect effect at -1 SD implicit bias
.24**
Perceived
Intent
.06*
.006, ns.
.13*
.11┼
.13*
Intention
Manipulation
.24**
Implicit
Bias
Explicit
Bias
Perceived
Validity
Perceived Validity
Indirect effect at +1 SD implicit bias
Indirect effect at -1 SD implicit bias
.05
.20**
Implicit
Bias
.06*
-.04, ns.
Figure 3. Perceived intent and perceived validity as mediators of the relationship between the manipulation of intent and implicit–explicit
attitude correspondence. All coefficients are unstandardized regression coefficients. All variables were standardized before analysis. yp < .10, *p
< .05, **p < .01, Study 2.
We also expanded the explicit measures in this study by
including a measure of internal motivations to control prejudice
toward gay men. Such motivations might be expected to influence the endorsement of prejudice (especially on the more
straightforward items we used to measure explicit prejudice
in Experiment 3) in two alternative ways. In studies of racial
prejudice, where many participants are highly motivated to
control prejudice, explicit prejudice levels are typically very
low and there is often little variability in explicit attitudes. If
attitudes toward gay men follow this pattern of ‘‘floor effects,’’
then we would expect to observe the effects of perceived intent
only among participants who are low in motivation to control
prejudice because only these participants would display substantial variability in explicit attitudes. In contrast to racial prejudice, however, studies of attitudes toward gay men tend to
find less evidence of strategic efforts to conceal bias (Jellison
et al., 2004). When the overall levels of explicit prejudice are
high, we might expect to see more variability among participants high in motivation to control prejudice. Participants who
are generally high in motivation to be unbiased may be the ones
most likely to use the feeling of intent as a cue that they should
explicitly endorse their intuitive feelings; conversely, they may
use the lack of perceived intent as a welcome cue that they
should not endorse such feelings.
Method
Participants
Amazon Mechanical Turk (Mturk) workers completed an
online survey in exchange for US$.50 (N ¼ 129, 64% female).
After indicating electronic agreement to an informed consent,
participants were warned that there would be an attention check
within the survey. Previous research indicates that attention
checks can improve data quality (Oppenheimer, Meyvis, &
Dayidenko, 2009).
Procedure and Materials
Participants then completed the AMP and then were randomly
assigned to the intentional or the unintentional group identical
to Study 2. Next, participants completed our explicit measure
of attitudes toward homosexuality. Of 9 items, 3 asked participants to report the degree to which they dislike, approve, or feel
negatively toward gay men on 1 (not at all) to 4 (very much)
scales. The other 6 items were unipolar semantic differential
scales with 7 points. These items asked participants the degree
to which they were approving of gay men, supportive of gay
men, and whether they perceive gay men to be disgusting,
offensive, immoral, or bad. For all items assessing explicit prejudice, we also recorded response times as a measure of attitude
accessibility. Next, participants responded to a 2-item measure
of attitude certainty (Tormala & Petty, 2002) and completed
Plant and Devine’s (1998) measure of internal motivations to
control prejudice (MCP) adapted for attitudes toward gay men.
Participants then reported their demographics.
Finally, before debriefing, all participants were given the
attention check. The attention check provided participants with
a series of statements (e.g., ‘‘this study was confusing’’) that
were placed next to 1 (strongly disagree) to 5(strongly agree)
scales. As a test of attention, the instructions at the top of the
Cooley et al.
page asked participants to ignore the items on the page and to
give an answer of strongly disagree to all statements.
Results and Discussion
Preliminary Analyses
In all, 23 participants failed the attention check and 7 participants were removed from analyses because of data recording
errors while taking the AMP over the Internet. As in the previous two experiments, we also eliminated 6 participants who
reported being able to read Chinese characters and 11 participants who pressed the same key across all trials in the AMP.
Our final sample was 82 participants.
Preliminary analyses indicated that the explicit measure was
highly reliable (a ¼ .97) as was the measure of MCP (a ¼ .89)
and the measure of attitude certainty (r ¼ .81). Additionally, as
in the previous experiments, participants displayed implicit
bias (a ¼ .87) such that the proportion of pleasant responses
to male–male primes (M ¼ 0.53, SD ¼ 0.25) was significantly
lower than the proportion of pleasant responses to male–female
primes, (M ¼ 0.65, SD ¼0.20), F(1, 81) ¼ 12.68, p ¼ .001. In
this sample, sexual orientation was significantly correlated
with implicit prejudice (r ¼ .29, p ¼ .009) and explicit prejudice (r ¼ .25, p ¼ .02); however, sex was not (r ¼ .10,
p ¼ .38; r ¼ .08, p ¼ .46). We controlled for both variables
to maintain consistency with models in Experiments 1 and 2.
Results were the same if we only controlled for sexual
orientation.
Main Analyses
Because MCP was measured after our manipulation, we first
tested whether our manipulation affected this variable. As
anticipated, MCP was not significantly affected by our manipulation, b ¼ .16, t ¼ 1.36, p ¼ .18. To test the main hypothesis, we ran a regression analysis using standardized variables
to test the effect of implicit bias, the intent manipulation, MCP,
and their interactions, on explicit bias. Main effects and controls were entered on the first step, followed by all two-way
interactions on a second step, and the three-way interaction
on the third step. There were significant main effects of implicit
bias and MCP. Higher levels of implicit bias were associated
with greater explicit bias, b ¼ .19, t ¼ 2.37, p ¼ .02, and higher
levels of MCP were associated with less explicit bias,
b ¼ .69, t ¼ 9.30, p < .001.
No two-way interactions were significant. However, the threeway interaction of implicit bias, the intent manipulation, and
MCP was significant, b ¼ .18, t ¼ 2.76, p ¼ .007.3 To probe this
three-way interaction, we tested the Implicit Bias Intent Manipulation interaction separately for those who were high in MCP
(þ1 SD) and for those who were low in MCP (1 SD). For those
low in motivations to control prejudice, the Implicit Bias Intent
Manipulation interaction was not significant, b ¼ .08, t ¼ .93,
p ¼ .36. Instead, there was simply a main effect of implicit bias
such that those with higher implicit bias also expressed greater
explicit prejudice, b ¼ .18, t ¼ 2.11, p ¼ .04.
505
Among those who were high in motivation to control prejudice, however, the Implicit Bias Intent Manipulation interaction was significant, b ¼ .31, t ¼ 2.30, p ¼ .03. We
decomposed the interaction by examining the implicit–explicit
association separately for the intentional and unintentional
groups. Implicit bias significantly predicted explicit attitudes
in the intentional group, b ¼ .60, t ¼ 3.07, p ¼ .003, but not
in the unintentional group, b ¼ .01, t ¼ .04, p ¼ .97. Thus,
the overall implicit and explicit attitudes were positively associated; the only case in which they were unrelated was among
highly motivated participants in the unintentional condition.
We also decomposed the interaction (among high-MCP participants) by testing the effects of the manipulation separately
for subjects high and low in implicit bias. When implicit bias
was high, the intentional group reported marginally more explicit prejudice than the unintentional group, b ¼ .35, t ¼ 1.89,
p ¼ .06. When implicit bias was low, the intentional group
reported marginally less explicit prejudice than the unintentional group, b ¼ .27, t ¼ 1.68, p ¼ .10.
In sum, the pattern of results from Study 2 replicated in
Study 3, but primarily among those high in motivation to control prejudice. Those low in motivation showed high explicit
prejudice overall and it was associated with individual differences in implicit bias. For those high in motivation, whether
implicit bias was endorsed explicitly depended on metacognitions of intent. Inferences of intent led people high in MCP
to endorse their implicit biases on an explicit measure.
We also investigated the alternative hypothesis that our
manipulation increased implicit–explicit attitude correspondence by increasing attitude strength rather than perceived
intent. Results indicated that our manipulation of intent had no
effect on attitude certainty, b ¼ .06, t ¼ .52, p ¼ .61. Thus,
attitude certainty is unlikely to explain our results. However, attitudes on the explicit measure were significantly more accessible
(expressed more quickly) in the intentional than in the unintentional group, b ¼ .23, t ¼ 2.11, p ¼ .04. However, accessibility
was unrelated to explicit attitudes, b ¼ .08, t ¼ .67, p ¼ .51 and
accessibility did not moderate the association between implicit
and explicit attitudes, b ¼ .18, t ¼ .73, p ¼ .47. Moreover, the
three-way interaction of Implicit Bias Intent Manipulation MCP remained significant even when controlling for both certainty and accessibility, b ¼ .18, t ¼ 2.69, p ¼ .009. In sum,
although the manipulation of metacognitions of intent affected
attitude accessibility, controlling for attitude accessibility and
attitude certainty did not change our results.
General Discussion
When participants construed their automatic evaluations to be
intentional, these evaluations indeed took on defining properties of explicit attitudes. Participants rated them as more intentional and became more willing to express those evaluations on
a self-report measure. By manipulating perceptions following
the implicit measure, we demonstrated that post hoc confabulations were sufficient to produce these changes in experienced
intent while eliminating any causal effects of intention.
506
The results are consistent with prior research, suggesting
that the subjective feeling of will does not always correspond
to its causal influence (Wegner, 2002). Participants’ subjective
perceptions of intentionality were malleable, leaving participants in no position to report accurately whether their evaluations were intended or unintended. Yet, these malleable
inferences were not simply passive observations. Perceived
intentions had downstream consequences for later behavior,
including explicitly endorsed prejudice.
Although each experiment has limitations, we attempted to
address the main limitations and rule out plausible alternative
explanations across the three studies using a variety of measures.
Study 1 measured but did not manipulate intent. This design limited experimental demand but was unable to differentiate
whether metacognitions of intent had their effect before or after
implicit bias was expressed. This weakness was addressed using
experimental manipulations in Experiments 2 and 3. Study2
showed that the effects of perceived intent were independent
of validity, and Study3 showed that the effects of perceived
intent were independent of attitude certainty and accessibility.
Our findings are consistent with dual process theories that
emphasize metacognitive inferences about implicit attitudes
(Gawronski & Bodenhausem, 2006; Petty et al., 2007). Those
theories have tended to emphasize metacognitions about the
validity of implicit responses, but the present data highlight that
a variety of other metacognitive inferences besides validity
may also be important (Loersch & Payne, 2011). Inferences
about validity essentially ask, ‘‘is my attitude correct?’’ We
have focused here on inferences of intent but other kinds of
inferences are also likely to be important. For example, people
might make judgments about whether they like or dislike their
automatic responses (e.g., ‘‘I feel negative affect toward gay
men but I wish I didn’t’’). Similar post hoc inferences might
also be made regarding whether the ideas that have come automatically to mind are one’s own attitudes versus cultural
knowledge (Gawronski, Peters, & Lebel, 2008).
A great deal of research has shown that people sometimes
confabulate intentions to explain unintended behavior. This
report extends those findings to the operation of implicit bias
with novel implications for stereotyping and discrimination
(see also Payne et al., 2013). Once activated, automatic biases
may be perceived as intentional, turning implicit bias into
explicit prejudice through an illusion of conscious ill will.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to
the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for
the research and/or authorship of this article: NSF grant 0924252.
Notes
1. Without controlling for sexual orientation and sex, results were in
the same direction and marginally significant, b ¼ .10, t ¼ 1.75,
p ¼ .08.
Social Psychological and Personality Science 5(4)
2. This interaction remains significant without controlling for sexual
orientation and sex, b ¼ .14, t ¼ 2.28, p ¼ .024.
3. The three-way interaction remains significant if we include those
who failed the attention check, b ¼ .138, t ¼ 2.58, p ¼ .012, or
if we do not control for sexual orientation and sex, b ¼ .14,
t ¼ 2.22, p ¼ .03.
References
Banse, R., Seise, J., & Zerbes, N. (2001). Implicit attitudes towards
homosexuality: Reliability, validity, and controllability of the IAT.
Zeitschrift für experimentelle Psychologie, 48, 145–160.
Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit
and explicit attitude change. Psychological Bulletin, 132, 692–731.
Gawronski, B., Peters, K. R., & LeBel, E. P. (2008). What makes mental associations personal or extra-personal? Conceptual issues in
the methodological debate about implicit attitude measures. Social
and Personality Psychology Compass, 2, 1002–1023.
Gazzaniga, M. S. (1983). Right hemisphere language following brain
bisection: A 20-year perspective. American Psychologist, 38,
525–537.
Gazzaniga, M. S. (1988). Brain modularity: Towards a philosophy of
conscious experience. In A. J. Marcel & E. Bisiach (Eds.), Consciousness in contemporary science (pp. 218–238). New York,
NY: Clarendon Press/Oxford University Press.
Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New
York, NY: Guilford Press.
Jellison, W. A., McConnell, A. R., & Gabriel, S. (2004). Implicit and
explicit measures of sexual orientation attitudes: Ingroup preferences and related behaviors and beliefs among gay and straight
men. Personality and Social Psychology Bulletin, 30, 629–642.
Kelley, C. M., & Jacoby, L. L. (1990). The construction of subjective
experience: Memory attributions. Mind & Language, 5, 49–68.
Kühn, S., & Brass, M. (2009). Retrospective construction of the judgement of free choice. Consciousness and Cognition, 18, 12–21.
Libet, B. (1985). Unconscious cerebral initiative and the role of conscious will in voluntary action. The Behavioral and Brain Sciences,
8, 529–566.
Libet, B., Gleason, C. A., Wright, E. W., & Pearl, D. K. (1983). Time
of conscious intention to act in relation to onset of cerebral activity
(readiness-potential): The unconscious initiation of a freely voluntary act. Brain, 106, 623–642.
Loersch, C., & Payne, B. K. (2011). The situated inference model: An
integrative account of the effects of primes on perception, behavior, and motivation. Perspectives on Psychological Science, 6,
234–252.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2, 175–220.
Oppenheimer, D. M., Meyvis, T., & Davidenko, N. (2009). Instructional manipulation checks: Detecting satisficing to increase statistical power. Journal of Experimental Social Psychology, 45, 867–872.
Payne, B., Cheng, C., Govorun, O., & Stewart, B. D. (2005). An
inkblot for attitudes: Affect misattribution as implicit measurement. Journal of Personality and Social Psychology, 89, 277–293.
Cooley et al.
Payne, B. K., Brown-Iannuzzi, J., Burkley, M., Arbuckle, N. L., Cooley,
E., Cameron, C. D., & Lundberg, K. B. (2013). Intention invention
and the affect misattribution procedure reply to bar-anan and nosek
(2012). Personality and Social Psychology Bulletin, 39, 375–386.
Petty, R. E., Briñol, P., & DeMarree, K. G. (2007). The Meta-Cognitive
Model (MCM) of attitudes: Implications for attitude measurement,
change, and strength. Social Cognition, 25, 657–686.
Plant, E., & Devine, P. G. (1998). Internal and external motivation to
respond without prejudice. Journal of Personality and Social
Psychology, 75, 811–832.
Raja, S., & Stokes, J. P. (1998). Assessing attitudes toward lesbians
and gay men: The modern homophobia scale. Journal of Gay,
Lesbian, and Bisexual Identity, 3, 113–134.
Sears, D. O., & Henry, P. J. (2005). Over thirty years later: A contemporary look at symbolic racism. Advances in Experimental Social
Psychology, 37, 95–150.
Tormala, Z. L., & Petty, R. E. (2002). What doesn’t kill me makes me
stronger: The effects of resisting persuasion on attitude certainty.
Journal of Personality and Social Psychology, 83, 1298–1313.
Wegner, D. M. (2002). The illusion of conscious will. Cambridge,
MA: MIT Press.
507
Wegner, D. M., & Wheatley, T. (1999). Apparent mental causation:
Sources of the experience of will. American Psychologist, 54,
480–492.
Author Biographies
Erin Cooley is a doctoral student in the Department of Psychology at
the University of North Carolina at Chapel Hill. Her research explores
implicit and explicit social cognition about individuals and groups of
people, intergroup processes more generally, and positive emotions in
the context of interracial interactions.
B. Keith Payne is an Associate Professor at the University of North
Carolina at Chapel Hill. He studies the unintended and the unconscious. More information on his research can be found at http://
www.unc.edu/~bkpayne.
Kimberly Jean Phillips contributed to this research for her undergraduate thesis at the University of North Carolina at Chapel Hill. She
is currently a research assistant for the Internet Tobacco Vendors
Study at Lineberger Comprehensive Cancer Center and intends to pursue a Ph.D. in social psychology.