Attitude Change

DESIRED ATTITUDES GUIDE ACTUAL ATTITUDE CHANGE 1
Desired Attitudes Guide Actual Attitude Change
Thomas I. Vaughan-Johnston, PhD1, 2

Leandre R. Fabrigar, PhD3
Ji Xia, MSc4
Ken DeMarree, PhD4
Jason Clark, PhD5
1
Corresponding author
2
Durham University, Department of Psychology
3
Queen’s University, Department of Psychology
4
University at Buffalo, Department of Psychology
5
University of Virginia, Department of Psychology
Word Count: 13,275 (all main text including references, excluding title and abstract)
Acknowledgements: This research was supported by a graduate fellowship (767-2018-1484) to
the first author and an Insight Grant (435-2015-0114) to the second author provided by the
Social Sciences and Humanities Research Council of Canada.
Data Access: All data have been made available to reviewers at
https://tinyurl.com/desiredattitude, and will be made publicly available in the event of
publication.
Abstract
Whereas actual attitudes represent people’s evaluations of specific objects as being good or bad,
desired attitudes represent the attitude positions that people wish they held. Previous work has
established that desired attitudes can have psychological consequences, but has not yet tested the
extent to which desired attitudes can predict actual attitude change. In five datasets involving a
variety of populations and procedural variations, we explored how political and moral factors
motivate people to form desired attitudes distinct from their actual attitudes. These desired
attitudes then predicted actual attitude change occurring after the formation of the desired
attitudes, even though participants received no new information about the object. The present
work demonstrates an additional value of the desired attitude construct: its ability to anticipate
how people’s attitudes will shift in the future.
Word Count: 131

Keywords: attitudes; attitude change; desired attitudes; mere thought; motivation
Desired Attitudes Guide Actual Attitude Change
In 2020, J. K. Rowling wrote a series of tweets expressing her viewpoints about sex and
gender, resulting in substantial public controversy (Duggan, 2022; Rowling, 2020). Rowling is
the bestselling author of the Harry Potter fantasy novel series, and doubtlessly many people held
positive views of her books when this controversy arose. However, many fans evidently saw
Rowling’s comments as expressing problematic values and/or beliefs. We suspect that
simultaneously liking the books and hating the author produced intense psychological conflict
for many people. Some initially Potter-liking people might have felt confident “separating the art
from the artist,” and for that reason might not desire to change their opinion of her books, but
others may have felt an intense public and/or private pressure to revisit their opinions of
Rowling’s novels. We are interested in whether people who form such desired attitudes might
attain these goals. Liking something while wanting to dislike it (or vice versa), as the present
work argues, may have important attitudinal implications.
The above is hardly an isolated example. People often fall in love with art or other
products, only to become alienated from the producers of those things, potentially creating
discrepancies between the opinions people actually hold (versus wish to hold) about objects.
More broadly, people often experience conflicts between their current opinions and their
identity-laden moral positions or broader goals. People may want to like a co-worker more than
they actually do, or desire to enjoy the flavor of healthy foods like kale or sardines. Researchers
have examined this interesting and commonplace phenomenon. Whereas actual attitudes,
people’s evaluations of things as being good and/or bad, have long been recognized as being a
central construct in social psychology (Eagly & Chaiken, 1993; Maio & Haddock, 2009; Ostrom,
1989; Petty & Cacioppo, 1981; Petty et al., 1997; Thurstone & Chave, 1929), desired attitudes
are a relatively novel construct that captures the attitudes that people wish they held (DeMarree
et al., 2014). A growing literature has explored desired attitudes’ conceptual distinctiveness from
existing constructs, their frequency, antecedents and consequences, and moderators (Carrera et
al., 2017; DeMarree & Rios, 2014; DeMarree et al., 2014, 2017).
Although desired attitudes relate to a range of important consequences, researchers have
yet to examine the extent to which they can influence a key concept in the attitudes literature:
attitude change. Given that desired attitudes are the opinions that people want to hold, this
absence of a direct connection to actual attitude change is surprising. We propose that people
adopt desired attitudes as internal goals that then can effectively predict the direction of their
subsequent attitude regulation efforts. These internal goals are important because self-persuasion
is often a cognitively demanding task (Maio & Thomas, 2007); therefore, people may benefit
from forming a target of what attitude they want (i.e., whether they want a more
positive/negative attitude, and how extremely the discrepancy from their present attitude is).
Previous work establishes that desired attitudes are related to a biased preference for, and
processing of novel, external information (DeMarree et al., 2017). However, we propose that
desired attitudes can shift people’s opinions even when they lack the opportunity to facilitate
attitude change with new information. Thus, we examine an impoverished situation in which
people cannot receive any novel information between their forming a desired attitude and their
subsequent actual attitude change. Nonetheless, we anticipate that even without new information,
people will self-persuade towards their desired attitude goal, by reflecting on and reorganizing
information that they already possess about an object (Briñol et al., 2012; Clarkson et al., 2011;
Maio & Thomas, 2007; Tesser, 1978; Tesser & Leone, 1977; Tesser & Conlee, 1975). No
previous work has examined whether these attitude regulation efforts guided by desired attitudes
effectively produce attitude change.
Desired Attitudes and Actual Attitude Change
The beliefs we wish we held often differ markedly from those that we actually hold. This
is recognized in the self literature, which has examined self-discrepancies as gaps between the
views people actually hold about themselves, versus the views people wish they held about
themselves (e.g., Higgins, 1987, 1989; also see Carver & Scheier, 1998). Past work has often
suggested that actual/desired self-discrepancies are aversive, and that people may effortfully
pursue these desired “self-guides.” However, discrepancies need not apply to self-evaluations.
DeMarree and colleagues (2014, 2017) described desired attitudes as evaluations that people are
motivated to hold about any discrete attitude object (i.e., person, thing, or concept; also see Maio
& Thomas, 2007). For instance, people may like Walmart (actual attitude) because it is cheap,
and has a wide variety of products, but may wish to dislike Walmart (desired attitude) because of
its associations with worker exploitation and deceiving the public (Benoit & Dorries, 1996).
Desired attitudes follow from people’s motivational drives, helping them to accomplish goals,
claim important identities, or maintain consistency with higher order values and ideologies
(Wheeler & DeMarree, 2019). However, desired attitudes are distinct from the drives that create
them, representing desires to hold particular opinions about particular attitude objects.
Desired attitudes are consequential for how people pursue attitude positions. For instance,
just as people’s actual attitudes can predict increased seeking and processing of attitude-
congenial information (e.g., Houston & Fazio, 1989; Lord et al., 1979), people’s desired attitudes
similarly predict evaluatively congruent information seeking and processing, over and above the
influence of actual attitudes (DeMarree et al., 2017). Additionally, people may modify the
properties of attitude objects to reach a desired attitude goal. For example, people who
anticipated drinking some coffee used more additives (i.e., sweeteners, creamers) as their desired
attitudes towards coffee became more positive (DeMarree et al., 2017, Study 4). In contrast,
those with positive actual attitudes towards coffee used less additives. Together, these findings
suggest that people are motivated to pursue their desired attitudes through various strategies,
dedicating significant effort to reconcile their actual attitudes with their desired attitude goals.
However, no existing empirical work has connected desired attitudes to actual attitude change.
Connecting desired attitudes to actual attitude change is important not only because such
attitude change effects are theoretically interesting, but also because extant research might
challenge the likelihood of such attitude change unfolding. For instance, people effortfully
pursue both their desired and their actual attitudes simultaneously (e.g., DeMarree et al., 2017).
Consequently, it is not obvious that people will shift their actual attitudes to match their desired
attitudes, rather than the reverse. Consider that when people are ambivalent about a topic, they
generally resolve this ambivalence by strengthening the relatively dominant (stronger) attitude
component rather than by bolstering the conflicting (weaker) attitude component (e.g., Clark et
al., 2008). Insofar as desired attitudes are a hypothetical or imagined state of mind (e.g., “it
would be good if I evaluated X positively”), it might be easier for people to change their desired
attitudes to match actual attitudes, rather than changing their actual (“real”) attitudes to match
desired states. However, past work indicates that people are often highly committed to their
‘hypothetical’ (desired) attitudes (DeMarree et al., 2017) so people may shift their actual
attitudes to match their desired attitudes, provided that the motivation that promotes the desired
attitudes is sufficiently powerful (e.g., driven by one’s political identity, fundamental moral
principles; Wheeler & DeMarree, 2019).

Desired Attitudes as Self-Persuasive Goals
For the above reasons, we propose that desired attitudes will often have consequences for
people’s actual attitude change. This idea is also consistent with cognitive dissonance theory,
which is based on the premise that people often experience tension when incongruency exists
between elements of their cognition and thus people work to minimize the resulting aversive
state of dissonance through several tactics including changing one or more of the discrepant
cognitive elements (Cooper, 2007; Festinger, 1957; Harmon-Jones & Mills, 2019; Simon et al.,
1995; see also Carver & Scheier, 1998; Heider, 1946). Thus, people could shift their opinions
over time to reconcile their actual and desired attitudes, thereby minimizing conflict between
their attitude positions. Given that a desired attitude is, by definition, an attitude position that
people want to hold, people might engage in considerable cognitive efforts to attain that attitude.
Desired attitudes may be initially stimulated by external information, but they may lead people to
feel an internal pressure to regain cognitive consonance between their actual and desired
attitudes. For instance, learning that a painting you initially liked was created by a reviled person
(e.g., a Nazi) doubtlessly will lead you to immediately like the painting less. But the reasons for
the originally positive evaluation (e.g., its pleasant aesthetic qualities) are not strictly
contradicted by this new information, and substantial cognitive effort may be required to talk
yourself into a satisfactorily negative opinion of the painting. Thus, the Nazi-related information
might also create a desire to dislike the painting, and that desire might be felt as an internally
driven pressure that shapes subsequent thinking and thus attitude change (i.e., self-persuasion;
Aronson, 1999; Maio & Thomas, 2007). The desired attitude is not just memory of the novel
information because it relies on an evaluation of the information (i.e., how much one hates
Nazis) and judgments of relevance and importance (e.g., should one judge a painting according
to its creator’s moral characteristics, and to what degree?), integrated into an estimate of “taking
it all in, what attitude do I want to obtain?” (i.e., a desired attitude).
If desired attitudes represent the motivation people have to like/dislike something, how
might they convince themselves to actually adopt their desired attitude positions? Psychologists
have often demonstrated strategies that people use to self-persuade, shaping their own opinions
towards an object (Briñol et al., 2012). Additionally, much of this past work supports that people
can change their minds without having to collect new information to justify attitude change.
Even without changing the content of thoughts, metacognitive processes (i.e., thoughts about
one’s thoughts) can change the impact of thoughts on attitudes (Petty et al., 2002). Validated
thoughts have larger influence over people’s attitudes, whereas invalidated thoughts have
reduced influence over attitudes (Briñol et al., 2018; Requero et al., 2020). Additionally, Maio
and Thomas (2007) pointed out a variety of metacognitive strategies people may use to self-
persuade towards situationally important goals by reorganizing or reweighting information (e.g.,
changing attributions, reweighting the importance of information). Thus interestingly, people
may show substantial attitude change towards an object even without learning anything new
about that object, and thus propose that people may be able to shift their actual attitudes to match
their desired attitudes even without any opportunity to modify the object or collect a biased set of
new information to facilitate that actual attitude change (i.e., the processes previously established
for desired attitudes). We see desired attitudes as likely to guide this process by providing people
with a target to direct their cognitive and metacognitive strategies. To maximize the likelihood of
self-persuasion, we provided a series of metacognitive prompts in Experiments 1-2. However,
we suspected that participants also would engage in metacognitive reflection without explicit
prompting, and we test this in Experiment 3.

Overview and Hypotheses
We investigated these questions with five experiments. Our main goal was to examine if
desired attitudes shape the direction of people’s actual attitude change. Because desired attitudes
indicate the actual attitude position that people are motivated to pursue, we reasoned that people
could persuade themselves into the attitude position indicated by their desired attitude.
Specifically, we hypothesized that the desired attitudes we created in these paradigms would
predict subsequent actual attitude change, even across a period when participants would have no
opportunity to change the object’s objective properties or acquire new information, thus creating
circumstances where participants could only self-persuade to reach their desired attitude position.
Experiment 1
The central goal in Experiment 1 was to provide an initial test of our hypothesis that
desired attitudes can predict subsequent actual attitude change, adjusting for the immediate
attitude change produced by information, and even in a context where established attitude-
relevant processes (physically modifying the object, biased information search) would be
impossible. To do this, we had people form an actual attitude about a fictitious target individual,
and then observed how desired attitudes formed when learning of the target’s political affiliation
(which matched or mismatched the participants’ own political stance). We reasoned that a
matching (mismatching) target would create positive (negative) desired attitudes, motivated by
participants’ fidelity to their social/political group (i.e., their social identity; Hogg & Ridgeway,
2003; Tajfel, 1974), and/or as an expression of participants’ core values (i.e., moral principles
leading them to identify as liberal or conservative in the first place; Graham et al., 2009).
Because our predictions were novel, we collected two samples in order to self-replicate, labeled
as Experiment 1a and 1b.

Methods
Participants. For Experiment 1a, we recruited 324 Canadian university students (Mage =
20.2 years, SDage = 6.8). We neglected to include demographic items, but this sample was drawn
from a population that is 78.9% women, 20.7% men, and 0.4% non-binary; and majority white
(73.3%; 15.5% East Asian, 5.3% South Asian, 2.7% Black, 1.8% Hispanic), according to a pre-
screening survey. For Experiment 1b, we recruited 700 students from an American university
(Mage = 19.1 years, SDage = 1.3; 62.6% women, 37.4% men, 0% non-binary). We used time-based
stopping rules, obtaining 80% power to identify small effects, r > .15 (r > .11), for individual
regression parameters in Experiment 1a (Experiment 1b; G*Power v. 3.1.9.4; Faul et al., 2009).
This compares favorably with the results of an internal meta-analysis of desired attitudes
predicting “pursuit-oriented outcomes” of desired attitudes; r = .19 (DeMarree et al., 2017).
Sample size was always determined before any data analysis. Experiment 1b was preregistered,
although the preregistration covers additional analyses not reported here (see footnote 1), and
some alterations were made compared to the preregistration (see “Changes from the
preregistration” section, below).
Procedure. Participants answered demographic questions including a political
orientation scale. Next, participants read a transcript from an ostensibly real interview between a
politician and an interviewer. The interview highlighted positive and negative attributes of the
politician (e.g., political experience, cigarette-smoking habit) but not her political stance. To
create moderate actual attitudes towards the politician (so that matching/mismatching
information could drive subsequent attitude change towards her), participants listed three
positive and three negative thoughts about the politician in counter-balanced order, and rated
their Time 1 attitude towards the politician. Because order had no effects across our experiments,
we do not discuss it further. We report or acknowledge all measures used in this research.
Figure 1.
Flow of the Experimental Procedure (Exp. 1).

Next, to manipulate desired attitudes, half of our participants learned that the politicians’
political views matched their own political views (i.e., she was liberal if the participant was
liberal, conservative if the participant was conservative), whereas the other half were given
information about the politician’s views that mismatched their own political views (i.e., she was
conservative for liberals; liberal for conservatives). Participants then rated their Time 2 desired
attitudes towards the politician. Participants also rated their actual attitudes at this time, so that
we could empirically distinguish between the immediate attitude change cultivated by the
matching/mismatching information, and participants’ later attitude changes. The format of our
Time 1, Time 2, and Time 3 actual attitude measures were distinct (having two, one, and eight
items, respectively; different endpoint values; different endpoint labels). These different formats
helped to avoid consistency effects that might obscure real attitude change (see Blankenship et
al., 2012, for a similar method). That is, if formatting was kept invariant, participants might have
felt “locked in” to earlier attitude questions when answering later attitude questions about the
same object (Downing et al., 1992; Petty & Cacioppo, 1981).
We next stimulated participants to reflect on their past thoughts with several
metacognitive prompts intended to give them an opportunity to self-persuade (Petty et al., 2002).
Participants read their positive and negative thoughts from earlier in the procedure and we
prompted them to consider how likeable, certain, and clear each thought was.1 Importantly, these
prompts only encouraged participants to think further, and did not provide any new information
about the politician. Participants then rated their Time 3 actual attitudes.2
1
In the present study, these three prompts were included to encourage participants to engage in meta-
cognitive reappraisals of their evaluative responses to the attitude object. Thus, they were not intended as
measures to test the current hypothesis, although responses to these prompts were recorded. Analyses of
these measures are pertinent to a separate research project examining different types of thought
perceptions, and will be reported in a future manuscript.
2
We also assessed Time 3 behavioral intentions towards the politician with six items, e.g., “I would vote
for [the politician] to be my political representative.” Results for behavioral intentions converged very
closely with results reported for actual attitudes, so to avoid redundancy while maintaining full
Materials. For this and future studies, additional information on the measures including
descriptive statistics and correlation tables between the variables appear in SOM-2.
Political orientation. We used a “forced-choice” single-item, six-point bipolar scale (1 =
very conservative, 6 = very liberal). In Experiment 1a, 25.4% of participants were conservative,
and 74.6% were liberal. In Experiment 1b, all but 11 participants provided their political
orientation. In total, 42.8% of participants were conservative, and 57.2% were liberal.
Experiment 1b’s American sample was significantly less liberal (M = 3.7, SD = 1.3) than
Experiment 1a’s Canadian sample (M = 4.3, SD = 1.2), t(1022) = -7.04, p < .001, d = -.47.
Because our politician’s political beliefs were very liberal / very conservative, we
examined political extremity as a potential moderator of the matching condition’s effect on
desired attitudes. We calculated political extremity as deviation from neutrality (i.e., 3.5), with a
+.5 adjustment to create whole number scores, such that a 1 on the extremity scale reflected
participants were “slightly” liberal/conservative, 2 reflected participants who were “moderately”
liberal/conservative, and 3 reflected participants who were “strongly” liberal/conservative. In
Experiment 1a, political extremity was significantly correlated with political ideology, r(321)
= .62, p < .001, unsurprisingly given that the sample was mostly liberal. In Experiment 1b,
political extremity was more weakly correlated with political ideology, r(687) = .14, p < .001;
the correlation differed across samples; r-to-z transformation test, z = 8.68, p < .001.
Time 1 actual attitude. Participants provided their actual attitude toward the politician
after reading the interview document. We used two items (1 = very unlikeable to 7 = very
likeable; 1 = dislike her to 7 = like her); these items were highly inter-correlated in each sample,
rStudy1a(322) = .80, p < .001, rStudy1b(695) = .77, p < .001, so we averaged these items in each
dataset, higher scores indicating more positive attitudes.
transparency we report details for these analyses in the Supplementary Online Materials (SOM-3).
Time 2 actual attitude. Participants provided their actual attitude toward the politician
after learning about her political views on one item ranging from 1 (strongly dislike) to 9
(strongly like).
Time 2 desired attitude. Consistent with past research (e.g, DeMarree & Rios, 2014, p.
203), we briefly explained desired attitudes by stating, “Sometimes the attitudes we have are
different from the attitudes we ideally would like to have or the attitudes we feel we should hold,
and sometimes these attitudes are the same.” Participants then indicated the attitude they
“IDEALLY would like to have” (ideal attitude) and “the attitude [they] SHOULD or OUGHT
to have” (ought attitude) from 1 (would IDEALLY / OUGHT TO strong dislike) to 9 (would
IDEALLY / OUGHT TO strong like). Because past literature finds that effects of ideal and ought
attitude effects converge (e.g., DeMarree et al., 2014), our interest was in desired attitudes
broadly, and ideal/ought attitudes correlated highly, we averaged ideal and ought attitudes into a
composite desired attitudes measure, only reporting ideal and ought attitude effects separately if
results differed meaningfully for these items.
Desired attitudes have often been captured either with this composite (i.e., averaging
ideal/ought attitudes) or through a subjective discrepancies measure (i.e., asking participants to
directly estimate how much more/less they want to like an attitude object). In the present
experiments we measured both. Results were similar for the two measures, and the two measures
were highly correlated, rStudy1a(322) = .69, p < .001, rStudy1b(694) = .65, p < .001. We therefore
report the subjective discrepancy measure’s results in SOM-4.
Time 3 actual attitude. Eight items (Crites et al., 1994) captured attitudes, each rated
from 1 (not at all) to 9 (definitely): good, bad (R), like, dislike (R), desirable, undesirable (R),
positive, and negative (R). After reversing (R) items, the scale was reliable (α = .97), so we
averaged items, higher scores indicating favourable attitudes.
Results
SOM-8 provides means and standard deviations for each dependent variable at all levels
of the experimental variable(s) for all experiments.
Manipulation check: Desired attitudes. Our first analysis examined if desired attitudes
formed based on identity-related pressures (the matching manipulation). Effects of matching /
mismatching condition on desired attitudes would demonstrate that desired attitudes can form
based on social/political goals, since the political identity of the target was the only variable
manipulated. Additionally, we included Time 2 actual attitude (i.e., the degree of actual attitude
change induced post-manipulation) as a covariate in this analysis. Obviously, people (actually)
prefer members of their political ingroups over political outgroups (Bruchmann et al., 2018;
Byrne, 1961), so by statistically controlling for this effect we isolated the unique effect that the
manipulation had on desired attitudes specifically. In all experiments, the desired attitude
manipulation also affected Time 2 actual attitudes; tstudy1a(322) = 12.06, p < .001, d = 1.34;
tstudy1b(694) = 10.22, p < .001, d = .78; tstudy2a(115) = 9.04, p < .001, d = 1.67; tstudy2b(512) = 7.65, p
< .001, d = .67; tstudy3(199) = 9.79, p < .001, d = 1.38. We also considered if the influence of
matching condition on desired attitude differed across levels of participant political extremity.
Because our conservative and liberal politician profiles were relatively extreme, we expected an
interaction of matching condition X political extremity. In other words, a “matching” (versus
“mismatching”) politician should match/mismatch most strongly for extreme-political
participants, and thus the politician’s ideology would dictate desired attitudes more powerfully
for people at the extremes of the political spectrum (versus moderates).

To test these ideas, we regressed Time 2 desired attitudes onto a contrast-coded matching
condition variable (-.5 = mismatching, +.5 = matching), Time 1 actual attitude, Time 2 actual
attitude, centered political extremity, and a matching condition X political extremity interaction
term. Table 1 shows the results for each of our two samples. Matching condition (i.e., the first
data row) had a large first-order effect on desired attitudes for people of average political
extremity in Experiment 1a. The unstandardized B indicates that at average political extremity,
the mean desired attitudes for people reading the matching condition passage was about one
scale unit more positive than the mean for their mismatching condition counterparts. Thus, for
people of average political extremity, learning that the politician matched (vs mismatched) their
political views led participants in Experiment 1a (but not Experiment 1b) to desire to like her.
We found no effect of Time 1 actual attitude on desired attitudes in either dataset. This
null result may reflect that participants’ Time 1 attitudes were captured before participants
learned the politician’s political orientation. This information might have been of reduced
relevance after learning that politician’s strong conservative or liberal views. However, Time 2
actual attitude was strongly related to Time 2 desired attitudes, consistent with past work that
identifies large correlations between actual/desired attitudes (DeMarree et al., 2014, 2017).
Most critically, we also found a significant interaction of matching condition X political
extremity in each dataset. We analysed these interactions by testing the effect of matching
condition across levels of political extremity. In Experiment 1a, matching condition had a
relatively modest effect on desired attitudes among the least extreme participants, but a
substantial effect among very extreme political participants. In Experiment 1b, similarly,
matching condition had a negligible effect on desired attitudes among less extreme participants,
but a significantly positive effect among more politically extreme participants. This moderation
qualifies the mixed main effect of manipulation on desired attitudes by highlighting a boundary
condition. That is, political moderates’ desired attitudes were minimally affected by the
politician’s political identity, but politically polarized participants’ desired attitudes were
significantly shaped by the politician’s political identity.
This effect also is depicted in Figure 1, the two panels displaying the two samples
separately. Both figures are very similar. Following political extremity from low (left part of
figure) to high (right part of figure), we see that the difference in desired attitudes (and subjective
discrepancies) prompted by the matching (versus mismatching) politician passage grows from
virtually non-existent into a quite large difference. This pattern reflects the same breakdown
narrated above: as political extremity increases, the desired attitudes expressed by participants
are increasingly dictated by the political stance of the politician.3
3
Separating ideal and ought attitudes slightly affected results. In Experiment 1a, the interaction effect
emerged both for ideal attitudes assessed individually, b = .90 [.38, 1.41], t(317) = 3.41, p < .001, and for
ought attitudes, b = .74 [.24, 1.25], t(317) = 2.89, p = .004. In Experiment 1b, the interaction effect
emerged for ideal, b = .35 [.04, .67], t(682) = 2.18, p = .030, but not for ought attitudes, b = .22
[-.09, .52], t(682) = 1.41, p = .160
Table 1. Moderation of Matching Condition on Time 2 Desired Attitude Formation by Participant Political Extremity. (Exp. 1).
Predictor Variable Parameter Coefficients (Exp. 1a) Parameter Coefficients (Exp. 1b)
Matching Condition b = 1.00 [.65, 1.34], b = .12 [-.06, .31],

t(317) = 5.65, p < .001, rsp = .17 t(682) = 1.29, p = .198, rsp = .03
Time 1 Actual Attitude b = -.04 [-.18, .10], b = -.004 [-.09, .08],

t(317) = -.54, p = .593, rsp = -.02 t(682) = -.09, p = .932, rsp = .00
Time 2 Actual Attitude b = .64 [.57, .72], b = .71 [.66, .77],

t(317) = 16.69, p < .001, rsp = .49 t(682) = 26.31, p < .001, rsp = .63
Participant Political Extremity b = -.01 [-.21, .19], b = -.08 [-.20, .05],

t(317) = -.12, p = .903, rsp = .00 t(682) = -1.20, p = .231, rsp = -.03
Matching Condition X b = .82 [.38, 1.25], b = .29 [.03, .55],

Participant Political Extremity t(317) = 3.70, p < .001, rsp = .11 t(682) = 2.15, p = .032, rsp = .05
Matching Condition Simple Slope: b = .42 [.02, .82], b = -.05 [-.28, .18],
Low (-1 SD) Political Extremity t(317) = 2.08, p = .039, rsp = .06 t(682) = -.41, p = .683, rsp = -.01
Matching Condition Simple Slope: b = 1.57 [1.05, 2.08], b = .32 [.05, .60],
High (+1 SD) Political Extremity t(317) = 6.00, p < .001, rsp = .18 t(682) = 2.29, p = .022, rsp = .05
Model Statistics F(4, 317) = 172.9, F(5, 682) = 216.6,

p < .001, R2 = .73 p < .001, R2 = .61
Note. Each row displays the statistics associated with a single independent variable. The left data column displays results for
Experiment 1a, the right data column for Experiment 1b. rsp indicates the semi-partial r effect size.
Figure 1. Desired Attitudes Influenced by Levels of Matching (of Participant with Target
Political Beliefs) and Participant Political Extremity.
A. Experiment 1a
B. Experiment 1b
Note. Errors bars reflect standard error values. Higher scores reflect more positive desired
attitudes. Time 1 and Time 2 actual attitudes were controlled in these analyses.
Desired attitudes predicting subsequent actual attitude change. We next tested if
desired attitudes predicted the actual attitude change that people next experienced. Recall that
participants received no new information between reporting their Time 2 and Time 3 actual
attitude. This implies that actual attitude change predicted by desired attitudes represents
people’s self-persuasion, that is, people shifting their actual attitudes to closer match their
desired attitudes merely by thinking further about the topic. This would demonstrate that desired
attitudes shape people’s actual attitude change, even in the absence of new information.
We regressed Time 3 actual attitudes onto Time 1 and Time 2 actual attitudes, and Time
2 desired attitudes. Results for both datasets are displayed in Table 2. The actual attitude
measures in these models were included as covariates so that Time 3 actual attitudes captured
attitude change. That is, the Time 3 actual attitude dependent variable here indicates the
favourability of people’s opinions towards the politician controlling for their opinion before
engaging in additional reflection. In both experiments (left and right data columns), Time 1 and
Time 2 actual attitudes related positively to Time 3 actual attitudes, suggesting that people’s
earlier attitudes influenced their later attitudes even after engaging in metacognitive reflection.
Critically, Time 2 desired attitudes positively predicted Time 3 actual attitudes in both samples.
In addition to how much people actually liked the politician, then, people’s desire to like her
accounted for additional attitude change. People thus pursued their desired attitudes by merely
reflecting on their own previous thoughts with a new metacognitive goal.4
4
We also examined if participants’ behavioral intentions towards the politician showed similar patterns to
the attitude change phenomena explained above. Results were very similar to the attitude change analysis
both in structure and results, so are detailed in SOM-3.
Table 2. Effects of Desired Attitudes on Changes in Actual Attitudes at Time 3. (Experiment 1).
Predictor Variables Parameter Coefficients (Exp. 1a) Parameter Coefficients (Exp. 1b)

t(320) = 5.42, p < .001, rsp = .14 t(692) = 2.71, p = .007, rsp = .06

t(320) = 13.06, p < .001, rsp = .34 t(692) = 20.66, p < .001, rsp = .44
Time 2 Desired Attitude b = .15 [.09, .21], b = .08 [.04, .13],

t(320) = 4.65, p < .001, rsp = .12 t(692) = 3.46, p = .001, rsp = .07

p < .001, R2 = .79 p < .001, R2 = .69
Note. rsp indicates the semi-partial r effect size.
23
Quality checks (Exp. 1b only). In our preregistration, we planned quality checks, re-
testing results but removing participants who (i) did not provide enough thoughts (n = 43), (ii)
revealed suspicion on our probes (n = 12), (iii) mentioned having technical issues (n = 5), or (iv)
did not correctly answer our attention check (n = 251). The first three re-analyses did not affect
our results, and the analyses remained unchanged: political extremity still moderated passage’s
effect on desired attitude; desired attitude predicted attitude change. Removing participants who
failed the attention check rendered both key analyses non-significant (p = .416, .247,
respectively). However, this is likely attributable to losing >1/3 of the sample, likely
undermining statistical power to detect our effect. Importantly, of the 251 removed participants,
195 (78%) failed because they left the attention check blank.5 If we re-run the analyses only
removing participants who actively answered the attention check incorrectly, both analyses
become significant again (ps < .037). A full breakdown of results is available in SOM-7.
Changes from the preregistration (Exp. 1b only). First, in the preregistration, we stated
that t-tests would be used for some analyses (e.g., effect of matching condition on Time 2 desired
attitude), but also that political extremity would be tested as a moderator of these analyses.
Ultimately, we foregrounded the moderated version of these analyses (i.e., in Table 1) because
lower-order effects such as the main effects of matching condition on desired attitudes are
already expressed in those regression models, and are qualified by the significant interaction with
political extremity. Second, in the preregistration we stated that the effects of our desired
attitudes manipulation (matching condition) would be used to predict Time-3 actual attitudes.
Indeed, this test supported our hypothesis, t(694) = 9.83, p < .001, d = .75 [.59, .90], with more
positive attitudes given a matching (M = 4.65, SD = 1.16) versus mismatching politician (M =
5
We are not certain why so many participants left this attention check item blank. One possibility is that
participants felt they had already demonstrated adequate attention by answering the suspicion item, and
felt this question was therefore redundant.
24
3.73, SD = 1.30). However, we instead foregrounded regression models in which Time-2 desired
attitudes and Time-1/Time-2 actual attitudes predicted Time-3 actual attitudes simultaneously
(i.e., Table 2). Our reasoning was that this analytic approach is more precise because it reveals
the unique effects attributable to the manipulation’s effect through desired attitudes specifically.
But either analysis supports our contentions.
Discussion
Our first experiments provided an initial empirical demonstration that desired attitudes
can shape subsequent attitude change, supporting our claim that people are not only motivated
but also capable of altering their opinions in a direction indicated by their desired attitude
standards. In essence, participants expressed a desire to hold a particular attitude position, and
their subsequent attitude change aligned with that desire. This shows that people at Time 2 had
some awareness of the direction that their actual attitudes would shift in by Time 3. Critically,
this actual attitude change then occurred even though participants learned nothing new about the
object between Time 2 and Time 3, suggesting a surprising self-awareness of participants. It is
noteworthy that actual attitude change occurred despite the likely consistency pressure that
participants may have felt to not contradict themselves by changing attitudes from Time 2 to
Time 3 with no new information provided between these intervals.
Furthermore, we created desired attitudes in these experiments in a manner different from
past research: by associating an attitude object with information relevant to people’s political
identities. Past theorizing suggests that identities can be powerful origins of desired attitudes
(Wheeler & DeMarree, 2019), but this idea had not been explicitly tested. Indeed, past research
that has manipulated desired attitudes has used manipulations that emphasize the utility of
particular viewpoints (DeMarree & Rios, 2014; DeMarree et al., 2014). We build on this past
25
work by showing desired attitudes forming based on broader factors that relate to people’s
ideological positions. This broadens our understanding of why people form desired attitudes.
However, as some exploratory follow-up analyses revealed, Experiments 1a/1b were
imperfect in that our matching/mismatching paradigm worked more effectively for liberals (who
almost universally preferred the liberal over the conservative candidate) than for conservatives
(who evaluated the liberal and the conservative candidate more equally). One explanation is that
our conservative profile was too extreme given that few of our conservatives (4%/11% in
Experiment 1a/1b) self-described as “extreme” conservatives. This asymmetry might have
worked against the construct validity of the political matching/mismatching manipulation.
Poignantly, this issue would in principle work against our hypothesized pattern. Regardless,
subsequent experiments avoid this issue because they expose participants to intensely moralized
stimuli that clearly match/mismatch the moral sensibilities of nearly any university student.
A second limitation is that the politician paradigm in Experiment 1 would ideally be
extended to other persuasion contexts. We wanted to enhance the construct validity and
generalizability of our findings beyond politicians, and therefore introduced a new paradigm in
Experiment 2-3. To increase the chance that all participants would be powerfully affected by our
manipulation, we employed a more radical approach involving artwork being associated with
extreme value-relevant information: association with Nazi victims versus Nazi sympathizers.
Experiment 2
In Experiment 2 we continued exploring the attitude change consequences of desired
attitudes in a new context involving extreme moral motivations. Whereas Experiment 1a/1b
related an attitude object with political beliefs directly related to the attitude object, from
Experiment 2 onwards we connected an attitude object (i.e., a painting) with morally provocative
26
information (i.e., that the painter had anti-fascist or pro-fascist sensibilities). Thus, Experiment 2-
3’s art paradigm is distinct from the political paradigm in two key regards. First, the information
manipulating desired attitudes is somewhat less directly connected with the object itself
(information about the object’s creator, rather than information about the attitude object itself).
Second, the information is about highly moralized associations of the object (rather than a
commonplace political group identity). Once again, we directly replicated our experiment,
reporting the results of these datasets together as Experiments 2a/2b.
Methods
Participants. For Experiment 2a, we recruited 119 undergraduates from a Canadian
University to complete the study online for partial course credit (79.8% women, 20.2% men, 0%
non-binary or prefer not to answer; Mage = 19.5 years, SDage = 5.2).6 For Experiment 2b, we
recruited 514 online volunteer American participants from ResearchMatch (74.0% women,
24.0% men, .8% non-binary, 1.2% prefer not to answer; 86.7% Caucasian/White, 3.5% African
American/Black, 3.1% “mixed,” 1.8% Latinx, 1.2% East Asian, 1.4% “other,” and 0.6% East
Indian; 2.2% did not answer). The latter group was substantially older than all previous samples,
Mage = 52.2 years, and with much greater heterogeneity in age, SD = 16.4. In each case, sample
sizes were determined by time-based stopping rules, and they permitted us with 80% statistical
power to detect effect sizes of r > .25 and .13, respectively. We reasoned that smaller samples
compared to Experiment 1 were reasonable given that (i) the manipulation is probably stronger
than the political paradigm because it involved highly moralized content (hatred of Nazis) versus
mere political preferences (e.g., liberal/conservative dislike of the other party), and because (ii)
our design did not require interaction tests, unlike Experiment 1a/1b.
6
We did not include a race item for Experiment 2b, but participants were drawn from a pool of
participants with 79.7% White; 13.9% East Asian, 4.7% South Asian, 1.7% Black.
27
Procedure. The structure of Experiment 2 was highly similar to the politician paradigm.
Participants first viewed an abstract-expressionist painting by Arshile Gorky (Agony). As before,
participants then listed three positive and three negative thoughts, in counterbalanced order, and
rated their Time 1 actual attitude regarding the painting. Next, participants were informed that
the artist who had created the painting was Bruno Steingard, who ostensibly worked in 1940s
Munich, Germany. In the moral painter condition Steingard was described as a Nazi resistor,
whereas in the immoral painter condition Steingard was described as a Nazi sympathizer. We
reasoned the painter resisting (supporting) Nazis would provide an intense, morally-grounded
desire to form positive (negative) desired attitudes, despite people already having adopted their
initial (Time 1) opinions based on the aesthetic merits of the painting.
Having learned this information, participants reported their post-manipulation attitudes
towards the painting (1 = strongly dislike to 7 = strongly like), desired attitude (ideal, ought), and
subjective discrepancies. Finally, participants engaged in the same metacognitive prompts that
we used in Experiment 1, and then we assessed their Time 3 attitudes towards the painting.
Because behavioral intentions showed very similar results to Time 3 actual attitudes in both
Experiment 1 samples (see SOM-3), we dropped it to avoid redundancy. We did not ask about
participants’ political orientations because there is strong evidence that Nazis are roundly
despised by most North Americans (e.g., McCarthy, 2019; Vonasch et al., 2018).
Creating indices. The Time 1 actual attitude items were highly inter-correlated in each
sample, rStudy2a(115) = .73, p < .001, rStudy2b(512) = .87, p < .001, so we averaged them into Time 1
attitude indices. Time 2 ideal and ought attitudes were highly correlated in each sample,
rStudy2a(115) = .82, p < .001, rStudy2b(512) = .79, p < .001, so we averaged these to represent desired
attitudes.
28
Results
Manipulation check: Desired attitudes. In Table 3, we examine several factors related
to the formation of desired attitudes towards the painting across the two experimental samples. In
Experiment 2a, Time 1 actual attitudes were not significantly related to desired attitudes at Time
2. However, Time 2 actual attitudes were significantly and substantially related to desired
attitudes at Time 2. This finding is unsurprising because people likely use their actual attitudes as
an anchor for determining their desired attitude although their pre-manipulation attitudes
presumably do not serve this same purpose. Examining the effect of painter’s identity on desired
attitudes, people wanted to like the painting more than two full scale units more when it was
made by a Nazi victim rather than a Nazi sympathizer - after adjusting for the actual attitude
change caused by this same information. Thus, although people immediately shifted their actual
opinions of the painting based on its maker’s moral identity, people also desired to like/dislike a
painting made by a Nazi victim/sympathizer. Further, these patterns were empirically distinct in
that the desired attitude difference held despite controlling for contemporaneous actual attitudes.
We also briefly consider the subjective discrepancy measure to further describe reactions
to the manipulation. In Experiment 2a (2b), in the immoral painter condition, 79.3% (66.0%)
wanted a more negative opinion than they held, 10.3% (28.2%) wanted to hold their current
attitude, and 10.3% (5.8%) wished to hold a more positive attitude. In the moral painter
condition, by contrast, 61.0% (60.4%) wanted a more positive opinion, 32.2% (31.4%) wanted to
hold their current attitude, and only 6.8% (8.2%) wanted a more negative opinion.
29
Table 3. Painter Identity Effects on Desired Attitude Formation. (Experiment 2).
Painter Identity b = 2.37 [1.68, 3.06], b = 2.59 [2.28, 2.89],

t(113) = 6.77, p < .001, rsp = .34 t(510) = 16.71, p < .001, rsp = .44
Time 1 Actual Attitude b = -.21 [-.44, .03], b = -.29 [-.43, -.14],

t(113) = -1.75, p = .083, rsp = -.09 t(510) = -3.95, p < .001, rsp = -.10

t(113) = 6.24, p < .001, rsp = .32 t(510) = 12.18, p < .001, rsp = .32

p < .001, R2 = .71 p < .001, R2 = .65
30
Desired attitudes predicting subsequent actual attitude change. We once again
regressed Time 3 attitudes onto Time 1 and Time 2 actual attitudes, and (Time 2) desired
attitudes (see Table 4). Both Time 1 and Time 2 actual attitudes related positively to Time 3
actual attitudes, although the standardized effect size of Time 2 actual attitudes was double that
of Time 1 actual attitudes. As in Experiments 1a-1b, this makes sense primarily because Time 2
actual attitudes reflect opinions at a moment closer to the Time 3 measurement, and here they
reflect opinions after (versus before) learning the painter’s identity.
Most importantly to our research questions, we examined whether desired attitudes were
able to predict actual attitude change. Indeed, desired attitudes were strongly related to actual
attitude change after participants had a chance to query their previous thoughts based on their
desired attitude goal. Compared to an average participant, each additional unit change in desired
attitude corresponded to a 0.3 (0.2) unit change in actual attitudes in Experiment 2a (Experiment
2b). Thus, people formed particular opinions about a painting initially based on its aesthetic
attributes. But then they self-persuaded to an actual attitude more congruent with their wished-
for actual attitude, even without any access to novel information about the painting itself. Thus,
the actual attitude change they eventually engaged in (by Time 3) was predictable based on the
attitude they considered desirable by Time 2. This finding is consistent with Experiments 1a/1b
but in a substantially different context. Specifically, the moral identity of a painter impacted
views of a painting, rather than the political position of a politician impacting views of the
politician herself.
PSYCHOLOGICAL MECHANISMS OF DESIRED ATTITUDES 31
Table 4. Effects of Desired Attitudes on Actual Attitude Change (Experiment 2).
Time 1 Actual b = .24 [.04, .44], b = .27 [.15, .38],

Attitude t(113) = 2.39, p = .019, rsp = .12 t(510) = 4.59, p < .001, rsp = .12
Time 2 Actual b = .41 [.25, .57], b = .46 [.37, .55],

Attitude t(113) = 4.95, p < .001, rsp = .25 t(510) = 9.70, p < .001, rsp = .25
Time 2 Desired b = .31 [.18, .45], b = .19 [.13, .24],

Attitude t(113) = 4.74, p < .001, rsp = .24 t(510) = 6.46, p < .001, rsp = .16

p < .001, R2 = .72 p < .001, R2 = .67
Discussion
Experiment 2 provided novel support for our hypotheses about the formation of desired
attitudes, and the effects of desired attitudes on self-persuasion, but in a very different, heavily
moralized, context. Once again, associating an attitude object with identity-relevant information
with minimal utilitarian implications prompted people to form desired attitudes towards that
object, above and beyond the manipulation’s immediate effect on their actual attitudes.
Interestingly, although the balance of desired/actual attitude discrepancies fell in the expected
directions, some participants (10.3%-32.2% depending on sample and condition) felt no desire to
hold a different attitude than their current one after exposure to the identity manipulation.
However, just as in the prior studies, desired attitudes were predictive of subsequent attitude
change towards the object after a period of cognitive reflection. Thus, although not everyone
desired to hold a different attitude than their actual one, the direction and magnitude of their
desired attitudes were predictive of their subsequent actual attitude change. Experiment 2 also
enhances the generalizability of our findings by extending them to a substantively different
context.
The two datasets provide a strong demonstration of our key claims and to some extent
address one another’s limitations. For example, Experiment 2a had a relatively small sample
size, making our statistical tests less precise than they would preferably be, but Experiment 2b
provided a much larger sample size for highly precise estimates. Furthermore, the two samples
were comprised of almost exclusively young adults (Experiment 2a) versus a broader sampling
of participants (Experiment 2b). Thus, replicating the findings very consistently in Experiment
2b’s larger, older sample enhanced certainty in our core findings.
Experiment 3
The first four datasets provide consistent support for our core hypothesis that desired
attitudes shape actual attitude change even without requiring additional information. However,
all studies encouraged people to reflect on their past thoughts through several metacognitive
prompts. The prompts were included to maximize the likelihood of attitude change occurring
after formation of desired attitudes. However, such direct external prompting may not be
necessary for desired attitudes to produce change presuming that the motivational drives
underpinning people’s desired attitudes are sufficiently strong. Indeed, the heavily moralized
context used in Experiments 2a/2b was created to produce just such a situation. Thus, it would
also be useful to know if the attitude change effects depended on participants being led to engage
in metacognitive reflection. Our primary goal in Experiment 3 was to provide conditions in
which prompts were not provided, to test if desired attitudes could still prompt subsequent
attitude change without any such encouragement. This would help to generalize the phenomenon
by demonstrating that people’s desired attitudes can still predict their subsequent attitude change
without the somewhat contrived stimulation of additional thinking.
Methods
Participants. We recruited 210 participants from a Canadian university (9 removed for
incomplete data). Of the remaining 201, most were young adults (Mage = 18.7, SDage = 1.6), and
most were women (72.1%, with 25.9% men, 0% non-binary, 2.0% prefer not to answer). Of
these, 78.6% were Caucasian/White, 11.9% East Asian, 3.0% African American/Black, 3.5%
East Indian, 2.5% Indigenous, 0.5% Latinx; 5.5% “other.” Sample size followed a time-based
stopping rule, giving us with 80% statistical power to detect effect sizes of r ≥ .19.
Procedure. The procedure matched Experiment 2a/2b, with one exception: half of
participants were given the metacognitive prompts and asked to review their thoughts (as
described in the Experiment 1 procedure), whereas the other half were not given these prompts.
Thus, Experiment 3 had a 2 (Painter Identity: Moral vs Immoral) X 2 (Order: Positive thoughts
first vs Negative thoughts first) X 2 (Metacognitive Prompts: Present vs Absent) between-
participants design. Thus, those in the Metacognitive Prompts Absent condition learned the
painter’s identity, recorded attitudes and desired attitudes at Time-2, considered attitude/desired
attitude discrepancies (see SOM-4), and then reported Time-3 actual attitudes. Order had no
main effects or interactive effects with other study variables, so we do not discuss it further.
Creating indices. The Time 1 actual attitude items were inter-correlated, r(199) = .72, p
< .001, so we averaged them into a Time 1 attitude index as in prior experiments. Time 2 ideal
and ought attitudes correlated, r(199) = .75, p < .001; their average represented desired attitudes.
Results
Manipulation check: Desired attitudes. As shown in Table 5, Time 1 actual attitudes
were negatively related to desired attitudes; a likely explanation is that this pattern is due to Time
2 actual attitudes also being in the model and correlating moderately with Time 1 actual
attitudes, r(199) = .34, p < .001. Indeed, this association vanishes when examined as a zero-order
correlation, r(199) = .06, p = .422. Thus, the ‘negative’ effect of Time 1 actual attitudes on (Time
2) desired attitudes reflects only Time 1 actual attitudes’ residual variance after accounting for
the effect of the more proximal Time 2 actual attitudes. Replicating all past studies, more
positive Time 2 actual attitudes were substantially related to positive desired attitudes at Time 2.
A substantial main effect of painter identity directly replicates Experiment 2. Like our prior
experiments, more positive desired attitudes were formed given a moral versus immoral artist.
Table 5. Painter Identity Effects on Desired Attitude Formation Controlling for Prior Actual Attitude. (Experiment 3).
Predictor Variable Parameter Coefficients
Painter Identity b = 2.31 [1.82, 2.81],

t(197) = 9.23, p < .001, rsp = .37
Time 1 Actual Attitude b = -.20 [-.38, -.03],

t(197) = -2.29, p = .023, rsp = -.09
Time 2 Actual Attitude b = .59 [.46, .71],

t(197) = 9.07, p < .001, rsp = .36
Model Statistics F(3, 197) = 145.30,

p < .001, R2 = .69
Actual attitude change. We next turned to the critical test of Experiment 3: an
interaction effect between desired attitudes and the presence/absence of metacognitive prompts.
We were primarily interested in whether the simple slope of desired attitudes on Time 3 actual
attitudes remained significant in the prompt-absent condition specifically. If so, this would
suggest that prompting participants was not necessary for actual attitude change. Results are
noted in Table 6. First, positive main effects of Time 1 and Time 2 actual attitude were detected,
which simply indicate that people who initially liked the painting also liked it later. We also
replicated the main effect of desired attitudes, suggesting that aggregating across metacognitive
prompt conditions, a typical participant ended up liking (disliking) the painting more when they
had earlier desired to like (dislike) it more. Most critically, the interaction effect is non-
significant. To probe our main question, we broke down the slopes of desired attitudes predicting
actual attitude change at each level of metacognitive prompting. Because the interaction test is
non-significant, there is not sufficient evidence to conclude that these slopes are different from
one another. Desired attitudes predicted actual attitude change whether metacognitive prompts
were present, b = .42 [.30, .53], t(195) = 7.29, p < .001, rsp = .29, or absent, b = .33 [.21, .44],
t(195) = 5.38, p < .001, rsp = .18. This strongly supports our contention that our decision to
include metacognitive prompts was not a necessary condition for people to self-persuade in a
direction prognosticated by their desired attitudes.7
7
Ideal and ought attitudes produced somewhat different results when analyzed separately. Ideal attitudes
X metacognitive prompts produced an interaction effect, b = .14 [.01, .26], t(195) = 2.18, p = .030, rsp
= .09, such that ideal attitudes predicted greater attitude change when prompts were included, b = .37
[.26, .47], t(195) = 6.94, p < .001, rsp = .29, versus when they were absent, b = .23 [.13, .33], t(195) =
4.41, p < .001, rsp = .19. Ought attitudes X metacognitive prompts led to no such interaction, b = .06
[-.07, .18], t(195) = .88, p = .382, rsp = .04. Supporting our main contention, however, desired attitudes,
whether measured as “ideal” or “ought,” robustly predicted attitude change in all analyses/breakdowns.
Table 6. Effects of Desired Attitudes on Actual Attitude Change (Experiment 3).
Independent Variables Parameter Coefficients

t(195) = 2.85, p = .005, rsp = .12

t(195) = 5.51, p < .001, rsp = .23
Time 2 Desired Attitude b = .37 [.27, .47],

t(195) = 7.57, p < .001, rsp = .31
Metacognitive Prompts b = -.01 [-.33, .31],

t(195) = -.05, p = .957, rsp = .00
Time 2 Desired Attitude X b = .09 [-.04, .22],

Metacognitive Prompts t(195) = 1.39, p = .165, rsp = .06
Model Statistics F(5, 195) = 78.42,

p < .001, R2 = .67
Discussion
Experiment 3 examined one objection that might be raised regarding the prior
experiments through a single procedural variation: half of participants undertook the
metacognitive prompts (as in earlier experiments) whereas another half did not complete these
prompts. This addresses the possibility that the self-persuasion effects documented in
Experiments 1-2 (people’s actual attitude change tending to follow their desired attitude position)
occurred only because we facilitated these effects with our metacognitive prompts. Strongly
contradicting this counter-explanation, our effect emerged significantly (ps < .001) even when all
metacognitive prompting was removed. These findings might suggest that people engage in
metacognitive reflections without prompting, or might suggest that people in this condition used
tactics other than metacognitive reflection to self-persuade. Maio and Thomas (2007) provide
examples of such processes, of which some (e.g., teleologic strategies) might be more easy to
complete in the narrow window of opportunity provided in the Metacognitive Prompts Absent
condition (e.g., concentrating on information facilitating pursuit of the desired attitude, e.g.,
Robbins, 1991) without using more difficult, epistemically-driven self-justifications.
What is less clear is whether metacognitive prompting at least contributed to more
attitude change, given that our focal interaction of prompting X desired attitudes on actual
attitudes was non-significant, but a significant interaction emerged for prompting X ideal
attitude. Although not a central issue for our present case, a larger sample size might have
generated a significant interaction, which would be interesting from the perspective of
understanding how these self-persuasion effects emerge most strongly. In sum, Experiment 3
suggests that given a sufficiently strong motivational foundation for desired attitudes, people
may self-persuade towards the position of their desired attitudes even without any prompting.
General Discussion
Five datasets provided consistent evidence that desired attitudes have important
consequences for the actual attitude change that people undergo. Desired attitudes can be
understood as attitude “guides.” That is, people seem able to persuade themselves towards their
desired attitude positions, even in the absence of compelling new information about the object
(DeMarree et al., 2017, Study 2), and even without the ability to modify the object to align it
with their desires (DeMarree et al., 2017, Study 4). Indeed, although most of our experiments
allowed participants to engage in metacognitive reflection about their thoughts (Experiments 1-
2), Experiment 3 revealed that such encouragement at most helped but was certainly not
necessary for participants to self-persuade towards their desired attitudes.
Theoretical Implications
Desired attitudes’ consequences for actual attitude change. The present experiments
shed light on several gaps in the growing literature on desired attitudes (for a review, see
Wheeler & DeMarree, 2019). Most crucially, our work addresses whether desired attitudes can
direct actual attitude change. The experiments provided consistent evidence that desired attitudes
indeed predict actual attitude change. Thus, in response to a situational pressure to like or dislike
an object, people immediately change their actual attitudes. However, the information may also
induce a separate desire to like/dislike the object, which prognosticates the actual attitude they
will reach after a few minutes of deliberation. Previous models have posited that people direct
their own attitude change towards desired attitudes (e.g., Maio & Thomas, 2007) but here we
directly demonstrate this effect. We show this result in two distinct ways: (1) in the main
manuscript, we show that Time-2 desired attitudes predict Time-3 actual attitudes adjusting for
Time-2 actual attitudes; and (2) in the supplementary SOM-4, people’s explicit comparisons
(how much more/less they want to like the object, compared to how much they actually do)
replicate these patterns.
Reviews often conclude that variables often have only small and unreliable effects in
shaping attitude change, even in laboratory contexts that afford substantial experimental control
(r = .10-.11; Albaraccin & Shavitt, 2018; Tannenbaum et al., 2015). Despite this, we found that
desired attitudes consistently could predict subsequent attitude change. Furthermore, this actual
attitude change was detected in contexts where people were not provided any additional
opportunity to attain actual attitude change by obtaining a biased collection of new information
or to engage in biased processing of additional information (see DeMarree et al., 2017); and
participants could not modify objective properties of the object to improve it (see Wheeler &
DeMarree, 2019). A skeptical reader may reason that between Time 2 and Time 3, in the present
paradigms, participants were continuing to react to the identity-relevant information from just
before the Time 2 attitude ratings. However, the consistent pattern of desired attitudes predicting
the direction and magnitude of that delayed reaction nonetheless remains interesting, suggesting
that those delayed reactions to the information are motivated, in that they literally are predictable
from participants’ stated desires. This extended self-persuasion even occurred when we provided
them with no help at all, that is, when we ceased to even ask them to reflect on their thoughts
(prompt-absent condition in Experiment 3). Despite all these barriers, people were robustly able
to self-persuade in a direction forecasted by their desired attitudes (rs = .07-.31 across five
datasets). In short, beyond the previously captured consequences of desired attitudes (that they
shape people’s pursuit of new information, or physical modifications of objects), mere thinking
can facilitate people’s changing their opinions towards their desired attitude positions.
Interestingly, then, our results suggest an important degree of self-awareness among
participants. Whereas past scholars have been clear that people are not aware of their self-
persuasion processes (e.g., that they will become more attitudinally extreme merely by thinking
more; Tesser, 1978; also see Dijksterhuis, 2004), we show that people can be consciously aware
of the direction that their future actual attitude change will move in. This is not as paradoxical as
it may sound. It has long been recognized that people being unaware of their attitude change
processes/tendencies (i.e., towards polarization; Tesser, 1978) does not entail that they are
unaware of what attitudes they hold (Nisbett & Wilson, 1977). Similarly, people may be quite
aware of their attitudinal goals (i.e., desired attitudes), which in fact may effectively predict the
direction of their subsequent attitude change (shown in all five present experiments), even if they
could not explain how they talked themselves into that attitude change. Indeed, this identification
of desired attitude goals may be quite instrumental to self-persuasion. Many self-directed attitude
change strategies (Maio & Thomas, 2007) and thought control strategies (Wenzlaff & Wegner,
2000) are cognitively demanding, so it makes sense that people’s organization of effective and
proportionate self-persuasion (e.g., to avoid “over-correcting,” e.g., Sommers & Kassin, 2001)
would benefit from a clear goal-state (i.e., “I want to like the painting slightly [versus
moderately; extremely] more.”).
One potential objection involves our examination of Time 2 desired attitudes as
predictors of Time 3 attitudes (in Experiments 1-3), in which we controlled for Time 2 actual
attitudes. Like all measures, attitude scales are prone to error – hence, the Time 2 actual attitude
measure did not capture all variance in opinions at Time 2. If we failed to adequately control for
Time 2 actual attitudes, Time 2 desired attitudes might have only appeared to predict Time 3
actual attitudes, but the effect might have been driven by Time 2 desired attitudes’ overlap with
Time 2 actual attitudes (insufficiently controlled because Time 2 actual attitudes were
imperfectly measured; see Bollen, 1989). However, several arguments render this inadequate
partialling objection less compelling. First, this measurement error criticism is not specific to the
present work and is applicable to any analysis that includes a covariate, or partials for variance in
one variable when examining the influence of a second variable (ANCOVAs, regressions, etc.).
Our studies are no more susceptible to this problem than the large part of the scientific literature
that makes claims based on this broad range of analyses. Second, our attitude measures were
generally very high in internal consistency (inter-item correlations rs > .70; αs > .90), making it
less likely that the existence of random error introduced substantial distortions to our estimates of
effects and that our attitude change effects were thus purely reflective of incomplete partialling
of effects. Finally, we conducted supplementary analyses using structural equation modeling to
further assess the potential impact of measurement error. In these analyses, we specified actual
and desired attitudes as latent variables so that we could estimate relations among these measures
after having removed the effects of random error. These analyses did not produce any
substantive differences in our findings, with all critical effects remaining significant (SOM-6).
The above arguments mostly assume random measurement error. One might instead
object that our briefer Time 1 and Time 2 actual attitude measures, combined with the longer
Time 3 actual attitude measure, artificially inflated actual attitude change at Time 3 through
systematic measurement error. For instance, if participants felt relatively constrained to give
similar responses to the “like”/“likeable” items at Time 1 and Time 2, the Time 3 measure’s
novel items (e.g., “positive”) might have allowed participants to express attitude change that
really occurred earlier. This interpretation implies that the actual attitude change should largely
evaporate if actual attitudes at Time 3 are measured with just the “like” item. We re-analysed the
self-persuasion with this substitution, using an integrative dataset that combined all of our
datasets. As predicted, desired attitudes still significantly predicted actual attitude change.
Therefore, the attitude change observed at Time 3 in our studies cannot be explained away by the
final attitude measure introducing new items.
One alternative interpretation of our findings involves attitude polarization from mere
thought (e.g., Tesser, 1978). After receiving the matching/mismatching information about the
politician (Time 2), people responded by immediately changing their actual attitudes to be
positive/negative, respectively. A polarization explanation might argue that later, given more
time to think (i.e., by Time 3), people might have polarized these actual attitudes due to the mere
thought phenomenon, with matching (mismatching) people becoming more positive (negative) in
their views. Desired attitudes might have shifted more quickly to be consistent with participants’
cognitive schemas (Tesser, 1978) about political groups, with actual attitudes taking longer to
become schema-consistent. Desired attitudes then predicted later attitude change because they
changed more extremely in the same direction that actual attitudes would later shift. However,
the data is not in line with this explanation. Although Time 2 desired attitudes shifted more
strongly than did Time 2 actual attitudes in some data sets (Experiment 2-3), Time 2 desired
attitudes changed less strongly than did Time 2 actual attitudes in other data (Experiment 1b),
and there was no difference in yet others (Experiment 1a). Thus, established processes of attitude
polarization cannot consistently account for our effects. We provide complete details in SOM-5.
Furthermore, it would be beneficial to replicate our effects among other populations. Our
experiments generalized effectively across Canadian and American University students, as well
as a broader (and older) sample of American volunteers on ResearchMatch. Thus, the modest
diversity of sampling we employed failed to undermine our effects, and we are optimistic that
our effects would emerge in other types of adult populations.
Identity-relevant desired attitudes. Additionally, both the political (Experiment 1) and
moral (Experiment 2-3) paradigms supported some conjectures about the origins of desired
attitudes. That is, past scholarship raised the idea that consistency with a person’s social
identities and values might be possible sources of desired attitudes (DeMarree et al., 2014;
Wheeler & DeMarree, 2019), but no empirical work has investigated these links directly. The
present studies provide experimental evidence that links between an attitude object and a
person’s social identity or relevant values does, in fact, affect people’s desired attitudes. Thus,
just as people form actual attitudes for a variety of motivated reasons (Herek, 1986, 1987; Katz,
1960; Petty & Wegener, 1998; Shavitt, 1990; Smith et al., 1956; Snyder & DeBono, 1985), both
utilitarian (DeMarree et al., 2017) and social/value relevant bases (the present work) can shape
desired attitude formation.
Limitations and Future Directions
Much work on desired attitudes (DeMarree et al., 2014, 2017; the present work) focuses
on situations where actual and desired attitudes diverge, and how people then reconcile those
differences. This research has successfully identified numerous objects (e.g., the self, African-
Americans, exercise) for which most people experience discrepancies, suggesting that people
cannot always reconcile actual and desired attitudes. Thus, the phenomenon we have identified
here clearly is not in universal operation, or else actual/desired discrepancies would always
eliminate themselves. The present paradigms were designed specifically to capture cases where
desired attitudes have the greatest potential to predict actual attitude change. We did this by
ensuring that actual attitudes were formed without deep connections to pre-existing values,
identities, or attitudes (e.g., an unfamiliar painting) but created desired attitudes richly connected
to existing attitudes (e.g., hatred of Nazis). Future research should address when desired attitudes
do (not) predict actual attitude change. Effects like ours should primarily be anticipated when
desired attitudes are “stronger” than actual attitudes. Additionally, predictions may be drawn
from the cognitive dissonance literature, in which conflicting cognitive elements may be
resolved via a range of processes including trivialization (Simon et al., 1995) or bolstering
(Sherman & Gorkin, 1980). For example, if one’s actual attitude is very strong, and one
encounters information entailing an opposite-valence desired attitude, people may trivialize the
new information (e.g., form a weak desired attitude) or bolster their original actual attitude,
instead of changing their actual attitude to match the desired attitude. Returning this to our
opening example, people with especially strong positive opinions of Rowling’s books may
trivialize her disliked Tweets (avoiding forming a negative desired attitude towards her books).
This mention of process also brings us to a second important limitation: we did not
directly provide a mechanism of how actual attitudes are changing in the present work, only that
desired attitudes predict such change. Maio and Thomas’s (2007) work on self-persuasion may
provide guidance here. Their epistemic-teleologic model maps two types of process by which
people change their own opinions: epistemic (attempting to form new valid attitude) and
teleologic processes (attitude change with minimal concerns for validity). For instance, epistemic
tactics include motivated interpretation (e.g., shifting judgment of a negative attribute so that it
may be viewed as a positive), whereas teleologic tactics include distraction and suppression (i.e.,
putting undesired information out of awareness). Importantly, Maio and Thomas (2007) posited
that deliberate self-persuasion is activated when attitude-holders recognize their own
actual/desired attitude discrepancies. Thus, their proposed processes may guide future research
that shows how people accomplish the attitudinal shifts that we have documented. Their model
also suggests that far from those more specific processes “replacing” desired attitudes as an
explanation, these variables should work in tandem (i.e., contexts elicit desired attitudes
discrepant from actual attitudes, and these desired attitudes guide contextually appropriate
epistemic and teleologic or other processes permit people to shift their actual attitudes; also see
Rucker et al., 2011, for a discussion of reconciling multiple mediators).
One methodological limitation is that although in Experiment 3 we eliminated all
metacognitive prompts for half of participants, these participants nonetheless completed the
subjective discrepancy measure (see SOM-5). It is possible that this single item, which asks
participants whether (or not) they want to hold a more positive or negative attitude than they
currently do, might have made this discrepancy more salient and accessible to participants. This
could be important because ambivalent attitudinal elements may be more psychologically
impactful for people when they are high (vs low) in accessibility (Newby-Clark et al., 2002).
In many contexts actual and desired attitudes will align, and the present findings
represent several circumstances in which people’s actual attitudes shift in directions indicated by
their desired attitudes. One interesting possibility is that such actual/desired attitude agreement
may itself be important. For example, actual attitudes that do shift to match desired attitudes may
confer strength properties (Krosnick & Petty, 1995) to the actual attitude, for example by
reducing the subjective ambivalence associated with that attitude (DeMarree et al., 2014). Thus,
stimulating people to align their actual and desired attitudes, or making salient such existing
overlaps, may bolster actual attitude strength. On the other hand, desired attitudes also may vary
in strength properties; that is, that some people may have more relatively more important,
certain, and/or accessible (strong) desired attitudes, relative to others (also see discussion of
desired attitude commitment; DeMarree and colleagues, 2017). Potentially, both actual and
desired attitudes may be weakened (strengthened) when the valence of those attitudes conflict
(agree). Past evidence already hints at this phenomenon for actual attitudes at least (DeMarree et
al., 2014, 2016), but such effects may be bolstered by making both types of attitudes salient to
people, and dampened by not making each attitude salient. Thus, future research on actual and
desired attitudes may stimulate richer understanding of both constructs, and how they relate in
nuanced and interesting ways.
Open Practices
For all experiments, we report how we determined our sample size, all data exclusions,
all manipulations, and all measures. Verbatim materials are included in the Supplementary
Online Materials (SOM-1), and data/syntax are available (https://tinyurl.com/desiredattitude).
Experiment 1b was preregistered: (https://osf.io/7wgrh/?
view_only=e85ef18892d64615afe1b9cf20852bbc).
References
Albarracin, D., & Shavitt, S. (2018). Attitudes and attitude change. Annual Review of
Psychology, 69(1), 299-327.
Benoit, W. L., & Dorries, B. (1996). Dateline NBC's persuasive attack on Wal‐
Mart. Communication Quarterly, 44(4), 463-477.
Blankenship, K. L., Wegener, D. T., & Murray, R. A. (2012). Circumventing resistance: Using
values to indirectly change attitudes. Journal of Personality and Social
Psychology, 103(4), 606-621.
Bollen, K. A. (1989). A new incremental fit index for general structural equation models.
Sociological Methods & Research, 17, 303-316.
Briñol, P., McCaslin, M. J., & Petty, R. E. (2012). Self-generated persuasion: effects of the target
and direction of arguments. Journal of Personality and Social Psychology, 102(5), 925-
940.
Briñol, P., Petty, R. E., Stavraki, M., Lamprinakos, G., Wagner, B., & Díaz, D. (2018). Affective
and cognitive validation of thoughts: An appraisal perspective on anger, disgust, surprise,
and awe. Journal of Personality and Social Psychology, 114(5), 693-718.
Bruchmann, K., Koopmann-Holm, B., & Scherer, A. (2018). Seeing beyond political affiliations:
The mediating role of perceived moral foundations on the partisan similarity-liking
effect. PloS One, 13(8), e0202101.
Byrne, D. (1961). Interpersonal attraction as a function of affiliation need and attitude
similarity. Human Relations, 14(3), 283-289.

Carrera, P., Caballero, A., Muñoz, D., & Fernández, I. (2017). Abstractness leads people to base
their behavioral intentions on desired attitudes. Journal of Experimental Social
Psychology, 70, 27–33.
Carver, C. S., & Scheier, M. F. (1998). On the self-regulation of behavior. Cambridge University
Press.
Clark, J. K., Wegener, D. T., & Fabrigar, L. R. (2008). Attitudinal ambivalence and message-
based persuasion: Motivated processing of proattitudinal information and avoidance of
counterattitudinal information. Personality and Social Psychology Bulletin, 34(4), 565-
577.
Clarkson, J. J., Tormala, Z. L., & Leone, C. (2011). A self-validation perspective on the mere
thought effect. Journal of Experimental Social Psychology, 47(2), 449-454.
Cooper, J. (2007). Cognitive dissonance: Fifty years of a classic theory. Sage.
Crites, S. L., Jr., Fabrigar, L. R., & Petty, R. E. (1994). Measuring the affective and cognitive
properties of attitudes: Conceptual and methodological issues. Personality and Social
Psychology Bulletin, 20(6), 619-634.
DeMarree, K. G., Clark, C. J., Wheeler, S. C., Briñol, P., & Petty, R. E. (2017). On the pursuit of
desired attitudes: Wanting a different attitude affects information processing and
behavior. Journal of Experimental Social Psychology, 70, 129-142.
DeMarree, K. G., & Rios, K. (2014). Understanding the relationship between self-esteem and
self-clarity: The role of desired self-esteem. Journal of Experimental Social
Psychology, 50, 202-209.
DeMarree, K. G., Wheeler, S. C., Briñol, P., & Petty, R. E. (2014). Wanting other attitudes:
Actual–desired attitude discrepancies predict feelings of ambivalence and ambivalence
consequences. Journal of Experimental Social Psychology, 53, 5-18.
DeMarree, K. G., Rios, K., Randell, J. A., Wheeler, S. C., Reich, D. A., & Petty, R. E. (2016).
Wanting to be different predicts nonmotivated change: Actual–desired self-discrepancies
and susceptibility to subtle change inductions. Personality and Social Psychology
Bulletin, 42(12), 1709-1722.
Dijksterhuis, A. (2004). Think different: the merits of unconscious thought in preference
development and decision making. Journal of Personality and Social Psychology, 87(5),
586-598.
Downing, J. W., Judd, C. M., & Brauer, M. (1992). Effects of repeated expressions on attitude
extremity. Journal of Personality and Social Psychology, 63(1), 17-29.
Duggan, J. (2022). Transformative readings: Harry Potter fan fiction, trans/queer reader
response, and JK Rowling. Children's Literature in Education, 53(2), 147-168.
Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Harcourt.
Faul, F., Erdfelder, E., Buchner, A., & Lang, A.-G. (2009). Statistical power analyses using
G*Power 3.1: Tests for correlation and regression analyses. Behavior Research
Methods, 41, 1149-1160.
Festinger, L. (1957). A theory of cognitive dissonance. Stanford University Press.
Graham, J., Haidt, J., & Nosek, B. (2009). Liberals and conservatives use different sets of moral
foundations. Journal of Personality and Social Psychology, 96, 1029–1046.
Harmon-Jones, E., & Mills, J. (2019). An introduction to cognitive dissonance theory and an
overview of current perspectives on the theory. In E. Harmon-Jones (Ed.), Cognitive

dissonance: Reexamining a pivotal theory in psychology (2nd ed., pp. 3-24). American
Psychological Association.
Heider, F. (1946). Attitudes and cognitive organization. Journal of Psychology, 21, 107–112.
Herek, G. M. (1986). The instrumentality of attitudes: Toward a neofunctional theory. Journal of
Social Issues, 42(2), 99-114.
Herek, G. M. (1987). Can functions be measured? A new perspective on the functional approach
to attitudes. Social Psychology Quarterly, 285-303.
Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review,
94, 319-340.
Higgins, E. T. (1989). Self-discrepancy theory: What patterns of self beliefs cause people to
suffer? In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 22, pp.
93-136). Academic Press.
Hogg, M. A., & Ridgeway, C. L. (2003). Social identity: Sociological and social psychological
perspectives. Social Psychology Quarterly, 97-100.
Houston, D. A., & Fazio, R. H. (1989). Biased processing as a function of attitude accessibility:
Making objective judgments subjectively. Social Cognition, 7(1), 51-66.
Katz, D. (1960). The functional approach to the study of attitudes. Public Opinion
Quarterly, 24(2), 163-204.
Krosnick, J. A., & Petty, R. E. (1995). Attitude strength: An overview. In R. E. Petty & J. A.
Krosnick (Eds.), Attitude strength: Antecedents and consequences (pp. 1–24). Erlbaum.
Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The
effects of prior theories on subsequently considered evidence. Journal of Personality and
Social Psychology, 37(11), 2098-2109.

Maio, G. R., & Haddock, G. (2009). The psychology of attitudes and attitude change. Sage.
Maio, G. R., & Thomas, G. (2007). The epistemic-teleologic model of deliberate self-persuasion.
Personality and Social Psychology Review, 11(1), 1–22.
McCarthy, M. (2019). If Nazi = Red, and Canadian = Red, does Red = Good or Bad? Testing the
effects of valenced perceptual cues on Implicit Association Test performance. Retrieved
from: https://psyarxiv.com/fsjba/
Newby-Clark, I. R., McGregor, I., & Zanna, M. P. (2002). Thinking and caring about cognitive
inconsistency: When and for whom does attitudinal ambivalence feel
uncomfortable?. Journal of Personality and Social Psychology, 82(2), 157-166.
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on
mental processes. Psychological Review, 84(3), 231-259.
Ostrom, T. M. (1989). Interdependence of attitude theory and measurement. In A. R. Pratkanis,
S. J. Breckler, & A. G. Greenwald (Eds), Attitude structure and function (pp. 11-36).
Lawrence Erlbaum.
Petty, R. E., Briñol, P., & Tormala, Z. L. (2002). Thought confidence as a determinant of
persuasion: the self-validation hypothesis. Journal of Personality and Social
Psychology, 82(5), 722-741.
Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and persuasion: Classic and contemporary
approaches. Brown.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion. In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 19, pp. 123-205).
Academic Press.
Petty, R. E., & Wegener, D. T. (1998). Matching versus mismatching attitude functions:
Implications for scrutiny of persuasive messages. Personality and Social Psychology
Bulletin, 24(3), 227-240.
Petty, R. E., Wegener, D. T., & Fabrigar, L. R. (1997). Attitudes and attitude change. Annual
Review of Psychology, 48(1), 609-647.
Requero, B., Briñol, P., & Petty, R. E. (2020). The impact of hope and hopelessness on
evaluation: A meta‐cognitive approach. European Journal of Social Psychology.
Advance Online Publication, retrieved from https://onlinelibrary-wiley-
com.proxy.queensu.ca/doi/abs/10.1002/ejsp.2726.
Robbins, A. (1991). Awake the giant within. Fireside.
Rowling, J. K. (2020). J.K. Rowling Writes About Her Reasons for Speaking Out on Sex and
Gender Issues. Accessed September 3, 2020 from https://www.jkrowling.com/opinions/j-
k-rowling-writesabout-her-reasons-for-speaking-out-on-sex-and-gender-issues/
Rucker, D. D., Preacher, K. J., Tormala, Z. L., & Petty, R. E. (2011). Mediation analysis in
social psychology: Current practices and new recommendations. Social and Personality
Psychology Compass, 5(6), 359–371.
Shavitt, S. (1990). The role of attitude objects in attitude functions. Journal of Experimental
Social Psychology, 26(2), 124-148.
Sherman, S. J., & Gorkin, L. (1980). Attitude bolstering when behavior is inconsistent with
central attitudes. Journal of Experimental Social Psychology, 16(4), 388-403.
Simon, L., Greenberg, J., & Brehm, J. (1995). Trivialization: the forgotten mode of dissonance
reduction. Journal of Personality and Social Psychology, 68(2), 247-260.
Smith, M. B., Bruner, J. S., & White, R. W. (1956). Opinions and personality. Wiley.
Snyder, M., & DeBono, K. G. (1985). Appeals to image and claims about quality: Understanding
the psychology of advertising. Journal of Personality and Social Psychology, 49(3), 586-
597.
Sommers, S. R., & Kassin, S. M. (2001). On the many impacts of inadmissible testimony:
Selective compliance, need for cognition, and the overcorrection bias. Personality and
Social Psychology Bulletin, 27(10), 1368-1377.
Tajfel, H. (1974). Social identity and intergroup behaviour. Social Science Information, 13(2),
65-93.
Tannenbaum, M. B., Hepler, J., Zimmerman, R. S., Saul, L., Jacobs, S., Wilson, K., &
Albarracín, D. (2015). Appealing to fear: A meta-analysis of fear appeal effectiveness
and theories. Psychological Bulletin, 141(6), 1178-1204.
Tesser, A. (1978). Self-generated attitude change. In Advances in experimental social
psychology (Vol. 11, pp. 289-338). Academic Press.
Tesser, A., & Conlee, M. C. (1975). Some effects of time and thought on attitude
polarization. Journal of Personality and Social Psychology, 31(2), 262.
Tesser, A., & Leone, C. (1977). Cognitive schemas and thought as determinants of attitude
change. Journal of experimental social psychology, 13(4), 340-356.
Thurstone, L. L., & Chave, E. J. (1929). The measurement of attitude. University of Illinois
Press.
Vonasch, A. J., Reynolds, T., Winegard, B. M., & Baumeister, R. F. (2018). Death before
dishonor: Incurring costs to protect moral reputation. Social Psychological and
Personality Science, 9(5), 604-613.

Wenzlaff, R. M., & Wegner, D. M. (2000). Thought suppression. Annual Review of
Psychology, 51(1), 59-91.
Wheeler, S. C., & DeMarree, K. G. (2019). Prevalence, antecedents and consequences of actual-
desired attitude discrepancies. In A. Reed II & M. Forehand (Eds.), Handbook of
Research on Identity Theory in Marketing (pp. 346-359). Edward Elgar Publishing.

This supplementary document contains verbatim materials for each study (SOM-1), additional
tables (SOM-2), analysis of behavior intention measures that closely matches the findings for
attitudes measures (SOM-3), analysis using an alternative desired attitude measure that closely
matches the findings for the desired attitude measure in the main text (SOM-4), analyzing a
“polarization” counter-explanation of our patterns (SOM-5), a re-analysis of most experiments
using structural equation modeling rather than regression (SOM-6), and complete statistics for
analyses with various quality control checks implemented (SOM-7). Finally, SOM-8 provides a
breakdown of all means and standard deviations for each level of each experimental variable, for
each experiment.
Contents
SOM-1: Verbatim Materials............................................................................................................4
SOM-1A: Verbatim Materials for Experiment 1a.......................................................................4
SOM-1B: Verbatim Materials for Experiment 1b.....................................................................17
SOM-1C: Verbatim Materials for Experiment 2a and 2b.........................................................22
SOM-1D: Verbatim Materials for Experiment 3.......................................................................31
SOM-2: Additional Tables............................................................................................................32
Table S2-1..................................................................................................................................32
Table S2-2..................................................................................................................................33
Table S2-3..................................................................................................................................34
Table S2-4..................................................................................................................................35
Table S2-5..................................................................................................................................36
SOM-3: Behavioral Intentions Measure........................................................................................37
Behavioral Intentions Measurement..........................................................................................37
Table S3-1..................................................................................................................................39
SOM-4: Subjective Desired/Actual Attitude Discrepancy Measure.............................................40
Subjective discrepancy measurement........................................................................................40
Experiment 1a Formation of Subjective Discrepancies.............................................................40
Experiment 1b Formation of Subjective Discrepancies............................................................41
Table S4-1..............................................................................................................................42
Figure S4-1............................................................................................................................43
Experiment 1a/1b Actual Attitude Change Effects...................................................................44
Table S4-2..............................................................................................................................45
Experiment 2a Formation of Subjective Discrepancies.............................................................46
Experiment 2b Formation of Subjective Discrepancies............................................................46

Table S4-3..............................................................................................................................47
Experiment 2a/2b Actual Attitude Change Effects...................................................................48
Table S4-4..............................................................................................................................49
Internal Meta-Analysis of Actual Attitude Change Incremental Effects (Experiment 1-2)......50
Figure S4-2............................................................................................................................52
Experiment 3 Formation of Subjective Discrepancies..............................................................54
Table S4-5..............................................................................................................................55
Experiment 3 Actual Attitude Change Effects..........................................................................56
Table S4-7..............................................................................................................................57
SOM-5: Analysis of the Polarization Explanation........................................................................58
Experiment 1a/1b.......................................................................................................................58
Experiment 2a/2b.......................................................................................................................59
Experiment 3..............................................................................................................................60
SOM-6: Structural Equation Modeling.........................................................................................61
SOM-6a: Experiment 1a............................................................................................................61
Table S6a-1............................................................................................................................61
Table S6a-2............................................................................................................................61
SOM-6b: Experiment 1b............................................................................................................62
Table S6b-1............................................................................................................................62
Table S6b-2............................................................................................................................62
SOM-6c: Experiment 2a............................................................................................................63
Table S6c-1............................................................................................................................63
Table S6c-2............................................................................................................................63
SOM-6d: Experiment 2b............................................................................................................64
Table S6d-1............................................................................................................................64
Table S6d-2............................................................................................................................64
SOM-6e: Experiment 3..............................................................................................................65
Table S6e-1............................................................................................................................65
Table S6e-2............................................................................................................................65
SOM-7: Do Effects Change if People are Removed for Quality-related Reasons?......................66
Effects of Manipulation on Desired Attitudes...........................................................................66
Effects of Desired Attitudes on Actual Attitude Change...........................................................67

SOM-8: Tables of Means and Standard Deviations for all Experiments......................................68
Experiment 1a............................................................................................................................68
Experiment 1b............................................................................................................................68
Experiment 2a............................................................................................................................69
Experiment 2b............................................................................................................................69
Experiment 3..............................................................................................................................70
SOM-1: Verbatim Materials

SOM-1A: Verbatim Materials for Experiment 1a
Initial Demographics.
What is your religious identity?

(Note that for the sake of this question, 'agnostic' or 'atheist' are also religious identities.)
{Text box}
What is your political orientation?

Very conservative (1)
Moderately conservative (2)
Slightly conservative (3)
Slightly liberal (4)
Moderately liberal (5)
Very liberal (6)
Please provide your age in the following box, as a numeral, e.g.: 24

{Text box}
Please provide a brief description of your personality in your own words. Try to give a few
examples of how you express your personality - in other words, 'who you are', and how you
come across to others.
{Text box}
Initial Information about Mrs. Bakker.
Welcome to a very different study. This study is about gaining an impression of a person.
In this study you will be asked to think about a particular individual from one of several
occupations or social roles. You will then be asked to give some thoughts about it.
Next, we would like you to read about: Mrs. Bakker, a political candidate.

The following is a transcript taken from an interview with Mrs. Bakker, a political
candidate in Ontario, Canada, conducted by Paul Simons, a journalist. Please read the
following information carefully, and then record your thoughts about this individual in the
space below.
SIMONS: Welcome to the Podcast, and uh - so for our listeners who may be less familiar with
your background, why don't we start with your personal background. Who is Jane Bakker?
BAKKER: Right! Isn't that the question - [they both laugh] - but that's what we are all trying to
figure out, no? Well, I am married. My husband's name is John, and we have two children
together. Flint is eight, Kyla seven. I am born and raised in Ontario, from a line of mostly
farmers, so I'm the first in my family to be getting into politics. Other than my father, on a
smaller scale.
SIMONS: Okay, awesome. And now, I guess in a similar vein, but who is Jane Bakker when it
comes to politics. Or what do you bring to -
BAKKER: Integrity.
SIMONS: Right, and so some people identify you with - well, for want of a better word,
cleaning up office, is that a fair characterization?
BAKKER: Well, certainly, I spent a lot of effort cleaning house, so to speak, in my first days in
office. I got involved in internal investigations of my party's deceptive practices and corruption.
A lot of small scale stuff, but I believe in a policy of paying attention to details.
SIMONS: They add up.
BAKKER: Well, there's that. And I have sharp instinct for dishonesty. I think that's in my
nature. So I was good at it.
SIMONS: And you are well-liked by voters, is that fair to say?
BAKKER: Well, I don't know about fair - [both laugh] - but I would like to say it. No, joking
aside, I do think I have a solid base at this point. People respond to honesty, is my point of view.
SIMONS: I understand you are interested in advocating for some policies in the educational
system.
BAKKER: That's right.
SIMONS: And so I know that one issue that people have brought up, and you'll forgive me for
saying this, but some people bring up that you are a smoker and -
BAKKER: Well -
SIMONS: - and so some people suggest that this does not set a very good role-model for
children. I wondered if you had any comment.
BAKKER: Well, not to that. I mean, I take a clear stance on smoking and, for example, drugs.
And having the occasional cigarette, as a choosing adult, is not relevant to my views on the
educational front.
SIMONS: But people suggest, in terms of role-modeling -
BAKKER:No, I hear what you are saying, but my personal life is a separate matter, and the fact
that I do it - well, the expression is "do as I say, not as I do," right? I don't claim to be perfect.
SIMONS: And what do you think you do that makes you a good role-model? For kids, I mean.
We have a lot of parents who tune -
BAKKER: - Well, I am known for contributing to charities, and working hard to make sure
these charities stay afloat. I think that trumps the odd cigarette. And people know I do not come
from family wealth. These are things I model. Caring about others.
SIMONS: One other issue that I think our listeners will care about. You have been noted in the
press at times, particularly by newspapers, as being antagonistic to journalists.
BAKKER: Right, you better watch out. [Simons laughs]. But seriously, I think that the press is
unnecessary in the modern era. Or at least in the typical sense. People can decide what they think
of issues for themselves, and the media's biases simply get in the way. On top of that, informal
news systems work just as well.
SIMONS: I know sometimes this has become a heated issue. So there was that incident
involving the journalist from ARP who claims that you pushed him.
BAKKER: Phew! That a hard day. You know, when the media crowds you, I don't normally
have a problem with it. It is just part of the process of being in the public limelight. But this
journalist was being unprofessional. He was pressing so close that his camera was touching me.
A very relentless interviewer, and frankly - he crossed professional lines there. I didn't push him.
I just kept walking, and he was getting into my path and collided with me. As far as I'm
concerned, he was acting very unprofessionally and I do blame him for that incident.
SIMONS: I see. Well, we are running out of time today, but any final thoughts to share with the
public?
BAKKER:Absolutely. I am excited to be able to represent Ontario. I have a lot of political and
legal knowledge that will enable me to do this skillfully, and my record clearly shows that I
would do it honestly and professionally. Ontario voters should keep the name Jane Bakker in
mind for the next election. And thank you Paul for this interview. Not bad, even for a biased
media - [both laugh].
SIMONS: But seriously, I appreciated your time today.
BAKKER: As did I. Thank you.
Time 1 Thought Listing and Attitudes.
Please write three negative thoughts about Mrs. Bakker.
1. ________________________________________________
2. ________________________________________________
3. ________________________________________________
Please write three positive thoughts about Mrs. Bakker.
1. ________________________________________________
2. ________________________________________________
3. ________________________________________________
Finally, considering both this person's strengths and weaknesses, how do you perceive this
person overall?
Very unlikeable (1)
(2)
(3)
(4)
(5)
(6)
Very likeable (7)
Finally, considering both this person's strengths and weaknesses, how do you perceive this
person overall?
Dislike him/her (1)
(2)
(3)
(4)
(5)
(6)
Like him/her (7)
Experimental manipulations: Liberal vs Conservative Politician.
In this next section, we would like you to read a passage about Mrs. Jane Bakker.

This is the same person who you just read about and commented on.

Please read the full, brief passage CLOSELY. Afterwards, you will be asked for your opinion of
the person being described.
Liberal Condition
Mrs. Jane Bakker is known to be a moderately "left-wing" politician, both in terms of social
policy and economics. This means that she strongly supports equality or egalitarianism, and
believes that society has many unjustified inequalities that should be fixed through political
action. To give some examples of her left-wing politics, here is a brief list of the political
initiatives that she has either directly supported, or alternatively, given public support for (i.e.,
during media appearances, or during campaign speeches).
 Strongly supported women's right to choose in debates about abortion and body rights,
i.e., she is pro-choice.
 Supported additional taxation on the rich, and in particular, stated she supported a
proposed policy to increase taxes on the top 1% most wealthy Canadians.
 Mrs. Bakker is opposed to initiatives regarding private insurance, and is instead a
supporter of a public health care system.
 Recently, a political candidate suggested that prisons should be reformed to be more
focused on rehabilitation, with decreased focus on punishment. The candidate stated that
"Prisons are getting out of control! Prisoners view it as a trap that keeps them in a cycle
of violence. We need to make prisons places for preparation for society". Mrs. Bakker
supported this statement and signed her name on a resultant petition to the Federal
government.
 During discussion involving indigeneous rights on Canadian campuses, Mrs. Bakker
suggested that faculty should be "encouraged or mandated" to take courses
concerning indigenous wisdom.
Conservative Condition
Mrs. Jane Bakker is known to be a moderately "right-wing" politician, both in terms of social
policy and economics. This means that she strongly supports order and tradition, and believes
that society necessarily has some forms of hierarchy that are healthy and natural. To give some
examples of her right-wing politics, here is a brief list of the political initiatives that she has
either directly supported, or alternatively, given public support for (i.e., during media
appearances, or during campaign speeches).

 Strongly supported restricting abortion within debates about abortion and body rights,
i.e., she is pro-life.
 Opposed additional taxation on the rich, and in particular, stated she opposed a proposed
policy to increase taxes on the top 1% most wealthy Canadians.
 Mrs. Bakker favors initiatives regarding private insurance, and opposes public health
care systems.
focused on punishment, with decreased focus on rehabilitation. The candidate stated that
"Prisons are getting out of control! Prisoners view it as an easy vacation with free meals.
We need to make prisons actually punishing again." Mrs. Bakker supported this
statement and signed her name on a resultant petition to the Federal government.
 During discussion involving indigeneous rights on Canadian campuses, Mrs. Bakker
suggested that faculty should "not be bullied and pressured" into taking courses
Time 2 Questionnaires
You will now be asked some questions about the politician, Mrs. Bakker. This is the same
person that you read about and commented on before.
Actual attitude.
Please consider your overall attitude towards Mrs. Bakker, and express it on the following
scale, where '1' indicates that you strongly dislike Mrs. Bakker, and '9' indicates that you
strongly like Mrs. Bakker.
(1) Strongly dislike
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) Strongly like
Desired attitude.
Sometimes the attitudes we have are different from the attitudes we ideally would like to have or
the attitudes we feel we should hold, and sometimes these attitudes are the same. For your
opinion of Mrs. Bakker, please indicate the attitude you IDEALLY would like to have. On this
scale, a '1' indicates that you would IDEALLY want to strongly dislike Mrs. Bakker; a '9'
indicates that you would IDEALLY want to strongly like Mrs. Bakker.
(1) Would IDEALLY strongly dislike
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) would IDEALLY strongly like
For your opinion of Mrs. Bakker, please indicate the attitude you SHOULD or OUGHT
to have. On this scale, a '1' indicates that you would SHOULD want to strongly dislike Mrs.
Bakker; a '9' indicates that you SHOULD want to strongly like Mrs. Bakker.
(1) OUGHT TO strongly dislike
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) OUGHT TO strongly like
Subjective discrepancies.
Comparing to how I actually feel about Mrs. Bakker, I want to:
(1) Dislike it much more
(2)
(3)
(4) Not dislike or like it any more
(5)
(6)
(7) Like it much more
Metacognitive Prompts.
In this next section, we would like you to consider the thoughts you previously listed about
the politician, Mrs. Bakker. Before, you were listing thoughts based on whether they were
negative thoughts or positive thoughts.
Now we would like you to consider how likeable your thoughts, by themselves, are to you.
Sometimes, people like to have specific thoughts ("I like the fact that I like myself"), sometimes
people dislike the thoughts they have ("I hate the fact that I doubted my friend"), and sometimes
they feel quite neutral about their thoughts. We would like you to rate how favorable
(enjoyable, nice to have) each of your thoughts is to you.
In other words, do you like having each thought, or do you dislike having each thought?
You said: {Computer feeds back first thought}
(1) I don’t like this thought about it
(2)
(3)
(4)
(5)
(6)
(7) I like this thought about it
{This question then repeats for the other five thoughts that participants listed.}
Now we would like you to consider how confident you are in your thoughts.
Specifically, thought confidence refers to how valid, accurate, and correct your thoughts are, and
as a result, how certain and sure you are in those thoughts.
For example, both person A and person B could think "self-esteem is a good thing". But person
A might have high confidence in this thought ("I'm 100% confident that self-esteem is a good
thing because I've read lots about it"), whereas person B might have low confidence in this
thought ("I think it's a good thing, but I don't know much about it so I'm not sure").
For the following questions, we would like you to rate how confident you are in each of your
thoughts.
(1) I'm not at all confident in this thought
(2)
(3)
(4)
(5)
(6)
(7) I'm highly confident in this thought
Now we would like you to consider how clear you feel each thought is to you.
Sometimes, people have very clear thoughts, sometimes people do not have very clear thoughts,
or are neutral about this.
We would like you to rate how clear each of your thoughts is to you.

(1) I do not feel at all clear about this thought
(2)
(3)
(4)
(5)
(6)
(7) I feel highly clear about this thought
Time 3 Actual attitude and behavioral intentions.

Please let us know your reactions to the politician, Mrs. Bakker. Read the following items, and
decide whether each describes your reactions to her, on a scale from 1 (not at all) to 7
(definitely).
1
7
Not at 2 3 4 5 6
Definitely
all
      
Good
      
Like
      
Positive
      
Favorable
      
Bad
      
Dislike
      
Negative
      
Unfavorable
Finally, please let us know about your intentions regarding the politician, Mrs. Bakker. Read the
following items, and decide whether you agree with each, on a scale from 1 (not at all) to 7
(definitely).
1
7
Not at 2 3 4 5 6
Definitely
all
I would vote
for Mrs.
Bakker to be
my political
      
representative.
I would sign a
petition
supporting       
Mrs. Bakker.
I would be
willing to
publicly state
that I support
      
Mrs. Bakker.
I would tell
people not to
vote for Mrs.       
Bakker.
I would sign a
petition
opposing Mrs.       
Bakker.
I would avoid
voting for
Mrs. Bakker
as a political
      
representative.
SOM-1B: Verbatim Materials for Experiment 1b
For Experiment 1b, we re-used the materials from Experiment 1a, except where otherwise noted
below.
Initial Information about Mrs. Bakker.
The current study is about gaining an impression of a person. In this study, you will be asked to
think about a particular individual from one of several occupations or social roles. You will then
be asked to give some thoughts about it.

This person may come from a number of backgrounds. When reading the materials associated
with your target person, just try to get an impression of who he/she is so that you can tell us your
thoughts after.
Next, we would like you to read about: Mrs. Bakker, a political candidate.

The following is a transcript taken from an interview with Mrs. Bakker, a political
candidate in New York, conducted by Paul Simons, a journalist. Please read the following
information carefully, and then record your thoughts about this individual in the space
below.

SIMONS: Welcome to the Podcast, and uh - so for our listeners who may be less familiar with
your background, why don't we start with your personal background. Who is Jane Bakker?

BAKKER: Right! Isn't that the question - [they both laugh] - but that's what we are all trying to
figure out, no? Well, I am married. My husband's name is John, and we have two children
together. Flint is eight, Kyla seven. I am born and raised in New York, from a line of mostly
farmers, so I'm the first in my family to be getting into politics. Other than my father, on a
smaller scale.

SIMONS: Okay, awesome. And now, I guess in a similar vein, but who is Jane Bakker when it
comes to politics. Or what do you bring to -

BAKKER: Integrity.

SIMONS: Right, and so some people identify you with - well, for want of a better word,
cleaning up office, is that a fair characterization?

BAKKER: Well, certainly, I spent a lot of effort cleaning house, so to speak, in my first days in
office. I got involved in internal investigations of my party's deceptive practices and corruption.
A lot of small scale stuff, but I believe in a policy of paying attention to details.

SIMONS: They add up.

BAKKER: Well, there's that. And I have sharp instinct for dishonesty. I think that's in my
nature. So I was good at it.

SIMONS: And you are well-liked by voters, is that fair to say?

BAKKER: Well, I don't know about fair - [both laugh] - but I would like to say it. No, joking
aside, I do think I have a solid base at this point. People respond to honesty, is my point of view.

SIMONS: I understand you are interested in advocating for some policies in the educational
system.

BAKKER: That's right.

SIMONS: And so I know that one issue that people have brought up, and you'll forgive me for
saying this, but some people bring up that you are a smoker and -

BAKKER: Well -

SIMONS: - and so some people suggest that this does not set a very good role-model for
children. I wondered if you had any comment.

BAKKER: Well, not to that. I mean, I take a clear stance on smoking and, for example, drugs.
And having the occasional cigarette, as a choosing adult, is not relevant to my views on the
educational front.

SIMONS: But people suggest, in terms of role-modeling -

BAKKER:No, I hear what you are saying, but my personal life is a separate matter, and the fact
that I do it - well, the expression is "do as I say, not as I do," right? I don't claim to be perfect.

SIMONS: And what do you think you do that makes you a good role-model? For kids, I mean.
We have a lot of parents who tune -

BAKKER: - Well, I am known for contributing to charities, and working hard to make sure
these charities stay afloat. I think that trumps the odd cigarette. And people know I do not come
from family wealth. These are things I model. Caring about others.

SIMONS: One other issue that I think our listeners will care about. You have been noted in the
press at times, particularly by newspapers, as being antagonistic to journalists.

BAKKER: Right, you better watch out. [Simons laughs]. But seriously, I think that the press is
unnecessary in the modern era. Or at least in the typical sense. People can decide what they think
of issues for themselves, and the media's biases simply get in the way. On top of that, informal
news systems work just as well.

SIMONS: I know sometimes this has become a heated issue. So there was that incident
involving the journalist from ARP who claims that you pushed him.

BAKKER: Phew! That a hard day. You know, when the media crowds you, I don't normally
have a problem with it. It is just part of the process of being in the public limelight. But this
journalist was being unprofessional. He was pressing so close that his camera was touching me.
A very relentless interviewer, and frankly - he crossed professional lines there. I didn't push him.
I just kept walking, and he was getting into my path and collided with me. As far as I'm
concerned, he was acting very unprofessionally and I do blame him for that incident.

SIMONS: I see. Well, we are running out of time today, but any final thoughts to share with the
public?

BAKKER:Absolutely. I am excited to be able to represent New York. I have a lot of political
and legal knowledge that will enable me to do this skillfully, and my record clearly shows that I
would do it honestly and professionally. New York voters should keep the name Jane Bakker in
mind for the next election. And thank you Paul for this interview. Not bad, even for a biased
media - [both laugh].

SIMONS: But seriously, I appreciated your time today.

BAKKER: As did I. Thank you.

Experimental manipulations: Liberal vs Conservative Politician.
In this next section, we would like you to read a passage about Mrs. Jane Bakker.

This is the same person who you just read about and commented on.


Liberal Condition
Mrs. Jane Bakker is known to be a moderately "left-wing" politician, both in terms of social
policy and economics. This means that she strongly supports equality or egalitarianism, and
believes that society has many unjustified inequalities that should be fixed through political
action. To give some examples of her left-wing politics, here is a brief list of the political
initiatives that she has either directly supported, or alternatively, given public support for (i.e.,
during media appearances, or during campaign speeches).

 Strongly supported women's right to choose in debates about abortion and body rights,

i.e., she is pro-choice.

 Supported additional taxation on the rich, and in particular, stated she supported a
proposed policy to increase taxes on the top 1% most wealthy Americans.

 Mrs. Bakker is opposed to initiatives regarding private insurance, and is instead a
supporter of a public health care system.

focused on rehabilitation, with decreased focus on punishment. The candidate stated that
"Prisons are getting out of control! Prisoners view it as a trap that keeps them in a cycle
of violence. We need to make prisons places for preparation for society". Mrs. Bakker
supported this statement and signed her name on a resultant petition to the Federal
government.

 During discussion involving indigenous rights on American campuses, Mrs. Bakker
suggested that faculty should be "encouraged or mandated" to take courses
Conservative Condition
Mrs. Jane Bakker is known to be a moderately "right-wing" politician, both in terms of social
policy and economics. This means that she strongly supports order and tradition, and believes
that society necessarily has some forms of hierarchy that are healthy and natural. To give some
examples of her right-wing politics, here is a brief list of the political initiatives that she has
either directly supported, or alternatively, given public support for (i.e., during media
appearances, or during campaign speeches).

 Strongly supported restricting abortion within debates about abortion and body rights,
i.e., she is pro-life.

 Opposed additional taxation on the rich, and in particular, stated she opposed a
proposed policy to increase taxes on the top 1% most wealthy Americans.

 Mrs. Bakker favors initiatives regarding private insurance, and opposes public health
care systems.

focused on punishment, with decreased focus on rehabilitation. The candidate stated that
"Prisons are getting out of control! Prisoners view it as an easy vacation with free meals.
We need to make prisons actually punishing again." Mrs. Bakker supported this
statement and signed her name on a resultant petition to the Federal government.

 During discussion involving indigenous rights on American campuses, Mrs. Bakker

suggested that faculty should "not be bullied and pressured" into taking courses

Quality check items
We included four quality control tests as follows:
1. Providing sufficient thoughts (minimum 1 negative, 1 positive)

2. Showing suspicion on a probe.
3. Mentioning issues with the survey.
4. Attention check item.
Item #1 has no wording: we simply examined whether they had provided enough thoughts to
fulfill the criterion.
Item #2: What do you think the current study is about?
Item #3: If you have any feedback regarding this research, please type in the space below (any
feedback will be greatly appreciated!).
Item #4: What did the politician, Mrs. Bakker and the reporter talk about in the interview
transcript that you read earlier (please write down anything that you remember from the
interview transcript)?

SOM-1C: Verbatim Materials for Experiment 2a and 2b

Welcome to a very different study. This study is about artwork.
In this study you will be asked to view a particular painting by a classic painter. You will then be
asked to give some thoughts about it.
{Page break}
In this task, you will be asked to view a painting.
You can imagine that you are in an art gallery, and this painting is on the wall.
As you view the painting, we ask that you provide your thoughts about it in the box beneath it.
Click "okay" if you understand.
{Page break}
Painting (“Agony” by Arshile Gorky).

Please write three negative thoughts about the painting.

1. ________________________________________________
2. ________________________________________________
3. ________________________________________________
Please write three positive thoughts about the above painting.

1. ________________________________________________
2. ________________________________________________
3. ________________________________________________
Finally, considering both the painting's strengths and weaknesses, how do you perceive this
painting overall?
(1) (very unlikeable)
(2)
(3)
(4)
(5)
(6)
(7) (very likeable)
Finally, considering both the painting's strengths and weaknesses, how do you perceive this
painting overall?
(1) (dislike it)
(2)
(3)
(4)
(5)
(6)
(7) (like it)
Painter Identity Manipulation Conditions.
In this next section, we would like you to read a passage about a painter who lived in the 1940s.
This is the same painter who created the painting you just viewed and commented on.
Immoral Painter Condition

The painter, Bruno Steingard, lived during the 1940s, around the Second World War.
He lived in Munich, Germany during World War II, supporting the Nazi party as far as possible.
Although few know about this part of history, the Nazi party often celebrated artists like Bruno
Steingard, perceiving abstract painting as “invigorating” and consistent with the party’s
motivations. Bruno was greatly celebrated by many leading Nazis, and at times actually
completed paintings in the households of members of the Nazi party’s leadership. During this
time period, Munich was often being bombed by the American air force, and the city was often
ravaged, with many buildings being destroyed. Thanks to the protection of Nazi leaders,
Steingard’s paintings were left intact during this period, protected in secure facilities. After the
American occupation of Munich he was able to have them restored of any damage by art experts,
many of whom were unfamiliar with their political history. Steingard kept silent after the war
about his history with the Nazis, but praised their philosophy and actions in private journals, and
expressed his regret that he had not participated more actively in the Nazi party’s actions. He
was thrilled by the Nazi’s worldview, as some Germans had been during this time period.
Moral Painter Condition

The painter, Bruno Steingard, lived during the 1940s, around the Second World War.
He lived in Munich, Germany during World War II, avoiding the Nazi party as far as possible.
Although few know about this part of history, the Nazi party often persecuted artists like Bruno
Steingard, perceiving abstract painting as “decadent” and disloyal to the party’s motivations.
Bruno managed to avoid detection by completing his works in secret, at times actually storing
canvases behind walls or under floorboards. During this time period, Munich was often being
bombed by the American air force, and the city was often ravaged, with many buildings being
destroyed. By chance, Steingard’s paintings were left intact during this period, and during the
American occupation of Munich he was able to have them restored of any damage. Steingard
spoke out after the war, criticizing the Nazi party, and expressing his regret that he had not done
more to resist their presence in Germany. He was intimidated into silence, as many Germans had
been during this time period.
Time 2 Questionnaires.
You will now be asked some questions about the painting made by Bruno Steingard. This is the
painting that you viewed and commented on before, that was created by the painter you just read
about.
Actual attitude.
Please consider your overall attitude towards Bruno Steingard's painting, and express it on the
following scale, where '1' indicates that you strongly dislike Bruno Steingard's painting, and '9'
indicates that you strongly like Bruno Steingard's painting.
(1) (strongly dislike)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) (strongly like)
Desired attitude.
Sometimes the attitudes we have are different from the attitudes we ideally would like to have or
the attitudes we feel we should hold, and sometimes these attitudes are the same.
For your opinion of Bruno Steingard's painting, please indicate the attitude you IDEALLY
would like to have. On this scale, a '1' indicates that you would IDEALLY want to strongly
dislike Bruno Steingard's painting; a '9' indicates that you would IDEALLY want to strongly
like Bruno Steingard's painting.
(1) (would IDEALLY strongly dislike)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) (would IDEALLY strongly like)
For your opinion of Bruno Steingard's painting please indicate the attitude you feel you
SHOULD or OUGHT to have. On this scale, a '1' indicates that you SHOULD strongly
dislike Bruno Steingard's painting; a '9' indicates that you SHOULD strongly like Bruno
Steingard's painting.
(1) (OUGHT TO strongly dislike)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9) (OUGHT TO strongly like)
Subjective discrepancies.
Comparing to how I actually feel about Bruno Steingard's painting, I want to
(1) (dislike it much more)
(2)
(3)
(4)
(5)
(6)
(7) (like it much more)
Metacognitive Prompts.
Bruno Steingard's painting. Before, you were listing thoughts based on whether they were
Now we would like you to consider how likeable your thoughts, by themselves, are to you.
Sometimes, people like to have specific thoughts ("I like the fact that I believe in liberalism /
conservatism"), sometimes people dislike the thoughts they have ("I hate the fact that I doubted
my friend"), and sometimes they feel quite neutral about their thoughts.
We would like you to rate how favorable (enjoyable, nice to have) each of your thoughts is
to you.
{Computer feeds back first thought}

In other words, do you like having each thought, or do you dislike having each thought?
(1) (I don’t like this thought about it)
(2)
(3)
(4)
(5)
(6)
(7) (I like this thought about it)
Now we would like you to consider how confident you are in your thoughts.
Specifically, thought confidence refers to how valid, accurate, and correct your thoughts are, and
as a result, how certain and sure you are in those thoughts.
For example, both person A and person B could think "universal healthcare is a good thing". But
person A might have high confidence in this thought ("I'm 100% confident that universal
healthcare is a good thing because I've read a lot about it"), whereas person B might have low
confidence in this thought ("I think it's a good thing, but I don't know much about it so I'm not
sure").
For the following questions, we would like you to rate how confident you are in each of your
thoughts.
{Computer feeds back first thought}
(1) (I’m not confident at all in this thought)
(2)
(3)
(4)
(5)
(6)
(7) (I’m highly confident in this thought)
Now we would like you to consider how clear you feel each thought is to you.
Sometimes, people have very clear thoughts, sometimes people do not have very clear thoughts,
or are neutral about this.
We would like you to rate how clear each of your thoughts is to you.

{One thought is listed here}
(1) (I do not feel at all clear about this thought)
(2)
(3)
(4)
(5)
(6)
(7) (I feel highly clear about this thought)
Time 3 Attitude Measures.

As of right now, please consider the following words, and decide whether or not they accurately
describe your attitude towards the Bruno Steingard painting. If you the word does ‘not at all’
describe your attitude towards the painting, select '1'. If it 'definitely' describes your attitude
towards the painting, select '9'.
{Item matrix as per Experiment 1a/1b}
To what extent did you feel confused about the questions you were asked in the previous sections
of the survey?
(1) Not at all
(2) -
(3) -
(4) -
(5) Extremely
In your understanding, what are the purposes of this study?
{Text box}
Did anything seem strange or unusual to you about this study? If so, please indicate below.
{Text box}
SOM-1D: Verbatim Materials for Experiment 3
{As per Experiment 2, except that half of participants were not exposed to the metacognitive
prompts.}
SOM-2: Additional Tables

Table S2-1.
Descriptive Information for Study Variables. (Experiment 1a).
1 2 3 4 5 6 7 8 M (SD)
1. T1 .89 4.04 (1.0

Actual
Attitude
2. T2 .33*** -- 4.73 (2.5
Actual
Attitude
3. T2 .22*** .84*** .90 5.10 (2.4
Desired
Attitudes
4. T2 .07 .63*** .69*** -- 4.04 (1.4
Subjective
Discrepancy
5. T3 .41*** .87*** .78*** .57*** .97 3.83 (1.6
Actual
Attitude
6. T3 .24*** .88*** .83*** .63*** .88*** .94 3.70 (1.8
Behavioral
Intention
7. Political .00 -.05 -.03 .01 -.04 -.04 -- 4.34 (1.1
Orientation
8. Political .06 -.05 -.03 -.04 -.08 -.03 .62*** -- 1.78 (.70
Extremity
Note. * p < .05, ** p < .01, *** p < .001. M = mean, SD = standard deviation. Values on the
diagonal reflect Cronbach’s alpha (internal consistency) coefficients. The -- symbol denotes
single-item scales for which internal consistency scores are not calculated.
Table S2-2.
Descriptive Information for Study Variables. (Experiment 1b).
1 2 3 4 5 6 7 8 M (SD)
1. T1 .87 4.37 (1.0

Actual
Attitude
2. T2 .34*** -- 4.93 (1.9
Actual
Attitude
3. T2 .26*** .78*** .84 5.24 (1.8
Desired
Attitudes
4. T2 .23*** .70*** .65*** -- 4.06 (1.3
Subjective
Discrepancy
5. T3 .34*** .83*** .69*** .68*** .95 4.19 (1.3
Actual
Attitude
6. T3 .16*** .79*** .68*** .63*** .83*** .89 3.88 (1.5
Behavioral
Intention
7. Political .07 -.09* -.04 -.10** -.12** -.10** -- 3.70 (1.2
Orientation
8. Political .01 -.04 -.07 -.06 -.09* -.10** -.14*** -- 1.60 (.69
Extremity
Table S2-3.
Descriptive Information for Study Variables. (Experiment 2a).
1 2 3 4 5 M (SD)
1. T1 Actual .84 3.67 (1.36)
Attitude
2. T2 Actual .56*** -- 4.51 (2.39)

Attitude
3. T2 Desired .26*** .75*** . 4.87 (2.55)

Attitudes
4. T2 .26*** .69*** .82*** -- 3.74 (1.88)

Subjective
Discrepancy
5. T3 Actual .50*** .81*** .75*** .67*** .90 4.29 (2.16)

Attitude
Table S2-4.
Descriptive Information for Study Variables. (Experiment 2b).
1 2 3 4 5 M (SD)
1. T1 Actual .93 3.31 (1.59)
Attitude
2. T2 Actual .77*** -- 3.98 (2.36)

Attitude
3. T2 Desired .27*** .60*** .88 4.73 (2.56)

Attitudes
4. T2 .13** .46*** .82*** -- 3.85 (1.87)

Subjective
Discrepancy
5. T3 Actual .65*** .80*** .59*** .46*** .93 4.25 (2.13)

Attitude
Table S2-5.
Descriptive Information for Study Variables. (Experiment 3).
1 2 3 4 5 M (SD)
1. T1 Actual .83 3.78 (1.25)
Attitude
2. T2 Actual .34*** -- 4.52 (2.07)

Attitude
3. T2 Desired .06 .72*** .85 4.82 (2.49)

Attitudes
4. T2 -.07 .58*** .80*** -- 3.91 (1.99)

Subjective
Discrepancy
5. T3 Actual .28*** .75*** .75*** .68*** .93 4.45 (1.97)

Attitude
SOM-3: Behavioral Intentions Measure
Behavioral Intentions Measurement

Participants in Experiments 1a and 1b also rated how likely they would be to engage in
three positive behaviors (voting for, signing a petition supporting, publicly stating approval of
the politician) and three negative behaviors towards the politician (telling others not to vote for,
signing a petition opposing, avoid voting for the politician). All items were rated from 1 (not at
all) to 7 (definitely). After reverse-coding negative behaviors, these items formed an internally
consistent (α = .94) scale, higher scores indicating supportive behavioral intentions towards the
politician. As we noted in the main text, results with the behavioral intentions mirrored the Time
3 actual attitudes results very closely, so we relegated these results to this supplementary section.
To assess determinants of participants’ Time 3 behavioral intentions, we regressed
behavioral intentions onto Time 1 and Time 2 actual attitudes, Time 2 desired attitudes, and
Time 2 subjective discrepancies. In addition, we also controlled for Time 3 actual attitudes in
this analysis to determine if desired attitudes could account for behavioral intentions above and
beyond the degree to which desired attitudes created actual attitude change. Starting with the first
data row, more positive Time 1 actual attitudes paradoxically predicted more negative behavioral
intentions. A simple explanation for this finding is that the Time 1 actual attitudes effect is only
the residual after adjusting for the more proximal actual attitudes measures (Time 2 and Time 3)
in this analysis.8 Time 2 and Time 3 actual attitudes were each positively related to Time 3
behavioral intentions, which is easily understandable: people who liked Mrs. Bakker (before and
after cognitive restructuring) were more likely to support her with their actions. But more
8
Indeed, a zero-order correlation test between Time 1 actual attitudes and Time 3 behavioral
intentions revealed a positive coefficient, r(322) = .24, p < .001.
critically, more positive Time 2 desired attitudes were significantly linked with more positive
behavioral intentions. Thus, desired attitudes helped to incrementally predict participants’ later
intentions to support versus oppose Mrs. Bakker through actions like voting, petitioning, and
word-of-mouth. As in the attitude change analysis, however, subjective discrepancies were not
incrementally linked with behavioral intentions. This regression analysis provides a conceptual
replication of past work linking desired attitudes (above/beyond actual attitudes) with behavioral
intentions. Once again it highlights desired attitudes rather than people’s subjective sense of how
much change they want, which relates to these intentions.

Table S3-1. Effects of Desired Attitudes and Subjective Discrepancies on Time 3 Behavioral Intentions. (Experiment 1a).
Independent Variables Experiment 1a Experiment 1b
Time 1 Actual Attitude B = -.17 [-.26, -.09], B = -.22 [-.28, -.17],

t(319) = -4.14, p < .001 t(691) = -7.53, p < .001
Time 2 Actual Attitude B = .25 [.18, .32], B = .23 [.17, .29],

t(319) = 6.77, p < .001 t(691) = 7.28, p < .001
Time 2 Desired Attitude B = .15 [.09, .22], B = .08 [.03, .13],

t(319) = 4.76, p < .001 t(691) = 3.12, p < .001
Time 3 Actual Attitude B = .54 [.43, .65], B = .67 [.59, .75],

t(319) = 9.82, p < .001 t(691) = 16.82, p < .001
Model Statistics F(4, 319) = 443.84, p < .001, R2 = .85 F(4, 691) = 511.81, p < .001, R2 = .75
Note. Multiple R2 coefficients reflect variance in the dependent variable accounted for by the entire regression model.
SOM-4: Subjective Desired/Actual Attitude Discrepancy Measure
Subjective discrepancy measurement
In Experiments 1-3, we also provided participants with an explicitly comparative item to
assess their subjective feeling about wanting a different attitude. We refer to this as participants’
subjective discrepancies because they are the participants’ direct judgment about how much
more/less they would want to like the object.
 In Experiment 1, participants responded to the statement, “Comparing to how I actually
feel about Mrs. Bakker, I want to…”, with response options: 1 (dislike her much more) to
4 (not dislike her or like her any more) to 7 (like her much more).
 In Experiments 2-3, participants responded to the statement, “Comparing to how I
actually feel about the painting, I want to…”, with response options: 1 (dislike it much
more) to 4 (not dislike it or like it any more) to 7 (like it much more).
Results were generally supportive of the same conclusions that we reached using the desired
attitude composite measure (i.e., averaging ideal and ought attitudes, as in much prior literature).
For space reasons, we thus moved these results here (to SOM-4) despite their supporting our
contentions.
Experiment 1a Formation of Subjective Discrepancies

We regressed subjective discrepancies on the same predictor variables as we examined in
the main text’s look at predictors of the desired attitude composite. Effects are displayed in Table
S4-1. We found about a one-unit difference between matching conditions (for participants of
average political extremity, represented by B = 1.18), with matching-passage participants
wanting to adopt more positive attitudes than mismatching-passage participants.
Crucially, just like with the desired attitude composite variable, matching condition and
political extremity interacted to qualify this effect, with matching condition having a relatively
weak effect on subjective discrepancies among the least extreme participants, B = .55 [.14, .96],
t(318) = 2.64, p = .009, rsp = .13, but a much larger effect among politically extreme participants,
B = 1.82 [1.41, 2.22], t(318) = 8.81, p < .001, rsp = .44. This matched our hypothesis. We also
provided figures for this interaction (for both Experiments 1a and 1b) below, noted as Figures
S4-1 and S4-2.
Experiment 1b Formation of Subjective Discrepancies
We also examined the same regression using the Experiment 1b data. A main effect of
matching condition revealed a four-fifths of a unit difference between matching conditions (B
= .80) for average-level politically extreme participants, with matching-passage participants
wanting to adopt more positive attitudes than mismatching-passage participants. The matching
condition and political extremity interaction was again significant with matching condition
having only a modest effect on subjective discrepancies among the least politically extreme
participants, B = .37 [.14, .61], t(683) = 3.09, p = .002, rsp = .11, and larger effects among
politically extreme participants, B = 1.29 [1.03, 1.54], t(683) = 10.02, p < .001, rsp = .35.
In short, matching our reasoning, people for whom the congruence of the politician’s
party with their own would have been more closely aligned and perhaps most identity-relevant
were, indeed, more impacted by the politician’s ideological views.

Table S4-1. Moderation of Matching Condition on Subjective Desired-Actual Attitude Discrepancy Formation by Participant
Political Extremity. (Exp. 1).
Matching Condition B = 1.18 [.90, 1.47], t(318) = 8.09, p < .001 B = .80 [.62, .98], t(683) = 8.82, p < .001
Time 1 Actual Attitude B = .07 [-.06, .20], t(318) = 1.02, p = .307 B = .28 [.19, .36], t(683) = 6.51, p < .001
Political Extremity B = -.10 [-.31, .11], t(318) = -.95, p = .341 B = -.08 [-.21, .05], t(683) = -1.26, p = .210
Matching Condition X B = .91 [.50, 1.32], t(318) = 4.34, p < .001 B = .71 [.45, .97], t(683) = 5.36, p < .001
Participant Political Extremity
Figure S4-1. Subjective Discrepancies Influenced by Levels of Political Matching and

Participant Political Extremity.
C. Experiment 1a
D. Experiment 1b
Note. Errors bars reflect standard error values. Higher scores reflect more positive desired
attitudes.
Experiment 1a/1b Actual Attitude Change Effects
As one additional analysis, we tried to test if there was any support for the idea that
desired attitudes as captured by the composite measure, versus desired attitudes as captured by
subjective discrepancies, might show incremental validity. This might indicate, for instance, that
the two map onto slightly different constructs, or at least that there is utility in considering both.
We therefore regressed Time 3 actual attitudes onto Time 1 and Time 2 actual attitudes, Time 2
desired attitudes, and Time 2 subjective discrepancies all at once. The actual attitude measures in
these models were included as covariates so that Time 3 actual attitudes captured attitude
change.
In both datasets, Time 1 and Time 2 actual attitudes (unsurprisingly) related positively to
Time 3 actual attitudes, suggesting that people’s pre-manipulation and post-manipulation
attitudes influenced their later attitudes even after some cognitive restructuring. More critically,
Time 2 desired attitudes were positively linked with Time 3 actual attitudes. Thus, in addition to
how much people actually liked the politician, how much they desired to like her explained
additional attitude change. Subjective discrepancies only predicted attitude change in
Experiment 1b, but not Experiment 1a.
There are several possible explanations for the inconsistent performance of subjective
discrepancies. Possibly the most crucial is that Experiment 1b (which detected an effect) had
more than double the sample size of Experiment 1a (which did not capture an effect).
Table S4-2. Effects of Desired Attitudes and Subjective Discrepancies on Changes in Actual Attitudes at Time 3. (Exp. 1).
Independent Variables Parameter Coefficients (Exp. 1a) Parameter Coefficients (Exp. 1b)
Time 1 Actual Attitude B = .22 [.14, .31], t(319) = 5.36, p < .001, B = .08 [.03, .13], t(691) = 2.90, p = .004
Time 2 Actual Attitude B = .40 [.34, .46], t(319) = 12.78, p < .001 B = .42 [.37, .47], t(691) = 16.91, p < .001
Time 2 Desired Attitude B = .15 [.08, .21], t(319) = 4.27, p < .001 B = .05 [.003, .10], t(691) = 2.08, p = .038,
Time 2 Subjective B = .01 [-.07, .09], t(319) = .17, p = .866 B = .18 [.12, .24], t(691) = 5.91, p < .001
Discrepancy
Experiment 2a Formation of Subjective Discrepancies

We also measured subjective discrepancies in Experiment 2. Once again, we began by
checking if our manipulation that was designed to influence desired attitudes (here, the painter’s
moral identity) would also affect subjective discrepancies. Unlike Experiment 1, this was tested
with a main effect rather than an interaction test (as explained in the main text).
Time 1 actual attitudes were only marginally related to Time 2 discrepancies (see Table
S4-3). More importantly, when people learned the painting was made by a Nazi victim, they
subjectively estimated that they wanted to hold an attitude about two-and-a-half scale units more
positive than when the painting was made by a Nazi sympathizer. This conceptually replicates
findings from the politician experiments but in the novel context of a moral contagion pressure.
It also replicates our own Experiment 2a effect wherein we showed that the painter’s moral
identity influenced the composite measure of desired attitudes, by showing a parallel effect on
subjective discrepancies.
Experiment 2b Formation of Subjective Discrepancies

We performed the same test on Experiment 2b’s data. Results were extremely similar to
Experiment 2a; indeed, the unstandardized coefficient tracking the impact of the painter’s moral
identity on subjective discrepancies (wanting to like the painting more/less) was almost exactly
the same as in Experiment 2a.

Table S4-3. Painter Identity Effects on Subjective Desired-Actual Attitude Discrepancy Formation (Exp. 2a).
Independent Variable Parameter Coefficients (Exp. 2a) Parameter Coefficients (Exp. 2b)
Passage B = 2.44 [1.93, 2.95], t(114) = 9.42, p < .001 B = 2.46 [2.22, 2.70], t(511) = 20.04, p < .001
Time 1 Actual Attitude B = .19 [-.001, .38], t(114) = 1.97, p = .051 B = .15 [.08, .23], t(511) = 3.93, p < .001
Experiment 2a/2b Actual Attitude Change Effects
Once again, we looked at how desired/subjective might mutually predict attitude change.
This is tracked in Table S4-4. The results indicated that although the composite measure of
desired attitudes robustly was related to subsequent actual attitude change, subjective
discrepancies were not.
However, it is somewhat notable that across all four datasets (Experiments 1a, 1b, 2a,
2b), subjective discrepancies’ usually non-significant relationship to attitude change was
directionally/descriptively positive. For this reason, we conducted an internal meta-analysis on
all four datasets.
Table S4-4. Effects of Desired Attitudes and Subjective Discrepancies on Actual Attitude Change (Experiment 2).
Time 1 Actual B = .24 [.04, .44], t(112) = 2.41, p = .017 B = .28 [.17, .40], t(500) = 4.77, p < .001
Attitude
Time 2 Actual B = .40 [.23, .56], t(112) = 4.70, p < .001 B = .45 [.36, .55], t(500) = 9.52, p < .001
Attitude
Time 2 Desired B = .28 [.11, .44], t(112) = 3.35, p = .001 B = .15 [.07, .23], t(500) = 3.55, p < .001
Attitude
Time 2 Subjective B = .07 [-.13, .27], t(112) = .71, p = .480 B = .07 [-.03, .17], t(500) = 1.37, p = .170
Discrepancy
Model Statistics F(4, 112) = 71.70, p < .001, R2 = .72 F(5, 499) = 208.88, p < .001, R2 = ..68
Internal Meta-Analysis of Actual Attitude Change Incremental Effects (Experiment 1-2)
The meta-analysis was designed to test if desired attitudes and/or subjective discrepancies
incrementally influenced actual attitude change. For desired attitudes the answer was obviously
affirmative because this coefficient was statistically significant in all experiments, but I reasoned
that estimating the effect size would provide a convenient comparison against subjective
discrepancies. Conveniently, the effect size we have used throughout, rsp or semipartial-r, is
specifically designed for meta-analysis involving regressions with varying numbers of
parameters, and its formulas also produce a variance statistic weighted by sample size (Aloe &
Becker, 2012). In each of the cases below, we used the metafor package (Viechtbauer, 2010) for
R (R Development Core Team, 2014). We used a fixed effects model approach due to the low
number of relevant studies per analysis, but the results with random effects modeling is
footnoted for interested readers.
Desired attitudes. Combining all effects, we found that the estimated rsp is .080, with a
95% confidence interval of [.053, .107]. Unsurprisingly, in other words, combining four
significant effects produced an overall significant meta-analytic effect. The effect has an
associated Z score of 5.80, and a p value < .0001.9 The effect is also shown in Panel A of Figure
S4-2, where each effect size is represented by a square “dot” with whiskers indicating the 95%
confidence interval for that particular study. As this figure makes clear, every experiment
supported the idea of desired attitudes positively impacting attitude change, and the diamond
shape at the bottom of the figure shows the estimation of the weighted effect size aggregating
across all experimental findings.
9
The random effects model, using maximum likelihood estimation, produces a Z score of 4.62, p < .0001.
Subjective discrepancies. Combining all four datasets, we found that the estimated rsp
is .060, with a 95% confidence interval of [.033, .086]. Not only is this effect significant, but it
has an associated Z score of 4.45, and a p value < .0001. Furthermore, the rsp effect size of .060 is
three-quarters that of the desired attitudes effect size (.080). In all cases, these effects partial out
the role of desired attitudes, so this meta-analysis reveals significant evidence that subjective
discrepancies explain attitude change incremental to the effect of desired attitudes. This finding
demonstrates the power of synthesizing heterogeneous study effects together (see also Lipsey &
Wilson, 2001). That is, by individual “vote counting,” we would have concluded that subjective
discrepancies are unrelated to attitude change (two nulls, one marginal, one significant effect).
This finding also is visually apparent in Panel B of Figure S4-2. Clearly the bulk of evidence,
when viewed individually, would be considered inconclusive when appraising the plausibility of
subjective discrepancies affecting attitude change. However, due to both the unequal sample
sizes (e.g., note that the significant effect had the largest sample size), and the absence of
contradictory evidence (i.e., no effect size fell on the “wrong side” of zero), we were able to
detect a significant, albeit subtle effect of subjective discrepancies on attitude change.10
10
A random effects model indicated significant, albeit weaker support, Z = 2.09, p = .036.
Figure S4-2. Forest Plot of Desired Attitudes’ Effect on Time 3 Attitude Change across Experiments 1-2.
A. Composite Measure of Desired Attitudes
Note. Each square dot represents a single effect size estimate, with the whiskers indicating the 95% confidence interval around that
estimate. The center of the diamond on the bottom represents the effect size estimate with the width of the diamond reflecting the 95%
confidence interval around that estimate.
B. Subjective Discrepancies
Note. Each square dot represents a single effect size estimate, with the whiskers indicating the 95% confidence interval around that
estimate. The center of the diamond on the bottom represents the effect size estimate with the width of the diamond reflecting the 95%
confidence interval around that estimate.
Experiment 3 Formation of Subjective Discrepancies
We also examined the formation of subjective discrepancies in Experiment 3. Effects are
displayed in Table S4-5. Here, we found a very similar effect to prior datasets (i.e., Experiment
2a, 2b), which is unsurprising because this study was a direct replication of those earlier studies
at this point (the metacognitive manipulation only coming into play later in the procedure). The
significant and substantial unstandardized B of 2.77 means that participants who learned the
painting was made by a Nazi victim subjectively estimated that they wanted to hold an attitude
approaching three full scale units more positive than when the painting was made by a Nazi
sympathizer.
Table S4-5. Painter Identity Effects on Subjective Discrepancies Formation Controlling for Time 1 Actual Attitude. (Experiment 3).
Independent Variable Parameter Coefficients

Passage B = 2.77 [2.38, 3.17], t(198) = 13.84, p < .001
Time 1 Actual Attitude B = -.09 [-.25, .07], t(198) = -1.16, p = .248
Model Statistics F(2, 198) = 96.84, p < .001

Multiple R2 .49
Experiment 3 Actual Attitude Change Effects
Finally, we examined actual attitude change in Experiment 3. As we reported in the main
text, our primary expectation was that desired attitudes (whether measured as desired attitude
composite, ideal attitude, ought attitude, or subjective discrepancies) should be positively related
to actual attitude change. However, an interaction effect indicating that metacognitive prompts
made this effect slightly larger would not be antithetical to our theory, and possibly would be
beneficial (because it would indicate that the actual attitude change was “real” rather than
feigned, by showing that a factor that might be expected to facilitate meaningful self-persuasion
in fact did so).
Indeed, the simple slopes indicating subjective discrepancies’ prediction of actual attitude
change was significant and positive whether metacognitive prompts were absent, B = .33
[.19, .46], t(195) = 4.78, p < .0001, or present, B = .52 [.39, .64], t(195) = 8.15, p < .0001.
However, in this instance we found a significant interaction, suggesting that the latter effect was
larger. This replicated the finding that we detected for ideal attitudes (explained in a footnote in
the main text), providing some additional reason to think that perhaps persuasion was facilitated
by stimulation towards metacognitive efforts being provided. However, once again, the highly
significant slopes even when metacognitive prompts were absent (p < .0001) overwhelmingly
indicates that such prompting was not a necessary part of our paradigm.
Table S4-7. Effects of Subjective Discrepancies and Metacognitive Prompts on Actual Attitude Change (Experiment 3).
Independent Variables Subjective Discrepancies
Time 1 Actual Attitude B = .24 [.10, .38], t(195) = 3.30, p = .001
Time 2 Actual Attitude B = .43 [.32, .54], t(195) = 8.01, p < .001
Time 2 Desired Attitude B = .33 [.19, .46], t(195) = 4.78, p < .001
Metacognitive Prompts B = -.69 [-1.38, .004], t(195) = -1.96, p = .051
Time 2 Desired Attitude X B = .19 [.03, .35], t(195) = 2.35, p = .020

Metacognitive Prompts
F(5, 195) = 84.29, p < .001, R2 = .68

Model Statistics
SOM-5: Analysis of the Polarization Explanation
Experiment 1a/1b. One alternative view of our findings involves attitude polarization
from mere thought (e.g., Tesser, 1978). After receiving the matching/mismatching information
about the politician, people responded by immediately changing their actual attitudes to be
positive/negative, respectively. A polarization explanation might argue that later, given more
time to think, people might have polarized these actual attitudes due to the mere thought
phenomenon, with matching (mismatching) people becoming more positive (negative) in their
views. In other words, desired attitudes might have shifted more quickly to be consistent with
participants’ cognitive schemas (Tesser, 1978), that is, their broader notions of what their
political ingroup versus outgroup members are like (good vs bad, respectively), with actual
attitudes taking longer to become schema-consistent. Desired attitudes then predicted later
attitude change because they changed more extremely in the same direction that actual attitudes
would later shift. Note that this would not render our results uninteresting because it would show
that people can self-report at Time 2 the degree of attitude change that they will reach at Time 3,
by means of an attitude they consider “desirable” at Time 2. This is particularly true because
mere thought processes are generally viewed as something participants are unaware of, with
participants unable to self-report that it has occurred (Tesser, 1978).
However, this counter-explanation presumes that desired attitudes were more schema-
consistent than actual attitudes (at Time 2) in response to the politician matching/mismatching
information. To test this, we ran a 2 (Matching Condition: Matching vs Mismatching) X 2 (Time
2 Attitude Type: Actual vs Desired) mixed ANOVA, including Time 1 actual attitudes as a
covariate. The critical interaction was non-significant in Experiment 1a, F(1, 321) = .01, p
= .926, ηp2 = .000, indicating that desired attitudes did not change more radically in response to
the matching versus mismatching information (Mdiff = 2.76) than did actual attitudes (Mdiff =
2.74). The interaction effect was significant in Experiment 1b, F(1, 693) = 8.73, p = .003, ηp2
= .012; however, this was driven by desired attitudes changing significantly less in response to
the matching/mismatching information (Mdiff = 1.16) compared with the larger difference
detected for actual attitudes at this timepoint (Mdiff = 1.44). Thus, the downstream attitude change
explained by desired attitudes in Experiment 1a/1b cannot be explained as a mere thought effect,
because desired attitudes were not more schema-consistent than actual attitudes were at Time 2.
Experiment 2a/2b. Like Experiment 1, we wanted to test the plausibility of a
polarization view of our findings. We used 2 (Painter Identity: Immoral vs Moral) X 2 (Time 2
Attitude Type: Actual vs Desired) mixed ANOVA, with Time 1 Actual Attitudes as a covariate.
Interestingly, both datasets yielded significant interactions. In Experiment 2a, the interaction,
F(1, 114) = 15.65, p < .001, ηp2 = .121 revealed that Time 2 desired attitudes indeed shifted more
radically in response to the painter’s identity (Mdiff = 3.81) than did Time 2 actual attitudes (Mdiff
= 2.66). In Experiment 2b, again, the interaction, F(1, 511) = 207.39, p < .001, ηp2 = .289 was
such that Time 2 desired attitudes shifted more radically in response to the painter’s identity
(Mdiff = 3.54) than did Time 2 actual attitudes (Mdiff = 1.53).
However, a polarization explanation would also seem to suggest that the effect of painter
identity on actual attitude should be larger at Time 3 than at Time 2. Unlike the above
comparison of Time 2 actual and desired attitudes, however, this comparison cannot be tested by
a mixed ANOVA model because Time 2 actual and Time 3 actual attitudes were measured on
different scalings. Nonetheless, simply comparing the effect size of painter identity’s influence
on Time 2 actual attitudes, t(115) = 9.04, p < .001, d = 1.67 [1.25, 2.09], versus the effect size of
painter identity’s effect on Time 3 actual attitudes, t(115) = 8.86, p < .001, d = 1.64 [1.22, 2.06],
does not provide support for a polarization view in Experiment 2a, because the influence of the
manipulation on actual attitudes did not grow more extreme at Time 3 than at Time 2. Similarly,
in Experiment 2b, the effect of painter identity on Time 2 actual attitudes, t(512) = 7.65, p
< .001, d = .67 [.50, .85], was quite similar to painter identity’s effect on Time 3 actual attitudes,
t(512) = 9.10, p < .001, d = .80 [.62, .98]. It might also be worth noting that even if attitudes
were polarizing from Time 2 to Time 3, it would still be intriguing that people would be able to
self-report this at Time 2 by means of their desired attitude.
Experiment 3. Once again, we tested a polarization view of our results with a 2 (Painter
Identity: Immoral vs Moral) X 2 (Time 2 Attitude Type: Actual vs Desired) mixed ANOVA,
with Time 1 actual attitudes as a covariate. Like Experiment 2a/2b, the interaction, F(1, 198) =
37.55, p < .001, ηp2 = .159 showed that Time 2 desired attitudes indeed shifted more radically in
response to the painter’s identity Mdiff = 3.71) than did Time 2 actual attitudes (Mdiff = 2.38).
Once again, a polarization explanation would also seem to suggest that the effect of painter
identity on actual attitude should be larger at Time 3 than at Time 2. The effect size of painter
identity’s influence on actual attitudes at Time 2, t(199) = 9.79, p < .001, d = 1.38 [1.07, 1.69],
grew slightly larger at Time 3, t(199) = 11.10, p < .001, d = 1.57 [1.25, 1.88].
SOM-6: Structural Equation Modeling

SOM-6a: Experiment 1a
Table S6a-1.
Fit Statistics for Structural Equation Models for Experiment 1a.

RMSEA SRMR CFI TLI
SEM Regression .113 [.101, .126] .025 .95 .93
(Full Model)
SEM Regression .125 [.112, .139] .027 .94 .92
(Fair Model)
Table S6a-2.
Standardized Regression Weights for Original (Linear Regression) Models vs Various Structural
Equation Models. (Experiment 1a).
Original Regression SEM Regression SEM Regression
(Full Model) (Fair Model)
Time-1 Actual .149 *** .153 *** .146 ***
Time-2 Actual .637 *** .524 *** .626 ***
Time-2 Desired .220 *** .352 *** .252 ***
SOM-6b: Experiment 1b
Table S6b-1.
Fit Statistics for Structural Equation Models for Experiment 1b.

RMSEA SRMR CFI TLI
SEM Regression .124 [.116, .133] .042 .92 .89
(Full Model)
SEM Regression .133 [.124, .142] .044 .91 .89
(Fair Model)
Table S6b-2.
Equation Models. (Experiment 1b).
Time-1 Actual .061 ** .095 ** .093 **
Time-2 Actual .715 *** .426 *** .465 ***
Time-2 Desired .116 *** .125 ** .073 **
SOM-6c: Experiment 2a
Table S6c-1.
Fit Statistics for Structural Equation Models for Experiment 2a.

RMSEA SRMR CFI TLI
SEM Regression .185 [.164, .206] .053 .85 .81
(Full Model)
SEM Regression .199 [.177, .222] .055 .85 .80
(Fair Model)
Table S6c-2.
Equation Models. (Experiment 2a).
Time-1 Actual .150 * .218 ** .174 **
Time-2 Actual .451 *** .327 ** .450 ***
Time-2 Desired .371 *** .486 *** .379 ***
SOM-6d: Experiment 2b
Table S6d-1.
Fit Statistics for Structural Equation Models for Experiment 2b.

RMSEA SRMR CFI TLI
SEM Regression .114 [.105, .124] .046 .93 .91
(Full Model)
SEM Regression .123 [.112, .133] .046 .93 .90
(Fair Model)
Table S6d-2.
Equation Models. (Experiment 2b).
Time-1 Actual .199 *** .307 *** .270 ***
Time-2 Actual .510 *** .410 *** .481 ***
Time-2 Desired .225 *** .287 *** .237 ***
SOM-6e: Experiment 3
Table S6e-1.
Fit Statistics for Structural Equation Models for Experiment 3.

RMSEA SRMR CFI TLI
SEM Regression .130 [.114, .146] .054 .90 .87
(Full Model)
SEM Regression .134 [.117, .152] .054 .90 .87
(Fair Model)
Table S6e-2.
Equation Models. (Experiment 3).
Time-1 Actual .129 ** .221 *** .163 **
Time-2 Actual .365 *** .170 n.s. .367 ***
Time-2 Desired .476 *** .683 *** .495 ***
SOM-7: Do Effects Change if People are Removed for Quality-related Reasons?
Effects of Manipulation on Desired Attitudes

Removal of People who Failed This Quality Check
Predictor Variables Invalid Thoughts Suspicious Technical Inattentive Inattentive (Only

Problems (Original check: exclude actively
no response = cut) wrong Ps)
Matching Condition b = .10, t(642) = b = .11, t(654) = b = .13, t(676) = b = .11, t(440) b = .13, t(626) =
1.01, p = .312 1.10, p = .270 1.36, p = .174 = .86, p = .392 1.27, p = .206
Time 1 Actual Attitude b = -.03, t(642) = b = -.01, t(654) = b = -.01, t(676) = b = -.07, t(440) = - b = -.03, t(626) =
-.72, p = .472 -.19, p = .854 -.32, p = .752 1.17, p = .245 -.54, p = .591
Time 2 Actual Attitude b = .73, t(642) = b = .71, t(654) = b = .72, t(676) = b = .73, t(440) = b = .72, t(626) =
25.80, p < .001 25.59, p < .001 26.31, p < .001 21.36, p < .001 25.32, p < .001
Participant Political b = -.08, t(642) = b = -.07, t(654) =- b = -.08, t(676) = b = -.08, t(440) = - b = -.05, t(626) =
Extremity -1.19, p = .236 1.04, p = .301 -1.17, p = .242 1.00, p = .318 -.78, p = .439
Matching Condition X b = .30, t(642) = b = .30, t(654) = b = .29, t(676) = b = .14, t(440) b = .30, t(626) =
Participant Political 2.12, p = .035 2.21, p = .027 2.18, p = .030 = .81, p = .416 2.10, p = .036
Extremity
Model Statistics F(5, 642) = F(5, 676) = F(5, 676) = F(5, 440) = 139.01, F(5, 626) = 201.62,
208.89, 203.19, 216.36, p < .001, R2 = .62
p < .001, R2 = .62 p < .001, R2 = .61 p < .001, R2 = .62 p < .001, R2 = .61
Effects of Desired Attitudes on Actual Attitude Change

Removal of People who Failed This Quality Check
Predictor Variables Invalid Thoughts Suspicious Technical Inattentive Inattentive (Only

Problems (Original check: exclude actively
no response = cut) wrong Ps)
Time 1 Actual Attitude b = .09, t(651) = b = .08, t(664) = b = .08, t(686) = b = .01, t(445) b = .07, t(636) =
3.05, p = .002 2.75, p = .006 2.81, p = .005 = .24, p = .814 2.36, p = .018
Time 2 Actual Attitude b = .48, t(651) = b = .48, t(664) = b = .48, t(686) = b = .55, t(445) = b = .49, t(636) =
19.60, p < .001 19.95, p < .001 20.46, p < .001 18.51, p < .001 20.04, p < .001
Time 2 Desired b = .09, t(651) = b = .08, t(664) = b = .08, t(686) = b = .04, t(445) = b = .08, t(636) =
Attitude 3.57, p < .001 3.37, p = .001 3.45, p = .001 1.16, p = .247 3.11, p < .001
Model Statistics F(3, 651) = F(3, 664) = F(3, 686) = F(3, 445) = 343.74, F(3, 636) = 479.53,
491.34, 478.36, 512.38, p < .001, R2 = .69
p < .001, R2 = .69 p < .001, R2 = .68 p < .001, R2 = .69 p < .001, R2 = .70
SOM-8: Tables of Means and Standard Deviations for all Experiments
Experiment 1a
Variable Mismatching Matching Difference (t-test)
Condition Mean Condition Mean
(SD) (SD)
Time 1 actual attitude 3.95 (1.00) 4.14 (1.15) t(322) = 1.51, p = .131
Time 2 actual attitude 3.29 (2.12) 6.16 (2.16) t(322) = 12.06, p < .001
Time 2 ideal attitude 3.81 (2.13) 6.62 (2.08) t(322) = 12.04, p < .001
Time 2 ought attitude 3.54 (2.13) 6.39 (2.03) t(322) = 12.30, p < .001
Time 2 desired 3.67 (1.99) 6.50 (1.90) t(322) = 13.09, p < .001
attitude
Time 2 subjective 3.44 (1.36) 4.63 (1.33) t(322) = 7.99, p < .001
discrepancy
Time 3 behavioral 2.56 (1.45) 4.83 (1.51) t(322) = 13.76, p < .001
intention
Experiment 1b
Variable Mismatching Matching Difference (t-test)
Condition Mean Condition Mean
(SD) (SD)
Time 1 actual attitude 4.39 (1.07) 4.34 (1.03) t(695) = -.58, p = .562
Time 2 desired 4.67 (1.82) 5.81 (1.67) t(694) = 8.60, p < .001
attitude
discrepancy
Time 3 behavioral 3.26 (1.48) 4.50 (1.32) t(694) = 11.70, p < .001
intention
Experiment 2a
Variable Immoral Moral Condition Difference (t-test)
Condition Mean (SD)
Mean (SD)
Time 1 actual attitude 3.41 (1.32) 3.92 (1.37) t(115) = 2.09, p = .039
Time 2 desired attitude 2.89 (1.76) 6.81 (1.49) t(115) = 13.03, p < .001
discrepancy
Experiment 2b
Condition Mean (SD)
Mean (SD)
discrepancy
Experiment 3
Condition Mean (SD)
Mean (SD)
discrepancy
Metacognitive Prompts Absent Metacognitive Prompts

Present
Immoral Moral Immoral Moral
Condition Condition Condition Condition
Mean (SD) Mean (SD) Mean (SD) Mean (SD)
Time 1 actual attitude 3.92 (1.22) 3.85 (1.28) n.s. 3.67 (1.30) 3.67 (1.22) n.s.
Time 2 actual attitude 3.67 (1.83) 5.94 (1.59) *** 2.96 (1.55) 5.47 (1.73) ***
Time 2 ideal attitude 3.50 (2.10) 6.77 (1.83) *** 2.89 (2.04) 6.64 (2.00) ***
Time 2 ought attitude 2.94 (1.83) 6.75 (1.52) *** 2.39 (1.73) 6.49 (2.17) ***
Time 2 desired attitude 3.22 (1.70) 6.76 (1.34) *** 2.64 (1.68) 6.57 (1.86) ***
Time 2 subjective 2.74 (1.65) 5.37 (1.10) *** 2.24 (1.29) 5.21 (1.51) ***
discrepancy
Time 3 actual attitude 3.55 (1.59) 5.73 (1.41) *** 2.86 (1.50) 5.59 (1.63) ***
Note. ‘n.s.’ and ‘***’ refer to comparisons of the denoted mean against the mean to the
immediate left, where ‘n.s.’ indicates non-significance (p > .80), and ‘***’ indicates significance
at the p < .001 level. In other words, these refer to comparisons of that row’s dependent variable,
between the immoral versus moral painter conditions, calculated separately for each
metacognitive prompt condition.

Attitude Change

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Attitude Change

Uploaded by

Copyright:

Available Formats

DESIRED ATTITUDES GUIDE ACTUAL ATTITUDE CHANGE 1

Desired Attitudes Guide Actual Attitude Change

Thomas I. Vaughan-Johnston, PhD1, 2

how people’s attitudes will shift in the future.

Word Count: 131

Desired Attitudes Guide Actual Attitude Change

Rowling’s comments as expressing problematic values and/or beliefs. We suspect that

work argues, may have important attitudinal implications.

Although desired attitudes relate to a range of important consequences, researchers have

effectively produce attitude change.

Desired Attitudes and Actual Attitude Change

principles; Wheeler & DeMarree, 2019).

Desired Attitudes as Self-Persuasive Goals

it all in, what attitude do I want to obtain?” (i.e., a desired attitude).

persuade towards situationally important goals by reorganizing or reweighting information (e.g.,

changing attributions, reweighting the importance of information). Thus interestingly, people

self-persuasion, we provided a series of metacognitive prompts in Experiments 1-2. However,

prompting, and we test this in Experiment 3.

Overview and Hypotheses

as Experiment 1a and 1b.

predicting “pursuit-oriented outcomes” of desired attitudes; r = .19 (DeMarree et al., 2017).

preregistration” section, below).

Procedure. Participants answered demographic questions including a political

Flow of the Experimental Procedure (Exp. 1).

same object (Downing et al., 1992; Petty & Cacioppo, 1981).

We next stimulated participants to reflect on their past thoughts with several

Political orientation. We used a “forced-choice” single-item, six-point bipolar scale (1 =

examined political extremity as a potential moderator of the matching condition’s effect on

participants were “slightly” liberal/conservative, 2 reflected participants who were “moderately”

liberal/conservative, and 3 reflected participants who were “strongly” liberal/conservative. In

dataset, higher scores indicating more positive attitudes.

results differed meaningfully for these items.

ideal/ought attitudes) or through a subjective discrepancies measure (i.e., asking participants to

report the subjective discrepancy measure’s results in SOM-4.

averaged items, higher scores indicating favourable attitudes.

of the experimental variable(s) for all experiments.

formed based on identity-related pressures (the matching manipulation). Effects of matching /

change induced post-manipulation) as a covariate in this analysis. Obviously, people (actually)

interaction of matching condition X political extremity. In other words, a “matching” (versus

“mismatching”) politician should match/mismatch most strongly for extreme-political

for people at the extremes of the political spectrum (versus moderates).

Most critically, we also found a significant interaction of matching condition X political

significantly shaped by the politician’s political identity.

are increasingly dictated by the political stance of the politician.3

Matching Condition b = 1.00 [.65, 1.34], b = .12 [-.06, .31],

Time 1 Actual Attitude b = -.04 [-.18, .10], b = -.004 [-.09, .08],

Time 2 Actual Attitude b = .64 [.57, .72], b = .71 [.66, .77],

Participant Political Extremity b = -.01 [-.21, .19], b = -.08 [-.20, .05],

Matching Condition X b = .82 [.38, 1.25], b = .29 [.03, .55],

Model Statistics F(4, 317) = 172.9, F(5, 682) = 216.6,

Desired attitudes predicting subsequent actual attitude change. We next tested if

reflecting on their own previous thoughts with a new metacognitive goal.4

Time 1 Actual Attitude b = .22 [.14, .30], b = .08 [.02, .13],

Time 2 Actual Attitude b = .40 [.34, .46], b = .48 [.44, .53],

Time 2 Desired Attitude b = .15 [.09, .21], b = .08 [.04, .13],

Model Statistics F(3, 320) = 391.21, F(3, 692) = 515.01,

positive attitudes given a matching (M = 4.65, SD = 1.16) versus mismatching politician (M =

But either analysis supports our contentions.

object between Time 2 and Time 3, suggesting a surprising self-awareness of participants. It is

Time 3 with no new information provided between these intervals.

Furthermore, we created desired attitudes in these experiments in a manner different from

However, as some exploratory follow-up analyses revealed, Experiments 1a/1b were