Content uploaded by Adelle X. Yang
Author content
All content in this area was uploaded by Adelle X. Yang on Dec 05, 2024
Content may be subject to copyright.
Framing Affects Postdecision Preferences Through Self-Preference
Inferences (and Probably Not Dissonance)
Adelle X. Yang
1
and Jasper Teow
2
1
Department of Marketing, National University of Singapore
2
Department of Business Innovation, The Business School, Royal Melbourne Institute of Technology University Vietnam
Psychologists have long beenintrigued by decision-induced changes in preferences where making a decision
strengthens one’s relative preference between more and less preferred options. This phenomenon has been
explained through two prominent theories: a dissonance account, which suggests that it results from the
decision maker’s attempt to minimize an unpleasant emotional–motivational state of “dissonance,”and an
inference account, which posits that it reflects a process of inferring and updating one’s“true”preferences. In
the current research, we investigate whether, how, and why framing a decision as a choice or a rejection
influences decision-induced preference modulation. Across 13 preregistered experiments, including seven
(N=6,248 participants from North America and Asia) reported in the main text, we find that reject-framed
decisions between attractive options induce greater postdecision preference modulation (i.e., a larger
preference gap between options) than choose-framed decisions, all else equal. Supporting the inference
account, the effect is moderated by attribute similarity and choice set valence while being mediated
consistently by perceived action diagnosticity. In contrast, purported moderators and process measures of the
dissonance account received no support when tested. Additionally, we systematically address potential
confounds associated with varying levels of “noise”in preference expression through decisions, an issue that
had encumbered previous paradigms on preference modulation. Our findings suggest that changes in
preference induced by ordinary day-to-day decisions primarily stem from an ongoing process of information
inference and updating rather than dissonance reduction. This research also provides insights into the
previously unforeseen consequences of framing interventions in policy and business.
Public Significance Statement
Decisions framed as opportunities to choose or reject are increasingly used by choice architects to
influence decision outcomes in business and policy. However, since decisions can alter subsequent
preferences (reviewed in Enisman et al., 2021), it is important to understand whether such framing
interventions may inadvertently influence postdecision preferences and attitudes beyond their intended
immediate impact. This research examines whether,how,andwhy framing impacts postdecision
consequences. We repeatedly show (in 13 total preregistered experiments, N=9,927 participants in North
America and Asia) that framing a decision between attractive options as a reject action intensifies
preferences more than framingthe same decision as a choose action.This framing effect is observed across
diverse stimuli (e.g., products, art prints, food, photos, and words), is mitigated when the options are
similar in attributes, and is reversed in direction when the decision is between unattractive options. Our
results corroborate an inference explanation for decision-induced preference modulation, based on key
insights from self-perception theory (Bem, 1967) and the theory of representatio nal exchange (Cushman,
2020): Individuals extract new information about their underlying preferences from observing their own
decisions, and reject framing tends to render decisions more self-diagnostic than choose framing.
Meanwhile, our results challenge the fit of a previously more popular dissonance explanation (Festinger,
1957) for preference modulation in ordinary day-to-day decisions. Therefore, our findings also suggest
that preference modulation may be more commonplace than previously implied. Beyond theoretical
contributions, this research also provides novel insights into how framing interventions can be leveraged
toward positive outcomes in practical domains including marketing, management, and policymaking.
Keywords: decision-induced preference modulation, choose/reject framing, self-perception theory, self-
preference inferences, cognitive dissonance theory
Supplemental materials: https://doi.org/10.1037/xge0001651.supp
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Abigail Sussman served as action editor.
Adelle X. Yang https://orcid.org/0000-0001-5817-7050
All data, original materials, and analysis code are shared on the Open
Science Framework at https://osf.io/qhcdy/. This research adheres to the
American Psychological Association’s ethical standards concerning partici-
pant treatment. This article has not been shared on any academic social
networks.
The authors have no conflicts of interest to disclose. Adelle X. Yang
received funding from the National University of Singapore. The authors
continued
Journal of Experimental Psychology: General
© 2024 American Psychological Association
ISSN: 0096-3445 https://doi.org/10.1037/xge0001651
1
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
The idea that decisions enhance subsequent preferences was
initially proposed by some of the most influential theories in
psychology (e.g., Bem, 1967,1972;Festinger, 1957,1962).
Consider John, who is faced with a choice between two preferred
art prints, a Monet and a Picasso. These theories predict that if John
selects the Monet, this decision will intensify his preference for the
Monet over the Picasso. This phenomenon, known as postdecision
preference modulation, has captivated psychologists for genera-
tions and has now been empirically substantiated after decades of
testing and corrections (see Enisman et al., 2021, for a review). The
accumulating evidence supporting this phenomenon challenges a
foundational assumption about the relationship between prefer-
ences and decisions—conventionally, preferences are assumed to
influence decisions but not the reverse.
While daily decisions are typically presented as opportunities to
choose, they can also be presented as opportunities to reject. From
customizing products to selecting medical treatments and planning
career paths, most decisions can be framed either as choices of
preferred options or rejections of nonpreferred ones. The power of
choose/reject framing in altering decision outcomes has been well-
documented (e.g., Dhar & Wertenbroch, 2000;Shafir, 1993).
However, limited research has examined the potential impact of
framing on postdecision consequences beyond the decision outcomes
per se, whereas no research has delved into how framing might
influence decision-induced preference modulation (or preference
modulation, for short).
This gap in the literature presents a unique opportunity to
advance both theory and practice. Theoretically, distinct explana-
tions for the phenomenon have been offered by influential
psychological theories including cognitive dissonance theory
(Festinger, 1957) and self-perception theory (Bem, 1967). Yet,
research has not resolved the theoretical ambiguity regarding the
phenomenon’s primary psychological mechanism(s), despite
significant recent progress in documenting the phenomenon and
addressing previous methodological issues (see review in Enisman
et al., 2021). To this end, examining how and when framing affects
preference modulation may give rise to new opportunities to
disambiguate mechanisms. Practically, the widespread adoption of
framing techniques in business and policy (Thaler & Sunstein,
2008) also calls for research into their downstream consequences.
A notable example is the increasing popularity of negative
advertising strategies in political campaigns, which encourage
voters to reject an alternative candidate as opposed to traditional
positive advertising that promotes support for a focal candidate. It
would be concerning if such a shift toward reject-framed strategies
inadvertently exacerbates political polarization in society beyond
merely impacting voting outcomes.
Motivated to address these questions, we examine whether, how,
and why decision framing impacts preference modulation. The
primary goal of this research was to test the effect of choose/reject
framing on postdecision preference modulation. A secondary goal
was to shed new light on its mechanisms toward addressing the
tension between alternative psychological explanations.
We first draw on self-perception theory (Bem, 1967,1972) and a
more recent framework of representational exchange (Cushman,
2020) to introduce a general information inference account. This
inference account suggests that observing one’s own decision prompts
people to infer new information about their “true”preferences and to
update their preference expressions accordingly. Based on this
account, we propose a series of hypotheses regarding how choose/
reject framing will affect preference modulation. The main hypothesis
is that reject framing, through heightening the perceived diagnosticity
of a decision, will induce more preference modulation than choose
framing, resulting in a larger preference gap between the preferred
option and the nonpreferred option.
We then distinguish this inference account from the dissonance
account offered by cognitive dissonance theory (Festinger, 1957,
1962). Dissonance theory posits that postdecision preference modula-
tion is caused by an aversive state of dissonance and an accompanied
motivation to reduce dissonance. Unlike the inference account, the
dissonance account does not provide a clear ex ante prediction on the
framing effect per se. Nevertheless, it furnishes flexible ex post
alternative explanations for our hypothesized framing effect and for
its opposite effect. Additionally, dissonance theory has articulated
necessary antecedents for observing preference modulation that differ
from those necessitated by the inference account. We will empirically
test between these mechanisms in the framing context.
We test our hypotheses in seven preregistered experiments
(N=6,248 participants from North America and Asia) and six
preregistered supplemental experiments (N=3,679 participants
from North America and Asia). We repeatedly find the proposed
framing effect, the diagnosticity-based process, and the moderating
role of attribute similarity and choice set valence on the framing
effect, all consistent with predictions of the inference account.
Meanwhile, our results do not support the process measures and
moderators of the dissonance account.
Notably, in testing the framing effect, our experiments do not
involve comparisons of pre- versus postdecision preferences (known
as the “spreading of preferences”or “preference spread”), a procedure
that had been shown to be susceptible to methodological confounds
(see critique and proof in Chen & Risen, 2010; see a review of
rectified studies in Enisman et al., 2021). Instead, we solely measure
and compare the relative strengths of postdecision preferences
between choose and reject frames. The original methodological
confounds of preference spread are hence not relevant for testing this
framing effect. Nonetheless, within the framing paradigm, we
systematically assess whether observed differences in postdecision
preferences were attributable to varying levels of “noise”in preference
expression through (choose and reject) decisions, addressing potential
confounds that are conceptually related to prior critiques.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
thank Tom Meyvis, Leif Nelson, Jane Risen, Klaus Werternbroch, Oleg
Urminsky, Clayton Critcher, Reid Hastie, and John Nash for providing
valuable comments on this research. The authors also acknowledge the
helpful feedback from the attendees at the 2020 Society of Judgement and
Decision Making conference and the Singapore Island Labs.
Adelle X. Yang played a lead role in conceptualization, funding
acquisition, methodology, writing–original draft, and writing–review and
editing, a supporting role in data curation, and an equal role in formal
analysis and project administration. Jasper Teow played a lead role in data
curation, a supporting role in conceptualization and writing–review and
editing, and an equal role in formal analysis and project administration.
Correspondence concerning this article should be addressed to Adelle
X. Yang, Department of Marketing, National University of Singapore, 15
Kent Ridge Drive, Singapore 119245. Email: adelle.yang@gmail.com
2YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Decision-Induced Preference Modulation
Why do decisions change subsequent preferences? Two of the
most influential theories in social psychology have offered distinct
answers to this question. One centers on an unpleasant state of
dissonance induced by the decision (e.g., Festinger, 1957,1962)
whereas the other emphasizes a spontaneous information inference
from the decision (e.g., Bem, 1967,1972;Cushman, 2020). These
theories broadly extend to the phenomenon of rationalization, the
observation that after a person performs an action, their subsequent
beliefs and attitudes tend to change in a manner that helps make
the initial action appear more rational (see Cushman, 2020, for a
review).
There is apparent tension between the dissonance account and the
inference account. However, because they involve separate sets of
psychological mechanisms, the two accounts typically do not
generate “unequivocally conflicting predictions”(Greenwald, 1975,
p. 491). As such, a growing consensus over the past few decades
suggests that the two accounts may offer complementary rather than
competing explanations (Cushman, 2020;Fazio et al., 1977;
Greenwald, 1975) and that they may apply primarily in different
domains (as reviewed by Cushman, 2020;Harmon-Jones & Mills,
1999). Indeed, whereas self-perception theory has enjoyed extensive
support in contexts involving attitude-congruent behaviors, such as
self-monitoring and self-signaling behaviors (e.g., Bodner & Prelec,
2003; Dhar & Wertenbroch, 2012;Rosenberg, 1979), evidence for
dissonance has been documented primarily in contexts involving
attitude-discrepant behaviors, such as attitude changes following
obviously counter-attitudinal actions or unexpectedly painstaking
efforts (Elliot & Devine, 1994;Snyder & Ebbesen, 1972).
Which account, then, better explains the observation of preference
modulation in ordinary day-to-day decisions, such as in the example
of John deciding between art prints? A brief glance through the
literature will reveal numerous casual references to such examples as
a“dissonance phenomenon,”implying dissonance reduction as its
primary explanation. However, this assumption has never been
empirically assessed. In fact, leading scholars in this area have
explicitly noted that existing research has never tested between the
two explanations in the context of decision-induced preference
modulation (e.g., Footnote 1 in Chen & Risen, 2010, p. 573; Footnote
4inRisen & Chen, 2010, p. 1161; see also Enisman et al., 2021,
p. 24). As of today, the answer to this question remains unclear.
This unresolved question regarding the psychological mechan-
ism(s) underlying preference modulation should not be conflated
with previous contentions regarding the validity of evidence for
preference modulation. In 2010, Chen and Risen (2010) identified
critical confounds in the free-choice paradigm (FCP; Brehm, 1956)
that had been the dominant paradigm used to study preference
modulation for more than half a century. Since then, researchers
used modified paradigms to correct for the confound and have
reestablished robust evidence for preference modulation (reviewed
in Enisman et al., 2021). Therefore, the key question in this research
pertains to why decisions alter subsequent preferences, not whether
they do.
1
Next, we consider the potential impact of (choose vs. reject)
framing on preference modulation based on each theoretical
perspective. We begin by introducing the inference account, which
generates a series of straightforward hypotheses on a framing effect
and its primary characteristics. Subsequently, we assess these
hypotheses from the dissonance perspective and discuss their
differences. As will be seen, these two accounts do not make
identical predictions on the presence, direction, and boundaries of
the framing effect. These differences will enable us to disentangle
the two accounts both conceptually and empirically.
The Inference Account
The gist of the information inference account is that people glean
insights into an actor’s underlying preferences through their actions.
Extensive research in social psychology has established a ubiquitous
process of attributional inference—people learn about others’
preferences by observing others’actions and attributing them to
internal or external factors (Jones & Davis, 1965;Kelley, 1973).
Self-perception theory (Bem, 1967,1972) builds on this premise and
takes one step further to contend that people, lacking perfect insight
into their own preferences, may also infer their own preferences
through the observation of their actions (Bem, 1972, p. 2).
Extending this key insight, a recently proposed and well-received
framework of representational exchange (Cushman, 2020) posits
that inferring preferences from one’s own actions is not only
plausible, but it is also broadly adaptive. Synthesizing over a century
of behavioral, developmental, and neurocognitive evidence on the
multiple processes underlying actions and decisions (e.g., Dolan &
Dayan, 2013;Kahneman, 2011; Thorndike, 1898), Cushman
(2020) suggested that one’s actions may reveal useful information
about their underlying preferences that are tied to their instincts,
reflexes, and habits, which would be largely inaccessible via
introspection and deliberation (see also Nisbett & Wilson, 1977).
Hence, preference modulation can be viewed as the result of new
preference information—which are revealed by the observable
decision—being integrated into the mental representations of one’s
own preferences. More concretely, in the earlier example with John,
this inference account suggests that John’s decision of acquiring the
Monet will prompt him to infer that he probably liked the Monet
more than the Picasso, which then influences how he rates the two
paintings subsequently.
Critically, not all actions are equally diagnostic of the actor’s
underlying preferences. In social attribution, negative actions (e.g.,
violating a social norm or writing negative reviews) have a larger
impact than positive actions (e.g., conforming to a social norm or
writing positive reviews) on observers’inferences about the actor’s
dispositional preference (Mizerski, 1982;Skowronski & Carlston,
1989). These findings have been attributed to multiple psychologi-
cal mechanisms, one of which is the greater attentional resources
evoked by the negative implications of actions than positive ones
(see Baumeister et al., 2001, for a review). Importantly, the reject
framing emphasizes the negative aspects of a decision (e.g.,
potential losses) whereas the choose framing emphasizes its positive
aspects (e.g., potential gains; Houston et al., 1991;Levin et al.,
1998;Shafir, 1993;Tversky & Kahneman, 1981). Therefore, if
observers tend to extract more information about another actor’s
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1
In fact, if choose/reject framing induces different degrees of preference
modulation, then preference modulation must be true. Hence, our empirical
tests for a framing effect on postdecision preferences may serve as an
independent validation for the existence of preference modulation without
having to examine the preference spread per se.
FRAMING AFFECTS POSTDECISION PREFERENCES 3
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
dispositional preference from negative actions than positive actions,
then they may likewise deem their own reject actions as more
diagnostic about their own underlying preferences than choose
actions, all else equal.
Indeed, lending support to the different impact of choose/reject
framing on the process of decision making, converging evidence
shows that a reject decision tends to evoke a more deliberative
judgment process than an otherwise equivalent choose decision
(Nagpal & Krishnamurthy, 2008;Laran & Wilcox, 2011;Sokolova &
Krishna, 2016). Therefore, based on existing associations between the
depth of information processing and decision frames, we propose that
people may treat their own reject actions as more informative about
their underlying preference than their choose actions—as if reject
actions tap “deeper”into their true preferences than choose actions.
Subsequently, when extracted preference information is integrated into
one’s existing knowledge and belief about their own preferences,
2
greater preference modulation should follow a reject-framed decision
than a choose-framed decision.
In summary, based on the inference account, we hypothesize that,
because reject actions are deemed more diagnostic than choose
actions about one’s underlying preferences, reject framing may
intensify postdecision preferences more than choose framing. Hence,
we have Hypotheses 1 and 2.
Hypothesis 1 (a framing effect): Reject framing will lead to greater
postdecision preference modulation (i.e., a larger postdecision
preference gap between the preferred and nonpreferred alter-
natives) than choose framing.
Hypothesis 2 (on action diagnosticity): The reject-framed
decision will be perceived by decision makers as more diagnostic
about their own underlying preferences than the choose-framed
decision, all else equal.
The Dissonance Account
Cognitive dissonance theory (Festinger, 1957,1962) explains
preference modulation as the result of an attempt to reduce decision-
induced dissonance. Dissonance, an aversive emotional–motivational
state, is theorized to occur when there is inconsistency between
cognitions, such as a discrepancy between preferences and decisions.
People are purportedly motivated to reduce dissonance, such as by
revising their postaction preferences to be more closely aligned with
the preceding decision. Classic demonstrations of dissonance include
individuals’evaluation of a boring task being elevated after having
complied to tell others that the task was enjoyable (Festinger &
Carlsmith, 1959). In such contexts, evidence for dissonance has been
obtained through measures of self-reported psychological discomfort
(Elliot & Devine, 1994) and physiological arousal (Croyle &
Cooper, 1983).
More concretely, for John in our example, dissonance theory
suggests that if John finds the two art prints equally attractive yet
decided to acquire the Monet only, the partial conflict between his
preference and his decision would produce dissonance. Consequently,
John would increase his liking of the Monet over the Picasso to
alleviate dissonance. This explanation is also plausible, while certainly
distinct from the inference explanation.
Now, in the framing context, we have proposed a larger
postdecision preference gap in the reject frame than in the choose
frame, based on the inference account. To the best of our knowledge,
the dissonance account does not yield a clear ex ante prediction on
the presence or the direction of such a framing effect.
3
Nonetheless,
it can flexibly furnish ex post explanations for a framing effect to be
observed in either direction. That is, dissonance theory remains a
plausible explanation as long as greater dissonance is observed
in the frame where a larger preference gap is found and if this
difference in dissonance accounts for at least some of the differences
in preference gaps between frames.
These are empirically testable. Therefore, we will collect
measures of both the perceived diagnosticity of the decision (per
the inference account) and the experience of dissonance induced by
the decision (per the dissonance account). We will then examine if
any observed differences in preference modulation are attributable
to these process measures, respectively. These tests will provide the
first set of evidence to assess the relative fit of theories.
Next, we again use the inference account to derive key theoretical
moderators of the framing effect: option similarity and choice set
valence. As will be seen, the two accounts’ranges of applicability
will be further differentiated by the potential boundary conditions.
Option Similarity
From the perspective of information inferences, the degree to
which a decision induces subsequent preference modulation is
inherently constrained by the amount of information extractable
from the decision. When options in the choice set share more
identical attributes, less new information can be inferred from the
decision. In essence, as the similarity of attributes between options
increases, the total preference information available decreases.
Consistent with this logic, studies have shown that choices made
from options with similar self-control connotations (e.g., “vice”vs.
“vice”) are perceived as less diagnostic of the decision maker’s
dispositional self-control than choices made from distinct options
(e.g., “vice”vs. “virtue”;Dhar & Wertenbroch, 2012). When little
new information about the decision maker’s preference can be
gleaned from a decision, preferences tend to remain relatively
unchanged before and after the decision. This limits the impact of
decision framing on preference modulation. Therefore, it follows
that, a high similarity between options serves as a natural boundary
condition for the framing effect.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
2
Bem (1972) originally proposed that preference updating is an updating
of one’s belief about own preferences whereas Cushman’s (2020) theorizing
is decidedly less skeptical, arguing for an updating of actual preferences. This
theoretical distinction between them does not affect our hypotheses in the
current research.
3
The total magnitude of dissonance is believed to be determined by the
relative strengths of dissonant versus consonant cognitions (Brehm, 1956;
Festinger & Carlsmith, 1959). Framing can impact both types of cognitions.
For example, the reject frame often draws more attentional resources to
information processing than the choose frame (Sokolova & Krishna, 2016),
potentially heightening the conflict between action and attitude (e.g.,
between “rejecting the Picasso”and “this Picasso painting is cool”).
Meanwhile, the reject frame may focus more attentional resources on
negative attributes than the choose frame (Houston et al., 1991;Shafir, 1993),
potentially strengthening the consonance between the reject action and the
negative attributes of the rejected option (e.g., “this Picasso painting uses dull
colors”). Without being able to quantify the relative magnitude of these
various factors, it is infeasible to predict how overall dissonance may differ
between frames.
4YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Hypothesis 3 (option similarity as a moderator): The framing
effect will attenuate when the options in the choice set are
highly similar in attributes.
In agreement with the above reasoning, dissonance theory has
also suggested that attribute dissimilarity between options is a
necessary antecedent for decision-induced dissonance and, hence,
preference modulation (Brehm & Cohen, 1959;Festinger, 1957).
However, dissonance theory has specified one more necessary
antecedent for dissonance—equal attractiveness of options. In fact,
since the inception of dissonance theory, dissonance researchers
have repeatedly stressed that dissonance emerges from decisions
between options perceived as “equally attractive”; deciding between
options that obviously differ in attractiveness is considered easy and,
hence, unlikely to provoke internal conflict (e.g., Brehm, 1956;
Festinger, 1962;Shultz et al., 1999). This is why the FCP always
mandates equating choice alternatives in attractiveness in its first step
(Brehm, 1956). Accordingly, dissonance researchers have reported in
several FCP studies where preference modulation no longer occurred
after a decision between options of unequal attractiveness (e.g.,
Brehm & Cohen, 1959;Harmon-Jones & Harmon-Jones, 2002;
Shultz et al., 1999).
4
In summary, according to dissonance theory, two prerequisites
must be met for dissonance-based preference modulation to occur:
options must be similar in attractiveness but dissimilar in attributes.
In contrast, the inference account only requires that options be
dissimilar in attributes. Therefore, if we still observe the framing
effect with options that are dissimilar in attractiveness, it would be
challenging for the dissonance account to explain it.
Valence of Choice Set
Thus far, our theorizing has focused on decisions involving
primarily attractive options. Notably, many framing effects initially
established with attractive options are reversed when tested with
unattractive options instead (e.g., Kessler et al., 1996;Meloy &
Russo, 2004;Nagpal & Krishnamurthy, 2008; Perfecto et al., 2017;
Wedell, 1997). This reversal pattern has been attributed to the
compatibility between the valence of the action and the valence of
the choice set, which can affect the metacognitive fluency of
information processing (Perfecto et al., 2017;Sokolova & Krishna,
2016;Wedell, 1997). Specifically, the act of choosing is deemed
relatively more compatible with attractive options whereas the act of
rejecting is deemed relatively more compatible with unattractive
options. In both cases, compatibility increases processing fluency,
which can modulate subsequent outcomes, such as enhancing
postdecision confidence (Perfecto et al., 2017; see also Meloy &
Russo, 2004).
The critical role of metacognitive fluency in information
processing suggests that the framing effect, as proposed based on
the inference account, may also depend on the valence of the choice
set. When the decision involves primarily attractive options, we
have suggested that people read deeper into the reject action as if it
reveals more underlying preferences than a comparable choose
action. These associations between actions and preference strengths
should be weakened or even reversed when decisions involve less
attractive options, particularly between strongly unattractive options,
where the reject action becomes morenatural and prevalent (Buiten &
Keren, 2009). Ultimately, it is conceivable that choosing between
unattractive options may signal even stronger underlying preferences
than rejecting between unattractive options.
Hypothesis 4 (choice set valence as a moderator): The framing
effect should be mitigated, and potentially reversed, when the
choice set comprises less attractive alternatives.
Once more, the dissonance account may again provide a plausible
ex post explanation of Hypothesis 4. However, that requires additional
speculations about how compatibility and fluency affected different
components of dissonance (see Footnote 2) and involves various
empirical questions, some of which will be addressed as we test our
hypotheses in order.
Experimental Paradigm
In each experiment, we randomly assigned participants to one of
two framing (choose vs. reject) conditions. Participants were presented
with the same binary choice set and were asked to either choose a
preferred option or reject a less preferred option, corresponding to the
assigned condition. After the decision, participants were asked to rate
their preference for each option on a numerical scale. Our key
dependent variable is the postdecision preference gap (Δ), computed
by subtracting the rating of the unwanted (i.e., unchosen/rejected)
option from the rating of the wanted (i.e., chosen/retained) option. All
our studies used hypothetical scenarios as preference modulation was
shown to occur similarly in hypothetical and real choices (e.g., Sharot
et al., 2010).
As indicated in Hypothesis 1, we expect a larger postdecision
preference gap in the reject condition than the choose condition.
However, to interpret this expected difference as different degrees of
preference modulation, we must rule out other factors that could
speciously produce the same result, including random error and
systematic error. Importantly, both decisions and preference ratings
are “noisy”expressions of underlying preferences (Chen & Risen,
2010). In fact, instances of self-contradictory decisions and
preferences are inevitable in psychological experiments, stemming
from various sources including vague initial preferences, changes of
mind, miscomprehensions, inattention, and so on. These instances
can lead to critical confounds if not carefully controlled.
Within our framing paradigm, two sources of such noise are
worthy of attention. First, it is possible that the two groups of
participants differ in their preference volatility or attentiveness at the
outset of the experiment. Although such randomization failures can
lead to spurious results (i.e., Type I error), their likelihood can be
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
4
These reported findings should be interpreted with caution due to their
faulty conceptual and empirical assumptions that have been discussed in an
influential critique (Chen & Risen, 2010). First, Chen and Risen identified
fatal methodological confounds arising from selection issues in the FCP
protocol used in all these studies. Chen and Risen explicitly specified that
these confounds undermine the validity of various claims in the quoted
studies, including the role of “equal attractiveness”(see this critique on
p. 587 and the mathematical proof on p. 593, Chen & Risen, 2010;seealso
Izuma & Murayama, 2013, p. 8). Second, none of the quoted studies
actually measured dissonance. Instead, their conclusions about the
characteri stics of dissonance were ba sed entirely on the implic it assumption
that dissonance caused preference modulation; yet, that assumption has
never been empirically verified (see this critique in Footnote 1 in Chen &
Risen, 2010, p. 573 and Footnote 4 in Risen and Chen, 2010, p. 1161), as
mentioned earlier.
FRAMING AFFECTS POSTDECISION PREFERENCES 5
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
significantly reduced by established practices, such as true random
assignment, large sample sizes, and repeated replications—all of
which we closely adhere to.
The second source lies in how the imposed frames might
differently affect the likelihood that decisions misrepresent existing
preferences during the experiment. For example, if a participant
chooses A but rates B higher, their calculated preference gap will be
negative in value (<0), thereby shrinking the average gap in
their condition. If one condition had significantly more of such
participants than the other condition, then we may observe the
predicted difference in preference gap even if framing had no effect
on preference modulation. To be clear, this possibility does not
undermine the empirical validity of the proposed framing effect or
its novelty. Rather, it suggests yet another psychologically meaningful
alternative explanation: Framing may affect the likelihood that a
decision misrepresents existing preferences. This possibility is
conceptually distinct from our proposal that choose and reject
decisions change subsequent preferences to different degrees.
Therefore, a clear theoretical interpretation of the observed effect
necessitates that we address whether and to what extent the
observed effect is attributable to such a possibility.
5
To address these possibilities, we use two methods to systemati-
cally assess the impact of self-contradictory decisions and preferences
on the framing effect. First, in each study, we compare the frequency
of decision-inconsistent preferences (DIP) between conditions, and
we report the framing effect both with and without these instances
included (the rationale and procedure of this method are detailed in
Study 1.). Table 1 displays the main results before DIP screening. Few
participants exhibited DIP, and the framing effect was highly similar
after applying the additional DIP screener in all studies (see
Supplemental Table S1.1). These results demonstrate the minimal
impact of DIP on the observed effect. Second, in Study 5, we adapted
a previously tested method to fit the framing paradigm and confirm
that potential confounds do not drive the framing effect.
Overview of Studies
We present seven experiments (N=6,248 participants from North
America and Asia) in the main text of this article and six supplementary
experiments (N=3,679) in the Supplemental Material.Wefirst tested
how decision framing affects preference modulation with a purchase
scenario in Study 1 (Hypothesis 1). We then replicated this effect in
Studies 2a and 2b while examining process variables to test between
two theoretical accounts (Hypotheses 1 and 2). Next, we examined the
role of option similarity (Hypothesis 3) in Study 3 and the role of
choice set valence (Hypothesis 4) in Studies 4a and 4b. Finally, we
conductedStudy5tohoneinontheminimalimpactofpotential
confounds on the framing effect.
Transparency and Openness
We preregistered all experiments and report all manipulations,
measures, and analyses. We share all data, unabridged survey
materials, analysis code, and the additional online material on the
Open Science Framework at https://osf.io/qhcdy/ (Yang & Teow,
2024). We report all primary results in the main text. We summarize
secondary analyses and results in the main text and report their full
details in the additional online material.
Sampling
Our target sample sizes followed the heuristic of at least 125
participants per between-subjects condition. Sample sizes were
doubled in replication studies when resources allowed (Studies 1
and 2b) and at least tripled in moderation studies (Studies 4a, 4b, and
5) except for Study 3, in which the sample size was capped to the
size of a subject pool.
Screening
We used the same standard screening procedure in all studies:
excluding all incomplete responses and duplicate internet protocol
addresses in online samples (and excluding all incomplete responses
for the subject pool sample). We also excluded responses that failed
an instructional manipulation check (Oppenheimer et al., 2009)
during periods when online platforms reported survey bots. All
analyses were conducted upon completion of standard screening.
These standard screening criteria were preregistered, and they
excluded similar numbers of participants from the choose and reject
conditions in all studies (Supplemental Table S1.2). Table 1
summarizes the main results on preference modulation, directly
corresponding to the preregistered analyses. The DIP screening
method, used for post hoc robustness tests, was not preregistered nor
intended as part of standard screening. The DIP-screened results
remain similar (see Supplemental Table S1.1).
Study 1: A Purchase Decision
Study 1 tested the framing effect in a purchase scenario and was
preregistered on AsPredicted.Org at https://aspredicted.org/g2u
q6.pdf.
Method
Participants
We recruited 550 Prolific participants and received 568 data
entries. After standard screening (see full screening details in
Supplemental Table S1.2), we obtained 528 valid responses from
230 male and 298 female participants with an average age of 34.
Procedure
Participants were asked to imagine making a purchase decision
between two preferred smartphone cases, both compatible with their
current phone. The two options were both in black and white, from the
same brand and priced the same, presented side by side (Figure 1). The
only difference was in their patterns: checkered versus striped.
We randomly assigned participants to the two framing (choose vs.
reject) conditions. In the choose condition, participants were instructed
to drag their preferred option into a box labeled “I choose …” in
response to the question “Which phone case would you choose?”.
In the reject condition, participants were instructed to drag their
nonpreferred option into a box labeled “Ireject…” in response to the
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
5
While this possibility may be reminiscent of the famous confounds in the
FCP, the original selection issues in the FCP are procedurally irrelevant for
testing the framing effect in this research. Again, this is because we examine
how framing impacts postdecision preference strengths alone without
measuring any predecision preferences or having to measure any.
6YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
question “Which phone case would you reject?”. Then participants
were asked to indicate how much they valued each option on a 10-
point scale from 1 (not at all)to10(very much). Participants first rated
the wanted (i.e., chosen/retained) option and then rated the unwanted
(i.e., nonchosen/rejected) option on a separate page. (Further procedural
details for the framing manipulations and preference measures in all
studies are summarized in Supplemental Table S1.3.)
Last, participants indicated their gender and age. Participants
were asked to select between “male”and “female”options in Studies
1 and 2a. A third “nonbinary/others”option was added in all
subsequent studies.
Results
The postdecision preference gap in the reject condition (Δ
reject
=
2.56, 95% confidence interval, CI [2.30, 2.81]) was larger than that in
the choose condition (Δ
choose
=2.05, 95% CI [1.80, 2.31]), t(526) =
2.71, p=.007, d=0.24, 95% CI [0.07, 0.41]; see also Table 1.Both
options received moderately positive ratings (see Table 2).
DIP Screening
We conducted an additional step of analysis to assess whether the
larger gap in the reject condition incidentally resulted from unequal
distributions of DIP between conditions. To illustrate the rationale
of this method, consider two participants: Ann, who rated her
wanted option at least as positively as her unwanted option, and Ed,
who rated his wanted option less positively than his unwanted
option. Most participants are Anns, and few are Eds. Ann’s decision
and preference ratings were internally consistent, whereas Ed’s were
internally inconsistent—his decision and preference ratings formed
a preference reversal.
As aforementioned, Ed’s preference reversal could result from
various sources of noise. Critically, if there were more Eds in one
condition than the other, the calculated preference gaps could be
skewed, inflating their difference between conditions. More specifi-
cally, this is because, in calculating the postdecision preference gap for
any participant, the minuend is always the “wanted”option and the
subtrahend is always the “unwanted”option, which are determined by
the decision (and not by the preference ratings). Therefore, Ann’s
postdecision preference gap will be ≥0, whereas Ed’swillbe<0. This
means each Ed will shrink the average gap in his condition, and how
much he shrinks the gap depends on the extremeness of his ratings. For
simplicity, if we assume that each of Ed’s ratings were identically
extreme and if more Eds emerged in the choose condition, then
including all Eds in the data would inflate the average gap in the reject
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 1
Framing Effect in the Test Conditions (Non-Shaded) and Boundary Conditions (Shaded)
Study Additional condition
Postdecision
preference gap (Δ)
and cell size (n) Choose frame Reject frame Framing effect
1Δ2.05 [1.80, 2.31] 2.56 [2.30, 2.81] t(526) =2.71, p=.007, d=0.24, [0.07, 0.41]
n262 266
2a Δ2.08 [1.67, 2.48] 3.02 [2.60, 3.43] t(254) =3.18, p=.002, d=0.40, [0.15, 0.65]
n130 126
2b Δ2.90 [2.66, 3.14] 3.36 [3.12, 3.59] t(629) =2.67, p=.008, d=0.21, [0.06, 0.37]
n317 314
3 Moderately similar options Δ1.74 [1.45, 2.04] 2.34 [2.03, 2.64] t(231) =2.75, p=.006, d=0.36, [0.10, 0.62]
n120 113
Highly similar options Δ1.49 [1.25, 1.72] 1.48 [1.25, 1.70] t(235) =−0.06, p=.95 d=−0.01, [−0.26, 0.25]
n113 124
4a Positive choice set Δ4.09 [3.64, 4.54] 5.32 [4.86, 5.78] t(675) =4.20, p<.001, d=0.32, [0.17, 0.48]
n344 333
Negative choice set Δ5.53 [5.08, 5.98] 5.44 [4.98, 5.90] t(681) =−0.25, p=.80, d=−0.02, [−0.17, 0.13]
n346 337
4b Positive choice set Δ3.04 [2.77, 3.30] 4.46 [4.19, 4.73] t(784) =7.39, p<.001, d=0.52, [0.39, 0.67
n398 388
Negative choice set Δ3.67 [3.44, 3.89] 3.39 [3.16, 3.62] t(783) =−1.66, p=.097, d=−0.12, [−0.26, 0.02]
n397 388
5 Decide–rate Δ1.62 [1.44, 1.80] 2.49 [2.31, 2.67] t(714) =6.68, p<.001, d=0.50, [0.35, 0.65]
n356 360
Rate–decide Δ1.60 [1.40, 1.81] 1.39 [1.18, 1.59] t(714) =−1.46, p=.15, d=−0.11, [−0.25, 0.04]
n356 360
Note. The framing effect was observed in all the test (nonshaded) conditions and attenuated or reversed in the proposed boundary (shaded) conditions.
Values in square brackets represent 95% confidence intervals. The results are similar before and after DIP screening. DIP =decision-inconsistent
preferences.
Figure 1
The Two Smartphone Cases Used in Study 1 and Study 2a
FRAMING AFFECTS POSTDECISION PREFERENCES 7
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
condition relative to the choose condition. Alternatively, if more Eds
emerged in the reject condition, then excluding all Eds from the data
would instead inflate the average gap in the reject condition relative to
the choose condition.
Therefore, to assess whether the larger preference gap observed in
the reject condition was inadvertently inflated by unequal distributions
of Eds’between conditions, we compare the framing effect before and
after excluding Eds in the data, namely, DIP screening. If we find
similar effects before and after DIP screening, then the framing effect
is unlikely a result of inflated gap calculations. However, if we find a
significant framing effect only before or only after DIP screening, then
further tests should be conducted to evaluate the extent to which
decision-caused preference modulation accounted for the effect, as
opposed to Eds’differential presence between frames. As such, the
DIP method errs on the side of false negatives (i.e., Type II error),
which is appropriate for its primary purpose of monitoring potential
confounds for further testing. We will discuss the uses and limitations
of the DIP method more thoroughly in the General Discussion section.
In Study 1, DIP screening excluded similar ratios of participants
in both conditions (2% vs. 2%), χ
2
(1) =.001, p=.98. The framing
effect persisted with a similar magnitude after DIP screening,
t(516) =2.85, p=.005, d=0.25, 95% CI [0.08, 0.42] (see also
Table 2 and Supplemental Table S1.1). Therefore, the effect is not
attributable to potential calculation confounds.
Discussion
Study 1 documents initial evidence for the hypothesized framing
effect: Reject framing intensified subsequent preferences more than
choose framing. Since very few participants had preference reversals
in either condition, their presence or absence barely affected the effect.
Thus, the framing effect is unlikely to be explained by unequal
distributions of decision-preference inconsistencies and associated
alternative explanations (e.g., the different likelihood that choose
and reject decisions misrepresent existing preferences). Instead, the
framing effect in the present study primarily reflects greater preference
modulation caused by the reject decision than the choose decision.
Some might wonder, did the preference gap differences between
frames emerge primarily around the wanted option or around the
unwanted option? Although our theorizing does not yield a prediction
on this second-order question, when we compared their differences,
we found that both contributed to the gap across all studies. These
additional results are summarized in the General Discussion section
with full details reported in Supplemental Table S4.1.
Currently,the results in Study 1 are compatible with bothinference
and dissonance accounts. We start to disentangle these two accounts
in subsequent studies, all of which include the DIP screening method
to monitor potential confounds.
Studies 2a and 2b: Diagnosticity or Dissonance?
We tested both Hypotheses 1 and 2 in these two studies. To reiterate,
the inference account suggests that decision makers perceive their own
reject action as more diagnostic of their underlying preferences
than choose action, which results in more preference updating. The
dissonance account suggests instead that, if more preference updating
is found in the reject condition, then it should be attributed to a stronger
dissonance experienced by the decision makers that is induced by the
reject action than the choose action. Correspondingly, we measured
both perceived action diagnosticity and experienced dissonance as
exploratory tests in Study 2a and then confirmed in Study 2b.
Study 2a
We preregistered this study on AsPredicted.Org at https://aspredi
cted.org/vx8g3.pdf.
Method
Participants
We recruited 280 Amazon Mechanical Turk participants and
received 307 data entries. After standard screening, we obtained 256
valid responses from 163 male and 93 female participants with an
average age of 36.
Procedure
Study 2a was based on the same stimuli and procedure in Study 1.
One minor difference from Study 1 was that participants were first
asked to rate the chosen/rejected option, followed by the unchosen/
retained option. All subsequent studies measured preference ratings
in this order.
After those, we included three sets of preregistered exploratory
variables. First and most pertinent to our theorization was a three-
item measure of perceived action diagnosticity, adapted from
Andersen (1984) and Touré-Tillery and Light (2018). Participants
were asked to indicate the degree to which they agreed that their
action of (choosing/not choosing/retaining/rejecting) each option
“reflects my real preference,”“…is an opportunity for me to express
myself,”and “… is a reflection of my identity,”from 1 (definitely
disagree)to10(definitely agree).
Second, we included a three-item measure of experienced
dissonance, adapted from prior research (Elliot & Devine, 1994;
Gerard, 1967;Izuma et al., 2010). Participants were asked to
indicate how “difficult,”“conflicted,”and “uncomfortable”it was
to (choose/not choose/retain/reject) each option from 1 (not at all)
to 10 (very).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 2
Preference Ratings by Option and by Frame in Study 1
Option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Wanted option 5.02 [4.71, 5.32] 5.42 [5.12, 5.72] 5.00 [4.70, 5.31] 5.42 [5.12, 5.73]
Unwanted option 2.96 [2.71, 3.21] 2.86 [2.61, 3.11] 2.88 [2.64, 3.12] 2.78 [2.54, 3.02]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
8YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Third, we asked participants how definite their actions of
(choosing/not choosing/retaining/rejecting) the options felt on a
10-point scale from 1 (not at all)to10(very). Based on prior findings
that postdecision satisfaction depends on perceived decision closure
(Gu et al., 2013), we included this measure to explore the possibility
that preference updating, if attributable to preference inferences from
own decisions, may also be facilitated when a stronger sense of closure
is perceived from the decision.
Last, we included two sets of individual difference measures to
explore if the framing effect was contingent on any personal traits,
such as decisiveness (measured with the “trait decisiveness”
subscale from the Need for Cognitive Closure Inventory in Webster
& Kruglanski, 1994) and self-esteem (Rosenberg, 1979).
Results
The postdecision preference gap in the reject condition (Δ
reject
=
3.02, 95% CI [2.60, 3.43]) was larger than that in the choose
condition (Δ
choose
=2.08, 95% CI [1.67, 2.48]), t(254) =3.18,
p=.002, d=0.40, 95% CI [0.15, 0.65] (Tables 1 and 3), replicating
the framing effect.
DIP screening resulted in similar additional attritions in the two
conditions (2% vs. 4%), χ
2
(1, N=256) =0.58, p=.45, which did
not significantly influence the results (Table 3 and Supplemental
Table S1.1). Next, we report the process findings based on data
before DIP screening. Results after DIP screening remain highly
similar and are reported in the Supplemental Section 5.
Action Diagnosticity (Cronbach’sα=.85)
Consistent with Hypothesis 2, participants in the reject condition
perceived their decision to be more diagnostic of their own
preferences than participants in the choose condition, M
reject
=6.12,
95% CI [5.73, 6.51] versus M
choose
=5.56, 95% CI [5.18, 5.94],
t(254) =2.05, p=.042, d=0.25, 95% CI [0.01, 0.50]. Action
diagnosticity positively predicted the size of the postdecision
preference gap (b
diagnosticity
=0.50, SE =0.06, p<.001) and
reduced the coefficient of framing on the preference gap when
included in the regression model (from b=0.94, p=.002 to b=
0.66, p=.013). Action diagnosticity mediated the framing effect
(indirect effect =−0.28, 95% CI [−0.57, −0.02]).
Experienced Dissonance (α=.91)
Participants in the reject condition reported less dissonance than
participants in the choose condition, M
reject
=2.14, 95% CI [1.82,
2.46] versus M
choose
=2.62, 95% CI [2.31, 2.94], t(254) =2.11,
p=.036, d=0.26, 95% CI [0.02, 0.51]. This is the opposite of what
would be expected if dissonance explained the observed framing
effect.
Decision Closure
Perceived decision closure moderated the framing effect, interaction
F(1, 252) =22.18, p<.001, η2
p=.08; M
closure
=7.79, J-Nvalue =
7.57, with a relatively stronger effect among participants who
perceived greater decision closure (conditional effect at M
closure
−1
SD =0.60, 95% CI [−0.14, 1.34]; conditional effect at M
closure
+1
SD =−1.91, 95% CI [−2.64, −1.18]). These results suggest that, in
this study, perceived decision closure amplified postdecision prefer-
ence updating—preference updating was more likely to follow a
decision that felt completed.
PID Measures
Neither trait decisiveness nor self-esteem moderated the framing
effect (Fs<1.57, ps>.21; see details in Supplemental Section 6). In
other words, the framing effect applied to participants irrespective of
individual differences in these traits.
Study 2b
We conducted Study 2b to further replicate the framing effect
with different stimuli and to provide a confirmatory test for the
process findings in Study 2a. The study was preregistered On
AsPredicted.Org at https://aspredicted.org/bp2v7.pdf.
Method
Participants
We preregistered the Pocock boundary method (Pocock, 1977)
for the framing effect on action diagnosticity, the effect size of which
was smaller than that of the preference gaps in Study 2a. This
ensured sufficient power to detect both effects. More specifically, we
planned to recruit 626 participants on Prolific for at least 80% power
to detect a between-participants difference in action diagnosticity
with a significance level of p<.05 (based on the effect size from
previously conducted studies using similar stimuli; see Supplemental
Studies S5 and S6). We specified that if the above analysis reached
the Pocock boundary of p<.0294 (Pocock, 1977), then we would
terminate data collection; if not, then we would continue data collection
until reaching a total sample size of 1,043 participants for 95% power
onthesameanalysis.
We received 667 data entries and obtained 631 valid responses
(M
age
=39, 327 male, 292 female, 12 nonbinary/others) after
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 3
Preference Ratings by Option and by Frame in Study 2a
Option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Wanted option 5.54 [5.09, 5.99] 5.83 [5.37, 6.28] 5.58 [5.13, 6.04] 5.91 [5.45, 6.37]
Unwanted option 3.46 [3.09, 3.84] 2.81 [2.43, 3.19] 3.40 [3.03, 3.77] 2.65 [2.27, 3.03]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
FRAMING AFFECTS POSTDECISION PREFERENCES 9
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
standard screening. The effect of framing on action diagnosticity
reached p=.022. Hence, we terminated data collection.
Procedure
We kept all procedures in Study 2b identical to those in Study 2a
except the stimuli. Participants were presented with two art prints:
“The Dessert”(1901) by Pablo Picasso and “Morning on the Seine
Near Giverny”(1897) by Claude Monet. The paintings were presented
without the titles or the names of the artists (Figure 2).
Results
The postdecision preference gap in the reject condition (Δ
reject
=
3.36, 95% CI [3.12, 3.59]) was larger than that in the choose
condition (Δ
choose
=2.90, 95% CI [2.66, 3.14]), t(629) =2.67,
p=.008, d=0.21, 95% CI [0.06, 0.37]. Again, this framing effect
was observed both with and without DIP screening (Tables 1 and 4).
DIP screening resulted in similar attritions in the two conditions (3%
vs. 2%), χ
2
(1, N=631) =1.13, p=.29, which did not affect the
results.
Action Diagnosticity (α=.81)
Again, perceived action diagnosticity was higher in the reject
condition than the choose condition, M
reject
=5.88, 95% CI [5.67,
6.09] versus M
choose
=5.53, 95% CI [5.33, 5.74], t(629) =2.30,
p=.022, d=0.19, 95% CI [0.03, 0.34], consistent with H2. Action
diagnosticity positively predicted the size of the preference gap
(b
diagnosticity
=0.43, SE =0.04, p<.001) and reduced the coefficient
of framing on the preference gap when included in the regression
model (from b=0.46, p=.008 to b=0.31, p=.052). Action
diagnosticity mediated the framing effect (indirect effect =−0.15,
95% CI [−0.29, −0.03]).
Experienced Dissonance (α=.90)
Different from Study 2a, participants in the reject condition
reported more dissonance than participants in the choose condition
(M
reject
=2.78, 95% CI [2.59, 2.97] versus M
choose
=2.35, 95% CI
[2.16, 2.54]), t(629) =3.14, p=.002, d=0.25, 95% CI [0.09, 0.41].
However, more dissonance predicted a smaller postdecision
preference gap (b
dissonance
=−0.40, SE =0.05, p<.001), and
including dissonance in the regression modelincreased the coefficient
of framing on the preference gap (from b=0.46, p=.008 to b=0.63,
p<.001). In other words, in this study, the framing effect was found
despite participants reporting stronger dissonance in the reject
condition than in the choose condition.
Decision Closure
Decision closure did not moderate the framing effect, interaction
F(1, 627) =0.30, p=.59, in this study.
Discussion
We replicated the framing effect in Studies 2a and 2b with different
stimuli, further supporting Hypothesis 1. Additionally, both studies
provide process evidence for the role of action diagnosticity underlying
the framing effect, in line with Hypothesis 2 and the inference account.
Meanwhile, the dissonance measures did not accord with the framing
effect.
Specifically, while experienced dissonance differed significantly
between the choose and reject conditions in both studies, they
differed in opposite directions in the two studies, yet they did not
contribute to the framing effect in either study. In Study 2b, in
particular, even when dissonance differed in the same direction as
would be expected by the dissonance account, the framing effect
occurred despite dissonance differences. In other words, even if
framing affects dissonance in some ways, the dissonance account
cannot explain the current framing effect. Last, it should be noted
that average dissonance was low (Ms<2.80) in all conditions, further
casting doubt on the relevance of dissonance to the framing effect. By
contrast, action diagnosticity was moderately high (Ms>5.50) in all
conditions while both were measured on 1–10 scales.
Last, on the role of perceived decision closure, our results in these
two studies were inconclusive. Decision closure amplified the
framing effect in Study 2a but not in Study 2b.
When we included all the above process variables in several
additional replications reported in the Supplemental Material,we
found similar patterns of results to those reported above. These
include consistent mediations by action diagnosticity, no consistent
results from experienced dissonance, and a directional moderating
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Figure 2
The Two Paintings in Study 2b
Note. See the online article for the color version of this figure.
Table 4
Preference Ratings by Option and by Frame in Study 2b
Option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Wanted option 7.39 [7.21, 7.58] 7.83 [7.64, 8.02] 7.46 [7.28, 7.64] 7.89 [7.71, 8.07]
Unwanted option 4.49 [4.26, 4.72] 4.47 [4.24, 4.70] 4.40 [4.17, 4.63] 4.44 [4.21, 4.67]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
10 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
role of perceived decision closure, which will be summarized and
further discussed in the General Discussion section.
The analyses in Studies 2a and 2b provide useful insights that help
adjudicate the two major accounts’relative fit with the current
framing effect. However, these analyses do not establish causal
evidence (Fiedler et al., 2011, 2018). In the subsequent studies, we
turn to testing causal evidence by manipulating theoreticalmoderators
(Spencer et al., 2005) of the inference account.
Study 3: Option Similarity
In Study 3, we manipulated attribute similarity between options to
test a key boundary condition of the framing effect (Hypothesis 3).
We preregistered the study on AsPredicted.Org and is accessible at
https://aspredicted.org/hf2qn.pdf.
Method
Participants
We recruited 533 undergraduate students from a large public
university for course credit and obtained 470 valid responses (M
age
=
21, 167 male, 302 female, one nonbinary/others) after standard
screening.
Procedure
Participants were randomly assigned to one of 2 (framing: choose
vs. reject) ×2 (option similarity: moderate vs. high) between-
subjects conditions. Each participant was presented with two
donburi bowls (i.e., rice topped with protein and vegetables) as
lunch options on a restaurant’s ordering page. We selected two pairs
of donburi bowls, a dish familiar to this population. Each bowl
contained seven ingredients. The moderately similar pair differed in
four ingredients; the highly similar pair differed in one ingredient
only (Table 5). In the choose conditions, participants were asked to
drag the preferred option into a box labeled “cart.”In the reject
conditions, participants were asked to drag the less preferred option
into a box labeled “bin.”After the decision, participants were asked
to indicate how much they liked each option on a 9-point scale from
1(very little)to9(very much).
Manipulation Check and Posttest
After rating both bowls, participants were asked to rate how
different the options were on a 10-point scale from 1 (very similar)
to 10 (very different). The results confirm our intended manipula-
tion, showing that the options in the moderate-similarity conditions
were deemed more different, M
moderately similar pair
=5.82, 95% CI
[3.45, 4.04] versus M
highly similar pair
=3.75, 95% CI [5.53, 6.12],
t(468) =9.71, p<.001, d=0.89, 95% CI [0.70, 1.08]. Additionally,
a posttest (N=253) found that the two bowls in the high-similarity
condition were also evaluated as more similar in attractiveness, on a
7-point scale from 1 (not similar)to7(very similar), than were those
in the moderate-similarity condition, M
moderately similar pair
=4.64,
95% CI [4.34, 4.93] versus M
highly similar pair
=5.48, 95% CI [5.18,
5.77], t(251) =3.96, p<.001, d=0.50, 95% CI [0.25, 0.75].
Exploratory Measures
We included three sets of exploratory measures. These measures
were conceptually consistent with those in Studies 2a and 2b but
were measured at the general decision level without specifying the
related actions and options. Since our main goal in this study was
to test a boundary condition, we made the wording changes to
shorten the study, reducing four questions for each variable down
to one (see full detail in Supplemental Table S1.4). However,
acknowledging that these changes might reduce the measures’
sensitivity, we treated and preregistered these shortened measures
as exploratory.
Results
As predicted, high option similarity mitigated the framing effect
in a two-way analysis of variance, F(1, 466) =4.95, p=.027, η2
p=
.01. The framing effect was replicated in the moderate-similarity
conditions, t(231) =2.75, p=.006, d=.36, 95% CI [0.10, 0.62],
and attenuated in the high-similarity conditions, t(235) =0.06,
p=.95 (Figure 3;Tables 1 and 6). Unsurprisingly, we also found a
main effect of framing (Δ
reject
=1.89, 95% CI [1.69, 2.08] vs.
Δ
choose
=1.62, 95% CI [1.42, 1.81]), F(1, 466) =4.60, p=.033,
η2
p=.01, and a main effect of manipulated option similarity, with a
generally larger overall preference gap in the moderate-similarity
conditions than in the high-similarity conditions (Δ
moderate similarity pair
=
2.03, 95% CI [1.84, 2.22] vs. Δ
high similarity pair
=1.48, 95% CI [1.29,
1.67]), F(1, 466) =16.78, p<.001, η2
p=.04.
DIP screening excluded similar ratios of participants in the two
frames (2% vs. 1%), χ
2
(1, N=470) =0.16, p=.69, without
significantly affecting the results. Again, high option similarity
mitigated the framing effect in a two-way analysis of variance, F(1,
459) =4.03, p=.045, η2
p=.01; the framing effect was replicated in
the moderate-similarity conditions, t(225) =2.59, p=.010, d=0.35,
95% CI [0.09, 0.61], and attenuated in the high-similarity conditions,
t(234) =0.06, p=.96 (Figure 3;Table 6,andSupplemental
Table S1.1).
Does this moderation merely reflect a “floor effect”in the high-
similarity conditions? That was not the case because the preference
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 5
Listed Ingredients in the Donburi Bowls
Choice set Ingredient
Moderately similar options Rice, teriyaki chicken, egg, sweet corn,mushroom,lettuce, sesame
Rice, grilled beef sirloin, egg, edamame,pickles,cucumber, sesame
Highly similar options Rice, teriyaki chicken, egg, sweet corn, mushroom, lettuce, sesame
Rice, teriyaki chicken, egg, sweet corn, mushroom, cucumber, sesame
Note. Ingredients that differ between options are in bold.
FRAMING AFFECTS POSTDECISION PREFERENCES 11
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
gaps were calculated as difference scores from the preference ratings
(in Table 6)—rather than measured on an artificially truncated scale.
In fact, as Table 6 shows, the smaller preference gaps in the high-
similarity conditions resulted from closer ratings between options,
all of which fell in the upper middle range of the 9-point (1–9) scale.
Further analyses confirmed that while most participants assigned
different ratings to the two options in both conditions, more
participants in the high-similarity conditions than in the moderate-
similarity conditions assigned the same rating to both options (19%
vs. 11%), χ
2
(1, N=470) =6.32, p=.012, consistent with the
posttest results.
The three sets of exploratory process measures did not produce
statistically meaningful results. In fact, we found across multiple
supplemental studies that measuring these variables at the decision
level without specifying the concrete actions and options was
generally ineffective (Supplemental Section 3;Supplemental Tables
S3.1 and S3.2). These results and their implications will be further
discussed in the General Discussion section.
Discussion
The framing effect diminished in a decision between options with
highly overlapping attributes, supporting Hypothesis 3 and further
corroborating the inference mechanism.
These results are not the most compatible with the dissonance
account. Dissonance theory has stressed an equal attractiveness
antecedent for postdecision dissonance. This antecedent was not
satisfied in the moderate-similarity conditions, in which the two
options were significantly dissimilar in attractiveness (as shown in the
posttest), yet the framing effect still occurred in those conditions. These
results further challenge the fit between the dissonance account and the
effect of framing on preference modulation. We also conducted
further analyses assessing the role of the equal attractiveness
antecedent on the framing effect across all our studies, which will
be reported and discussed in the General Discussion section.
Studies 4a and 4b: The Valence Moderator
Next, in line with Hypothesis 4, we examined choice set valence
as another moderator of the framing effect. The framing effect has
emerged in decisions involving primarily attractive options, and we
expected it to be mitigated or even reversed in decisions involving
unattractive options. We tested these predictions using images in
Study 4a and words in Study 4b. Both used a factorial design with 2
(framing: choose vs. reject) ×2 (choice set valence: positive vs.
negative) between-subjects conditions. The studies were preregis-
tered on AsPredicted.Org at https://aspredicted.org/uy9ce.pdf and
https://aspredicted.org/67yx6.pdf, respectively.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Figure 3
Framing Effect Attenuated Between Highly Similar Options
1.74
1.49
2.34
1.48
0
1
2
3
4
Moderately Similar Options Highly Similar Options
Preference Gap (Calculated)
Before DIP Screening
Choose Frame Rejec t Fr ame
1.85
1.49
2.40
1.50
0
1
2
3
4
Moderately Similar Options Highly Similar Options
Preference Gap (Calculated)
After DIP Screening
Choose Frame Rejec t Frame
Note. Error bars represent 95% confidence intervals. DIP =decision-inconsistent preferences.
Table 6
Preference Ratings by Option and by Frame in Study 3
Condition and option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Moderately similar options
Wanted option 7.12 [6.90, 7.33] 7.31 [7.09, 7.53] 7.16 [6.95, 7.38] 7.32 [7.10, 7.55]
Unwanted option 5.38 [5.05, 5.70] 4.97 [4.64, 5.31] 5.31 [4.99, 5.64] 4.93 [4.60, 5.26]
Highly similar options
Wanted option 6.78 [6.55, 7.01] 7.15 [6.93, 7.37] 6.78 [6.55, 7.01] 7.15 [6.93, 7.37]
Unwanted option 5.29 [4.99, 5.60] 5.67 [5.38, 5.96] 5.29 [4.99, 5.59] 5.65 [5.36, 5.94]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
12 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Study 4a
Method
Participants
We recruited 1,400 Prolific participants, received 1,441 entries,
and obtained 1,360 valid responses (M
age
=40, 640 male, 696
female, 24 nonbinary/others) after standard screening.
Procedure
We selected four photographs from the Open Affective
Standardized Image Set (Kurdietal.,2017), an open-access
stimulus set that provides 900 color images with validated valence
and arousal ratings. Figure 4 displays the four photographs, all
selected under the “Scene”category and with similar arousal
ratings (between 4.23 and 4.43 out of 7). For their valence (positive
vs. negative) ratings, the pair in the positive conditions were rated
6.12 and 6.16, and the pair in the negative conditions were rated
2.24 and 2.29 (out of 7; all images and ratings are available at
https://db.tt/yYTZYCga;Kurdi et al., 2017).
Participants were randomly assigned to one of four conditions.
Each participant was presented with a pair of photographs, both
positive or both negative. In the choose condition, participants were
asked to click on their preferred photograph. In the reject condition,
participants were asked to click on their less preferred photograph.
After making the decision, participants were asked to rate each
photograph on a bipolar scale from −10 (dislike this photograph
very much)to10(like this photograph very much). No exploratory
measures were included.
Results
First, before DIP screening, we found the Predicted Framing ×
Valence interaction, F(1, 1356) =8.06, p=.005, η2
p=.01. The
framing effect was replicated in the positive-images conditions,
t(675) =4.20, p<.001, d=0.32, 95% CI [0.17, 0.48], and
diminished in the negative-images conditions, t(681) =−0.25, p=.80
(Tables 1 and 7;Figure 5).
The DIP screener resulted in similar attritions in the choose and
reject frames (2% vs. 3%), χ
2
(1, N=1,360) =0.59, p=.44. The
results held after DIP screening, including the Framing ×Valence
interaction, F(1, 1320) =6.08, p=.014, η2
p=.01, the framing effect
in the positive-images conditions, t(660) =4.50, p<.001, d=0.35,
95% CI [0.20, 0.50], and the diminished effect in the negative-
images conditions, t(660) =0.62, p=.54 (Table 7 and Supplemental
Table S1.1;Figure 5).
Study 4b
Method
Participants
We recruited 1,600 Prolific participants, received 1,651 entries,
and obtained 1,571 valid responses (M
age
=39, 858 male, 688
female, 25 nonbinary/others) after standard screening.
Procedure
We used 10 pairs of words (Table 8). All were taken from Perfecto et
al. (2017), who matched up pairs of “extremely negative,”“slightly
negative,”“slightly positive,”and “extremely positive”words based
on valence ratings initially validated in Bellezza et al. (1986).Weonly
used the extreme pairs (and not the slightly positive and slightly
negative pairs), expecting that the more extremely evaluated stimuli
may facilitate the detection of a stronger moderation—with the framing
effect potentially reversed in the “extremely negative”conditions.
Participants were randomly assigned to one of the four conditions.
Each participant was asked to make five decisions, each involving
a pair of words. The five pairs were sequentially presented in
randomized order, one pair per page. In the choose condition,
participants were asked to mark their preferred word within each
pair with a green tick “✓.”In the reject condition, participants were
asked to mark their less preferred word within each pair with a red
cross “×.”After the decision, participants were asked to rate each
word on a bipolar scale similar to that in Study 4a. Each participant’s
postdecision preference gap was averaged from ratings of all five
pairs of words. No exploratory measures were included.
Results
First, before DIP screening, we found the predicted Framing ×
Valence interaction, F(1, 1567) =44.93, p<.001, η2
p=.03. The
framing effect was again replicated in the positive-words conditions,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Figure 4
Photographs in Study 4a
egamievitageNegamievitisoP
Note. See the online article for the color version of this figure.
FRAMING AFFECTS POSTDECISION PREFERENCES 13
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
t(784) =7.39, p<.001, d=0.52, 95% CI [0.39, 0.67], and reversed in
direction in the negative-words conditions, t(783) =−1.66, p=.097,
d=−0.12, 95% CI [−0.26, 0.02] (Figure 6;Tables 1 and 9).
Since each participant made five decisions, the DIP screener
excludes any participant with decision preference inconsistency in any
word pair. This led to higher attrition ratios in both frames (11% vs.
9%), χ
2
(1, N=1,571) =2.30, p=.13 compared with earlier studies.
The results held after DIP screening, including the Framing ×Valence
interaction, F(1, 1409) =56.12, p<.001, η2
p=.04, the framing effect
in the positive-words conditions, t(724) =7.20, p<.001, d=0.54,
95% CI [0.39, 0.69], and the reversed effect in the negative-words
conditions, t(685) =−3.11, p=.002, d=−0.24, 95% CI [−0.39,
−0.09] (Figure 6;Table 9 and Supplemental Table S1.1).
Discussion
Similar results were found in both Study 4a, using images, and Study
4b, using words. While the original framing effect was consistently
replicated in the positive conditions, it no longer emerged in the
negative conditions—when decisions involving primarily unattractive
options. These results support the moderating role of choice set valence
to the effect, in line with Hypothesis 4. Additionally, in an earlier, less
powered version of Study 4b employing the same word pairs and a
different cover story (in which the words were introduced as temporary
hand tattoo options), we found principally similar results to those in
Study4b(seeSupplemental Study S2). Together, these results indicate
that the valence moderator is robust to idiosyncratic differences in
stimuli and contexts.
The effect sizes in Studies 4a and 4b provide additional insights.
As can be seen in Table 7 and 9, the positive and negative stimuli in
Study 4b received more extreme ratings than those in Study 4a.
Correspondingly, both the framing effect in the positive conditions
(d=.52) and its directionally reversed effect in the negative
conditions (d=−0.12) in Study 4b were more sizable than those in
Study 4a (d=.32 and d=−0.02). In fact, in the negative conditions,
the original framing effect was only attenuated in Study 4a whereas
it was reversed in direction in Study 4b. These patterns coalesce with
previous framing studies (e.g., on post-decision confidence, Study 2
in Perfecto et al., 2017), in which the reversal of a framing effect was
facilitated by stimuli of more extreme valence.
We have now obtained empirical evidence for the four hypotheses
derived from the inference account. Although dissonance was not
measured in Studies 4a and 4b, it would be improbable for the
dissonance account to provide a coherent explanation for the current
results. This is because our earlier results from Studies 2a and 2b
already revealed that the simple framing effect in positive-valence
conditions was incompatible with the dissonance account, let alone
its reversal in negative-valence conditions.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 7
Preference Ratings by Option and by Frame in Study 4a
Condition and option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Positive images
Wanted option 6.70 [6.31, 7.09] 7.29 [6.90, 7.69] 6.74 [6.35, 7.13] 7.41 [7.01, 7.81]
Unwanted option 2.61 [2.21, 3.02] 1.98 [1.57, 2.39] 2.50 [2.11, 2.88] 1.90 [1.50, 2.29]
Negative images
Wanted option −1.15 [−1.54, −0.75] −0.70 [−1.09, −0.30] −1.06 [−1.45, −0.67] −0.62 [−1.02, −0.22]
Unwanted option −6.67 [−7.08, −6.27] −6.14 [−6.55, −5.73] −6.81 [−7.20, −6.43] −6.58 [−6.97, −6.19]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
Figure 5
The Framing Effect Diminished When Both Images Were Less Desirable
4.09
5.53
5.32 5.44
1
2
3
4
5
6
7
8
Posi tiv e Images Negative Images
Preference Gap (Calculated)
Before DIP Screening
Choo se Frame Rejec t Frame
4.25
5.76
5.52 5.96
1
2
3
4
5
6
7
8
Posi tiv e Images Negativ e Images
Preference Gap (Calculated)
After DIP Screening
Choose Frame Rejec t Fr ame
Note. All error bars represent 95% confidence intervals. DIP =decision-inconsistent preferences.
14 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Study 5: An Alternative Design to Assess Potential
Confounds
Thus far, we have replicated the framing effect in six studies using
true randomization, large samples and diverse stimuli. These results
indicate that the effect was robust to random errors (e.g., participants’
idiosyncratic differences in preference volatility or inattentiveness).
Additionally, the framing effect remained mostly unchanged before
and after DIP screening in all six studies, indicating that the effect was
not significantly affected by systematic errors either (e.g., inflated gap
calculations due to unequal distributions of preference reversals).
In Study 5, we sought to hone in on the role of systematic error on
the framing effect using yet another method. This method was
inspired by one of the tested solutions offered by Chen and Risen
(2010) to address previous confounds in the FCP (Brehm, 1956).
We modified their solution to fit our framing paradigm. Specifically,
we added a new set of control conditions, in which the order between
the decision and preference ratings reversed such that the decision
cannot possibly influence the preference ratings in these conditions.
If the framing effect primarily reflects different degrees of decision-
induced preference modulation, as we theorized, then it should not
emerge in these control conditions. In contrast, if framing systemati-
cally affected the extent to which a participant’s decision would
misrepresent their existing preferences, then the effect should emerge
similarly in both experimental and control conditions. We preregistered
this study and its full analysis plan on AsPredicted.Org at https://aspre
dicted.org/ej4uz.pdf.
Method
Participants
We recruited 1,400 Prolific participants, received 1,484 entries,
and obtained 1,432 valid responses (M
age
=38, 623 male, 774
female, 35 nonbinary/others) after standard screening.
Procedure
Participants were randomly assigned to 2 (framing: choose vs.
reject) ×2 (order: decide–rate vs. rate–decide) between-subjects
conditions. Each participant was presented with a pair of words,
“pleasure”and “vacation,”taken from Study 4b. The decide–rate
(D-R) conditions were similar to the positive-valence conditions in
Study 4b. The rate–decide (R-D) conditions mirrored the D-R
conditions. The only difference was that participants were first asked
to rate the two words, and then to make a decision. In other words, in
the R-D conditions, the decision followed preference ratings, rather
than preceding them. Decision-induced preference modulation could
only occur in the D-R conditions. However, if framing influences the
likelihood of preference reversal, then it should impact both D-R and
R-D conditions.
Results
Following all three sets of preregistered analyses, the framing
effect was observed only in the D-R conditions and not in the R-D
conditions. Below, we report them in order.
First, before DIP screening, the framing effect emerged in the D-R
conditions, t(714) =6.68, p<.001, d=0.50, 95% CI [0.35, 0.65],
and not in the R-D conditions, t(714) =−1.46, p=.15. The
mitigation of the framing effect in the R-D conditions was validated
by an interaction between framing and (D-R vs. R-D) order, F(1,
1428) =30.20, p<.001, η2
p=.02 (Figure 7 left panel; Tables 1 and 10).
Moreover, the framing effect was primarily driven by gap differences
within the reject conditions, simple effect between the black bars:
t(718) =7.68, p<.001, d=0.57, 95% CI [0.42, 0.72]. The two gaps in
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 8
Word Pairs in Study 4b
Extremely positive word Extremely negative word
joy vs. kiss murderer vs. tumor
pleasure vs. vacation poison vs. slaughter
family vs. laughter war vs. maggot
paradise vs. sunrise cancer vs. funeral
romantic vs. love lice vs. suicide
Figure 6
The Framing Effect Was Reversed When Both Words Were Negative in Study 4b
3.04
3.67
4.46
3.39
1
2
3
4
5
6
Positi ve Word s Negat iv e Wor ds
Preference Gap (Calculated)
Before DIP Screening
Choose Frame Rejec t Frame
3.19
4.09
4.62
3.58
1
2
3
4
5
6
Positi ve Word s Negati ve Word s
Preference Gap (Calculated)
After DIP Screening
Choose Frame Rejec t Fr ame
Note. All error bars represent 95% confidence intervals. DIP =decision-inconsistent preferences.
FRAMING AFFECTS POSTDECISION PREFERENCES 15
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
the choose conditions did not significantly differ, simple effect between
the white bars: t(710) =0.17, p=.87.
Second, DIP screening excluded similar ratios of participants in
choose and reject conditions (3% vs. 3%), χ
2
(1, N=1,432) =0.26,
p=.61. The results remained similar including the framing effect in the
D-R conditions, t(703) =6.52, p<.001, d=0.49, 95% CI [0.34, 0.64],
its absence in the R-D conditions, t(684) =−1.44, p=.15 (Table 10
and Supplemental Table S1.1), and the interaction between framing and
order, F(1, 1387) =28.93, p<.001, η2
p=.02 (Figure 7, middle panel).
Third, we computed an alternative “high–low gap”by subtracting
the lower rating from the higher rating between options without DIP
screening. In this calculation, the minuend and subtrahend were
solely determined by preference ratings, independent of the decision
outcome (which determined the wanted and unwanted options in all
previous calculations of preference gaps). This “high–low gap”
reflects the face-value preference gap between options irrespective
of the decision outcome, since the minuend was always the higher
rating. This means that each bar in the R-D conditions represents a
baseline preference gap plus random error whereas each bar in the
D-R conditions represents a preference gap caused by having made a
(choose vs. reject) decision, on top of the baseline preference gap
plus random error. Again, all our results remained, including the
framing effect in the D-R conditions, t(714) =6.61, p<.001, d=0.49,
95% CI [0.34, 0.64], its absence in the R-D conditions, t(714) =−1.46,
p=.15, d=−0.11, 95% CI [−0.25, 0.04] (Tables 1 and 10), and the
significant interaction between framing and decision-rating order,
F(1, 1428) =30.13, p<.001, η2
p=.02 (Figure 7, right panel).
Discussion
Study 5 confirms that the framing effect primarily reflects more
preference changes caused by the reject (vs. choose) decision
instead of potential calculation confounds associated with unequal
distributions of preference reversals between frames. These results
corroborate our previous results using the DIP screening method.
Besides Study 5, we also report Supplemental Study S3, which used
the same design as Study 5, the same smartphone cases in Study 1,
and yielded principally similar results; although it was underpow-
ered to detect a significant two-way interaction between framing and
order, it replicated both the framing effect in the R-D conditions and
the null effect in the D-R conditions.
General Discussion
Across seven preregistered experiments, we reported that framing
a decision as a rejection intensified postdecision preferences
compared to framing the same decision as a choice. This framing
effect was observed across diverse positive stimuli including
products, art prints, food, photos, and words. The larger preference
gap in the reject (vs. choose) frame was consistent with the greater
perceived diagnosticity of the reject (vs. choose) action. Moreover,
the framing effect was moderated by two theoretical moderators:
option similarity and choice set valence. Together, our findings
support our hypotheses derived from the inference account and, by
necessity, support the inference account for preference modulation.
In contrast, the dissonance account received little support when
tested in these studies.
The framing effect was further replicated in another six preregistered
supplemental studies (N=3,679) reported: Supplemental Studies S1,
S2, and S3 were similar to Studies 1, 4b, and 5, respectively, with
minor procedural differences. Supplemental Studies S4, S5, and S6
extended the effect’s robustness to larger choice sets in which two out
of four art prints were chosen or rejected (Supplemental Study S4)and
to procedures that required participants to write down their preferences
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 9
Preference Ratings by Option and by Frame in Study 4b
Condition and option
Before DIP screening After DIP screening
Choose frame Reject frame Choose frame Reject frame
Positive words
Wanted option 7.63 [7.40, 7.86] 7.16 [6.93, 7.40] 7.80 [7.57, 8.03] 7.29 [7.05, 7.52]
Unwanted option 4.60 [4.25, 4.94] 2.70 [2.36, 3.05] 4.61 [4.25, 4.96] 2.67 [2.30, 3.03]
Negative words
Wanted option −3.52 [−3.88, −3.18] −3.55 [−3.90, −3.20] −3.39 [−3.76, −3.03] −3.62 [−3.97, −3.26]
Unwanted option −7.19 [−7.47, −6.92] −6.94 [−7.22, −6.67] −7.48 [−7.75, −7.22] −7.20 [−7.46, −6.94]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
Table 10
Preference Ratings by Option and by Frame in Study 5
Condition and
option
Before DIP screening After DIP screening “High–low gaps”
Choose frame Reject frame Choose frame Reject frame Option Choose frame Reject frame
Decide–rate
Wanted option 8.47 [8.33, 8.61] 8.91 [8.78, 9.05] 8.51 [8.37, 8.65] 8.92 [8.79, 9.06] Higher rated option 8.51 [8.37, 8.64] 8.91 [8.78, 9.05]
Unwanted option 6.84 [6.63, 7.06] 6.42 [6.21, 6.63] 6.82 [6.60, 7.03] 6.40 [6.19, 6.62] Lower rated option 6.83 [6.62, 7.04] 6.41 [6.19, 6.62]
Rate–decide
Wanted option 8.97 [8.83, 9.11] 8.99 [8.86, 9.13] 9.06 [8.92, 9.19] 9.06 [8.93, 9.20] Higher rated option 9.02 [8.89, 9.15] 9.06 [8.93, 9.19]
Unwanted option 7.37 [7.13, 7.61] 7.61 [7.37, 7.84] 7.33 [7.09, 7.58] 7.55 [7.31, 7.79] Lower rated option 7.29 [7.06, 7.53] 7.55 [7.31, 7.78]
Note. Values in square brackets represent 95% confidence intervals. DIP =decision-inconsistent preferences.
16 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
before the decision (Supplemental Studies S5 and S6). Their main
results are summarized in Supplemental Table S2.1. Across all 13
studies, we did not find the size of the framing effect to depend on
specific procedural operationalizations of the framing manipulation
(e.g., dragging, clicking, or marking options; summarized in
Supplemental Table S1.3).
Other Decision Consequences
Next, we report findings of second-order theoretical interest. First,
as briefly mentioned after Study 1, we examined (as preregistered)
whether the framing effect was driven more by the wanted option or
the unwanted option. Across all 13 studies, we found that both
options contributed to the preference modulation, and neither option
had a consistently larger contribution (Supplemental Table S4.1).
Second, we examined whether framing affected choice shares
(i.e., the binary decision outcome) and average preference ratings.
We did not find the choice shares of options to differ systematically
between the choose and reject conditions (Supplemental Table S4.3),
nor did the postdecision preference ratings systematically differ
between choose and reject conditions (Supplemental Table S4.2).
Therefore, the current framing effect was not contingent on framing,
altering the decision outcomes, or shifting the average evaluation.
Third, returning to the equal attractiveness antecedent proposed
by dissonance theory, we explored whether the similarity in
attractiveness between options affected the size of the framing
effect. Since we did not equate option attractiveness within each pair
of options in most studies, their differences varied considerably. We
thus computed an “attractiveness difference score”in each study
based on the average preference ratings of each option (see
Supplemental Table S4.4). Then, we plotted the score in each study
against the observed size of the framing effect in Figure 8. If equal
attractiveness was indeed an antecedent for preference modulation,
and hence a necessary condition to observing the framing effect,
then observed effect sizes should cluster on the far left of the x-axis
where the options are equal in attractiveness. Moreover, average effect
size should quickly decline as the options diverge in attractiveness,
with a sharp downward slope representing a negative correlation
between attractiveness difference scores and effect sizes.
However, as Figure 8 shows, the framing effect occurred in a wide
range of attractiveness difference scores, mostly away from zero.
Their effect sizes did not correlate with the attractiveness difference
score, Pearson’sr=−0.165, p=.59. These results suggest that
preference modulation did not depend on the equal attractiveness
antecedent, consistent with the results in Study 3. The results are
highly similar when we computed attractiveness differences from
choice shares.
What do these results suggest for dissonance theory? There are
two interpretations. One, let us assume that dissonance theorists
were right about the “equal attractiveness”antecedent (again,
although prior studies did not adequately test this assumption; see
Footnote 3), and then perhaps the options in our studies do not meet
the stringent criteria of being “equal”in attractiveness while
sufficiently “dissimilar”in attributes to trigger dissonance—a rather
narrow range of stimuli in real-world settings. This suggests that
dissonance is not as broadly applicable as has been presumed.
Alternatively, perhaps the equal attractiveness antecedent was a
faulty and unnecessary assumption for dissonance theory. In that
case, the results reported in Study 3 and Figure 8 are no longer
evidence against the dissonance account, even though the dissonance
account remains incompatible with the framing effect based on results
in Studies 2a and 2b (and more supplemental studies finding null results
of dissonance, as summarized below). Either way, the dissonance
account cannot explain the current framing effect. Dissonance
researchers may reconsider and/or reexamine the necessary conditions
of dissonance and its boundaries in light of our findings.
Additional Results on Process Variables
In addition to Studies 2a and 2b, we measured perceived action
diagnosticity, experienced dissonance, and perceived decision
closure in five of the six supplemental studies. Analyses on these
measures revealed consistent patterns, which we summarize here
using an internal meta-analyses (IMA) conducted on the test
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Figure 7
Postdecision Preference Gaps in Study 5
1.62
1.60
2.49
1.39
0
1
2
3
4
5
6
Deci de-Rat e
Conditions
Rate-Dec id e
Conditions
Before DIP Screening
Choose Frame Rejec t Fr ame
1.69 1.73
2.52
1.52
0
1
2
3
4
5
6
Deci de-Rat e
Conditions
Rate-Dec id e
Condit io ns
After DIP Screening
Choose Frame Reject Fr ame
1.69 1.71
2.51
1.51
0
1
2
3
4
5
6
Deci de-Rat e
Conditions
Rate-Dec id e
Conditions
"High-Low" Gaps
Choose Frame Reject Fram e
Note. Each panel corresponds to one set of preregistered analysis. The framing effect was expected in the D-R conditions and never expected
in the R-D conditions. Positive preference gaps (bars above 0) were expected in all conditions irrespective of framing or the order between
decision and ratings. D-R =decide–rate; R-D =rate–decide; DIP =decision-inconsistent preferences.
FRAMING AFFECTS POSTDECISION PREFERENCES 17
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
conditions of all eight studies in which these variables were
measured (N=5,852). While we report all our studies in the article
and the Supplemental Material, we acknowledge and agree with
important critiques that the IMA can inadvertently inflate effect sizes
(see Vosgerau et al., 2019). Thus, we intend to use the IMAs to
succinctly highlight consistent patterns in our findings, not to claim
specific effect sizes.
First, we found that perceived action diagnosticity was
significantly higher in the reject frame than the choose frame in
the test conditions of every study when diagnosticity was measured
at the action and option level (i.e., in reference to the concrete
actions and corresponding options; in Studies 2a, 2b, Supplemental
Studies S1, S5, and S6;Supplemental Tables S1.4 and S3.1)
whereas null results were found when this variable was measured at
the decision level (in Studies 3, Supplemental Studies S2, and S4;
Supplemental Table S3.1). We surmise that the generic wording in
the decision-level measurement failed to engage participant attention
effectively. Overall, perceived diagnosticity differed significantly
between framing conditions across all eight studies (average
diagnosticity difference =−0.26, SE =0.08, 95% CI [−0.42,
−0.11], Z=−3.34, p<.001) and mediated the framing effect
(indirect effect =−0.10, SE =0.04, 95% CI [−0.17, −0.03],
Z=−2.70, p=.007), with sizable heterogeneity across studies,
I
2
=57%, 95% CI [10%, 93%].
Second, experienced dissonance did not consistently differ
between frames regardless of measurement type (Supplemental
Table S3.2), nor did it mediate the framing effect (indirect effect =
−0.01, SE =0.05, 95% CI [−0.09, 0.08]), Z=−0.14, p=.89. These
results corroborate our earlier results to show that dissonance was
unlikely the primary mechanism for the framing effect. Given
that the self-report measures of dissonance produced statistically
significant yet contradictory results (rather than null effects) between
Studies 2a and 2b, measurement sensitivity issues alone cannot
explain these discrepant results. If alternative measures of dissonance
(e.g., physiological arousal; Croyle & Cooper 1983)weretobeused
in future studies, researchers should address not only potential
measurement sensitivity issues but also the discrepancies inthe results
across studies.
Third, we found that perceived decision closure amplified the
framing effect in four out of the eight studies in which it was
measured (Study 2a, Supplemental Studies S1, S2, and S4;
Supplemental Section 3.3). Overall, the framing effect was stronger
at higher levels of decision closure (conditional effect at M
closure
+1
SD =−0.76, SE =0.16, 95% CI [−1.08, −0.45], Z=−4.75,
p<.001) and weaker at lower levels of decision closure (conditional
effect at M
closure
−1SD =−0.27, SE =0.20, 95% CI [−0.65, 0.12],
Z=−1.37, p=.17). Taken together, this moderation pattern
suggests that preference updating more readily occurs when a decision
is deemed complete. These additional findings are consistent with the
inference account. They do not help adjudicate the applicability of the
dissonance account because perceived closure may increase or
decrease dissonance (e.g., Carmon et al., 2003;Stalder, 2010).
Last, we explored decision time differences, a measure imbedded
on the decision page in Studies 2b, 3, 4, Supplemental Studies S2,
S4, and S5. In all the test conditions of these studies (N=3,122), we
found that the reject decision took longer time than the choose
decision (average difference =1.63, SE =0.36, 95% CI [0.91, 2.34],
Z=4.47, p<.001), consistent with the greater processing depth
in the reject (vs. choose) frame reported in prior research (e.g.,
Sokolova & Krishna, 2016). However, this difference emerged not
only in the test conditions but also in the boundary conditions (e.g.,
Study 3 and Supplemental Study S2;Supplemental Table S3.6),
in which framing did not affect preference gaps. Indeed, decision time
differences did not mediate the framing effect (indirect effect <0.001,
SE =0.006, 95% CI [−0.01, 0.01], Z=0.02, p=.98). Based on these
results and related critiques on decision time as a rather crude measure
of psychological processes (Evans et al., 2015), we surmise that
longer decision time may accompany the reject frame as a concomitant
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Figure 8
No Correlation Between Observed Effect Sizes and Attractiveness Differences Between Options In the Test
Conditions of All 13 Studies (r =−0.165, p =.59)
Study 1
Study 2a
Study 2b
Study 3
Study 4b
Study 5
Study S1
Study S2
Study S3
Study S4
Study S5
Study S6
-0.40
-0.20
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00
Effect Size (Cohen's d)
Aracveness Difference Score
Note. See the online article for the color version of this figure.
18 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
but not the most reliable process measure nor a unique explanation for
the framing effect.
Theoretical, Methodological, and Practical Implications
This research makes several contributions. First, empirically, we
present the first evidence that choose/reject framing affects
postdecision preference modulation. This finding extends previous
research, which had focused on the impact of framing on decision
outcomes (Dhar & Wertenbroch, 2000;Shafir, 1993). We show that
framing can influence postdecision preferences beyond decision
outcomes—even without affecting decision outcomes per se.
Moreover, by establishing that reject framing intensifies prefer-
ences more than choose framing, we join increasing research on the
multifaceted effects of reject framing, such as in eliciting
deliberative judgments (Sokolova & Krishna, 2016) and boosting
actions’social influence (Nan et al., 2023). The reject frame and its
potential implications merit more research in the future.
Second, we address a long-standing theoretical ambiguity
surrounding the psychological mechanisms of preference modula-
tion. Despite common references to preference modulation as “a
dissonance phenomenon”in the literature, the dissonance account
lacked support when put to the test in the current context whereas the
inference account readily predicted and explained all our results.
These findings show that preference modulation can often occur
without invoking dissonance. These findings also suggest that
preference modulation may be more commonplace than previously
conjectured based on the assumed dissonance explanation. To be clear,
our results should not be interpreted as refuting dissonance as a
plausible explanation for all preference modulation or all phenomena
of rationalization. Rather, our results suggest that dissonance is
probably not the best explanation for preference changes induced by
ordinary daily decisions—such as those examined in our studies. Such
quotidian decision contexts are arguably unlikely to elicit from people
extraordinary emotional or motivational conflicts after all.
Indeed, many psychologists consider the inference account more
parsimonious (see Bem, 1967;Laurin & Jettinghoff, 2020) for it
does not necessitate any emotional–motivational dynamics. The
inference account is based on the simple premise that spontaneous
observation-based inferences may operate similarly in intrapersonal
and interpersonal processes. As such, our findings align with
converging insights that the dissonance and inference accounts may
differ in their primary domains of application. Drawing from our own
and prior findings, we surmise that the inference mechanism may be
generally more relevant to common ordinary decisions whereas the
dissonance mechanism may be more relevant to more extraordinary
situations (or “flagrant behavioral situations,”Simon & Holyoak, 2002,
p. 284; see also Abelson, 1983)—in which overtly attitude-
incongruent, painstaking, or embarrassing actions do evoke intense
emotions and motivational conflicts.
Another smaller theoretical contribution may stem from our
findings on the self-diagnositicity of a decision, which was bolstered
by the reject frame. These findings cast a spotlight on the
underexplored role of self-diagnosticity (Bodner & Prelec, 2003;
Dhar & Wertenbroch, 2012) in modulating the relationship between
decisions and preferences. Future research may find it fruitful to
investigate situational factors that determine the diagnosticity of
decisions and actions, with potential implications on cognitive
consistency, identity formation, and behavior change.
Third, this research offers methodological contributions.
Traditionally, research on preference modulation relied on measur-
ing pre- versus postchoice preference spreading in FCP and its
variants, which often involved laborious and sometimes confounded
protocols. Our approach circumvents the need to examine preference
spreading. We employ a simple framing paradigm, examining only
postdecision preference gaps. Within this paradigm, different levels of
noise in preference expression through decisions were ruled out based
on converging evidence from DIP screening and additional control
conditions in the R-D versus D-R design. Future studies adopting our
framing paradigm should also use these methods to address systematic
error. Additionally, future research should adhere to rigorous practices
(e.g., sufficient statistical power, preregistered replications) to minimize
random error—another major cause of spurious findings in this
literature. We hope our novel methods can help researchers overcome
previous methodological challenges and facilitate more theory testing
on the phenomenon of preference modulation.
Last, this research provides rich insights into the uses and
consequences of framing interventions in practical domains. In
today’s political climate, for example, negative advertising strategies
frequently replace positive advertising strategies, reframing choices
into rejections. Our findings suggest that the impact of such strategies
on the voters’future political attitudes is a valid concern, and its
impact will in part depend on the desirability of the candidates.
Between two desirable candidates, a reject-framed campaign or ballot
design may exacerbate the polarization of political attitudes among
voters whereas the opposite may be true when both candidates are
primarily undesirable. Policymakers should take these potential
ramifications into account when adopting framing interventions.
Our findings also suggest means to proactively leverage framing
for positive business and social outcomes. Marketers aiming to
enhance customers’postpurchase satisfaction and brand loyalty may
highlight the “reject”aspects of the shopping journey. Conversely,
health practitioners may highlight the “choose”aspects of patients’
decisions involving unpleasant treatment options to bolster
posttreatment evaluation, potentially reducing patient attrition in
long-term treatment plans.
The Information Inference Account: Clarifications,
Limitations, and Future Directions
To reiterate, the inference account suggests that people make
attributional inferences upon observing an action, which helps them
understand the actor’s dispositions and motives either when the actor is
another individual or oneself. As such, the framing effect rests on
general associations between actions and their diagnosticity—reject
actions are typically treated as more diagnostic of an actor’s
preferences than are choose actions. As such, an accurate perception
of the decision process is not necessary for the effect. That is, the actual
diagnosticity of a decision process and its perceived diagnosticity are
separate issues. Further, our findings do not imply that the information
inference process following reject decisions will necessarily engage
more cognitive resources than those following choose decisions. The
postdecision processes of inference and updating may have distinct
psychological characteristics than the decision-making process per se.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
FRAMING AFFECTS POSTDECISION PREFERENCES 19
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
On a related note, some readers may be inclined to interpret the
process of self-preference inference and updating primarily as a
deliberate and controlled process. This impression could arise as
“attribution”and “inference”are often discussed as components of
a highly effortful and controlled process of causal reasoning.
However, we do not presume that self-preference inferences are
necessarily effortful and controlled, nor do our findings rely on such
assumptions. In fact, we think it is entirely possible that such
processes are spontaneous and habitual given how common and
potentially adaptive it is (see Cushman, 2020).
Third and related, some might expect self-preference inferences
to be contingent on knowing uncertainty about one’s existing
preferences, a need to justify one’s decision to others, or high
identity relevance of decision stimuli. While such conjectures have
been raised in prior self-perception research, we are skeptical about
them as well. We agree that these factors can enhance one’s
motivation to engage in self-relevant attributional inferences,
especially if the inference process primarily reflects conscious
and goal-directed efforts of reasoning. However, their relevance
would be diminished if the inference process is spontaneous and
habitual. Given our agnosticism about the nature of this process, we
do not view the above factors as necessary boundary conditions to
the framing effect. In our limited explorations of such factors in the
supplemental studies and additional IMA analyses, no supportive
evidence for them emerged. Nonetheless, more data are required
before one can conclude whether these factors affect the framing
effect and preference modulation in general.
Finally, there are interesting questions awaiting future research.
For instance, different interpretations exist on the nature of
preference modulation within the inference camp: Does it reflect
a change in actual preferences (per Cushman, 2020) or, rather, a
change in preference beliefs (per Bem, 1972)? Our current studies
cannot address these nuanced points. There are also other emerging
accounts of preference changes based on sampling of mental
representations (Regenwetter et al., 2011) that are beyond the scope
of our empirical investigations. Other questions that we wish to
examine in follow-up work include how long the framing effect will
persist, whether it will extend beyond immediate self-reported
preferences, and if it will influence subsequent incentive-compatible
behaviors.
Constraints on Generality
Our results were obtained from 13 studies involving 8,782 North
American adults from screened online study platforms and 1,145
Asian undergraduate students from a large public university. Similar
results were found across common decision contexts with diverse
choice stimuli, based on generic instructions for decisions and
preference ratings. Preference ratings were measured immediately
following decisions. We expect these results to generalize to a broad
range of populations and situations. Future replications should
consider identified boundary conditions for the framing effect such
as high attribute similarity and unattractive choice sets. It remains
possible that context changes affect effect size, even though prior
studies found similar magnitudes of preference modulation in
hypothetical and incentivized studies, sometimes with a significant
time lag between measures (e.g., Sharot et al., 2010). We have no
reason to believe that the results depend on other characteristics of
the participants, materials, or context.
Context
The initial impetus of this research was formed when the first
author attended the doctoral seminars offered by Jane Risen and
Nick Epley at the University of Chicago Booth School of Business.
This interest was further developed into the current research when
the second author joined the first author to explore novel framing
effects a few years later. We acknowledge that this article benefited
tremendously from recent advancements in the literature, particu-
larly the methodological and theoretical critiques by Chen and Risen
(2010) on previous research on preference modulation and the
expanded framework of rationalization proposed by Cushman
(2020). We hope that our findings, insights, and methods will further
inspire and facilitate research on the important phenomenon of
decision-induced preference modulation.
References
Abelson, R. P. (1983). Whatever became of consistency theory? Personality
and Social Psychology Bulletin,9(1), 37–54. https://doi.org/10.1177/
0146167283091006
Andersen, S. M. (1984). Self-knowledge and social inference: II. The
diagnosticity of cognitive/affective and behavioral data. Journal of
Personality and Social Psychology,46(2), 294–307. https://doi.org/10
.1037/0022-3514.46.2.294
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001).
Bad is stronger than good. Review of General Psychology,5(4), 323–370.
https://doi.org/10.1037/1089-2680.5.4.323
Bellezza, F. S., Greenwald, A. G., & Banaji, M. R. (1986). Words high and
low in pleasantness as rated by male and female college students. Behavior
Research Methods, Instruments, & Computers,18(3), 299–303. https://
doi.org/10.3758/BF03204403
Bem, D. J. (1967). Self-perception: An alternative interpretation of cognitive
dissonance phenomena. Psychological Review,74(3), 183–200. https://
doi.org/10.1037/h0024835
Bem, D. J. (1972). Self-perception theory. Advances in Experimental Social
Psychology,6,1–62. https://doi.org/10.1016/S0065-2601(08)60024-6
Bodner, R., & Prelec, D. (2003). Self-Signaling and diagnostic utility in
everyday decision making. The Psychology of Economic Decisions,
1(105), 26. https://economics.mit.edu/sites/default/files/2022-09/bodner_
prelec_self_signaling_52001.pdf
Brehm, J. W. (1956). Postdecision changes in the desirability of alternatives.
Journal of Abnormal Psychology,52(3), 384–389. https://doi.org/10
.1037/h0041006
Brehm, J. W., & Cohen, A. R. (1959). Re-evaluation of choice alternatives
as a function of their number and qualitative similarity. Journal of
Abnormal Psychology,58(3), 373–378. https://doi.org/10.1037/h0
040493
Buiten, M. V., & Keren, G. (2009). Speakers’choice of frame in binary
choice: Effects of recommendation mode and option attractiveness.
Judgment and Decision Making,4(1), 51–63. https://doi.org/10.1017/
S193029750000070X
Carmon, Z., Wertenbroch, K., & Zeelenberg, M. (2003). Option attachment:
When deliberating makes choosing feel like losing. The Journal of
Consumer Research,30(1), 15–29. https://doi.org/10.1086/374701
Chen, M. K., & Risen, J. L. (2010). How choice affects and reflects
preferences: Revisiting the free-choice paradigm. Journal of Personality
and Social Psychology,99(4), 573–594. https://doi.org/10.1037/a0020217
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
20 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Croyle, R. T., & Cooper, J. (1983). Dissonance arousal: Physiological
evidence. Journal of Personality and Social Psychology,45(4), 782–791.
https://doi.org/10.1037/0022-3514.45.4.782
Cushman, F. (2020). Rationalization is rational. Behavioral and Brain
Sciences,43, Article e28. https://doi.org/10.1017/S0140525X19001730
Dhar, R., & Wertenbroch, K. (2000). Consumer choice between hedonic and
utilitarian goods. Journal of Marketing Research,37(1), 60–71. https://
doi.org/10.1509/jmkr.37.1.60.18718
Dhar, R., & Wertenbroch, K. (2012). Self-signaling and the costs and benefits
of temptation in consumer choice. Journal of Marketing Research,49(1),
15–25. https://doi.org/10.1509/jmr.10.0490
Dolan, R. J., & Dayan, P. (2013). Goals and habits in the brain. Neuron,
80(2), 312–325. https://doi.org/10.1016/j.neuron.2013.09.007
Elliot, A. J., & Devine, P. G. (1994). On the motivational nature of cognitive
dissonance: Dissonance as psychological discomfort. Journal of Personality
and Social Psychology,67(3), 382–394. https://doi.org/10.1037/0022-3514
.67.3.382
Enisman, M., Shpitzer, H., & Kleiman, T. (2021). Choice changes
preferences, not merely reflects them: A meta-analysis of the artifact-
free free-choice paradigm. Journal of Personality and Social Psychology,
120(1), 16–29. https://doi.org/10.1037/pspa0000263
Evans, A. M., Dillon, K. D., & Rand, D. G. (2015). Fast but not intuitive,
slow but not reflective: Decision conflict drives reaction times in social
dilemmas. Journal of Experimental Psychology: General,144(5),
951–966. https://doi.org/10.1037/xge0000107
Fazio, R. H., Zanna, M. P., & Cooper, J. (1977). Dissonance and self-
perception: An integrative view of each theory’s proper domain of
application. Journal of Experimental Social Psychology,13(5), 464–479.
https://doi.org/10.1016/0022-1031(77)90031-2
Festinger, L. (1957). A theory of cognitive dissonance. Stanford
University Press.
Festinger, L. (1962). Cognitive dissonance. Scientific American,207(4), 93–
102. https://doi.org/10.1038/scientificamerican1062-93
Festinger, L., & Carlsmith, J. M. (1959). Cognitive consequences of forced
compliance. Journal of Abnormal Psychology,58(2), 203–210. https://
doi.org/10.1037/h0041593
Fiedler, K., Harris, C., & Schott, M. (2018). Unwarranted inferences from
statistical mediation tests–An analysis of articles published in 2015.
Journal of Experimental Social Psychology,75,95–102. https://doi.org/10
.1016/j.jesp.2017.11.008
Fiedler, K., Schott, M., & Meiser, T. (2011). What mediation analysis can
(not) do. Journal of Experimental Social Psychology,47(6), 1231–1236.
https://doi.org/10.1016/j.jesp.2011.05.007
Gerard, H. B. (1967). Choice difficulty, dissonance, and the decision
sequence. Journal of Personality,35(1), 91–108. https://doi.org/10.1111/j
.1467-6494.1967.tb01417.x
Greenwald, A. G. (1975). On the inconclusiveness of “crucial”cognitive tests of
dissonance versus self-perception theories. Journal of Experimental Social
Psychology,11(5), 490–499. https://doi.org/10.1016/0022-1031(75)90051-7
Gu, Y., Botti, S., & Faro, D. (2013). Turning the page: The impact of choice
closure on satisfaction. The Journal of Consumer Research,40(2), 268–
283. https://doi.org/10.1086/670252
Harmon-Jones, E., & Harmon-Jones, C. (2002). Testing the action-based
model of cognitive dissonance: The effect of action orientation on
postdecisional attitudes. Personality and Social Psychology Bulletin,
28(6), 711–723. https://doi.org/10.1177/0146167202289001
Harmon-Jones, E., & Mills, J. (1999). An introduction to cognitive
dissonance theory and an overview of current perspectives on the theory.
In E. Harmon-Jones & J. Mills (Eds.), Cognitive dissonance: Progress on
a pivotal theory in social psychology (pp. 3–21). American Psychological
Association. https://doi.org/10.1037/10318-001
Houston, D. A., Sherman, S. J., & Baker, S. M. (1991). Feature matching,
unique features, and thedynamics of the choiceprocess: Predecision conflict
and postdecision satisfaction. Journal of Experimental Social Psychology,
27(5), 411–430. https://doi.org/10.1016/0022-1031(91)90001-M
Izuma, K., Matsumoto, M., Murayama, K., Samejima, K., Sadato, N., &
Matsumoto, K. (2010). Neural correlates of cognitive dissonance and
choice-induced preference change. Proceedings of the National Academy
of Sciences of the United States of America,107(51), 22014–22019.
https://doi.org/10.1073/pnas.1011879108
Izuma, K., & Murayama, K. (2013). Choice-induced preference change in
the free-choice paradigm: A critical methodological review. Frontiers in
Psychology,4, Article 41. https://doi.org/10.3389/fpsyg.2013.00041
Jones, E. E., & Davis, K. E. (1965). From acts to dispositions the process in
person perception. Advances in Experimental Social Psychology,2,
219–266. https://doi.org/10.1016/S0065-2601(08)60107-0
Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.
Kelley, H. H. (1973). The processes of causal attribution. American
Psychologist,28(2), 107–128. https://doi.org/10.1037/h0034225
Kessler, E. H., Ford, C. M., & Bailey, J. R. (1996). Object valence as a
moderator of the framing effect on risk preference. Journal of Economic
Behavior & Organization,30(2), 241–256. https://doi.org/10.1016/S0167-
2681(96)00860-8
Kurdi, B., Lozano, S., & Banaji, M. R. (2017). Introducing the open affective
standardized image set (OASIS). Behavior Research Methods,49(2), 457–
470. https://doi.org/10.3758/s13428-016-0715-3
Laran, J., & Wilcox, K. (2011). Choice, rejection, and elaboration on
preference-inconsistent alternatives. The Journal of Consumer Research,
38(2), 229–241. https://doi.org/10.1086/659040
Laurin, K., & Jettinghoff, W. M. (2020). What kind of rationalization is
system justification? Behavioral and Brain Sciences,43, Article e39.
https://doi.org/10.1017/S0140525X19002243
Levin, I. P., Jasper, J. D., & Forbes, W. S. (1998). Choosing versus rejecting
options at different stages of decision making. Journal of Behavioral
Decision Making,11, 193–210. https://doi.org/10.1002/(SICI)1099-
0771(199809)11:3<193::AID-BDM297>3.0.CO;2-G
Meloy, M. G., &Russo, J. E. (2004). Binarychoice under instructions to select
versus reject. Organizational Behavior and Human Decision Processes,
93(2), 114–128. https://doi.org/10.1016/j.obhdp.2003.12.002
Mizerski, R. W. (1982). An attribution explanation of the disproportionate
influence of unfavorable information. The Journal of Consumer Research,
9(3), 301–310. https://doi.org/10.1086/208925
Nagpal, A., & Krishnamurthy, P. (2008). Attribute conflict in consumer
decision making: The role of task compatibility. Journal of Consumer
Research,34(5), 696–705. https://doi.org/10.1086/521903
Nan, L. X., Park, S. K., & Yang, Y. (2023). Rejections are more contagious
than choices: How another’s decisions shape our own. The Journal of
Consumer Research,50(2), 363–381. https://doi.org/10.1093/jcr/ucad007
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know:
Verbal reports on mental processes. Psychological Review,84(3), 231–
259. https://doi.org/10.1037/0033-295X.84.3.231
Oppenheimer, D. M., Meyvis, T., & Davidenko, N. (2009). Instructional
manipulation checks: Detecting satisficing to increase statistical power.
Journal of Experimental Social Psychology,45(4), 867–872. https://
doi.org/10.1016/j.jesp.2009.03.009
Perfecto, H., Galak, J.,Simmons, J. P., & Nelson, L. D. (2017). Rejectinga bad
option feels like choosing a good one. Journal of Personality and Social
Psychology,113(5), 659–670. https://doi.org/10.1037/pspa0000092
Pocock, S. J. (1977). Group sequential methods in the design and analysis of
clinical trials. Biometrika,64(2), 191–199. https://doi.org/10.1093/biome
t/64.2.191
Regenwetter, M., Dana, J., & Davis-Stober, C. P. (2011). Transitivity of
preferences. Psychological Review,118(1), 42–56. https://doi.org/10
.1037/a0021150
Risen, J. L., & Chen, M. K. (2010). How to study choice-induced attitude
change: Strategies for fixing the free-choice paradigm. Social and
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
FRAMING AFFECTS POSTDECISION PREFERENCES 21
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
Personality Psychology Compass,4, 1151–1164. https://doi.org/10.1111/j
.1751-9004.2010.00323.x
Rosenberg, M. (1979). Conceiving the self. Basic Books. https://books.goo
gle.com/books?id=nUJqAAAAMAAJ
Shafir, E. (1993). Choosing versus rejecting: Why some options are both
better and worse than others. Memory & Cognition,21(4), 546–556.
https://doi.org/10.3758/BF03197186
Sharot, T., Velasquez, C. M., & Dolan, R. J. (2010). Do decisions shape
preference? Evidence from blind choice. Psychological Science,21(9),
1231–1235. https://doi.org/10.1177/0956797610379235
Shultz, T. R., Léveillé, E., & Lepper, M. R. (1999). Free choice and cognitive
dissonance revisited: Choosing “lesser evils”versus “greater goods.”
Personality and Social Psychology Bulletin,25(1), 40–48. https://doi.org/
10.1177/0146167299025001004
Simon, D., & Holyoak, K. J. (2002). Structural dynamics of cognition: From
consistency theories to constraint satisfaction. In R. R. Vallacher, S. J.
Read, & A. Nowak (Eds.), The dynamic perspective in personality and
social psychology (pp. 283–294). Psychology Press.
Skowronski, J. J., & Carlston, D. E. (1989). Negativity and extremity biases
in impression formation: A review of explanations. Psychological
Bulletin,105(1), 131. https://doi.org/10.1037/0033-2909.105.1.131
Snyder, M., & Ebbesen, E. B. (1972). Dissonance awareness: A test of
dissonance theory versus self-perception theory. Journal of Experimental
Social Psychology,8(6), 502–517. https://doi.org/10.1016/0022-1031(72)
90076-5
Sokolova, T., & Krishna, A. (2016). Take it or leave it: How choosing
versus rejecting alternatives affects information processing. The Journal
of Consumer Research,43(4), 614–635. https://doi.org/10.1093/jcr/
ucw049
Spencer, S. J., Zanna, M. P., & Fong, G. T. (2005). Establishinga causal chain:
Why experiments are often more effective than mediational analyses in
examining psychological processes. Journal of Personality and Social
Psychology,89(6), 845–851. https://doi.org/10.1037/0022-3514.89 .6.845
Stalder, D. R. (2010). Competing roles for the subfactors of need for closure
in moderating dissonance-produced attitude change. Personality and
Individual Differences,48(6), 775–778. https://doi.org/10.1016/j.paid
.2010.01.028
Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about
health, wealth, and happiness. Yale University Press.
Thorndike, E. L. (1898). Animal intelligence: An experimental study
of the associative processes in animals. The Psychological Review:
Monograph Supplements,2(4), 1–109. https://doi.org/10.1037/107
80-000
Touré-Tillery, M., & Light, A. E. (2018). No self to spare: How the cognitive
structure of the self influences moral behavior. Organizational Behavior
and Human Decision Processes,147,48–64. https://doi.org/10.1016/j
.obhdp.2018.05.002
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the
psychology of choice. Science,211(4481), 453–458. https://doi.org/10
.1126/science.7455683
Vosgerau, J., Simonsohn, U., Nelson, L. D., & Simmons, J. P. (2019). 99%
impossible: A valid, or falsifiable, internal meta-analysis. Journal of
Experimental Psychology: General,148(9), 1628–1639. https://doi.org/10
.1037/xge0000663
Webster, D. M., & Kruglanski, A. W. (1994). Individual differences
in need for cognitive closure. Journal of Personality and Social
Psychology,67(6), 1049–1062. https://doi.org/10.1037/0022-3514.67
.6.1049
Wedell, D. H. (1997). Another look at reasons for choosing and rejecting.
Memory & Cognition,25(6), 873–887. https://doi.org/10.3758/BF03211332
Yang, A. X., & Teow, J. (2024, May 26). Framing affects post-decision
preferences through self-preference inferences (and probably not
dissonance).https://osf.io/qhcdy
Received May 8, 2022
Revision received June 24, 2024
Accepted July 1, 2024 ▪
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
22 YANG AND TEOW
Template Version: 28 Sep 2024 ▪3:39 pm IST XGE-2022-0275_format_final ▪7 November 2024 ▪9:31 pm IST
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Experimental Psychology: General
This content is subject to copyright. Terms and conditions apply.
