Content uploaded by Nicholas Epley
Author content
All content in this area was uploaded by Nicholas Epley on May 14, 2021
Content may be subject to copyright.
Overcoming Intuition: Metacognitive Difficulty Activates Analytic
Reasoning
Adam L. Alter and Daniel M. Oppenheimer
Princeton University
Nicholas Epley
University of Chicago
Rebecca N. Eyre
Harvard University
Humans appear to reason using two processing styles: System 1 processes that are quick, intuitive, and
effortless and System 2 processes that are slow, analytical, and deliberate that occasionally correct the
output of System 1. Four experiments suggest that System 2 processes are activated by metacognitive
experiences of difficulty or disfluency during the process of reasoning. Incidental experiences of
difficulty or disfluency—receiving information in a degraded font (Experiments 1 and 4), in difficult-
to-read lettering (Experiment 2), or while furrowing one’s brow (Experiment 3)—reduced the impact of
heuristics and defaults in judgment (Experiments 1 and 3), reduced reliance on peripheral cues in
persuasion (Experiment 2), and improved syllogistic reasoning (Experiment 4). Metacognitive experi-
ences of difficulty or disfluency appear to serve as an alarm that activates analytic forms of reasoning that
assess and sometimes correct the output of more intuitive forms of reasoning.
Keywords: fluency, disfluency, dual-system processing, reasoning, judgment
Few psychological theories enjoy the longevity of William
James’s (1890/1950) suggestion that human reasoning involves
two distinct processing systems: one that is quick, effortless,
associative, and intuitive and another that is slow, effortful, ana-
lytic, and deliberate. When deciding whether it is more dangerous
to travel by car or airplane, for instance, people may quickly
generate horrific images of airline disasters and (erroneously)
conclude that it is more dangerous to fly than to drive. Alterna-
tively, they may think more analytically about the total number of
automobile versus airline accidents, the number of miles driven
versus flown per accident, or the possibility that automobile acci-
dents are underreported whereas airline accidents command media
headlines and conclude (accurately) that they are safer in an
airplane.
Although not without controversy (see Kruglanski & Thomp-
son, 1999; Osman, 2004),
1
dual-process theories have been used
widely by developmental, cognitive, and social psychologists to
explain such diverse phenomena as persuasion (e.g., Chaiken,
1980; Petty & Cacioppo, 1986), social cognition (Epley, Keysar,
Van Boven, & Gilovich, 2004; Keysar & Barr, 2002), self-
perception (Schwarz, 1998), causal attribution (Gilbert, 1989),
stereotyping (Bodenhausen, Macrae, & Sherman, 1999), overcon-
fidence (Griffin & Tversky, 1992), higher order reasoning (Evans,
2003; Evans & Over, 1996; Sloman, 1996), various memory
phenomena (Jacoby, Kelley, & McElree, 1999; Jones & Jacoby,
2001; Whittlesea & Leboe, 2003), and a long list of non-Bayesian
biases in judgment and decision making (e.g., Kahneman & Fred-
erick, 2002). These dual-process theories enable understanding of
diverse phenomena because they predict qualitatively different
judgments depending on which reasoning system is used. In par-
ticular, deliberate and analytical systems of reasoning (System 2)
can override or undo intuitive and associative (System 1) re-
sponses. Understanding when System 2 reasoning is likely to be
used is therefore critical for understanding human judgment and
decision making.
Although most dual-process models provide an extensive de-
scription of each system, few devote much attention to exactly
when people will adopt each approach to information processing.
Those that do typically argue that System 2 will be activated when
1
Osman (2004) proposed a unitary process model that nonetheless
acknowledges that processing occurs along a gradient from explicit to
implicit, which mirrors System 1 and System 2 processing, respectively.
Kruglanski and Thompson (1999; the unimodel) adopted a subtly different
approach. They recognized that people make use of different types of
information but subsume them under a single umbrella of “persuasive
cues.” They emphasized that people are ultimately trying to make sense of
the world, regardless of which type of cue they use in a given situation.
However, like Osman’s unitary process model, the unimodel acknowledges
that some forms of processing are more complex than others.
Adam L. Alter, Psychology Department, Princeton University; Daniel
M. Oppenheimer, Psychology Department and the Woodrow Wilson
School of Public and International Affairs, Princeton University; Nicholas
Epley, Graduate School of Business, University of Chicago; Rebecca N.
Eyre, Department of Psychology, Harvard University.
This research was funded by National Science Foundation Grants
051811 and SES0241544. We thank Sara Etchison, Shane Frederick, Geoff
Goodwin, Leif Holtzmann, Daniel Kahneman, Eugenia Mamikonyan,
Melissa Miller, Manish Pakrashi, Cordaro Rodriguez, Yuval Rottenstreich,
Joe Simmons, Alex Todorov, and Erin Whitchurch for helpful assistance.
Correspondence concerning this article should be addressed to Adam L.
Alter, Psychology Department, Princeton University, Princeton, NJ 08540.
E-mail: aalter@princeton.edu
Journal of Experimental Psychology: General Copyright 2007 by the American Psychological Association
2007, Vol. 136, No. 4, 569–576 0096-3445/07/$12.00 DOI: 10.1037/0096-3445.136.4.569
569
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.
people have both the capacity and the motivation to engage in
effortful processing. Existing research demonstrates that errors in
System 1 reasoning are less likely to be corrected when people are
under cognitive load or respond quickly (e.g., Bless & Schwarz,
1999; Chaiken, 1980; Petty & Cacioppo, 1986), but they are more
likely to be corrected when people are accountable for their deci-
sions (Tetlock & Lerner, 1999) and when the outcome is person-
ally relevant (Ajzen & Sexton, 1999; Chaiken, 1980; Petty &
Cacioppo, 1986). This existing research does not address, how-
ever, when people will recognize that System 1 processes might be
producing faulty output that requires more analytical thought.
Accordingly, we investigated the novel question of when people
are compelled to use System 2 processes in the first place. We
predicted that people’s use of more elaborate reasoning processes
would be based on experiential cues that more elaborate reasoning
processes are required and that System 2 processes would there-
fore be activated by cues that suggest a simple System 1 judgment
might be faulty.
Confidence in the accuracy of intuitive judgment appears to
depend in large part on the ease or difficulty with which
information comes to mind (Gill, Swann, & Silvera, 1998;
Kelley & Lindsay, 1993) and the perceived difficulty of the
judgment at hand. If information is processed easily or fluently,
intuitive (System 1) processes will guide judgment. If informa-
tion is processed with difficulty or disfluently, however, this
experience will serve as a cue that the task is difficult or that
one’s intuitive response is likely to be wrong, thereby activating
more elaborate (System 2) processing. The ease or difficulty
experienced while processing information is therefore used as a
cue to guide one’s subsequent processing styles. We thus pre-
dicted that experienced difficulty or disfluency would function
as a signal that a simple and intuitive judgment was insufficient
and that more elaborate cognitive processing would be neces-
sary, thereby increasing System 2 processing. Indeed, neuro-
scientific evidence suggests that disfluency triggers the anterior
cingulate cortex (Boksman et al., 2005), an alarm that activates
the prefrontal cortex responsible for deliberative and effortful
thought (Botvinick, Braver, Carter, Barch, & Cohen, 2001;
Lieberman, Gaunt, Gilbert, & Trope, 2002; see also Goel,
Buchel, Frith, & Dolan, 2000).
Previous research has shown that people rate disfluent stimuli
more negatively than fluent stimuli across a range of domains.
For example, people believe that disfluently named stocks will
perform more poorly than will fluently named stocks (Alter &
Oppenheimer, 2006), that disfluent prose is written by an author
that is less intelligent than an author of fluent prose (Oppen-
heimer, 2006), and that disfluent aphorisms are less likely to be
true than fluent aphorisms (McGlone & Tofighbakhsh, 2000).
However, the role of disfluency in the current research is novel.
Unlike existing research in which disfluency served as a direct
cue to judgment, we investigate disfluency as an indirect cue
that serves as a metacognitive signal to prompt more systematic
processing. We predicted that experiencing difficulty or disflu-
ency during the course of reasoning would trigger System 2
processes and decrease the frequency of responses consistent
with System 1 processes. We present four experiments, across
a range of domains, that are consistent with this hypothesis.
Experiment 1—Intuitive Defaults
Experiment 1 was designed to provide initial evidence that
people adopt a systematic approach to reasoning when they expe-
rience cognitive disfluency. Participants completed the Cognitive
Reflection Test (CRT; Frederick, 2005). This test consists of three
items for which the gut reaction, or intuitive default, is incorrect
but that respondents can correctly answer through deliberate re-
consideration. A correct answer on each item suggests that the
respondent engaged systematic processing to correct the intuitive
response. Frederick (2005) administered the CRT to 3,400 partic-
ipants across 35 studies and 11 samples and showed that scores on
the CRT were highly correlated with a variety of measures asso-
ciated with analytic thinking, including intelligence (Stanovich &
West, 2000). If disfluency initiates systematic processing, then
participants should perform better on the test if they experience
disfluency while generating their answers. We manipulated disflu-
ency in this experiment by printing the questions in either a
difficult-to-read font (disfluent condition) or an easy-to-read font
(fluent condition). We predicted that participants would answer
more of the CRT items correctly when they were printed in a
difficult-to-read font than when they were printed in an easy-to-
read font.
Method
We recruited 40 Princeton University undergraduate volunteers
at the student campus center to complete the three-item CRT
(Frederick, 2005). Participants were seated either alone or in small
groups, and the experimenter ensured that they completed the
questionnaire individually. Those in the fluent condition com-
pleted a version of the CRT written in easy-to-read black Myriad
Web 12-point font, whereas participants in the disfluent condition
completed a version of the CRT printed in difficult-to-read 10%
gray italicized Myriad Web 10-point font. Participants were ran-
domly assigned to complete either the fluent or the disfluent
version of the CRT. Previous research has shown that similar font
manipulations effectively influence fluency (e.g., Oppenheimer,
2006; Werth & Strack, 2003). Consistent with these studies, a
separate sample of 13 participants rated (on a 5-point scale) the
disfluent font (M⫽3.08, SD ⫽0.76) as being more difficult to
read than the fluent font (M⫽1.54, SD ⫽0.87), t(12) ⫽3.55, p⬍
.01,
2
⫽.51.
Results and Discussion
As predicted, participants answered more items on the CRT
correctly in the disfluent font condition (M⫽2.45, SD ⫽0.64)
than in the fluent font condition (M⫽1.90, SD ⫽0.89), t(38) ⫽
2.25, p⫽.03,
2
⫽.12. Whereas 90% of participants in the fluent
condition answered at least one question incorrectly, only 35% did
so in the disfluent condition,
2
(1, N⫽40) ⫽12.91, p⬍.001,
Cramer’s V⫽.57. Finally, participants in the fluent condition
provided the incorrect and intuitive response more often (23% of
responses) than did participants in the disfluent condition (10% of
responses), Z⫽1.96, p⫽.05,
2
⫽.07. When the CRT was
difficult to read, participants appeared to engage in systematic
processing and overcame their invalid intuitions to answer more
questions correctly. These results provide preliminary evidence
570 ALTER, OPPENHEIMER, EPLEY, AND EYRE
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.
that disfluency initiates systematic processing. We sought con-
verging evidence for this hypothesis in the subsequent experiments
and also addressed a variety of alternative interpretations for the
results of Experiment 1.
The most obvious alternative interpretation of Experiment 1 is
that presenting the test items in a disfluent font simply slowed
participants down, thereby forcing them to process the information
more carefully. This exogenous, task-specific mechanism is some-
what less interesting than our proposed endogenous mechanism, so
we attempted in the remaining experiments to rule out the possi-
bility that these effects merely reflected task-imposed constraints
on processing speed. In Experiment 2, we adopted a fluency
manipulation that did not force participants to process the infor-
mation slowly and sought evidence for our proposed mechanism
that people interpret disfluency as a signal to exert extra cognitive
effort to adequately complete a task. In addition, although the CRT
is a domain-general test of systematic reasoning, we wanted to
ensure that the effects of disfluency on processing depth also
applied to other concrete domains of judgment. Accordingly, we
examined whether disfluency would lead people to focus on sys-
tematic processing cues when reading a review of a new MP3
player.
Experiment 2—Persuasion
The dominant models of persuasion, the heuristic–systematic
model (Chaiken, 1980) and the elaboration likelihood model (Petty
& Cacioppo, 1986), propose two distinct types of processing cues.
Systematic or central cues generally involve the evidentiary qual-
ity of an argument, whereas heuristic or peripheral cues generally
involve contextual factors irrelevant to an argument’s quality (e.g.,
a target’s appearance of competence). In Experiment 2, we tested
whether experienced disfluency would increase people’s reliance
on systematic processing cues when evaluating a persuasive com-
munication.
Participants read a fabricated review of a new MP3 player
accompanied by a picture of either a competent-looking person
discussing unimportant features (positive heuristic–negative sys-
tematic condition) or an incompetent-looking person discussing
important features (negative heuristic–positive systematic condi-
tion). We manipulated processing fluency by presenting the mast-
head of the review in either difficult-to-read (disfluent condition)
or easy-to-read (fluent condition) type. We predicted that, consis-
tent with the results of Experiment 1, participants in the disfluent
condition would rely more heavily on the systematic cue than on
the heuristic cue.
Method
Pilot Experiments
Stimulus selection: Heuristic cues. A separate sample of 42
participants (Willis & Todorov, 2006) judged the apparent com-
petence of a series of faces taken from a database of faces (Lund-
qvist, Flykt, & O
¨hman, 1998). The faces judged, on average, to be
the most competent and the most incompetent from the database
served as the targets of our heuristic-cue manipulation. Given that
the physical appearance of competence is considered a heuristic
cue, the competent-looking face constituted a strong (convincing)
heuristic cue, and the incompetent-looking face constituted a weak
(unconvincing) heuristic cue.
Stimulus selection: Systematic cues. In partial fulfillment of a
course requirement, 10 Princeton University undergraduates re-
ported the three most important and the three least important
features of an MP3 player. We manipulated the strength of the
systematic cue by using either the three most commonly men-
tioned important features (strong systematic cue: price, storage
capacity, and battery life) or the three most commonly mentioned
unimportant features (weak systematic cue: variety of colors, pop-
ular with celebrities, used by “everyone”). The resulting reviews
are depicted in Figure 1.
Assessed difficulty: Underlying mechanism. Twenty Princeton
University undergraduate volunteers at the campus student center
completed a questionnaire designed to investigate whether our
manipulation of fluency did indeed serve as a cue that greater
cognitive effort would or would not be needed in the task. The
Figure 1. Stimuli from two of the four conditions in Experiment 1. The left panel shows the disfluent masthead
and strong heuristic cue (competent-looking face) condition, and the right panel shows the fluent masthead and
strong systematic cue (review listing important features) condition. The masthead fluency was switched in the
other two conditions, which were otherwise identical.
571
FLUENCY AND PROCESSING DEPTH
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.
questionnaire began with the fluent masthead for 10 participants
and the disfluent masthead for the remaining 10 participants (see
Figure 1 for mastheads). We asked participants to read the mast-
head and to imagine that they were going to read the remainder of
the review. They rated how much effort they expected to have to
expend to understand the contents of the review and the difficulty
of reading the masthead (both on 5-point scales). As we expected,
the fluency manipulation check showed that the disfluent masthead
was considered more difficult to read (M⫽2.30, SD ⫽0.67) than
the fluent masthead (M⫽1.20, SD ⫽0.42), t(18) ⫽4.37, p⬍
.001,
2
⫽.52. Perhaps more important, participants who read the
disfluent masthead expected to need to expend more cognitive
effort to understand the remaining contents of the review (M⫽
2.30, SD ⫽0.67) than did those who read the fluent masthead
(M⫽1.40, SD ⫽0.70), t(18) ⫽2.93, p⬍.01,
2
⫽.32. This
preliminary result suggests that people rely on the ease with which
they process initial information to determine how much effort they
will require to process subsequent information.
Main Experiment
Having shown that participants interpret disfluency as a cue to
engage greater cognitive resources, we investigated whether dis-
fluency also increases participants’ reliance on systematic cues.
Forty Princeton University volunteers at the student campus center
read a short review of a new MP3 player, ostensibly printed from
a Web site called Techbiz.com. Participants were seated alone and
in small groups, and the experimenter ensured that participants in
groups completed the questionnaire individually. The masthead on
the review was written either in easy-to-read typeset (fluent con-
dition) or in a difficult-to-read combination of letter-resembling
symbols (disfluent condition; see Figure 1). Participants in the
positive heuristic/negative systematic condition saw the highly
competent-looking face used for the reviewer’s photo paired with
a review praising unimportant features of the MP3 player. Partic-
ipants in the negative heuristic/positive systematic condition saw
the opposite: the incompetent-looking face paired with a review
praising important features of the MP3 player. In both conditions,
the overall review of the MP3 player was positive. We randomly
assigned participants to read one of the four versions of the review.
Note that, contrary to the fluency manipulation in Experiment 1,
this fluency manipulation did not alter the information on which
participants based their judgments. This manipulation instead al-
tered the fluency of the masthead at the top of the review rather
than the fluency of the information in the review itself. The actual
content and format of the reviews were identical between condi-
tions. Participants in the disfluent condition were therefore not
compelled to read the review itself more slowly. Although it is
possible that participants in the disfluent condition read the mast-
head more slowly and then maintained this slower processing
speed through the rest of the materials, we suspect that the thou-
sands of hours of practice our participants spent reading text in
normal fonts outside our experimental context render this partic-
ular alternative quite unlikely. Nonetheless, this manipulation im-
proves on many fluency manipulations as the actual content and
format of the reviews were identical between conditions, and
participants in the disfluent condition were therefore not required
to read the review itself more slowly.
After reading the review, participants rated the competence of
the reviewer, rated the quality of the MP3 player, and estimated
how much they would like the experience of owning the MP3
player on separate 7-point scales. Scores on the three scales were
highly correlated (Cronbach’s ␣⫽.82), so we averaged scores to
create a composite favorability rating.
Results and Discussion
As predicted, participants’ favorability ratings were more
heavily influenced by the systematic cue (the quality of the argu-
ments) in the disfluent condition than by the systematic cue in the
fluent condition. Specifically, participants in the disfluent condi-
tion preferred the MP3 player when the systematic cue was per-
suasive (M
systematic
⫽4.50, SD ⫽0.82, vs. M
heuristic
⫽3.47, SD ⫽
1.62), whereas those in the fluent condition preferred the MP3
player when the heuristic cue was persuasive (M
heuristic
⫽3.63,
SD ⫽0.72, vs. M
systematic
⫽3.00, SD ⫽1.14), F
interaction
(1, 39) ⫽
5.44, p⬍.03,
2
⫽.13 (see Table 1 for results of each component
of the composite favorability rating).
2
Neither follow-up simple
effect comparison reached significance, although participants’ rat-
ings were marginally more favorable toward the strong systematic
cue review than the strong heuristic cue review when the masthead
was disfluent, F(1, 19) ⫽3.24, p⬍.10,
2
⫽.15. These results
suggest that disfluency induces a more systematic processing style.
It is important to note that this study used a manipulation that did
not require participants to spend more time reading information in
the disfluent condition than in the fluent condition. Furthermore,
the pilot study shows that people expect to need additional cog-
nitive resources to process information after the experience of
disfluency. This strongly suggests that the apparent need for more
systematic processing activated by the disfluent cues was at least
a contributing if not the primary mechanism that led to System 2
processing.
Nonetheless, it is still possible that participants in Experiment 2
who read the review with the disfluent masthead also processed
subsequent information more slowly. We therefore adopted a
fluency manipulation in Experiment 3 that did not alter the prop-
erties of the stimuli at all, thereby eliminating the possibility that
participants processed information more slowly because of
changes in processing speed that originated in the stimuli. We also
conducted Experiment 3 in a third domain of judgment to further
demonstrate that the effects of disfluency on processing depth
2
On closer inspection, the interaction in this experiment was driven by
the strong systematic cue condition, in which participants gave signifi-
cantly higher favorability ratings when the masthead was disfluent than
when it was fluent, F(1, 19) ⫽18.89, p⬍.001,
2
⫽.51. In contrast, the
masthead had little effect on favorability ratings in the strong heuristic cue
condition, F⬍1. One possible explanation for this effect is that in the
strong systematic cue condition, the disheveled appearance of the reviewer
acted as a negative cue that directly contrasted with the compelling content
of the actual review. Thus, when participants paid relatively more attention
to the content of the review, their ratings were commensurately more
favorable. In contrast, although the review in the strong heuristic condition
was not particularly compelling, it still praised the MP3 player. As such, in
the strong heuristic condition, the two cues did not run in opposition to one
another—they were both positive. Thus, in the strong heuristic condition,
participants’ favorability ratings might not have been swayed strongly by
whether they attended to the reviewer or the content of his review.
572 ALTER, OPPENHEIMER, EPLEY, AND EYRE
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.
occur across a broad array of domains, in this case, investigating
reliance on simple heuristics in judgment.
Experiment 3—Representativeness Heuristic
In reviewing dual-process accounts in judgment and decision
making, Kahneman and Frederick (2002) discussed the represen-
tativeness heuristic (Kahneman & Tversky, 1973) as a prototypical
case of System 1 reasoning. Using this heuristic when making
judgments of category inclusion entails categorizing exemplars
primarily on the basis of the extent to which they are representative
of a given category, ignoring more cognitively demanding cues
such as base rates. Because representativeness is so strongly as-
sociated with the heuristic branch of dual-process reasoning, in
Experiment 3, we examined whether the experience of disfluency
might reduce people’s reliance on the representativeness heuristic.
Extending the generality of the effect, we manipulated fluency by
asking participants to adopt a facial expression consistent with the
exertion of cognitive effort (furrowed brow) or a control expres-
sion that did not imply the exertion of effort (puffed cheeks; see
Stepper & Strack, 1993). As in Experiment 2, this fluency manip-
ulation did not necessarily slow participants down, which elimi-
nated the possibility that participants in the disfluent condition
were merely forced to process the stimuli more slowly.
Method
Pilot Study
As in Experiment 2, we first investigated whether our manipu-
lation of fluency—in this case, one’s facial expression— did in-
deed serve as a cue that greater cognitive effort would or would not
be needed in the task. Because heuristic reasoning represents a
default intuitive judgment that systematic reasoning would over-
come, we investigated whether furrowing one’s brow influenced
confidence in one’s judgment. Low confidence in judgment would
serve as a cue that more cognitive processing would be necessary
to arrive at the correct answer, and we expected those in the
disfluent condition (who were furrowing their brows) would be
less confident in their judgments than those in the fluent condition
(who were puffing their cheeks).
Twenty Harvard University undergraduates participated in a lab
study in which they answered a series of trivia questions and
indicated their confidence that each of their answers was correct
(from 0% to 100% confident). The experimenter told participants
that the study was designed to examine how people answer ques-
tions under distracting conditions, and participants learned that
they were going to answer questions while making various poten-
tially distracting body movements. Participants were randomly
assigned to adopt one of two facial expressions. Half of the
participants were instructed to puff out their cheeks, whereas the
other half were instructed to furrow their brows (Stepper & Strack,
1993). The experimenter demonstrated each expression until the
participant held it correctly. Furrowing one’s brow is associated
with difficult mental effort and concentration, whereas puffing
one’s cheeks is a neutral expression that is equally difficult to
maintain but is not associated with mental effort (Tourangeau &
Ellsworth, 1979).
As predicted, participants who puffed their cheeks were signif-
icantly more confident in their responses (M⫽65%, SD ⫽14%)
than were participants who furrowed their brows (M⫽52%, SD ⫽
12%), t(18) ⫽2.07, p⬍.05,
2
⫽.21. It is important to note that
this difference in confidence did not reflect any difference in the
actual accuracy of participants’ responses among those who puffed
their cheeks (M⫽36% correct, SD ⫽13%) versus those who
furrowed their brows (M⫽38% correct, SD ⫽10%), t(18) ⫽
0.37, ns; for the interaction between confidence and accuracy, F(1,
18) ⫽4.85, p⬍.05,
2
⫽.22. These findings suggest that people
are less confident in their judgments when they adopt facial
expressions commonly associated with cognitive effort.
Main Experiment
One hundred fifty Harvard University undergraduates partici-
pated in a lab study in exchange for course credit. The materials for
this experiment were taken from one of the original demonstra-
tions of the representativeness heuristic: the Tom W. scenario
(Kahneman & Tversky, 1973). Participants in this original dem-
onstration read a description of a fictitious person—Tom W.—who
was described in a way intended to seem similar to the widely
recognized stereotype of an engineer.
As in the original experiment, one group of participants (n⫽51)
read a description of Tom W. and estimated on an 11-point scale
how similar he was to a typical student with one of nine under-
graduate majors (library science, social science and social work,
business administration, computer science, humanities and educa-
tion, law, medicine, engineering, and physical and life sciences).
The mean similarity rating for each major functioned as a measure
of how representative Tom W. was of a typical student with each
major.
A second group of participants (n⫽55) did not read a descrip-
tion of Tom W. but instead estimated the percentage of students on
campus who were studying each of the nine majors. The mean
proportion represented an estimate of the base rates associated
with each of the nine majors.
Table 1
Experiment 1: Mean Ratings of Predicted Liking of MP3 Player,
Estimated Quality of MP3 Player, and Reviewer Competence
Rating dimension and
masthead fluency
Strong
heuristic cue
Strong
systematic cue
MSDMSD
Liking
Fluent 3.90 1.37 3.50 0.97
Disfluent 4.10 2.08 5.00 1.25
Quality
Fluent 4.30 1.42 3.40 0.84
Disfluent 3.60 1.90 4.40 1.17
Competence
Fluent 2.70 1.16 2.10 0.88
Disfluent 2.70 1.16 4.10 1.37
Note. Ratings were made on separate 7-point scales. Although the data
adhere to similar patterns across the three measures, the interaction is
significant on the competence dimension, F(1, 39) ⫽7.50, p⫽.01,
2
⫽
.17; marginally significant on the quality dimension, F(1, 39) ⫽3.75, p⫽
.06,
2
⫽.09; but nonsignificant on the liking dimension, F(1, 39) ⫽1.94,
p⫽.17,
2
⫽.05.
573
FLUENCY AND PROCESSING DEPTH
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.
Finally, a third group of students (n⫽44) ranked the likelihood
(from 1 ⫽most likely to 9 ⫽least likely) that Tom W. actually
studied each of the nine majors. As in the pilot test, we randomly
allocated half of these participants to puff their cheeks (fluent
condition) and the other half to furrow their brows (disfluent
condition) while rendering their judgments.
Results and Discussion
We correlated each participant’s nine likelihood ratings with the
mean representativeness and base-rate evaluations from the first
two groups of participants. These two correlations provide an
estimate of how strongly each participant in the third group relied
on representativeness and base-rate information when assessing
the likelihood that Tom W. studied each of the nine majors.
We began by reverse scoring participants’ rankings so that the
most likely major was scored a 9 and the least likely was scored a
1. A positive correlation between this recoded ranking variable and
representativeness scores therefore indicates that the participant
relied heavily on the representativeness heuristic, whereas a neg-
ative correlation indicates that the participant strongly neglected
base rates.
As predicted, participants who furrowed their brows relied less
on the representativeness heuristic (mean r⫽.43, SD ⫽0.46) than
did participants who puffed their cheeks (mean r⫽.74, SD ⫽
0.19), t(42) ⫽2.84, p⬍.01,
2
⫽.16. Similarly, participants who
furrowed their brows exhibited less base-rate neglect (mean r⫽
⫺.12, SD ⫽0.30) than did participants who puffed their cheeks
(mean r⫽⫺.36, SD ⫽0.33), t(42) ⫽2.60, p⬍.01,
2
⫽.14.
Thus, participants whose facial expressions induced a feeling of
disfluency were less likely to rely on System 1 processes and were
more inclined to use System 2. As the pilot study suggests,
participants who furrowed their brows felt less confident in their
judgments. These results suggest that participants dealt with the
experience of lowered confidence by adopting a more careful,
systematic approach to the task.
Although Experiments 1–3 demonstrate that experienced disflu-
ency leads to more systematic processing, none of the experiments
address the potential influence of mood. People in negative mood
states— especially sadness—tend to process information more sys-
tematically (Schwarz, Bless, & Bohner, 1991), and it is at least
possible that experienced disfluency worsens people’s transient
mood states. We therefore designed Experiment 4 to extend Ex-
periments 1–3 into a new domain of reasoning while simulta-
neously investigating the possible role of mood in producing the
observed results.
Experiment 4 —Syllogistic Reasoning
Participants in Experiment 4 attempted to deduce logical con-
clusions from a series of two-statement syllogisms printed in either
easy-to-read or difficult-to-read font. Syllogistic reasoning is one
of the most widely studied processes in cognitive psychology (e.g.,
Johnson-Laird & Bara, 1984; Rips, 1994) and is often used as a
case study in higher order reasoning for dual-process models of
cognition (e.g., Evans & Over 1996; Sloman, 1996). We expected
people to answer more questions correctly when the syllogisms
were written in difficult-to-read font, which would be consistent
with the results of Experiments 1–3. We also sought to investigate
whether these effects were driven by differential mood effects of
the fluency conditions.
Method
Pilot Study
As in Experiments 2 and 3, we first investigated whether our
manipulation of fluency—in this case, the fonts of the items—
could serve as a cue that greater cognitive effort would or would
not be needed in the task. Accordingly, 69 Princeton University
undergraduate volunteers at the student campus center read two
syllogism questions printed in either an easy-to-read (fluent) font
or a difficult-to-read (disfluent) font. The fonts were identical to
those used in Experiment 1. Without actually answering the ques-
tions, participants rated how confident they were that they would
be able to answer them correctly and estimated their difficulty
(each on a 5-point scale). As we expected, participants who read
the syllogisms printed in fluent font were more confident that they
would be able to answer the questions correctly (M⫽4.86, SD ⫽
1.39, vs. M⫽4.00, SD ⫽1.39), t(67) ⫽3.03, p⬍.01,
2
⫽.12,
and believed the task was less difficult (M⫽2.54, SD ⫽1.17, vs.
M⫽3.29, SD ⫽1.27) than did participants who read the syllo-
gisms printed in disfluent font, t(67) ⫽2.58, p⬍.05,
2
⫽.09.
Thus, the experience of reading the syllogisms printed in disfluent
font lowered participants’ confidence in their ability to answer the
questions correctly and led them to believe that the questions were
more difficult. Notice that in contrast to Study 3, these measures of
confidence and difficulty were taken before answering the ques-
tions rather than after, thereby providing convergent evidence for
our proposed mechanism.
Main Study
Forty-one Princeton University undergraduates at the student
campus center volunteered to complete a questionnaire that con-
tained six syllogistic reasoning problems. The experimenter ap-
proached participants individually or in small groups but ensured
that they completed the questionnaire without the help of other
participants. The syllogisms were selected on the basis of accuracy
base rates established in prior research (Johnson-Laird & Bara,
1984; Zielinski, Goodwin, & Halford, 2006). Two were easy
(answered correctly by 85% of respondents), two were moderately
difficult (50% correct response rate), and two were very difficult
(20% correct response rate). The easy and very difficult items were
omitted from further analyses because the ceiling and floor effects
obscured the effects of fluency on processing depth. Shallow
heuristic processing enabled participants to answer the easy items
correctly, whereas systematic reasoning was insufficient to guar-
antee accuracy on the difficult questions. Participants were ran-
domly assigned to read the questionnaire printed in either an
easy-to-read (fluent) or a difficult-to-read (disfluent) font, the same
fonts that were used in Experiment 1.
Finally, participants indicated how happy or sad they felt on a
7-point scale (1 ⫽very sad;4⫽neither happy nor sad;7⫽very
happy). This is a standard method for measuring transient mood
states (e.g., Forgas, 1995).
Results and Discussion
As expected, participants in the disfluent condition answered a
greater proportion of the questions correctly (M⫽64%) than did
574 ALTER, OPPENHEIMER, EPLEY, AND EYRE
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.
participants in the fluent condition (M⫽43%), t(39) ⫽2.01, p⬍
.05,
2
⫽.09. This fluency manipulation had no impact on
participants’ reported mood state (M
fluent
⫽4.50 vs. M
disfluent
⫽
4.29), t⬍1,
2
⬍.01; mood was not correlated with performance,
r(39) ⫽.18, p⫽.25; and including participants’ mood as a
covariate did not diminish the impact of fluency on performance,
t(39) ⫽2.15, p⬍.05,
2
⫽.11. The performance boost associated
with disfluent processing is therefore unlikely to be explained by
differences in incidental mood states.
General Discussion
Dual-process models of human judgment are fundamental to
much research in cognitive psychology, social psychology, and
judgment and decision making. People make different decisions
depending on whether they adopt systematic processing or rely on
intuitive, heuristic processing. Understanding the output of human
judgment and decision making therefore hinges on the ability to
predict when people will activate these reasoning systems.
This research provides evidence that experienced difficulty or
disfluency is one cue that leads people to adopt a systematic
approach to information processing. This effect emerged with
three manipulations of fluency across four domains of reasoning (a
domain-general test of systematic reasoning, persuasion, person
perception, and higher order reasoning). Our results suggest that
participants in each experiment who experienced difficulty or
disfluency while reasoning believed the tasks were more difficult
and therefore engaged in more analytical processing than did those
who did not. Pilot tests conducted for Experiment 2, 3, and 4
confirmed that our manipulations of fluency increased the apparent
difficulty of the task and decreased participants’ confidence in
their accuracy of judgment. Although people are usually content to
rely on heuristic processing, experienced difficulty or disfluency
appears to act as a cue that the problem in question may require
more elaborate thought and one’s simple or intuitive response is
likely to be wrong. These results are consistent with research
demonstrating that people are less likely to choose a default option
when their confidence is weakened (Simmons & Nelson, 2006).
In addition to describing the effect of fluency on processing
depth, we also ruled out the alternative possibilities that stimulus-
driven delays in processing speed (Experiments 2 and 3) and
negative mood (Experiment 4) rather than disfluency per se in-
duced systematic processing. Attending to stimuli more carefully
and processing them more slowly are both hallmarks of System 2
processes, but our results were not created merely by constraints
inherent in our stimuli that required more careful attention or
slower processing. Instead, our results appear to have been pro-
duced because disfluency acts as a cue that more deliberate pro-
cessing is required.
Existing fluency research has shown that people interpret stim-
uli depending on how easy those stimuli are to process (for a
review, see Schwarz, 2004), but our research makes the distinct
theoretical point that processing fluency can also influence judg-
ment indirectly by serving as a cue to engage in deeper reasoning.
Until now, fluency researchers have not provided participants with
cues at a shallow, heuristic level that systematically contradict cues
at a deeper, systematic level (e.g., an MP3 player reviewer who
looks incompetent but writes a compelling review and vice versa).
As a result, it has not been possible to determine whether fluency
affects how deeply people process available information. Our
findings are therefore novel because they show that fluency is used
not only directly as a cue for judgment but also indirectly as a
mechanism for strategy selection.
Moreover, these findings suggest a resolution for inconsisten-
cies in the fluency literature. For instance, some research suggests
that fluent stimuli are perceived as being more familiar than
disfluent stimuli (Monin, 2003), whereas other research suggests
that fluent stimuli are judged as being less familiar than disfluent
stimuli (Guttentag & Dunn, 2003). In addition, although disfluency
implies low quality at a superficial level (e.g., Reber, Winkielman,
& Schwarz, 1998), it also activates System 2 processing that might
lead people to look beyond a target’s superficial characteristics.
Fluency might exert opposing influences when people use cogni-
tive difficulty as a direct cue (e.g., “I don’t like the target”) or as
an indirect cue (e.g., “I should think more carefully about my
judgment”). Thus, the direct effects of fluency on evaluation and
the indirect effects of fluency on subsequent processing might, in
certain contexts, have opposing influences on judgment.
More generally, these findings bring us closer to determining
what activates System 2 processing. To be viable accounts of
human reasoning, dual-system theories need to explain not only
what the two systems are but also what activates the use of each
system. This research suggests that processing fluency may be one
important factor that determines when people will overcome their
intuitive responses to engage more systematic reasoning.
References
Ajzen, I., & Sexton, J. (1999). Depth of processing, belief congruence, and
attitude– behavior correspondence. In S. Chaiken & Y. Trope (Eds.),
Dual-process theories in social psychology (pp. 117–138). New York:
Guilford Press.
Alter, A. L., & Oppenheimer, D. M. (2006). Predicting short-term stock
fluctuations using processing fluency. Proceedings of the National
Academy of Sciences, USA, 103, 9369 –9372.
Bless, H., & Schwarz, N. (1999). Sufficient and necessary conditions in
dual-process models: The case of mood and information processing. In
S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychol-
ogy (pp. 423– 440). New York: Guilford Press.
Bodenhausen, G. V., Macrae, C. N., & Sherman, J. W. (1999). On the
dialectics of discrimination: Dual processes in social stereotyping. In S.
Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology
(pp. 271–290). New York: Guilford Press.
Boksman, K., The´berge, J., Williamson, P., Drost, D. J., Malla, A.,
Densmore, M., et al. (2005). A 4.0-T fMRI study of brain connectivity
during word fluency in first-episode schizophrenia. Schizophrenia Re-
search, 75, 247–263.
Botvinick, M. M., Braver, T. S., Carter, C. S., Barch, D. M., & Cohen, J. D.
(2001). Conflict monitoring and cognitive control. Psychological Re-
view, 108, 624 – 652.
Chaiken, S. (1980). Heuristic versus systematic information processing and
the use of source versus message cues in persuasion. Journal of Per-
sonality and Social Psychology, 39, 752–766.
Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective
taking as egocentric anchoring and adjustment. Journal of Personality
and Social Psychology, 87, 327–339.
Evans, J. S. B. T. (2003). In two minds: Dual-process accounts of reason-
ing. Trends in Cognitive Sciences, 7, 454 – 459.
Evans, J. S. B. T., & Over, D. E. (1996). Rationality and reasoning. Hove,
UK: Psychology Press.
575
FLUENCY AND PROCESSING DEPTH
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.
Forgas, J. P. (1995). Mood and judgment: The affect infusion model
(AIM). Psychological Bulletin, 117, 39 – 66.
Frederick, S. (2005). Cognitive reflection and decision making. Journal of
Economic Perspectives, 19, 25– 42.
Gilbert, D. T. (1989). Thinking lightly about others: Automatic compo-
nents of the social inference process. In J. S. Uleman & J. A. Bargh
(Eds.), Unintended thought (pp. 189 –211). New York: Guilford Press.
Gill, M. J., Swann, W. B., Jr., & Silvera, D. H. (1998). On the genesis of
confidence. Journal of Personality and Social Psychology, 75, 1101–
1114.
Goel, V., Buchel, C., Frith, C., & Dolan, R. J. (2000). Dissociation of
mechanisms underlying syllogistic reasoning. Neuroimage, 12, 504 –
518.
Griffin, D., & Tversky, A. (1992). The weighing of evidence and the
determinants of confidence. Cognitive Psychology, 24, 411– 435.
Guttentag, R., & Dunn, J. (2003). Judgments of remembering: The reve-
lation effect in children and adults. Journal of Experimental Child
Psychology, 86, 153–167.
Jacoby, L. L., Kelley, C. M., & McElree, B. D. (1999). The role of
cognitive control: Early selection vs. late correction. In S. Chaiken & Y.
Trope (Eds.), Dual-process theories in social psychology (pp. 383– 400).
New York: Guilford Press.
James, W. (1950). The principles of psychology. New York: Dover. (Orig-
inal work published 1890)
Johnson-Laird, P. N., & Bara, B. G. (1984). Syllogistic inference. Cogni-
tion, 16, 1– 61.
Jones, T. C., & Jacoby, L. L. (2001). Feature and conjunction errors in
memory: Evidence for dual-process theory. Journal of Memory and
Language, 45, 82–102.
Kahneman, D., & Frederick, S. (2002). Representativeness revisited: At-
tribute substitution in intuitive judgment. In T. Gilovich, D. Griffin, &
D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive
judgment (pp. 49 –81). New York: Cambridge University Press.
Kahneman, D., & Tversky, A. (1973). On the psychology of prediction.
Psychological Review, 80, 237–251.
Kelley, C. M., & Lindsay, D. S. (1993). Remembering mistaken for
knowing: Ease of retrieval as a basis for confidence in answers to
general knowledge questions. Journal of Memory and Language, 32,
1–24.
Keysar, B., & Barr, D. J. (2002). Self anchoring in conversation: Why
language users do not do what they “should.” In T. Gilovich, D. W.
Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology
of intuitive judgment (pp. 150 –166). Cambridge, UK: Cambridge Uni-
versity Press.
Kruglanski, A. W., & Thompson, E. P. (1999). Persuasion by a single
route: A view from the unimodel. Psychological Inquiry, 10, 83–109.
Lieberman, M. D., Gaunt, R., Gilbert, D. T., & Trope, Y. (2002). Reflec-
tion and reflexion: A social cognitive neuroscience approach to attribu-
tional inference. In M. P. Zanna (Ed.), Advances in experimental social
psychology (Vol. 34, pp. 199 –249). New York: Academic Press.
Lundqvist, D., Flykt, A., & O
¨hman, A. (1998). The Karolinska directed
emotional faces. Stockholm, Sweden: Karolinska Institute, Department
of Clinical Neuroscience, Psychology Section.
McGlone, M. S., & Tofighbakhsh, J. (2000). Birds of a feather flock
conjointly (?): Rhyme as reason in aphorisms. Psychological Science,
11, 424 – 428.
Monin, B. (2003). The warm glow heuristic: When liking leads to famil-
iarity. Journal of Personality and Social Psychology, 85, 1035–1048.
Oppenheimer, D. M. (2006). Consequences of erudite vernacular utilized
irrespective of necessity: Problems with using long words needlessly.
Applied Cognitive Psychology, 20, 139 –156.
Osman, M. (2004). An evaluation of dual-process theories of reasoning.
Psychonomic Bulletin and Review, 11, 988 –1010.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of
persuasion. In L. Berkowitz (Ed.), Advances in experimental social
psychology (Vol. 19, pp. 123–205). New York: Academic Press.
Reber, R., Winkielman, P., & Schwarz, N. (1998). Effects of perceptual
fluency on affective judgments. Psychological Science, 9, 45– 48.
Rips, L. J. (1994). The psychology of proof: Deductive reasoning in human
thinking. Cambridge, MA: MIT Press.
Schwarz, N. (1998). Accessible content and accessibility experiences: The
interplay of declarative and experiential information in judgment. Per-
sonality and Social Psychology Review, 2, 87–99.
Schwarz, N. (2004). Metacognitive experiences in consumer judgment and
decision making. Journal of Consumer Psychology, 14, 332–348.
Schwarz, N., Bless, H., & Bohner, G. (1991). Mood and persuasion:
Affective states influence the processing of persuasive communications.
In M. Zanna (Ed.), Advances in experimental social psychology (Vol. 24,
pp. 161–197). New York: Academic Press.
Simmons, J. P., & Nelson, L. D. (2006). Intuitive confidence: Choosing
between intuitive and nonintuitive alternatives. Journal of Experimental
Psychology: General, 135, 409 – 428.
Sloman, S. A. (1996). The empirical case for two systems of reasoning.
Psychological Bulletin, 119, 3–22.
Stanovich, K. E., & West, R. F. (2000). Individual differences in reasoning:
Implications for the rationality debate? Behavioral and Brain Sciences,
23, 645– 665.
Stepper, S., & Strack, F. (1993). Proprioceptive determinants of emotional
and nonemotional feelings. Journal of Personality and Social Psychol-
ogy, 64, 211–220.
Tetlock, P. E., & Lerner, J. S. (1999). The social contingency model:
Identifying empirical and normative boundary conditions for the error-
and-bias portrait of human nature. In S. Chaiken & Y. Trope (Eds.),
Dual-process theories in social psychology (pp. 571–585). New York:
Guilford Press.
Tourangeau, R., & Ellsworth, P. (1979). The role of the face in the
experience of emotion. Journal of Personality and Social Psychology,
37, 1519 –1531.
Werth, L., & Strack, F. (2003). An inferential approach to the knew-it-all-
along phenomenon. Memory, 11, 411– 419.
Whittlesea, B. W. A., & Leboe, J. P. (2003). Two fluency heuristics (and
how to tell them apart). Journal of Memory and Language, 49, 62–79.
Willis, J., & Todorov, A. (2006). First impressions: Making up your mind
after a 100-ms exposure to a face. Psychological Science, 17, 592–598.
Zielinski, T., Goodwin, G. P., & Halford, G. S. (2006). Complexity of
categorical syllogisms: An integration of two metrics. Manuscript sub-
mitted for publication.
Received August 4, 2006
Revision received March 23, 2007
Accepted March 26, 2007 䡲
576 ALTER, OPPENHEIMER, EPLEY, AND EYRE
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.
