Perceptual anchoring and adjustment

Abstract

Anchoring and adjustment, a ubiquitous heuristic process in judgment and decision making, has been vastly demonstrated in the numerical domain. We, with the help of four studies, demonstrate the anchoring and adjustment bias in perceptual domains. Our results show that anchoring and adjustment can bias our judgments at relatively low levels of cognition. Additionally, we outline a process by which anchoring and adjustment biases individuals' judgments in perceptual domains. Our results indicate a process wherein individuals search for an answer by testing plausible answers, the search being biased by the anchor question. We show that this movement is dominated by adjustments to adjacent possible responses indicating a search process constrained by selective accessibility. This process account explains the extant data in numerical domains as well—thus providing a way for a potential resolution to the disagreement among different existing process accounts for the anchoring phenomenon.

Full text

RESEARCH ARTICLE
Perceptual anchoring and adjustment
Gaurav Jain
1
| Dhananjay Nayakankuppam
2
| Gary J. Gaeth
2
1
Lally School of Management, Rensselaer
Polytechnic Institute, Troy, NY, USA
2
Department of Marketing, Tippie College of
Business, University of Iowa, Iowa, IA, USA
Correspondence
Gaurav Jain, Lally School of Management,
Rensselaer Polytechnic Institute, Troy, NY
12180, USA.
Email: gauraj@rpi.edu
Abstract
Anchoring and adjustment, a ubiquitous heuristic process in judgment and decision
making, has been vastly demonstrated in the numerical domain. We, with the help of
four studies, demonstrate the anchoring and adjustment bias in perceptual domains.
Our results show that anchoring and adjustment can bias our judgments at relatively
low levels of cognition. Additionally, we outline a process by which anchoring and
adjustment biases individuals' judgments in perceptual domains. Our results indicate
a process wherein individuals search for an answer by testing plausible answers, the
search being biased by the anchor question. We show that this movement is domi-
nated by adjustments to adjacent possible responses indicating a search process con-
strained by selective accessibility. This process account explains the extant data in
numerical domains as well—thus providing a way for a potential resolution to the dis-
agreement among different existing process accounts for the anchoring
phenomenon.
KEYWORDS
anchoring, insufficient adjustment, perceptual anchor, scale distortion, selective accessibility
1|INTRODUCTION
Think of the last two digits of your social security number. Now,
would you like to pay this amount in dollars for a color-changing coffee
mug? How much would you like to pay for such a mug? Although the
two are unrelated, the two-digit number that you initially thought of can
impact the price you are willing to pay for the mug. In fact, Ariely
et al. (2003) did a similar experiment and found that individuals whose
social security number ended with a higher number were willing to pay
between 60% and 120% more for products such as wine and coffee
mugs than the individuals whose social security number ended with a
lower number.
Since Tversky and Kahneman (1974) first demonstrated the
anchoring phenomenon with numerical estimates and judgments,
anchoring has been shown to influence both laboratory and real-world
judgments in various domains, including legal decisions (Englich &
Mussweiler, 2001), negotiations (Galinsky & Mussweiler, 2002), con-
sumer purchases (Wansink et al., 1998), and general trivia (Tversky &
Kahneman, 1974).
Most studies of anchoring and adjustment rely upon the num-
ber line as the vehicle for the response—that is, anchors and
adjustments are numerical. While the number line has excellent
properties that encourage continuous response metrics, it is subject
to the criticism of relying upon higher-order cognition. LeBoeuf
and Shafir (2006) showed that participants' estimates of the length
of a drawn line, the weight of a cup full of pennies, and the
volume of music were biased by earlier encounters with a line of a
particular length, a cup of a particular weight, and the volume of a
sound heard initially. Oppenheimer et al. (2008) find that individ-
uals evaluated the temperature in Honolulu to be lower when
they were made to draw short lines versus when they were made
to draw long lines. But in these studies, the final estimates
were still a numeric measurement of a target entity calling for
higher-order metric-based cognitions from the individuals making
the estimates. That is, a perception still has to be cognitively repre-
sented and mapped onto a number line. Is it possible that anchor-
ing only applies when this mapping to a numeric response is
required?
In our work described here, we demonstrate that anchoring does
indeed occur in varying non-numeric domains such as haptics, sounds,
and grayscales, where participants are exposed to a non-numeric
anchor. Most importantly, we keep the modality consistent so that
Received: 15 April 2020 Revised: 25 December 2020 Accepted: 27 December 2020
DOI: 10.1002/bdm.2231
J Behav Dec Making. 2021;1–12. wileyonlinelibrary.com/journal/bdm © 2021 John Wiley & Sons, Ltd. 1

decision-makers are exposed to the non-numeric anchor and are
asked for a similar non-numeric estimate. Hence, participants are not
reliant on a mental representation of the number line. Thus, with the
help of novel methodologies across multiple studies, our work extends
anchoring to non-numeric domains (such as domains of haptics and
sounds) and shows that anchoring can take place in perceptual
domains as well.
In Study 1, we create a tone scale to demonstrate anchoring in
the domain of sounds. In Study 2, we demonstrate the anchoring phe-
nomenon in the domain of textures. In Study 3, we use a grayscale to
show that anchoring and adjustment can bias individuals' judgments
in the visual domain as well. In addition, we use process-level experi-
mental techniques to examine the mechanisms underlying the demon-
strated perceptual anchoring. In Study 4, we check whether the
proposed mechanism can explain the anchoring phenomenon in the
numerical context as well.
There are existing process accounts that explain numerical
anchoring, including the three most common: insufficient adjustment
(Epley & Gilovich, 2001, 2004, 2006; Epley et al., 2004;
Quattrone, 1982; Tversky & Kahneman, 1974), selective accessibility
(Chapman & Johnson, 1999; Mussweiler & Strack, 1999, 2000,
2001b; Strack & Mussweiler, 1997), and scale distortion theory
(Frederick & Mochon, 2012; Mochon & Frederick, 2013). After the ini-
tial demonstration Studies 1 and 2, we discuss these existing process
accounts and make a case for a common process account for percep-
tual and numerical anchoring.
2|STUDY 1
Using auditory perception, Study 1 demonstrates the anchoring and
adjustment phenomenon in a numeric-free, perceptual domain.
2.1 |Method
Individuals who were employees or students at a large US university
were invited to participate in the study through a university-approved
mass email. Sixty-eight participants agreed to take part in a single-
factor between-subject design study. We used the anticipated effect
size of Cohen's f= 0.4 at an α= 0.05 and β= .95. The calculated sam-
ple size using G*Power (Faul et al., 2007) was 84. Thus, we had put
the limit at 90 participants—however, all the slots were not filled.
A tonal scale was presented on the computer to all participants,
consisting of 21 unlabeled buttons, each of which produced a pure
sinusoidal tone (i.e., no timbre) of increasing frequency, in steps of
100 Hz, ranging from 900 to 3900 Hz. Participants were allowed to
familiarize themselves with the scale by pressing any of the buttons to
generate the associated sound as many times as they wanted. Subse-
quently, they heard a reference tone of 1600 Hz. They were then
shown the tone scale but with one of the buttons enlarged and were
instructed to press the enlarged button, which served as the anchor
with a frequency either above the reference tone or below it, and
judge whether it matched the reference tone they had just heard. No
other button at this stage was functional except the enlarged button
(approximately half of the participants were provided the tonal scale
with the fourth button enlarged, and the remaining were provided the
tonal scale with the fifteenth button enlarged. The correct answer
was the ninth button). After this judgment, they were instructed to
judge which button produced a tone that matched the reference tone
they had heard earlier. At this point, all the buttons worked, including
the enlarged button, and they could press any button any number of
times.
2.2 |Analysis and results
For the purpose of the analysis, the buttons were coded in the
ascending order of frequency—the lowest frequency button was
coded “1,”and the highest frequency button was coded as “21”(note
these numbers were not present in the stimuli to preserve their non-
numeric nature). The participants provided with a lower frequency
anchor picked a significantly lower tone as the final response
(M= 8.38) compared to participants provided with a higher frequency
anchor (M= 10.31), (F(1, 67) = 5.24, p= .025, η
p2
= .074). Thus, the
anchoring and adjustment phenomenon extended to the auditory
domain.
3|STUDY 2
Study 2 utilized the sense of touch as the stimulus and response
domain; we investigated anchoring and adjustment mechanisms in the
domain of textures. Sandpapers of varying degrees of coarseness
were utilized. This is a domain that is much further divorced from the
number line when compared to line lengths, and so forth, that have
been used in the past (line lengths are closer to assessments of magni-
tude compared to textures, which are more qualitative in nature).
3.1 |Method
Individuals who were employees or students at a large US university
were invited to participate in the study through a university-approved
mass email. Sixty-seven individuals agreed to participate and were
randomly assigned to one of two conditions in a single-factor,
between-subjects' design. We used the anticipated effect size of
Cohen's f= 0.4 at an α= 0.05 and β= .95. The calculated sample size
using G*Power (Faul et al., 2007) was 84. Thus, we had put the limit
at 90 participants—however, all the slots were not filled.
After signing a consent form, participants were shown a folder,
inside which was a piece of sandpaper. Instructions on the folder
asked them not to open the folder themselves. They were requested
to close their eyes and ask the research assistant to open the folder
for them. They were then instructed to feel the sandpaper inside the
folder with their eyes closed and were informed that they would be
2JAIN ET AL.

asked questions about the same later. They were asked to tell the
research assistant to close the folder before opening their eyes.
Subsequently, they proceeded to another room where 16 pieces of
sandpaper of various levels of coarseness were arranged in decreasing
order of coarseness on a poster board placed on a table. One of the
16 sandpapers had two pointers, one red and one blue, above it and
served as the anchor. Participants were asked to judge whether this
sandpaper with the pointers above it was of the same coarseness as
the one they had felt earlier in the previous room. If so, they could
leave the blue pointer where it was; else, they were instructed to
move the blue pointer to the sandpaper they judged to be the same
as they had felt earlier. The red pointer could not be moved. Partici-
pants were told that they could touch all the sandpapers as many
times as they wanted in the process of making their judgment. A
camera was used to record their decision processes and, in particular,
the instances of touching the various grades of sandpaper placed on
the board. The recording was reviewed and scored at a later time by
the primary investigator. An external coder naïve to the hypotheses
recoded 10 of the videos randomly to check the accuracy of the
coded data. There was a hundred percent agreement with the
originally coded data.
The sixth sandpaper sample in the set was actually identical to
the one they had felt earlier in the previous room. However, the red
pointer which served as the anchor was placed over either the second
sandpaper (which was coarser) or over the tenth sandpaper (which
was finer). Initially, the blue pointer was placed with the red pointer as
well, but, as mentioned above, it could be moved by the participants.
Thus, this was a single-factor (coarseness of the anchor) between-
subjects design with two levels. Both anchors were four positions
away from the “correct”answer.
3.2 |Analysis and results
The camera recording was transcribed to code for the order in which
participants touched the various sandpapers before making their final
judgment. For the purposes of the analysis, the coarsest sandpaper
was coded 1, and the smoothest was coded 16 as per the industrial
scaling system, where a higher rating on the sandpaper reflects a finer
grade of abrasive particles.
We considered the influence of the anchor on the final judgment
and found that the anchoring and adjustment phenomenon was repli-
cated. Participants who started with the coarser (No. 2) anchor judged
a coarser sandpaper (M= 5.15) to be similar to the target sandpaper
they had touched earlier, compared to participants who started with
the finer (No. 10) anchor (M= 7.14), (F(1, 66) = 27.69, p< .001,
η
p2
= .296).
Studies 1 and 2 allow us to draw the conclusion that anchoring
extends to non-numeric domains as well. Also, it should be noted that
judgments of sound and granularity of sandpaper are more basic
perceptions—thus, the data suggest that the anchoring and
adjustment bias does not depend on a numeric representation of the
stimulus or response scale.
4|DISCUSSION OF STUDIES 1 AND 2: A
PEEK INTO THE “ADJUSTMENT”PROCESS
Due to the use of non-numeric modalities, and unlike a response on a
number line, we could actually surreptitiously observe the process
individuals follow to get to the final response, which allows us to deci-
pher the processes underlying perceptual anchoring. Also, we can
make important inferences from our data that shed light on the mech-
anisms underlying numerical anchoring, which has been a topic of
conflict in the literature.
4.1 |The conflicting process accounts for
numerical anchoring
Although numerical anchoring as a phenomenon is reliable and easy
to demonstrate, there is less agreement about the underlying pro-
cesses (Furnham & Boo, 2011; Turner & Schley, 2016). We know the
anchor has an impact but are less certain how the adjustment process
works. In Tversky and Kahneman's original demonstration of this phe-
nomenon (see also Epley et al., 2004; Epley & Gilovich, 2001, 2004,
2006; Quattrone, 1982), the anchor was seen to bias subsequent
judgments, prompting the conclusion that adjustments away from the
anchor were insufficient. However, Mussweiler and Strack (1999,
2000, 2001b) advanced an account that would appear to subsume the
anchoring and adjustment phenomenon under an accessibility
account. Specifically, they argue that individuals engage in confirma-
tory testing for the anchor, which makes information consistent with
the anchor more available in the later judgment, resulting in the sub-
sequent judgments assimilating towards the anchor. In other words,
adjustment is not a necessary step in this account. Rather, an anchor
biases the construction of the eventual judgment by making some
information more accessible than others. Epley and Gilovich (2001,
2006) have argued that this is inconsistent with the influence of
implausible, self-generated, or patently random anchors and argue for
the continued role of insufficient adjustment.
A third proposed account often referred to as the “scale
distortion theory”(Frederick & Mochon, 2012; Mochon &
Frederick, 2013) suggests that the “true answer”looks extreme rela-
tive to the anchor that was just considered, prompting an adjustment
towards the anchor. In other words, it proposes a contrast effect as
driving the adjustment.
The extant data cannot resolve this existing conflict. What is
observed empirically in a typical anchoring and adjustment experiment
are final estimates that are biased by answering an initial question that
makes one particular anchor salient—that is, judgments assimilate
towards an anchor. This phenomenon is perfectly consistent with the
idea of insufficient adjustment from the anchor (or adjustment towards
the anchor as proposed by the scale distortion account), but it is also
consistent with the idea of an answer constructed from information,
some of which was made preferentially accessible by answering the
anchor question. Moderating variables could help to resolve this debate,
but finding one that does so unequivocally has been problematic.
JAIN ET AL.3

Creative explanations and counter-explanations can be generated for
each moderating variable affecting construction processes as well as
adjustment processes. For example, it has been argued that implausible
anchors generating anchoring effects are incompatible with the con-
struction account. However, as Mussweiler and Strack (2001a) have
argued, an implausible anchor may generate adjustment away from the
anchor till a plausible anchor is reached, but upon reaching the realm of
plausibility, positive hypothesis testing still occurs, and the final judg-
ment is thus still constructed. This retains the adjustment component of
the process, but it is relegated to explaining movement from implausibil-
ity and is divorced from the anchoring phenomenon.
Unlike previous research, we do not use numeric domains. The
use of numerical domains does not allow “observing”the process by
which individuals get to the final response after being biased by the
anchor. This is because individuals responding in the context of
numerical anchoring and adjustment demonstrations rely upon a men-
tal number line as the vehicle for the response. The mental number
line, being an abstract concept in mind, makes it difficult to observe
any kind of process. The use of alternate non-numeric domains for
anchoring and adjustment demonstrations allows us to surreptitiously
observe how individuals are getting to their final response after being
exposed to the anchor.
4.2 |Inferences from our data about the
underlying mechanism
Additional analyses on the data that we had collected as part of
Studies 1 and 2 allowed us greater insight into how individuals get
from the anchor to their final answer.
In Study 1, the order in which participants pressed buttons while
deciding on the correct response was recorded and analyzed. Every
button tried by the participant counted as a step, and the size of the
step from the previous button was recorded. We conducted an analy-
sis of the frequency of various sized steps on the data aggregated to
the level of the individual. Analysis of the frequency of steps of
various sizes as a within-subject factor revealed a significantly larger
number of unit-sized steps (M
number of steps of size 1
= 6.19) compared
to all non-unit-sized steps combined (M
number of steps >size 1
= 1.58), (F
(1, 68) = 106.51, p< .001, η
p2
= .614). In other words, the process of
“adjustment”appears to be dominated by moving to adjacent
positions.
As mentioned earlier, in Study 2, the camera recording was tran-
scribed to code for the order in which participants touched the vari-
ous sandpapers before making their final judgment. For the purposes
of the analysis, the coarsest sandpaper was coded 1, and the
smoothest was coded 16. An analysis of the frequency of steps of
various sizes as a within-participant factor revealed that participants
took many more unit-sized steps (i.e., steps to adjacent possibilities)
(M
number of steps of size 1
= 17.69) than all non-unit-sized steps
combined (M
number of steps >size 1
= 2.26), (F(1, 67) = 250.25, p< .001,
η
p2
= .789), again suggesting that “adjustment”appears to be domi-
nated by moving to adjacent positions.
Additionally, in Study 2, we modeled whether the size of the nth
step taken by a participant is influenced by the value of n. This tests
for whether participants make initial large “adjustments”followed by
smaller ones. The size of the step taken was modeled as a function of
the number of prior steps taken, with random effects at the individual
level to allow for heterogeneity (variation in intercepts (or starting
points) and slopes (the rate at which step size varied as a function of
steps). Table 1 provides the fit statistics for the various models that
were tested. The best-fitting model allowed for heterogeneity in inter-
cepts and slopes (hardly surprising given individuals were given differ-
ent anchor values and adjusted in different directions). With this
model formulation, analyses revealed that step size significantly
decreased as a function of the number of prior steps. Table 2 provides
the results for the best-fitting model. It is worth noting that partici-
pants in the coarse sandpaper condition started towards the left end
of the scale (their anchor was sandpaper No. 2, which was the second
sandpaper from the left), but participants in the smoother sandpaper
condition started from the mid-right region of the scale (their anchor
was sandpaper No. 10, which was the seventh sandpaper from the
right). Thus, participants in the coarser sandpaper condition had a big-
ger part of the scale to take larger steps if they wanted in the initial
stages of their search. This could be a potential drawback of the study
if we did a between-groups analysis. The data for the above analysis,
though, was aggregated across groups (as we do not expect the initial
anchor to influence the search process), and the utilization of the hier-
archical nature of the data allows us to control for such potential
effects.
4.3 |Our proposed process account
How well do the extant process accounts explain the data presented
here? The traditional adjustment-based perspective would argue that
individuals would begin at the anchor and then start adjusting. This
account, though, is silent on the specifics of adjustment. Importantly,
as per this account individuals would stop far too early, generating an
“insufficient adjustment”explanation for the anchoring phenomenon.
The “scale distortion”account would specify a move from the anchor
TABLE 1 Fit statistics for the models tested for Study 1
Model
−2 Res Log-
Likelihood
Base model with random intercept 5024.9
STSIZE = β
1
STORD with random intercept 5029.6
STSIZE = β
1
STORD + β
2
STORD
2
with random
intercept
5038.8
STSIZE = β
1
STORD with random intercept and
linear term
5023.1
STSIZE = β
1
STORD + β
2
STORD
2
with random
intercept, linear and quadratic term
5020.0
Abbreviations: STSIZE, size of a particular step; STORD, order of a
particular step.
4JAIN ET AL.

to the answer, followed by adjustment away from the answer towards
the anchor. Finally, the accessibility and construction-based perspec-
tive would argue that no adjustment takes place, but there would be
positive hypothesis testing in answering the anchor question. The
subsequent judgment would be constructed, the construction being
dominated by information made accessible in answering the prior
anchor question. So, there should be no adjustment (unless presented
with an implausible anchor, in which case, one would adjust to the
realm of the plausible, after which there is still a positive hypothesis
testing which takes place). As such, from the realm of the plausible
(which is reached by the presentation of an anchor or by adjustment),
there should be one move to the constructed answer. In other words,
there should be one move from a plausible anchor or a series of
moves from an implausible anchor followed by one move to the
constructed final answer. In any case, these moves should be domi-
nated by the relatively large last adjustment. Our data are clearly not
explained completely by all these predictions. What we see is a
dominance of steps of size 1 and a reduction in the size of step as the
process continues. In other words, participants appear to make large
leaps in one direction followed by a search in the adjacent space sur-
rounding where they land.
We suggest that the anchoring and adjustment phenomenon rep-
resents searching an ordered space for a response. Given the anchor
is usually rejected as the final judgment and roundly rejected in the
case of implausible anchors, the absence of a resultant contrast effect,
rather than the observed assimilation effect, suggests that what is
being observed need not (always) be the construction of a judgment.
We posit that the process resembles searching for an answer in the
related space (which has typically been a number space). This search,
however, can be biased by information rendered accessible in
assessing the anchor, much as Mussweiler and Strack have argued
(Mussweiler & Strack, 1999, 2001b also, Chapman & Johnson, 1999;
Strack & Mussweiler, 1997; see also Simmons et al., 2010 for an argu-
ment that the competing anchoring theories can operate simulta-
neously). The search process will be dominated by adjustments to
adjacent possible responses, implying a search process constrained by
selective accessibility. In other words, as suggested by the adjustment
proponents, there is indeed adjustment and not construction, but the
adjustment itself is constrained by positive hypothesis testing as
suggested by the construction proponents. Thus, “search for a
response”may be a more accurate description rather than
adjustment—individuals are searching a hypothesis space in the wake
of having rejected an anchor, and the cognitive and affective reaction
to the anchor constrains the search in predictable ways.
This begs the following questions: How would the search be
biased? We argue that the bias would be in the range of the search.
Confirmatory hypothesis testing is likely to generate an affective
response, that is, a “feeling”of the answer being near at hand as it is
shown that people feel higher arousal when they feel they are near
the answer to a question (Litman et al., 2005). So, we should see sea-
rch constrained to the space close to the anchor, in what we term
“the adjacent possible.”If presented with an implausible anchor or an
anchor that feels “wrong,”individuals may take a large “leap”in the
direction of an expected answer. Then this potential search region
would contain responses that look reasonable compared to the
implausible or “wrong”anchor one encountered a moment ago; it
should again generate a “feeling”of the answer being close at hand
and constrain the search. In any case, we should see individuals
searching more intensively in “the adjacent possible”space rather
than more far-flung areas of the hypothesis space barring the first
step away from an implausible response option. This is in contrast to
the more normative rule of search (Knuth, 1973) wherein one would
establish the boundaries and work inwards from the boundaries by
halving the search space with each subsequent hypothesis.
Our data are consistent with this account and suggests that sea-
rch is constrained by hypothesis confirmation and would be domi-
nated by the movement to the “adjacent possible”explaining the
large number of small steps seen in the two studies. The search should
start with a few large steps which would get one to a region that feels
reasonable compared to the anchor, and search would then rapidly
converge to a large number of small steps whose size is of a
diminishing, negative exponential form, as seen in the studies.
Before we move on to the next study, it is important to note that
although we argue that the use of non-numeric domains allows us to
eliminate scale-based explanations of the adjustment process, we
need to address alternate arguments that can account for the pattern
of data observed in Studies 1 and 2. First, one can argue that partici-
pants may have been trying to familiarize themselves with the scale
(to see if it was linear, etc.), which may have increased the number of
unit-sized steps. It should be noted that we had a familiarization stage
at the beginning of Study 1, where participants could press any sound
button any number of times to get themselves acquainted with the
sound scale. The scale familiarization account would argue that indi-
viduals with less practice (and thus less familiarity with the scale)
would be inclined to explore the scale during the judgment phase of
the task and would press more keys. The average number of buttons
pressed during this familiarization period was 13.71 button presses,
providing some face validity to the idea that individuals did expend
effort familiarizing themselves with the scale. The number of button
presses in the scale familiarization stage ranged from 2 to 51. A fur-
ther analysis using this measure as an indicator of the amount of scale
familiarization yielded a significant interaction (F(1,66) = 3.050,
p= 0.005, η
p2
= .112) with the focal effect. The number of non-unit-
sized steps did not differ as a function of the amount of practice. The
number of unit-sized steps did differ as a function of practice. Data
were analyzed at 1 SD above and below the mean. Examining the
number of non-unit-sized and unit-sized steps 1 SD above and below
TABLE 2 Results for the best-fitting model for Study 1
Effect Estimate Standard error DF tvalue Pr > jtj
Intercept 1.8265 0.1137 67 16.06 <.001
STORD −0.0468 0.0178 67 −2.63 .011
STORD
2
0.0011 0.0001 67 2.10 .040
Abbreviation: STORD, order of a particular step.
JAIN ET AL.5

the mean amount of practice confirmed that there was no difference
in the number of non-unit-sized steps (1.72
at Mean –1SD
vs.
1.45
at M+ 1SD
;t(66) = −.56, p> 0.250), while there was a difference
in the number of unit-sized steps (5.09
at Mean –1SD
vs. 7.28
at M+ 1SD
;
t(66) = 2.53, p= 0.014). Note, however, that this difference is in the
opposite direction to the scale familiarization account. The scale famil-
iarization account would argue the individuals with less practice (and
thus, less familiarity with the scale) who would be inclined to explore
the scale during the judgment phase of the task. What was observed,
however, was exactly the opposite, with increased practice resulting
in a greater number of unit-sized steps (std. b= .297). Thus, the scale
familiarization account is incompatible with these data and is unlikely
as an explanation for the results of the prior studies.
The second argument can be that participants took a large num-
ber of unit-sized steps as they were trying to experience each sandpa-
per or sound button since it is arguably almost costless to look at or
feel a particular square. To take care of this issue, we repeated Study
2 with a minor change—we introduced the notion of a cost to the
adjustment process.
The procedure for this version of Study 2 was substantially
the same as the original Study 2, except that participants were
now informed that while they could touch as many pieces of sand-
paper as they wanted to, their task was to provide as accurate an
answer as possible while minimizing the number of touches. It was
explained to participants that touching a piece of sandpaper a sec-
ond time after touching other sandpapers would be counted as a
separate touch.
The traditional anchoring and adjustment effect was again repli-
cated as participants who had the coarse anchor (sandpaper No. 2)
chose a coarser sandpaper as their final answer (M= 5.52) compared
to those who had a smooth anchor (sandpaper No. 10) (M= 6.90), (F
(1, 42) = 6.06, p= .018, η
p2
= .126). Moreover, participants used a sig-
nificantly higher number of unit-sized steps (M= 1.98) compared to all
non-unit-sized steps combined (M= 1.11), (F(1, 41) = 11.096,
p= .002, η
p2
= .205).
The next study, Study 3, was done with two purposes in mind:
(1) to demonstrate the phenomenon in an additional, numeric-free,
perceptual domain, namely, in visual perception; and (2) to test
the process account we have proposed for the anchoring
phenomenon.
5|STUDY 3
Study 3 utilized shades of gray as the stimuli. This affords an added
advantage apart from being another domain that is not dependent on
the number line—by allowing us to utilize eye tracking as a process
measure of the search process. While participants were performing
the task, an eye tracker was utilized to track their eye fixations. Thus,
we could track the sequence of eye fixations, which gave us a covert
way of examining whether there was a tendency for small adjust-
ments to the “adjacent possible”rather than large steps in exploring
the hypothesis space.
5.1 |Method
Individuals who were employees or students at a large US university
were invited to participate in the study through a university-approved
mass email. One hundred and three individuals agreed to participate.
We used the anticipated effect size of Cohen's f= 0.4 at an α= 0.05
and β= .95. The calculated sample size using G*Power (Faul
et al., 2007) was 112. Thus, we had put the limit at 115 participants—
however, all the slots were not filled.
Participants were seated at a computer with an eye tracker. A
five-point calibration was performed to calibrate the eye tracker to
the particular individual's eyes. Participants were randomly assigned
to one of four “anchor”conditions in a single-factor, between-
subjects design. Twenty-six shades of gray from near black (G1) to
near white (G26) were utilized for the response scale. An ordered
array of 26 shades of gray from near white to near black was created
on the computer. Participants were shown a shade of gray that served
as an anchor (which was either G2 (almost black), G15 (dark gray),
G19 (light gray), or G24 (near white), depending on the anchor condi-
tion they were in) and were asked if that was the shade of the moon.
They were then shown all 26 shades and asked to pick the shade that,
in their opinion, was the same as that of the moon. Shades of gray
were developed by additive mixing of the three primary colors (R, G,
B) such that (0, 0, 0) corresponded to black. The next lighter shade is
utilized (10, 10, 10) and so on to the lightest shade of gray (250, 250,
250). The idea was to have a scenario where people are not aware of
the correct answer in order to demonstrate the anchoring phenome-
non. There is no single correct response to the “shade of moon”ques-
tion being asked here—hence, the scenario was deemed appropriate
for demonstrating an anchoring phenomenon.
Binocular eye movement data were recorded using a noninvasive
Tobii X2-60 Eye Tracker that records the accuracy of eye movements
to 0.4of visual angle. The eye tracker measures visual scanning by
computing the pupil-corneal reflection at a sampling rate of 60 Hz
(i.e., 60 gaze data points are collected per second for each eye) based
on the reflection of near-infrared light from the cornea and pupil.
5.2 |Analysis and results
The anchoring and adjustment phenomenon was replicated in this
novel domain (M
anchor = G2
= 6.536, M
anchor = G15
= 17.85,
M
anchor = G19
= 20.00, M
anchor = G24
= 22.75; F(3, 99) = 6.98, p< .001,
η
p2
= .175). Participants' final ratings of the moon's shade were signifi-
cantly impacted by the anchor they had been given.
The pattern of fixations for the eye-tracking data was coded to
indicate the pattern of movement from one shade of gray to the next.
Each individual could take step sizes ranging from 1 (to the adjacent
shade of gray) to a maximum step size of 25.
Our position is that the number of unit-sized steps would be
more than the number of non-unit-sized steps. Moreover, participants
would take bigger steps in the beginning, converging to smaller steps
as they closed in on a final response. Again, because the data are
6JAIN ET AL.

nested (sequential steps taken nested within participants), and there
are independent variables at the individual and group levels
(e.g., anchor at the level of the individual, number of prior steps at the
level of the steps taken), the use of multi-level modeling permits pars-
ing the variance at both levels (participant level heterogeneity and
step level), utilizing the appropriate statistical tests and testing for the
crucial cross-level interaction.
Just like in Studies 1 and 2, we modeled whether the size of the
nth step taken by a participant was influenced by the value of n. The
size of the step taken was modeled as a function of the number of
prior steps taken, with random effects at the individual level to allow
for variation in intercepts (or starting points) and slopes (the rate at
which step size varied as a function of steps). Table 3 provides the fit
statistics for the various models that were tested. The best-fitting
model allowed for heterogeneity in intercepts and slopes (hardly sur-
prising given individuals were given different anchor values). With this
model formulation, analyses revealed that consistent with our predic-
tion; step size did decrease as a function of the number of prior steps.
Table 4 provides the results for the best-fitting model.
We aggregated the data to the level of the individual and ana-
lyzed the frequency of steps of various sizes. This analysis revealed
that participants took significantly more unit-sized steps (M= 7.52)
compared to all larger than unit-sized steps combined (M= 6.34) (F
(1, 102) = 7.49, p= .007, η
p2
= .068). This was not moderated by the
anchor. Figure 1 provides the heat maps, and Appendix A illustrates
the order of fixations across different conditions.
The eye-tracking data permit an additional set of analyses to test
a further aspect of our prediction. Note that we have argued that sea-
rch in the “adjacent possible”is due to a feeling of the answer being
close at hand which results in a positive affect (Litman et al., 2005).
Such a feeling is likely to result in longer gaze duration at potential
hypotheses as one assesses whether it could be the answer. Thus, we
would predict that longer gaze duration at an area of interest (AOI)
(indicating greater consideration of that “gray”representing the
answer) should be followed by smaller movements (i.e., a greater ten-
dency to move to the “adjacent possible”). A multi-level model con-
firmed this to be the case—the gaze duration time at an AOI predicted
the size of the next step (b=−0.002, F(1, 1367) = 8.36, p= .004). In
other words, a long gaze duration (indicative of an increased feeling of
the answer being close at hand) was followed by a greater tendency
to move to a nearby option.
It should also be noted that the previous step size also predicted
gaze duration (b=−14.17, F(1, 1406) = 25.12, p< .001), suggesting
that smaller step sizes presumably served as a signal of a potentially
likely answer, which should exacerbate positive hypothesis testing
tendencies.
In Study 3, the idea was to have a scenario where people are
not aware of the correct answer in order to demonstrate the
anchoring phenomenon. There is no single correct response to the
“shade of moon”question being asked here—hence, the scenario
was deemed appropriate for demonstrating an anchoring phenome-
non. However, it can be argued that as the moon does not have
any specific color, the participants may not have a specific answer
to think of; that is, a single participant may think of multiple shades
of the moon as contenders for an appropriate response. Thus, we
ran another study (Study 3b) with a different context. One hundred
and thirty American individuals participated in a single-factor (lower
anchor vs. higher anchor) between-subjects study (51.5% females,
average age: 33.9 years). Participants, after given an initial cover
story, were told, “Australia is the second-largest producer of zinc.
The natural zinc extracted in Australia has a specific color—a partic-
ular shade of gray. How likely do you think the circled shade repre-
sents the color of natural zinc extracted in Australia?”. On the same
screen, participants were shown a grid of shades of gray (25 shades
with the lightest shade numbered as “1”and the darkest as “25”)
with one of the shades circled—fourth shade in case of lower
anchor condition and a fourth-last shade (i.e., 21st shade) in case of
higher anchor condition. Participants had to give their responses on
a five-point Likert-type scale (extremely unlikely—extremely likely).
On the next screen, participants were shown the grid again (with no
shade circled) and were asked to select the shade that they think
represents the natural zinc extracted from Australia. Participants
responded by clicking a shade on the grid. Participants' gaze was
recorded on this screen.
The results were similar to the previous study. The anchoring
phenomenon was replicated (M
Lower anchor
= 7.30, M
Higher
anchor
= 14.77; F(1, 128) = 61.477, p< .001, η
p2
= .324), that is, partici-
pants' final responses of the zinc's shade were significantly impacted
by the anchor they had been given. We analyzed the fixation patterns,
just like the previous eye-tracking study, and found the results to be
in a similar direction and form—that is, step size did decrease as a
function of the number of prior steps.
TABLE 3 Fit statistics for the models tested for Study 3
Model
−2 Res Log-
Likelihood
Base model with random intercept 8095.3
STSIZE = β
1
STORD with random intercept 8084.5
STSIZE = β
1
STORD + β
2
STORD
2
with random
intercept
8091.8
STSIZE = β
1
STORD with random intercepts and
slopes
8082.8
STSIZE = β
1
STORD + β
2
STORD
2
with random
intercepts and slopes
8046.7
Abbreviations: STSIZE, size of a particular step; STORD, order of a
particular step.
TABLE 4 Results for the best-fitting model for Study 3
Effect Estimate Standard error DF tvalue Pr > jtj
Intercept 5.303 0.427 102 12.43 <.001
STORD −0.415 0.094 99 −4.40 <.001
STORD
2
0.017 0.005 99 3.63 <.001
Abbreviation: STORD, order of a particular step.
JAIN ET AL.7

In the studies so far, we have demonstrated the anchoring phe-
nomenon in three different domains. Moreover, the studies find evi-
dence in support for the proposed underlying mechanism by which
anchors bias individuals' responses in the perceptual domains. We had
suggested earlier that it is possible a similar mechanism may hold true
for numerical anchoring as humans access magnitudes in different
domains in a similar manner (Walsh, 2003). The next study, Study
4, examines this possibility. Although it is a small step in the pursuit of
“looking”at the process by which individuals reach the final response
from the anchor, it is the first study in our knowledge that utilizes eye
tracking in the domain of numerical anchoring.
6|STUDY 4
6.1 |Method
One hundred and seven American individuals participated in a single-
factor (lower anchor versus higher anchor) eye-tracking study (54.2%
females, average age: 33.6 years). All participants, after being given an
initial cover story, were shown a screen with a question and numerical
grid (going from “100”to “490”in steps of 10). The question asked
the participants in the lower anchor condition: “How likely do you
think it is that the parking lot in the Jorpati Aquarium was designed to
have 160 parking spaces?”. The anchor was “430”for the higher
anchor condition. Participants responded on a five-point Likert-type
scale (extremely unlikely—extremely likely). On the next screen, partic-
ipants were again shown the grid and were asked: “Exactly how many
parking spaces, do you think, the parking lot at the Jorpati Aquarium
was designed to have?”. Participants responded by clicking a number
on the grid. Participants' gaze was recorded on this screen at a sam-
pling rate of 30 Hz (i.e., 30 gaze data points are collected per second
for each eye) using Realeye eye tracking. The anchors were equidis-
tant from the ends of the number grid.
6.2 |Analysis and results
As expected, the anchoring phenomenon was replicated. Participants'
final responses varied significantly between the two “anchor”groups
(M
Lower anchor
= 224.13, M
Higher anchor
= 402.91; F(1, 105) = 142.362,
p< .001, η
p2
= .576).
The purpose of Study 4 was to investigate whether participants
will follow the same process as they do in other perceptual domains—
thus, we carried out the same analyses as in Study 3. First, participants
took significantly more unit-sized steps (M= 2.02) compared to all
larger than unit-sized steps combined (M= 1.44) (F(1, 106) = 8.277,
p= .005, η
p2
= .072).
Just like in the previous study, we modeled whether the size of
the nth step taken by a participant was influenced by the value of n.
The size of the step taken was modeled as a function of the number
of prior steps taken, with random effects at the individual level to
allow for variation in intercepts (or starting points) and slopes (the rate
at which step size varied as a function of steps). Just like previous
studies, the best-fitting model allowed for heterogeneity in intercepts
and slopes (−2 Res Log-Likelihood = 2222.1). With this model formu-
lation, analyses revealed that consistent with our prediction, step size
did decrease as a function of the number of prior steps. Table 5 pro-
vides the results for the best-fitting model.
The results of Study 4 support our position that the underlying
mechanism for perceptual anchoring may hold true for numerical
anchoring as well. The results lay the ground for future work to look
for an exhaustive process account that can explain all the existing data
pertaining to numeric as well as non-numeric anchoring bias.
7|GENERAL DISCUSSION
We demonstrate the anchoring phenomenon in several perceptual
domains. Theoretically, these findings suggest that anchors can
FIGURE 1 The heat maps for
Study 3 illustrate the duration of a
typical participant's fixations on
different parts of the grayscale when
the anchor was G24, G19, G15, or G2.
In Figure 1a–c, where the anchors
were G24, G19, and G15,
respectively, it can be seen that the
fixation duration was maximum near
the anchor. However, as expected and
as seen in Panel (d), this was not the
case when the anchor was G2 as it
was a highly implausible anchor. The
anchors are circled in the below figure
only for illustrative purposes (and
were not circled in the actual stimuli)
8JAIN ET AL.

influence our judgments at a much lower level of cognition as well,
showing that the anchoring and adjustment phenomenon is not
dependent on one's numerical understanding and awareness.
The findings have practical implications in the consumption
domain. Although numerical anchoring is widely applied in marketing
(e.g., Ariely et al., 2003; Wansink et al., 1998), we do not see percep-
tual anchoring being used. Individuals, while making a decision in the
marketplace, consider multiple features of available options where
many of the features are non-numerical modes of perception
(e.g., consumers may contemplate whether the crockery is “heavy”
enough; is the laptop “light weight”; is the curry too “salty”; is the cof-
fee “warm”enough). This work provides a systematic framework that
can be utilized by marketers to anchor consumers to a particular taste,
color, weight, and other perceptual factors.
It is noteworthy that the use of novel non-numeric stimuli in
these studies gave us an added advantage—allowing us to surrepti-
tiously observe the process. We obtain process evidence by examin-
ing how individuals look for a response to a question asked after
being influenced by the anchor in the context of perceptual anchoring.
Because the way we assess magnitudes in different domains utilizes
essentially the same neural circuitry (ATOM—A Theory Of Magnitude,
Walsh, 2003), the data can contribute to resolving the disagreement
in the anchoring literature about the mechanisms underlying anchor-
ing and adjustment. Our results indicate a search process dominated
by adjustments to adjacent possible responses, implying a search pro-
cess constrained by selective accessibility. In other words, as
suggested by the adjustment proponents, there is indeed adjustment
and not construction, but the adjustment itself is constrained by posi-
tive hypothesis testing as suggested by the construction proponents.
Thus, the process account proposed here paves the way to integrate
the previous conflicting accounts. At the very least, the data reported
here provide the most precise description yet of movement from an
anchor to an answer and should thus assist to develop sharper theo-
retical formulations.
As mentioned above, even though we have used non-numeric
modalities, the underlying mechanism that we propose may hold true
for numerical anchoring as individuals use similar cortical metrics to
assess magnitudes of different domains (Walsh, 2003). The last study
does provide limited support for the position that our process account
may explain the extant numerical anchoring data as well. Although
there is research which shows humans share similar cortical metrics
for different domains (e.g., Rugani, Vallortigara, & Priftis, 2015;
Schwiedrzik Bernstein, & Melloni, 2016), more future work is required
to devise ways in which it can be confirmed whether the same
mechanism underlies numerical domain as well. It would be interesting
to find direct evidence for our process account in the numerical
domain in the future.
FUNDING INFORMATION
This research did not receive any specific grant from funding agencies
in the public, commercial, or not-for-profit sectors.
ORCID
Gaurav Jain https://orcid.org/0000-0002-6461-8286
Dhananjay Nayakankuppam https://orcid.org/0000-0003-0147-
3566
Gary J. Gaeth https://orcid.org/0000-0001-9514-8531
REFERENCES
Ariely, D., Loewenstein, G., & Prelec, D. (2003). Coherent arbitrariness: Stable
demand curves without stable preferences. Quarterly Journal of Econom-
ics,118,73–105. https://doi.org /10.116 2/00335530360535153
Chapman, G. B., & Johnson, E. J. (1999). Anchoring, activation, and the
construction of values. Organizational Behavior and Human Decision
Processes,79,1–39.
Englich, B., & Mussweiler, T. (2001). Sentencing under uncertainty:
Anchoring effects in the courtroom. Journal of Applied Social Psychol-
ogy,31, 1535–1551.
Epley, N., & Gilovich, T. (2001). Putting adjustment back in the A&A heu-
ristic: Differential processing of self-generated and experimenter-
provided anchors. Psychological Science,12, 391–396. https://doi.org/
10.1111/1467-9280.00372
Epley, N., & Gilovich, T. (2004). Are adjustments insufficient? Personality
and Social Psychology Bulletin,30(4), 447–460. https://doi.org/10.
1177/0146167203261889
Epley, N., & Gilovich, T. (2006). The anchoring-and-adjustment heuristic:
Why the adjustments are insufficient. Psychological Science,17(4),
311–318. https://doi.org/10.1111/j.1467-9280.2006.01704.x
Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective taking as
egocentric anchoring and adjustment. Journal of Personality and Social Psy-
chology,87, 327–339. https://doi.org/10.1037/0022-3514.87.3.327
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flex-
ible statistical power analysis program for the social, behavioral, and
biomedical sciences. Behavior Research Methods,39, 175–191. https://
doi.org/10.3758/BF03193146
Frederick, S. W., & Mochon, D. (2012). A scale distortion theory of anchor-
ing. Journal of Experimental Psychology: General,141(1), 124–133.
https://doi.org/10.1037/a0024006
Furnham, A., & Boo, H. C. (2011). A literature review of the anchoring
effect. The Journal of Socio-Economics,40(1), 35–42. https://doi.org/
10.1016/j.socec.2010.10.008
Galinsky, A., & Mussweiler, T. (2002). First offers as anchors: The role of
perspective-taking and negotiator focus. Journal of Personality and
Social Psychology,81, 657–669.
Knuth, D. E. (1973). Sorting and searching. In The art of computer program-
ming (Vol. 3, pp. 551–575). Reading, MA: Addison-Wesley.
LeBoeuf, R. A., & Shafir, E. (2006). The long and short of it: Physical
anchoring effects. Journal of Behavioral Decision Making,19(4),
393–406. https://doi.org/10.1002/bdm.535
Litman, J., Hutchins, T., & Russon, R. (2005). Epistemic curiosity, feeling-
of-knowing, and exploratory behaviour. Cognition & Emotion,19(4),
559–582. https://doi.org/10.1080/02699930441000427
Mochon, D., & Frederick, S. (2013). Anchoring in sequential judgments.
Organizational Behavior and Human Decision Processes,122(1), 69–79.
https://doi.org/10.1016/j.obhdp.2013.04.002
TABLE 5 Results for the best-fitting model for Study 4
Effect Estimate Standard error DF tvalue Pr > jtj
Intercept 8.320 0.756 93 11.00 <.001
STORD −2.0264 0.409 275 −4.95 <.001
STORD
2
0.1146 0.048 275 2.43 =.016
Abbreviation: STORD, order of a particular step.
JAIN ET AL.9

Mussweiler, T., & Strack, F. (1999). Hypothesis-consistent testing and
semantic priming in the anchoring paradigm: A selective accessibility
model. Journal of Experimental Social Psychology,35, 136–164. https://
doi.org/10.1006/jesp.1998.1364
Mussweiler, T., & Strack, F. (2000). The use of category and exemplar
knowledge in the solution of anchoring tasks. Journal of Personality and
Social Psychology,78, 1038–1052. https://doi.org/10.1037/0022-
3514.78.6.1038
Mussweiler, T., & Strack, F. (2001a). The semantics of anchoring. Organiza-
tional Behavior and Human Decision Processes,86, 234–255. https://
doi.org/10.1006/obhd.2001.2954
Mussweiler, T., & Strack, F. (2001b). Considering the impossible:
Explaining the effects of implausible anchors. Social Cognition,19,
145–160. https://doi.org/10.1521/soco.19.2.145.20705
Oppenheimer, D. M., LeBoeuf, R. A., & Brewer, N. T. (2008). Anchors
aweigh: A demonstration of cross-modality anchoring and magnitude
priming. Cognition,106(1), 13–26. https://doi.org/10.1016/j.cognition.
2006.12.008
Quattrone, G. A. (1982). Overattribution and unit formation: When behav-
ior engulfs the person. Journal of Personality and Social Psychology,42,
593–607. https://doi.org/10.1037/0022-3514.42.4.593
Rugani, R., Vallortigara, G., Priftis, K., & Regolin, L. (2015). Number-space
mapping in the newborn chick resembles humans’mental number line.
Science,347(6221), 534–536.
Schwiedrzik, C. M., Bernstein, B., & Melloni, L. (2016). Motion along the
mental number line reveals shared representations for numerosity and
space. Elife,5, e10806.
Simmons, J. P., LeBoeuf, R. A., & Nelson, L. D. (2010). The effect of accu-
racy motivation on A&A: Do people adjust from provided anchors?
Journal of Personality and Social Psychology,99(6), 917–932. https://
doi.org/10.1037/a0021540
Strack, F., & Mussweiler, T. (1997). Explaining the enigmatic anchoring
effect: Mechanisms of selective accessibility. Journal of Personality and
Social Psychology,73, 437–446. https://doi.org/10.1037/0022-3514.
73.3.437
Turner, B. M., & Schley, D. R. (2016). The anchor integration model: A
descriptive model of anchoring effects. Cognitive Psychology,90,1–47.
https://doi.org/10.1016/j.cogpsych.2016.07.003
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuris-
tics and biases. Science,185, 1124–1131. https://doi.org/10.1126/
science.185.4157.1124
Walsh, V. (2003). A theory of magnitude: common cortical metrics of time,
space and quantity. Trends in cognitive sciences,7(11), 483–488.
Wansink, B., Kent, R. J., & Hoch, S. J. (1998). An anchoring and adjustment
model of purchase quantity decisions. Journal of Marketing Research,
35,71–81. https://doi.org/10.1177/002224379803500108
AUTHOR BIOGRAPHIES
Dr. Gaurav Jain is an Assistant Professor of Marketing at The Lally
School of Management at Rensselaer Polytechnic Institute (RPI) in
New York. Dr. Jain holds a Ph.D. in Marketing and Consumer-
Behavior from The Tippie College of Business at The University of
Iowa. He also holds a Bachelor's degree in Engineering and an
MBA in Marketing and Economics.
Dr. Dhananjay Nayakankuppam is Professor and DEO of Market-
ing at the Henry B Tippie College of Business at the University of
Iowa. He received his Ph.D. in Marketing at the University of
Michigan in 2000. His areas of expertise include decision making,
evaluative processes, and social cognitive and judgment.
Dr. Gary J Gaeth is Professor of Marketing and Cedar Rapids Area
Business Chair at the Henry B Tippie College of Business at the
University of Iowa. He received his Ph.D. in Experimental Psychol-
ogy from Kansas State University in 1984. His areas of expertise
include consumer decision making.
How to cite this article: Jain G, Nayakankuppam D, Gaeth GJ.
Perceptual anchoring and adjustment. J Behav Dec Making.
2021;1–12. https://doi.org/10.1002/bdm.2231
10 JAIN ET AL.

APPENDIX A
Order of fixations for a typical participant in different anchor conditions.
Panel (a) illustrates the order of fixations for a typical participant when the anchor was G24. The numbers inside the circles represent the
order of the particular fixation, and the size of the circle represents the duration of that particular fixation. Panels (b), (c), and (d) illustrate the same
when the anchors were G19, G15, and G2, respectively. The anchors are marked by arrows in the figure below only for illustrative purposes (the
arrows were not part of the actual stimuli).
JAIN ET AL.11

12 JAIN ET AL.