You are on page 1of 8

Available online at www.sciencedirect.

com

Journal of Experimental Social Psychology 44 (2008) 904–911


www.elsevier.com/locate/jesp

Eye-tracking the cancellation and focus model for preference judgments


Bernadette Sütterlin, Thomas A. Brunner *, Klaus Opwis
Institut für Psychologie, University of Basel, Missionsstrasse 60/62, CH-4055 Basel, Switzerland

Received 8 May 2006; revised 13 July 2007


Available online 5 October 2007

Abstract

By means of a relatively new eye-tracking method that allows for a test situation much closer to reality, we recorded and examined
gaze time and fixation number within the cancellation and focus paradigm, a feature-matching model for preference judgments between
two alternatives. In line with the cancellation and focus model we found that when subjects encountered the second option in each pair,
shared features were canceled out and thus given less consideration whereas unique features were focused on more. We also investigated
the role of feature attractiveness as a second important factor in preference judgments and found a U-shaped relationship between attrac-
tiveness and visual consideration intensity; that is, attractive and unattractive features received more attention than did those of inter-
mediate attractiveness. Finally, we tested the ability of two models, Franklin’s rule and the WReSt (Weighted Recalled Stepwise
Comparing) heuristic, to predict the preference ratings.
Ó 2007 Elsevier Inc. All rights reserved.

Keywords: Feature matching; Direction of comparison; Preference judgments; Feature attractiveness; Franklin’s rule; WReSt heuristic

‘‘Should I take the car to go buy some milk or should I at the impact of feature uniqueness and attractiveness on
cover the 15 min walking distance to the mall by foot?’’ the attribution of attention. Furthermore, we test two mod-
These are the kinds of choice situations that we face in els, Franklin’s rule and the WReSt (Weighted Recalled
everyday life. Usually, the two options in question have Stepwise Comparing) heuristic, in regard to their capability
some features in common (in both cases you have to go of predicting preferences.
and buy some milk) and they offer some unique attributes
(driving by car would be easier but walking to the mall Cancellation and focus
would be healthier). Shared features offer no distinguishing
information and therefore are of no interest in the decision To explain the processes that take place during reaching
process. This means that in order to reach a preference a preference decision between two objects, Houston, Sher-
judgment, attention is primarily drawn to the unique attri- man, and Baker (1989,1991) defined a feature-matching
butes. However, in addition to uniqueness, these features model for preference judgments, the cancellation and focus
also differ in their level of attractiveness. In view of this, model, following the feature-matching model for similarity
we investigate whether the different feature qualities judgments introduced by Tversky (1977). The model con-
(unique/shared, unattractive/attractive) influence the way sists of two components: (a) cancellation of features shared
the attributes are processed. by the two alternatives, and (b) focusing on the remaining
This study is based on the cancellation and focus model, unique features of the option serving as the target for the
a feature-matching model for preference judgments. By comparison. In sequential presentations, features are iden-
means of the eye-tracking technique we take a closer look tified as unique or shared as soon as the second item is
encountered and thus, the second option usually serves as
*
Corresponding author. Fax: +41 61 267 3526. the target (Houston et al., 1989). The material Houston
E-mail address: brunner.thomas@link.ch (T.A. Brunner). et al. (1989) used in order to test their model consisted of

0022-1031/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jesp.2007.09.003
B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911 905

options that either shared all bad and possessed unique direction-of-comparison effect is quite stable and has
good features (unique good pairs), or shared all good been replicated with many different object categories
and possessed unique bad features (unique bad pairs). (e.g., Chernev, 2001; Dhar & Simonson, 1992; Hodges,
We employed a similar technique, with two unique good 1997,1998; Hodges, Bruininks, & Ivy, 2002; Mantel &
pairs (Jobs 1/2 and Jobs 3/4) and two unique bad pairs Kardes, 1999; Slaughter & Highhouse, 2003; see also
(Jobs 1/3 and Jobs 2/4) (Table 1). Brunner & Wänke, 2006).
If two equivalent alternatives are examined one after
another in a sequential rather than in a simultaneous Differences in consideration of unique and shared features
presentation mode (Houston & Sherman, 1995), then
the order in which the objects are encountered influences Based on the assumption that shared features are can-
the preference outcome: People show a general tendency celed and thus exert no influence on the preference judg-
to favor the second option in unique good pairs but the ment, one would expect that this different emphasis
first in unique bad pairs. This direction-of-comparison placed on unique and shared attributes would also be
effect can be explained by the cancellation and focus reflected in a distinct visual consideration intensity. Shared
model. It assumes that in a first step people cancel the features should receive less attention and consequently, less
features shared by both options because of their non- time should be spent gazing at them compared with unique
diagnosticity and in the next stage they focus on the features. Analogous results should be found regarding the
unique features of the second option. In the case of number of fixations (how many times a feature is looked
unique good pairs this means a cancellation of the at). However, these differences in gaze time and fixations
shared bad features and the final judgment is based only between unique and shared features should only occur in
on the unique good features. In comparisons of unique relation to the second option. There should be no difference
bad pairs, in contrast, canceling the shared good features for the first option because all features are perceived as
leads to a choice on the basis of only the unique bad fea- unique. In the second option, however, shared features
tures. The focus is primarily on the second option, and should attract less attention because they are recognized
therefore it is the valence of the unique features of the as common and no longer play a role in the decision-mak-
second option that is crucial in determining preference. ing process and this should be reflected in a shorter gaze
Thus, for unique good pairs focusing on the unique fea- time and fewer fixations. The unique features of the second
tures of the second option should make it appear more object, on the other hand, after having been identified as
attractive and thus lead to it being preferred relative to differing from the first option and thus classified as crucial
the first option; in unique bad pairs the focus on the for the judgment, are focused on more and for a longer per-
unique bad features of the second option makes it seem iod of time.
less appealing and rejection is the consequence. This The varying importance of shared and unique features
for choices has been shown in several ways (for an over-
view see Sherman, Houston, & Eddy, 1999) using method-
ologies such as tagging features wished to see again, listing
Table 1 the most influential features (Weeks & Houston, 1998), or
Unique good and unique bad pairs studied in the cancellation and focus think-aloud procedures (Dhar & Sherman, 1996). A further
paradigm
methodology is the Mouselab, a computer-based process-
Job 1 Job 2 Job 3 Job 4 tracing technique permitting assessment of feature
Good team Interesting Good team Interesting importance in the form of gaze time (Dhar & Nowlis,
Secure position Opportunity for Secure position Opportunity for 1999; Dhar, Nowlis, & Sherman, 1997,1999). Another
development development
method of extracting gaze time and fixation number (one
Benefits Opportunity for Benefits Opportunity for
promotion promotion that is much closer to reality) is the eye-tracking method
Near the Bonus Near the Bonus (for a detailed discussion about the eye-tracking method’s
domicile domicile advantages over the Mouselab, see Lohse & Johnson,
Known Fixed working Known Fixed working 1996). This method has already been applied in studies
company hours company hours
investigating multiattribute binary choice settings (Russo
Unfriendly boss Unfriendly boss Bad working Bad working
atmosphere atmosphere & Dosher, 1983) but to date has never been used to analyze
Competitiveness Competitiveness Inadequate Inadequate the differential treatment of unique and shared features in
installations installations the cancellation and focus model. One advantage of the
Over challenged Over challenged Under Under eye-tracking methodology is that it permits a differentiated
considerable considerable
examination of the first and the second option, an aspect
pressure pressure
Few vacation Few vacation Dress code Dress code that the studies mentioned in the previous section did
No canteen No canteen Long working Long working not take into account. In contrast to other evaluation tech-
hours hours niques such as recall, where assignment of the shared
Note. Jobs 1/2 and 3/4 represent a unique good pair; Jobs 1/3 and 2/4 features to the first and the second options is methodolog-
represent a unique bad pair. ically impossible, the eye-tracking method manages to
906 B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911

record data for each option separately on the basis of gaze ferent importance, then the option with the more impor-
time and fixation. tant feature should be chosen in cases of unique good
pairs whereas it should be rejected in cases of unique bad
Feature attractiveness pairs. If there is no difference in importance between the
two features, the second most important feature in each
Features differ not only in their uniqueness but also in option is considered, and the judgment is reached in the
their levels of attractiveness. Because people tend to look same manner as outlined above. The procedure is repeated
for very attractive features, avoid unattractive ones and in the same manner until a decision is reached.
neglect neutral attributes, feature attractiveness should also The dependent measures, gaze time (how long a fea-
have an impact on preference judgments. Based on recall ture is looked at) and fixation number (how often a fea-
data Brunner, Reimer, and Opwis (2005) were able to pro- ture is looked at), used in this study are of a different
vide evidence for the major role feature attractiveness plays nature to those used by Brunner and Opwis (in press).
in the formation of preferences. They found a U-shaped These authors tested Franklin’s rule and the WReSt heu-
relationship between recall and attractiveness; more attrac- ristic based on people’s memory, i.e. only features that
tive and more unattractive features were better recalled than were recalled by the person were considered. Further-
relatively neutral ones. When considered separately for more, the recalled features were weighted according to
unique and shared features the U-shaped relationship was how important they had been rated by the participants
maintained for unique features but not for shared ones. in a separate part of the experiment. The eye-tracking
However, our use of eye-tracking methodology—where fea- data are a more direct measure because feature impor-
ture importance (induced by the level of attractiveness) is tance, as a derivation from the attractiveness level, is
reflected in gaze time and fixation number—allows an addi- directly reflected in gaze time and fixation number.
tional distinction between the two choice options, over and Hence, a feature that plays a more important role in
above this differentiated examination of unique and shared the decision is looked at for a longer period of time
features. In this study we are thus also to consider sepa- and more frequently than is a less important one. Thus,
rately the impact of attractiveness for both the first and there is no need to identify features crucial for the choice
the second options. by means of recall or to rate their importance. In the
case of Franklin’s rule this means that simply the gaze
Preference prediction time or fixation number of all unique features of each
option have to be added up. If unique good pairs are
So far we have only discussed the different gaze time and presented, the option with the higher sum should be
fixation patterns that we would expect to find depending on selected whereas for unique bad pairs it should be
the uniqueness and attractiveness of the features, but rejected. Applying the WReSt heuristic, the procedure
another crucial consideration is whether it is possible to is even more trivial. In this case only the feature with
predict people’s preference ratings based on these data. the longest gaze time or the highest fixation number of
To learn more, we tested two models in regard to their abil- each option is considered. The option that contains the
ity to predict preferences, Franklin’s rule and the WReSt feature attracting the longer gaze time or a higher num-
heuristic. ber of fixations should be preferred in unique good pairs
In a letter to the scientist Joseph Priestley, Benjamin and rejected in unique bad ones. Considering the various
Franklin (1772/1987) illustrated the classical view of aspects regarding the cancellation and focus model, we
rational judgment, thereafter known as ‘‘Franklin’s rule’’ postulate three hypotheses:
(see also Gigerenzer, Todd, & The ABC Research Group,
1999). Applied to the cancellation and focus paradigm, H1: There is a significant difference in gaze time and fixa-
Franklin’s rule assumes that, in order to make a preference tion number for shared and unique features. Unique
judgment, people take into account all available features of features receive more attention than shared ones.
an option, weigh each one according to its importance, add However, this discrepancy only occurs in the second
them up, and finally, comparing the summations of the two option. In short, an interaction between option and
options, come to a decision. Of course, shared features feature uniqueness (unique/shared) is to be expected.
receive equal weight and therefore should be excluded from H2: A U-shaped relationship exists between feature
summation. attractiveness and the two consideration measures,
Based on the assumption made by other decision-mak- gaze time and fixation number. Attractive and unat-
ing paradigms that people do not integrate all the available tractive features should be looked at for a longer per-
information but just focus on the main aspects (e.g. Tver- iod of time and more often than neutral ones.
sky, 1972; Payne & Bettman, 2001; for an overview see H3: The two models based on feature-importance, Frank-
Gigerenzer, 2004), Brunner and Opwis (in press) proposed lin’s rule and the WReSt heuristic, manage to predict
a non-compensatory heuristic, the WReSt heuristic. participants’ preference judgments. The WReSt heu-
According to this, people only look for the most important ristic possesses about the same predictive power as
unique feature in each option. If the two features are of dif- Franklin’s rule.
B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911 907

Methods as wearing an unnatural helmet or limitation of head move-


ment. Thus, the experimental situation is much closer to a
Participants real choice situation.
After an initial calibration the participants saw the fol-
Forty students of the University of Basel (24 women and lowing instructions on the eye-tracker screen: ‘‘Imagine
16 men, mean age M = 24.65 years, SD = 5.25) partici- you just finished school and are now looking for a job.
pated in this study. They received either course credit or But at the moment the unemployment rate is quite high.
a chocolate bar for their participation. The eye-tracking Which one of the following two jobs would you prefer?
data of two students could not be evaluated because they You have 30 s to look at each of the following two jobs.’’
were cross-eyed, leaving 38 participants for the analyses The description of the choice context was added to
based on gaze time and fixation number. increase the importance of the decision. After participants
Stimulus materials. A pilot study was conducted in had read through the instructions, the two options were
order to assess the attractiveness of the different job fea- sequentially presented on the screen, each one for 30 s.
tures. In exchange for course credit 24 students rated Pairs and their valence as well as options were counterbal-
the features in regard to their attractiveness on an 8-point anced between the test subjects. Each participant was pre-
scale (from very unattractive to very attractive). By means sented with one choice pair, either unique good or unique
of the pretested features two comparable sets of five good bad. Having studied the two options, participants were
features and two comparable sets of five bad features were asked to indicate their preference on an 8-point scale
constructed as shown in Table 1. In composing the sets ranging from one, meaning ‘‘I strongly prefer the first
care was taken that the final description consisted of fea- option’’, to eight, representing ‘‘I strongly prefer the sec-
tures from the whole range of the attractiveness scale, ond option.’’ Finally, after the eye tracking, participants
thus features varying from very attractive to neutral to were given a list containing all the features they had seen,
very unattractive were selected (attractiveness ranged and independent of the previous preference judgment they
from 1.50 to 7.58). Furthermore, in order to avoid an had to judge each feature in regard to its attractiveness on
advantage of recall depending on word length, all features an 8-point scale from ‘‘very unattractive’’ to ‘‘very
consisted of an equal number of syllables (in the German attractive.’’
language). The final options were constructed by combin-
ing each of the two good feature sets with each of the two Results
bad feature sets. In this way, four job options were cre-
ated each consisting of five good and five bad features. No significant effects were found for the different quali-
The order of the features listed in the experiment was ties (good/bad) of the features mentioned first in the pairs
determined randomly with the restriction that no more and in the diverse pair versions (the two unique good and
than two features of the same quality (good or bad) fol- the two unique bad pairs) and thus, preference scores were
lowed in succession. For each option two different ver- collapsed across these factors.
sions existed; the option was described once beginning
with a good feature and once beginning with a bad one. Preference
Accordingly, the options of the same choice pair began
once with a positive feature and once with a negative fea- According to the cancellation and focus model, a higher
ture. Furthermore, the two versions of an option differed score on preference is to be expected for unique good pairs
in the order in which the features were listed, in that no than for unique bad pairs. By means of an independent
feature was presented in the same position. Using this samples t-test the preference scores of unique good and
procedure, anchoring effects based on the quality of the unique bad pairs were compared with each other. As pre-
first encountered feature and sequence effects were con- dicted, the mean preference score for unique good pairs,
trolled to some extent. Pairing each option with the two M = 5.16 (SD = 1.54), was significantly higher than the
other options with which it had half of its features in mean for unique bad pairs, M = 4.14 (SD = 1.32),
common, two pairs with shared good features (unique t(38) = 2.25, p < .05.
bad pairs) and two pairs with shared bad features (unique
good pairs) were constructed. By composing the options Differences in consideration of unique and shared features
in this way every feature appeared as a unique feature
in some options and as a shared one in others. In order to test the different emphases placed on unique
and shared features in the second option compared to the
Procedure and design first one, a two-factorial ANOVA with the two within-sub-
jects factors option and feature uniqueness was conducted.
The experiment was conducted using a Tobii 1750 eye- For this purpose each of the twenty features, differentiated
tracker. This relatively new eye-tracking system allows according to the four possible configurations (first vs. sec-
the participant to be in a completely natural environment ond option and unique vs. shared), was averaged across
during the experiment because there are no restraints such participants resulting in four measures per feature.
908 B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911

Gaze time. As expected, we found a significant interac- attractiveness and the two consideration measures were
tion between option and uniqueness, F(1, 37) = 26.13, averaged across subjects.
p < .001. Participants spent more time looking at the Gaze time. As predicted, a quadratic relationship existed
unique features than at the shared ones in the second between attractiveness and gaze time, which explained with
option whereas for the first option no such pattern was DR2 = .187 a significant part of the variance,
detected. There was also a main effect for uniqueness, F(1, 16) = 6.62, p < .05. As far as uniqueness was con-
which was due to the difference between unique and shared cerned, the U-shaped relationship did not reach signifi-
features in the second option [F(1, 37) = 4.32, p < .05]. As cance, neither with DR2 = .126 for unique features,
expected due to the equal presentation time, the main effect F(1, 16) = 3.40, p = .084, nor with DR2 = .108 for shared
for option did not reach significance (F < 1). Broken down ones, F(1, 16) = 2.51, p = .133. This leads to the assump-
in paired-samples t-tests all single comparison effects corre- tion that there must be a difference between the two options
sponded with our predictions (see Table 2). regarding this aspect. In fact, the quadratic relationship
Fixation number. The interaction between option and between attractiveness and gaze time held true for the sec-
uniqueness was significant with F(1, 37) = 27.52, p < .001: ond option with an explained variance of DR2 = .352,
Unique features of the second option were focused on more F(1, 16) = 11.33, p < .01, but not for the first one,
often than shared ones whereas the first option did not F(1, 16) = .01, p = .914, DR2 = .000. Broken down even
show this difference. The main effect for uniqueness was further, the U-shaped relationship stayed significant for
again significant [F(1, 37) = 5.73, p < .05], whereas that the unique features, F(1, 16) = 6.03, p < .05, DR2 = .213,
for option did not reach significance [F(1, 37) = 2.49, and the shared features, F(1, 16) = 6.24, p < .05,
p = .123]. Paired-samples t-tests were again in line with DR2 = .270, of the second option but neither for unique,
the assumptions (see Table 2). p = .721, nor for shared ones, p = .531, of the first option.
Fixation number. A U-shaped relationship relating to
Feature attractiveness feature attractiveness was also found for fixation number,
F(1, 16) = 5.78, p < .05, DR2 = .187. Contrary to gaze time,
Attractiveness data were examined by conducting a hier- the quadratic relationship for unique features reached sig-
archical regression analysis with gaze time or fixation num- nificance, F(1, 16) = 7.10, p < .05, DR2 = .276. For shared
ber as the dependent variable. Gaze time and fixation features, once again no significant results emerged,
number of a feature depend to some extent on the position, F(1, 16) = 1.34, p = .263, DR2 = .054. As before, a
so that features listed first are considered more intensely U-shaped relationship between attractiveness and fixation
than the subsequent features. In fact, we found significant number only occurred in the second option,
correlations between feature position and gaze time F(1, 16) = 10.49, p < .01, DR2 = .336, but not in the first
(r(20) = .549, p < .05) as well as between feature position one, F(1, 16) = .06, p = .809, DR2 = .003. Differentiating
and fixation number (r(20) = .541, p < .05). In order to between unique and shared features for each option, again
control this confounding variable, feature position was the quadratic relationship stayed significant just for
entered as a dummy variable into the regression model in unique, F(1, 16) = 8.00, p < .05, DR2 = .303, and for shared
the first step. The linear aspect of attractiveness was added features, F(1, 16) = 4.97, p < .05, DR2 = .207, of the second
second and as a final predictor the quadratic aspect of fea- option encountered, but not for the unique (p = .544) and
ture attractiveness was entered. For each feature position, for shared features (p = .399) of the first option.

Preference prediction
Table 2
Eye-tracking data for gaze time and fixation number differentiated
according to option (vertical comparison) and feature uniqueness (hori- In order to test whether gaze time and fixation number
zontal comparison) could explain additional variance in preference prediction
Option Uniqueness Significance besides the variance explained by feature valence, a gaze
time and a fixation number index were computed in the
Unique Shared
same way that Sanbonmatsu, Kardes, and Gibson (1991)
Gaze time Option 1 2232.85 2409.58 t(37) = 1.69,
created their recall index. That is, in order to test Franklin’s
(506.10) (497.94) p = .100
Option 2 2602.28 2108.63 t(37) = 5.13, rule, the gaze time of unique features of the first option was
(457.39) (471.23) p < .001 subtracted from the gaze time of unique features of the sec-
Significance t(37) = 5.26, t(37) = 3.86, ond option. In an analogous manner the fixation number
p < .001 p < .001 index was computed. For the WReSt heuristic the two indi-
Fixation Option 1 6.03 (1.30) 6.39 (1.10) t(37) = 1.62,
ces were computed simply by subtracting the highest score
number p = .113
Option 2 7.13 (1.45) 5.81 (1.14) t(37) = 4.63, (gaze time or fixation number) of the first option from the
p < .001 highest score of the second option. Higher scores on these
Significance t(37) = 4.83, t(37) = 2.60, indices refer to greater consideration for the unique features
p < .001 p < .05 of the second option. Accordingly, for unique good pairs
Note. Gaze time is indicated in milliseconds per feature. the higher the score the more the second option should be
B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911 909

preferred (a positive correlation) and for unique bad pairs 8


the higher the score the more the second option should be
rejected and the first one should be preferred (a negative 7
correlation). In short, we expected an interaction between
the valence of the unique features and the gaze time index 6
as well as the fixation number index.
A hierarchical regression analysis was conducted with 5
preference judgments as the dependent variable. In a first
step the valence of the unique features was entered into 4
the regression model. Second, the computed index (for gaze
time or fixation number) was added, and finally in the third 3
step the interaction of valence and index was entered.
Franklin’s rule. Testing the prediction capability of 2
unique good
Franklin’s rule in regard to gaze time, we found no signif- unique bad
1
icant interaction between valence and gaze time index,
-7 -5 -3 -1 1 3 5 7 9
F(1, 34) = .86, p = .361, DR2 = .022. Taking a separate
look at the correlations, the results headed in the expected Difference of fixation number
direction but did not reach significance; we found a slightly Fig. 2. Preference prediction according to WReSt heuristic based on
positive tendency for unique good pairs, r(19) = .16, difference in fixation number between the most considered feature of the
p = .254 (one tailed), and a slightly negative one for unique first and second options and valence of unique features. The higher the
difference in fixation number the more often the unique feature of the
bad pairs, r(19) = .15, p = .268 (one tailed).
second option is looked at relative to the first option. The preference
Quite different results emerged in the evaluations that indicated by participants ranges from 1, ‘‘I strongly prefer the first
included the fixation number. Now the interaction was sig- option,’’ to 8, ‘‘I strongly prefer the second option.’’
nificant, F(1, 34) = 6.62, p < .05, DR2 = .145 (Fig. 1).
Although valence alone, i.e. the cancellation and focus
model, explained 11% of the variance, Franklin’s rule WReSt heuristic. Analyzing the ability of the WReSt
added another 15% resulting in 26% explained variance heuristic to predict preferences on the basis of the gaze time
for the complete model. Compared with the gaze time data measures, as in the case of Franklin’s rule, no significant
the correlation for unique good pairs was stronger and interaction between valence and gaze time index emerged,
became significant, r(19) = .44, p < .05 (one tailed). Unique F(1, 34) = .79, p = .380, DR2 = .020. Neither the correla-
bad pairs yielded a marginally significant result, tion of unique good pairs, r(19) = .03, p = .447 (one tailed),
r(19) = .35, p = .072 (one tailed). nor that of unique bad pairs, r(19) = .25, p = .148 (one
tailed), was significant, but the tendencies in the right direc-
tion could still be retrieved.
8 As before, the hierarchical regression analysis based on
fixation number yielded better results. The interaction
7 between valence and fixation number index reached signif-
icance, F(1, 32) = 10.97, p < .01, DR2 = .2301 (Fig. 2). The
6 complete model explained 33% of the variance in the data,
which is more than the 26% of Franklin’s rule, despite the
5 fact that the WReSt heuristic works with considerably less
information regarding the features. Separate correlation
4 analyses showed a marginally significant positive correla-
tion for unique good pairs, r(18) = .38, p = .063 (one
3 tailed), and a significant negative correlation for unique
bad pairs, r(18) = .64, p = .01 (one tailed).
2
unique good
unique bad
1 General discussion
-2 -1 0 1 2 3 4 5
Difference of fixation number The present study was able to replicate the direction-of-
comparison effect within the cancellation and focus
Fig. 1. Preference prediction according to Franklin’s rule based on paradigm. Furthermore, by means of the eye-tracking
difference in fixation number between first and second options, and
valence of unique features. The higher the difference in fixation number
1
the more often the unique features of the second option are looked at For this analysis the data of two participants had to be excluded
relative to the first option. The preference indicated by participants ranges because they were identified as outliers according to the mahalanobis
from 1, ‘‘I strongly prefer the first option,’’ to 8, ‘‘I strongly prefer the distance of preference and index. The results including the two outliers did
second option.’’ also reach significance but the effect was weaker.
910 B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911

methodology we were able to reveal some of the processes more confounding variables. Therefore, we conclude that
taking place during decision-making. In fact, the assump- fixation number and gaze time cannot be treated as analo-
tions of the cancellation and focus model were found to gous measures.
be directly reflected in people’s gaze time and fixation num- Using the eye-tracking methodology allowed us to
ber. Thus, the eye-tracking data provide further support investigate the processes taking place during decision-mak-
for the assumptions of the cancellation and focus para- ing, free from external influences. Furthermore, we were
digm, in that they reflect exactly its predictions. able to take a separate look at the two options in an objec-
Furthermore we examined another important factor for tive manner and thus found interesting results concerning
preference judgments, the level of attractiveness, in regard the processing of the first option. However, several ques-
to its influence on consideration intensity. As seen in gaze tions remain open for future studies such as neglecting
time and fixation number data, people pay more attention the level of attractiveness in the processing of the first
to very unattractive and very attractive features than they option, the difference between fixation number and gaze
do to neutral ones. However, using the eye-tracking meth- time in regard to preference prediction, and the generaliza-
odology, we were also able to consider the two options sep- tion of the eye-tracking data to other object categories.
arately, and an interesting result emerged. The quadratic
relationship between attractiveness and people’s consider- Acknowledgment
ation intensity only persisted for the second option and
not for the first one. This means that features more crucial We thank Laura Wiles for helpful suggestions regarding
for the evaluation of the object are not automatically the English.
focused on more intensely when an option is encountered
but only when an alternative is available. Thus, similar to References
unique features, very attractive and very unattractive fea-
tures only attract special attention when the second option Brunner, T. & Opwis, K. (in press). The WReSt heuristic: The role of
is present and the comparison process has begun. At this recall as well as feature-importance in and beyond the cancellation and
focus model. Social Cognition.
point their impact on decision-making is highlighted and Brunner, T., Reimer, T., & Opwis, K. (2005). Cancellation and focus: The
they become important. This result is interesting because impact of feature attractiveness on recall. In K. Opwis & I. K. Penner
it shows that the importance of feature attractiveness is (Eds.), Proceedings of KogWis05. The German Cognitive Science
somewhat restricted to the second option attaching even Conference 2005 (pp. 27–32). Basel, Switzerland: Schwabe.
Brunner, T., & Wänke, M. (2006). The reduced and enhanced impact of
more value to the focus component. Further studies pro-
shared features on individual brand evaluations. Journal of Consumer
viding a single object without the context of a choice Psychology, 16, 101–111.
should be conducted to assess whether this neglect of the Chernev, A. (2001). The impact of common features on consumer
feature attractiveness level of an option is a general phe- preferences: A case of confirmatory reasoning. Journal of Consumer
nomenon or whether it is restricted to situations where Research, 27, 475–488.
an additional comparison alternative is expected. Dhar, R., & Nowlis, S. M. (1999). The effect of time pressure on consumer
choice deferral. Journal of Consumer Research, 25, 369–384.
Finally, two models, Franklin’s rule and the WReSt Dhar, R., Nowlis, S.M., Sherman, S.J. (1997). Similarity effects on choice.
heuristic, incorporating the decisive influence of unique Paper presented at Colorado Decision-making Camp.
feature valence (as predicted by the cancellation and focus Dhar, R., Nowlis, S. M., & Sherman, S. J. (1999). Comparison effects on
model) as well as the attractiveness factor (in the sense of preference construction. Journal of Consumer Research, 26, 293–306.
importance for the decision), were tested on the basis of Dhar, R., & Sherman, S. J. (1996). The effect of common and unique features
in consumer choice. Journal of Consumer Research, 23, 193–203.
the eye-tracking data in regard to their ability to predict Dhar, R., & Simonson, I. (1992). The effect of the focus of comparison on
preferences. Franklin’s rule and the WReSt heuristic man- consumer preferences. Journal of Marketing Research, 29, 430–440.
aged to predict—with 26% and 33%, respectively—people’s Gigerenzer, G. (2004). Fast and frugal heuristics: The tools of bounded
preference judgments to a somewhat larger extent than did rationality. In D. Koehler & N. Harvey (Eds.), Blackwell Handbook of
the cancellation and focus model, which explained only Judgment and Decision Making (pp. 62–88). Oxford, UK: Blackwell.
Gigerenzer, G., & Todd, P.The ABC Research Group. (1999). Simple
11% of the variance. Surprisingly, the WReSt heuristic heuristics that make us smart. New York: Oxford University Press.
explained more variance than Franklin’s rule even though Hodges, S. D. (1997). When matching up features messes up decisions:
the WReSt heuristic works with an extremely reduced The role of feature matching in successive choices. Journal of
amount of information. Thus, the advantage is clearly on Personality and Social Psychology, 72, 1310–1321.
the side of the more economical WReSt heuristic, which Hodges, S. D. (1998). Reasons for the referent: Reducing direction of
comparison effects. Social Cognition, 16, 367–390.
is simply based on the two most important features and Hodges, S. D., Bruininks, P., & Ivy, L. (2002). It’s different when I do it:
therefore requires less cognitive capacity. Feature matching in self-other comparisons. Personality and Social
However, there is an important caveat regarding the Psychology Bulletin, 28, 40–53.
ability to predict the preferences of Franklin’s rule and Houston, D. A., & Sherman, S. J. (1995). Cancellation and focus: The role
of shared and unique features in the choice process. Journal of
the WReSt heuristic: The significant results are restricted
Experimental Social Psychology, 31, 357–378.
to the evaluations based on fixation number. The perceived Houston, D. A., Sherman, S. J., & Baker, S. M. (1989). The influence of
importance of unique features seems to be rather reflected unique features and direction of comparison on preferences. Journal of
in fixation quantity whereas gaze time incorporates other Experimental Social Psychology, 25, 121–141.
B. Sütterlin et al. / Journal of Experimental Social Psychology 44 (2008) 904–911 911

Houston, D. A., Sherman, S. J., & Baker, S. M. (1991). Feature matching, Sanbonmatsu, D. M., Kardes, F. R., & Gibson, B. D. (1991). The role
unique features, and the dynamics of the choice process: Predecision of attribute knowledge and overall evaluations in comparative
conflict and postdecision satisfaction. Journal of Experimental Social judgment. Organizational Behavior and Human Decision Processes,
Psychology, 27, 411–430. 48, 131–146.
Lohse, G. L., & Johnson, E. J. (1996). A comparison of two process Sherman, S. J., Houston, D. A., & Eddy, D. (1999). Cancellation and
tracing methods for choice tasks. Organizational Behavior and Human focus: A feature-matching model of choice. European Review of Social
Decision Processes, 68, 28–43. Psychology, 10, 169–197.
Mantel, S. P., & Kardes, F. R. (1999). The role of direction of comparison, Slaughter, J. E., & Highhouse, S. (2003). Does matching up features mess
attribute-based processing, and attitude-based processing in consumer up job choice? Boundary conditions on attribute-salience effects.
preference. Journal of Consumer Research, 25, 335–352. Journal of Behavioral Decision Making, 16, 1–15.
Payne, J. W., & Bettman, J. R. (2001). Preferential choice and adaptive Tversky, A. (1972). Elimination by aspects: A theory of choice. Psycho-
strategy use. In G. Gigerenzer & R. Selten (Eds.), Bounded Rationality. logical Review, 79, 281–299.
The Adaptive Toolbox (pp. 123–145). Cambridge, Massachusetts: MIT Tversky, A. (1977). Features of similarity. Psychological Review, 84,
Press. 327–352.
Russo, J. E., & Dosher, B. A. (1983). Strategies for multiattribute binary Weeks, M., & Houston, D. A. (1998). Feature screening in the cancellation-
choice. Journal of Experimental Psychology: Learning, Memory and and-focus model of choice. Paper Presented at the Midwestern Psycho-
Cognition, 9, 676–696. logical Association Seventieth Annual Meeting, Chicago, IL (May).

You might also like