Journal of Economic Behavior & Organization
Vol. 39 (1999) 41–58
Explaining how preferences change across joint versus
Max H. Bazermana, Don A. Moorea,∗, Ann E. Tenbrunsel b,
Kimberly A. Wade-Benzonia, Sally Blountc
aJ. L. Kellogg Graduate School of Management, Northwestern University, 2001 Sheridan Road, Evanston
60208, IL, USA
bCollege of Business Administration, Notre Dame University, South Bend IN, USA
cGraduate School of Business, University of Chicago, Chicago IL, USA
Received 21 August 1997; received in revised form 14 July 1998; accepted 20 July 1998
Thispaperexamineshowpreferences foroutcomes changeacross jointversus separate evaluation
ofalternatives.Injoint evaluation,two (or more)options are presentedand evaluatedsimultaneously.
In separate evaluation, each option is presented and evaluated separately. We review a growing body
of evidence demonstrating this type of preference shift and discuss how it is different from existing
biases and preference reversals documented in the literature. We then review and integrate three
competing explanations for this type of preferential inconsistency. ©1999 Elsevier Science B.V.
All rights reserved.
JEL classiﬁcation: Experimental; Theoretical treatment ; Organizational behavior 
Keywords: Decision making; Preference reversals; Preferences
A central tenet within economics is that the individual acts to maximize expected util-
ity (von Neumann and Morganstern, 1947). An on-going aim within cognitive psychol-
ogy is to study and document how utility is actually constructed in decision making and
bargaining. Consider, for example, research on prospect theory (Kahneman and Tversky,
1979) and preference reversals (Tversky et al., 1990), along with numerous other studies
that have shown that subtle differences in elicitation can have a profound effect on stated
preferences (e.g. Slovic and Macphillamy, 1974; Slovic et al., 1990; Hershey et al., 1982).
0167-2681/99/$ – see front matter ©1999 Elsevier Science B.V. All rights reserved.
42 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
These differences can change people from being risk-averse to risk-seeking (Kahneman and
Tversky, 1979), from positive time discounting to negative time discounting (Loewenstein
andPrelec, 1993),and reversingtheirpreferences across anarray ofmulti-attribute prospects
(Tversky et al., 1988).
This paper reviews recent psychological ﬁndings regarding a new elicitation effect,
namely, preference shifts between joint evaluation and separate evaluation. Joint evalu-
ation, as we use the term, refers to situations in which two concrete options are evaluated
simultaneously, while separate evaluation refers to situations in which the options are pre-
sented and evaluated one option at a time (Bazerman et al., 1992; Hsee et al., 1997).
The original demonstration of changes in preference across joint versus separate evalua-
tion was provided by Bazerman et al. (1992), and focused on the choice between absolute
and relative payoffs. Participants read a description of a dispute between two neighbors and
then evaluated different potential resolutions of the dispute. The dispute involved splitting
either sales revenue or a tax liability associated with the ownership of a vacant lot between
the neighbors’ houses. Participants were asked to take the perspective of one homeowner
and to evaluate ﬁve possible settlements associated with receiving revenue and ﬁve possible
settlementsassociated with payingtaxes. Each settlement was expressedin terms ofa payoff
(or liability) to oneself and a payoff (or liability) to the neighbor. Across outcomes, the au-
thorsvaried boththe absolute payoffto oneself and whether the neighbor wouldbe receiving
thesame as or morethan the respondent.As an example, consider the followingtwo options:
Money Option: $600 for self and $800 for neighbor
Fair Option: $500 for self and $500 for neighbor
Note: we have labeled the options in the text (e.g. money and fair). Participants did not see
In joint evaluation, participants were presented with pairs of options, such as the one
listed above and were asked to indicate which of the two options was more acceptable or
more satisfying. In separate evaluation,1participants were presented with these options
one at a time and were asked to indicate on a rating scale how acceptable or how satisfying
each option was. In joint evaluation, 75 percent of the participants preferred the Money
Option. In separate evaluation, 71 percent preferred the Fair Option.
Many other demonstrations of preference shifts across joint versus separate evaluation
have been documented. We review these in the next section. Following our review of recent
empirical evidence, we present three competing explanations for these results, which have
been offered by psychologists. In our ﬁnal section, we develop an integrated framework for
conceptualizing preference shifts across joint versus separate evaluation.
Our goal in presenting this evidence and integration to an economics audience is to
contribute to a dialog with those who model decision making and game theory. Our ﬁndings
suggest structural elicitation effects, which emanate from the way that related decisions are
1In this study, we employed a within-subjects design—the same subjects assessed the different options one-at-
a-time in the separate condition. In other studies reported in this paper, different subjects were used for the two
different separate decisions. For example, participants in the Blount and Bazerman (1996) study that follows only
made one choice each. Since our main purpose in this paper is to provide a conceptual integration, we will not
provide a methodological discussion of the merits of different procedures.
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 43
temporally sequenced and/or combined. While prior work has shown that some of these
shiftsin preferencescan be predictedby mathematicalmodels of individualdecision making
(see e.g. appendix of Bazerman et al., 1992), not all of the evidence is adequately explained
by such an approach. The fact that the temporal sequencing of decisions may inconsistently
change preferences across outcomes creates several questions for economists. We discuss
these questions further in our integration section.
2. The evidence
Building off of the work of Bazerman et al. (1992) and Bazerman et al. (1994) sought
to demonstrate joint versus separate preference shifts in a context that was more familiar
to participants. In their study, MBA students were presented with a series of potential job
offers and asked whether they would accept the offers. Similar to the Bazerman et al. (1992)
study, in joint evaluation, participants were presented with a pair of hypothetical job offers,
and asked which offer, if either, they would accept. In separate evaluation, participants were
presented with offers separately and asked on a case-by-case basis whether they would
accept each offer. Typical offers included:
Percent selecting job, by evaluation mode2
Money Job: $85,000 for self and
$95,000 for other MBAs 55% 64%
Fair Job: $75,000 for self and
$75,000 for other MBAs 69% 13%
In joint evaluation, participants tended to select the Money Job. In separate evaluation,
however, participants preferred the Fair Job. As predicted, the disparity in compensation
and the resulting social comparison effects were more powerful when jobs were evaluated
separately and the actual levels of compensation were more powerful when participants
could compare job offers.
The power of social comparison effects was demonstrated again by Blount and Baz-
erman (1996), in an experiment in which subjects were recruited for a colleague’s study
and measured rates of volunteering. Participants in the joint evaluation condition had the
option of participating in the Fair Study that included equal payments of $7 for all subjects
or the Money Study that paid $8 and paid other subjects $10. Participants in the separate
evaluation condition were only presented with one of the studies (either the Fair Study or
the Money Study) and were asked whether or not they would be interested in participating
in that particular experiment.
2Note that percents will not add to 100. In the joint condition, 64 percent of participants chose the Money job,
13 percent chose the Fair Job, and 23 percent chose to accept neither job. Also, in the separate condition, where
participants reviewed each job on its own and reported whether they would accept it, 12 percent chose the Money
job only, 26 percent chose the Fair job only, 43 percent chose to accept both jobs, and 19 percent chose to accept
44 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
Percent choosing to participate, by evaluation mode
Money Study: $8 for self and
$10 for others 54% 56%
Fair Study: $7 for self
$7 for others 72% 16%
The results were consistent with prior evidence (Bazerman et al., 1992, 1994): the partici-
pation rate was higher for the Fair Study than the Money Study in separate evaluation, but
this pattern was reversed in joint evaluation.
Bazermanet al. (1994) extendedthese ﬁndings beyondthe scope of comparisonsbetween
absolute and relative payoffs. Here, MBA students were again asked to report their interest
in accepting various job offers. However, in this experiment, participants were presented
with jobs that varied in salary and the extent to which the organizational culture was de-
scribed as democratic or hierarchical. Again there were two unlabeled offers in each pair.
We label these here the Fair Culture Job and the Money Job. The Fair Culture Job offered a
corporate culture in which the junior staff had a voice in their assignments and in important
organizational decisions. At the Money Job, junior hires were offered a higher salary but
were not encouraged to participate in important decisions and were assigned to jobs and
projects by senior management.
Percent selecting job, by evaluation mode
Money Job: $75,000 salary/
hierarchical culture 33% 55%
Fair Culture Job: $60,000 salary/
democratic culture 48% 41%
The results, like those shown here, found that culture was weighted more heavily in separate
than in joint evaluation. Separately, more participants said that they would take the Fair
Culture Job but, in joint evaluation, the Money Job was preferred.
In another job context study, Hsee (1994) asked his undergraduate participants to report
how likely they would be to accept different 4-year jobs after graduating from college. The
following two job offers were presented:
Money Option: $27K $26K $25K $24K
Increasing Pay Option: $23K $24K $25K $26K
In separate evaluations, participants were asked to report how likely it would be that they
would accept each of the offers. Here, the Increasing Pay Option received more favorable
ratings. However, in the joint condition, participants tended to favor the Money Option.
Shifts in preference across joint and separate evaluation have been extended to a number
of other decisions and stimulus items. A number of these studies have asked participants to
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 45
evaluate consumer products. Hsee (1996, 1998), for example, asked participants in both the
separate and joint evaluation conditions to report their willingness-to-pay (WTP) price for
various products. By holding constant the nature of the responses, this measure addresses
thecriticism that the nature of thequestion in previous researchwas different inthe joint and
separate evaluation conditions. In the separate evaluation condition, participants reported
WTP for each of the two alternatives. In the joint evaluation condition, participants reported
WTP for the two alternatives presented simultaneously.
In Hsee’s (1998) third experiment, participants were asked to report WTP for two dif-
ferent servings of ice cream. Participants received drawings of the ice cream servings with
the sizes of the servings and containers marked on the drawings. One alternative, which
we label here the Overﬁlled Portion, was depicted pictorially in a small 5 ounce container
which was overﬁlled, almost overﬂowing, with 7 ounces of ice cream. The other alternative,
whichwe label the Large Portion, offered8 ounces of ice cream,but it was depicted asbeing
served in a 10 ounce container that was not ﬁlled to capacity. Hsee found a clear shift in
preferences across joint versus separate evaluation:
Willingness-to-pay for servings of ice
cream, by evaluation mode
Large Portion: 8 ounces of ice cream in a
10 ounce container $1.80 $2.02
Overﬁlled Portion: 7 ounces of ice cream in a
5 ounce container $2.23 $1.81
Similarly, in Hsee’s (1998) fourth experiment, participants were asked to evaluate two sets
of dinnerware for sale in a store that was having a clearance sale. Participants reported their
WTP for the following sets, either separately or together:
Large Set Unbroken Set
includes 40 PCs includes 24 PCs
dinner plates: 8, in good condition 8, in good condition
soup/salad bowls: 8, in good condition 8, in good condition
dessert plates: 8, in good condition 8, in good condition
cups: 8, 2 of them are broken
saucers: 8, 7 of them are broken
Note that the Large Set includes all the items of the Unbroken Set, plus more. Nevertheless,
WTP reports indicated a shift: Willingness-to-pay for dinnerware,
by evaluation mode
Large Set: 31 pieces plus 9 broken pieces $23.25 $32.03
Unbroken Set: 24 pieces $32.69 $29.70
Forboth ofthese experiments,Hsee (1998) arguedthat inseparate evaluation,the dominated
alternative was rated lower because it fell short of the reference standard that it evoked. For
46 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
the ice cream, the underﬁlled container was judged against the same container, overﬁlled.
For the dinnerware, the large set was judged against a complete, unbroken large set.
Hsee (1996) found similar reversals in WTP for music dictionaries. Participants were
asked to imagine that they were in the market for a music dictionary. The options presented
and reported WTP are as follows: Willingness-to-pay for music dictionaries,
by evaluation mode
Large Dictionary: 20,000 entries
and a torn cover $20 $27
Intact Dictionary: 10,000 entries
and an intact cover $24 $19
Theresults areconsistent withthe twostudies from Hsee(1998) discussedabove.In separate
evaluation, the torn cover is overweighted. This weighting reverses itself in joint evaluation,
where the number of entries becomes more important.
Nowlis and Simonson (1997) also asked participants to rate various consumer prod-
ucts, such as toasters, microwave ovens, and batteries. They found that brand names were
weighted heavily in separate evaluation, but that price became more important when partic-
ipants could compare different products. They found a signiﬁcant shift in reported intention
to purchase for the following pair of products:
Percent preferring each, by evaluation mode
Cheap Phone: Cobra brand/$49 price tag 26% 42%
Quality Phone: Sony brand/$69 price tag 74% 58%
Nowlis and Simonson (1997) were also able to demonstrate a shift in preference between
price and number of features, such that reported intention to purchase was driven by the
number of features in the separate evaluation condition, but that price was more important
in the joint evaluation condition.
Irwin et al. (1993) demonstrated strong shifts in preferences between consumer products
andenvironmental improvements.They mailed surveysto residents inthe Denver,Colorado
metro area. Participants in the separate evaluation condition were asked to reportWTP for
improvementsin Denverair quality and a newVCR. Participants in the joint evaluation con-
dition were asked to choose between air quality improvements and the VCR by identifying
the one for which they would be willing to pay more. The results were as follows:
Percent willing to pay more,
by evaluation mode
Better Environment: Improved air quality 22% 58%
Better Possessions: Get a new VCR 70% 21%
It is interesting to note that unlike other experiments where participants in the separate
condition only rate one of the two alternatives, participants in this study all saw the same
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 47
set of stimulus items. That is, participants in the separate condition rated their WTP for air
quality and the VCR. Despite being presented with the same option set, individuals in the
separate condition indicated a preference for better possessions, and individuals in the joint
condition indicated a preference for a better environment.
In a similar vein, Kahneman and Ritov (1994) were able to show preferential inconsisten-
cies for different types of environmental or social issues. They presented participants with
various headlines that highlighted speciﬁc problems, and asked the participants to either
report their support for government intervention for such a cause (separate condition), or
choose between causes by indicating the one they would support more (joint condition).
The results of one were as follows:
Percent reporting greater support,
by evaluation mode
Human Issue: Skin cancer in farm workers 29% 70%
Attractive Issue: Australian mammal species 32% 30%
In separate evaluations, people leaned toward ‘attractive’ causes such as spotted owls, coral
reefs, and toxic spills. When choosing between causes, however, participants tended to
prefer human causes such as skin cancer, multiple myeloma, and lead-based paint.
Hsee (1996) also found reversals in WTP when he presented participants with a hiring
decision. Each participant was asked to assume that he or she was the owner of a consulting
ﬁrm and was looking for a computer programmer. Participants were asked how much they
wouldbe willing topay to hirea programmer towrite programs in anew computerlanguage.
Like Hsee’s other studies using WTP (Hsee, 1998), this experiment held response mode
constant by having participants in both joint and separate evaluation conditions report WTP.
A shift was revealed, demonstrating that individuals were willing to pay more for the smart
candidate than the experienced candidate in separate evaluation but that this pattern was
reversed in joint evaluation.
Willingness-to-pay for computer
programmers, by evaluation mode
Experienced Candidate: 70 programs/
3.0 out of 5.0 GPA $26,800 $33,200
Smart Candidate: 10 programs/
4.9 out of 5.0 GPA $32,700 $31,200
Finally, Lowenthal (1993) applied joint versus separate evaluation to the arena of political
science. She found that separate evaluations of individual candidates reversed themselves
when it came to pairwise evaluations and to voting behavior.
48 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
by evaluation mode
Jobs Candidate: would bring 10,000 new jobs/
followed by tax evasion rumor 14% 56%
Clean Candidate: would bring 5,000 new jobs/
no rumors of misconduct 61% 31%
Participantsrated theJobs Candidate worsethan the CleanCandidate in separateevaluations
but better than the Clean Candidate in joint evaluation. Consistent with their joint evaluation
of the candidates, when asked how they would vote, 62% of participants reported that they
would vote for the Jobs Candidate.
3. Alternative explanations
The above examples document a growing body of evidence which demonstrates incon-
sistencies in expressed preferences across joint versus separate evaluation. In interpreting
these examples, we begin by noting their shared structure. They all involve outcome pairs,
which are distinguished along two prominent attributes. (It is generally implied that they
are equivalent across all other undescribed attributes.) One of these attributes is preferred
in separate evaluation, and the other attribute is preferred in joint evaluation. Observing
this structure, researchers have provided three different explanations for the joint versus
separate shifts reviewed above. These include: norm theory, the want/should proposition,
and the evaluability hypothesis. Below, we brieﬂy review each.
3.1. Norm theory
Ritov and Kahneman (1997) suggest that these reversals can be explained by norm theory
(Kahnemanand Miller,1986). Accordingto thisperspective,when individualsare presented
with a single item to evaluate, they struggle to make sense of it. To do so, they evoke a
set of available, internal referents for comparison and evaluate the item in the context of
these referents. When individuals are presented with more than one item to evaluate, the
alternatives themselves provide the comparison set for evaluation. Ritov and Kahneman
argue that, in joint evaluation, if there are differences in category membership across the
two alternatives, reconciling these category differences will dominate assessment.
For example, when asked to report support for governmental intervention to remedy
threatened extinction of Australian mammal species, people generate a set of other dying
speciesas theirinternally evokedcomparisoncategory.Theythen decidewhether thecurrent
itemrates favorablyorunfavorablywithinthis category.Typically,cuteAustralian mammals
compare favorably with less attractive species like condors, reptiles, or plants, and the
prospect of saving them from extinction will be positively valued. In contrast, skin cancer
in farm workers evokes a comparison category of other human health crises, on which
skin cancer might rate comparatively low, relative to heart disease, AIDS, or famine. Thus,
under separate evaluation, the prospect of preventing skin cancer may not be highly valued.
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 49
However, when confronted with Australian mammals and farm workers simultaneously,
the individual must evaluate across categories, rather than within categories. Here, human
health problems will typically be valued higher than animal extinctions.
Thus,Ritovand Kahnemanexplain inconsistenciesacross jointversus separateevaluation
based on the contrast between within category versus across category assessments. The
rating of the alternative within its category and the rating of the overall category are both
prominentattributes across the outcomes.In separate evaluation,the rating of thealternative
within its category is more salient than the rating of the overall category. In joint evaluation,
the rating of the overall category is more salient. Ritov and Kahneman’s theory is consistent
with their data. It also accounts nicely for Irwin et al.’s ﬁnding that people value private
goods over public goods in separate evaluation, but select public goods over private goods
in joint evaluation. Yet, while insightful and compelling, little direct evidence exists at this
point in research development in favor of norm theory as an explanation for the broader set
3.2. The want/should proposition
The want/should proposition offered by Bazerman et al. (1998) sees a tension between
what an individual wants to do versus what the individual thinks that he or she should
do. They argue that we often have an emotional desire to engage in behaviors that are
inconsistent with the behaviors in which we believe we should engage. Under separate
evaluation, lacking a counterbalancing alternative, we lean toward what we ‘want’ to do.
However, in joint evaluation, we tend to select the most justiﬁable option—the one that we
think that we ‘should’ choose. Thus, the direct comparison between options puts the ‘want’
self in check, in favor of the ‘should’ self.
For example, people want the consumer good over the environmental public good but
believe that they should select the environmental public good. Similarly, cute Australian
animals are more vivid and create a greater desire (want) to help than does the problem of
skincancer, yet peoplebelieve that humans should be valued overanimals. Thus, in separate
evaluation, the emotional aspect of the ‘want’ attribute overshadows the pragmatism of the
‘should’ attribute. In joint evaluation, the tradeoff costs of wanting the ‘want’ are brought
to light, and greater reason prevails. This explanation is compatible with Schelling’s (1984)
discussion of the multiple self problem, Thaler and Shefrin’s (1981) model of internal
competition between a planner and a doer, and Loewenstein’s (1996) assessment of the role
of transient preferences.
Recent empirical evidence supports the viability of the want/should proposition (Ten-
brunsel et al., 1998). Using similar and sometimes identical materials used in earlier re-
search, these researchers asked participants to predict how a fellow student would rate the
alternative behaviors or options in terms of the want/should distinction. The evidence was
overwhelmingly consistent: the alternative preferred in separate evaluation was the one
people wanted to choose and the alternative preferred in joint evaluation was the one that
people felt should be chosen. For example, consistent with expectations, Kahneman and
Ritov’s (1994) headlines were rated differently on the want and should dimensions. When
participants were asked how fellow students would feel that they should donate $10, human
50 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
health issues tended to be rated higher than wildlife causes. The same respondents indicated
thatothers, however,would wantto donate the $10 to the wildlife cause.Similar adaptations
were made of the stimuli used in studies by Irwin et al. (1993) concerning public goods
versus consumer goods and Hsee’s (1994) work on job choice.
A related study by O’Connor et al. (1998) lends some insight into the difference between
want and should. In their study, participants were asked to retrospectively think about a
real situation where they had been involved in conﬂict. Post hoc, participants were asked
to report what they wanted to do in that situation, what they thought they should do, and
what they actually did. Most participants reported that they did what they wanted to do,
rather than what they should have done. This pattern existed, even though in many cases,
the participants were aware that they would have been better off overall had they done
what they should have. Further, the actions that participants reported having wanted to
do were often more emotional and more volatile in the context of the conﬂict. Consistent
with Loewenstein’s (1996) hypotheses regarding the immediacy of visceral motivators,
participants reported that they were more emotional when they were in the middle of a
conﬂict than when they were either looking forward toward a conﬂict or looking back at
one. Thus, the degree of emotional involvement appears to be a key element differentiating
want versus should behaviors.
Bazerman et al. (1998) offer the want/should explanation to account for joint–separate
PRsin which people prefer jobsin which they are treatedfairly over jobs that pay them more
onlyin the separate condition(Bazerman et al., 1994). Toapply the want/shouldexplanation
here, one has to argue that somehow participants want to be treated fairly but believe that
theyshould select the job with higherpay.This reasoning runscounter to thecommonly held
opinionthat the selection ofjobs based onmoney is ahedonistic and self-indulgent decision.
We would argue that the subject population in which this ﬁnding was obtained resolves this
apparentinconsistency: MBAstudents are likely to sharea consensus that maximizingone’s
own economic well-being is the rational course of action. Other, less quantiﬁable factors
like fairness constitute reasons that may be more important to the individual, but may be
more difﬁcult to justify or explain.
3.3. The evaluability hypothesis
The evaluability hypothesis (Bazerman et al., 1992; Hsee, 1996; Hsee et al., 1998)
offers another cognitive explanation of separate versus joint preference reversals. Namely,
it suggests that these reversals are driven by differences in the ‘evaluability’ of attributes. It
argues that when two options involve trading off between a hard-to-evaluate attribute and
an easy-to-evaluate attribute, the hard-to-evaluate attribute will have less impact in separate
evaluation than in joint evaluation. Whether an attribute is hard or easy to evaluate can be
measured empirically by asking people whether they have an idea of how good a given
value is on that attribute (Hsee, 1998).
The evaluability hypothesis argues that in separate evaluation people have difﬁculty
assessing the desirability of an option on the hard-to-evaluate attribute, which means that
thehard-to-evaluate attribute has littleinﬂuence on decision making.For example, inHsee’s
(1994) hiring study, those assessing only one of the two candidates would not know how
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 51
to judge the candidate’s KY programming experience. With no available internal metric
for this attribute, people would tend to base their evaluation on the candidate’s GPA, for
which a familiar metric is available. In joint evaluation, however, the two options’ values
on the hard-to-evaluate attribute provide additional information, increasing the attribute’s
evaluability. Having written 70 programs is clearly much better than having written 10. As a
result, the ‘experienced’ candidate receives a substantial boost in desirability on the basis of
this comparison. In this sense, joint evaluation increases the evaluability of hard-to-evaluate
attributes, but does not increase the evaluability of easy-to-evaluate attributes.
Hsee (1996) demonstrated the power of evaluability with two direct studies. In the ﬁrst
study, participants were given a choice between descriptions of two CD players. The two
players varied in the number of CDs they could hold and in the sound quality they produced,
as measured by total harmonic distortion (THD):
Willingness-to-pay for CD players,
by evaluation mode
Fidelity Player: 5 CDs, THD of 0.003% $212 $228
Capacity Player: 20 CDs, THD of 0.01% $256 $204
These items generated a shift in joint versus separate evaluation. In separate evaluation,
measures of THD are not meaningful for most people and thus CD capacity was more
important. When choosing between the two, it becomes clear that the one has 3 times the
harmonic distortion of the other, and the impact of this difference carries more weight.
However, the reversal of preference was eliminated when subjects were provided with
information that made harmonic distortion evaluable in separate evaluation. Subjects who
were informed that harmonic distortion ranged between 0.002% for the best CD players
and 0.012% for the worst did not demonstrate a shift across joint versus separate evaluation
on this item. Essentially, the range information made the information easy to evaluate under
both separate and joint assessments.
Hsee et al. (1997) provided similarly compelling evidence that hard-to-evaluate informa-
tion will only be underweighted in separate assessments when no other external information
is available to make sense of this data. In the context of eyeglasses, Hsee et al. showed that
when additional information was provided to make the hard-to-evaluate attribute evaluable,
two concrete options were not necessary for the hard-to-evaluate attribute to inﬂuence the
From an economic perspective, the evaluability hypothesis can be thought of as an infor-
mational effect. When more information about the value of an attribute is available in joint
evaluation, people weight that information more heavily. What the evaluability hypothesis
adds to an economics of information explanation is a directional prediction, which suggests
a systematic cognitive bias. When insufﬁcient information is available about an attribute
in separate evaluation, that attribute does not, on average, receive a randomized weighting
in assessment. Instead, the evaluability hypothesis predicts that it will systematically be
underweighted. This is because in the presence of ambiguous or confusing information the
decision maker will always overweight familiar information; that is, the relatively more
evaluable attribute value.
52 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
3.4. Comparison of accounts
Table 1 provides an overview of experimental ﬁndings and the theoretical explanations
account for them. The three theories are: (1) Norm theory; (2) Want/should proposition; and
(3) Evaluability hypothesis. The three theoretical accounts have been given credit conserva-
tively in Table 1. That is, X’s only appear where, in our opinion, the theoretical explanation
did such a good job at explaining the result that it could have been used to predict it. We
expect others to disagree with our Xs, since deﬁnitive data are not yet available. However,
we provide this information to clarify our view of the current literature.
The evaluability hypothesis has been credited elsewhere (Hsee, 1996) with explaining
more than we give it credit for here. The limitation of the evaluability hypothesis is that it
is sometimes difﬁcult to identify the relative evaluability of two attributes. For example, in
Blount and Bazerman’s (1996) subject recruitment study, to apply evaluability, one must
make the argument that a $8 payoffs is not evaluable on its own and becomes evaluable only
when contrasted with the $7 alternative. Certainly it is the case that $8 looks better when
compared with $7. However, it is more difﬁcult to make the argument that participants were
unable to evaluate how much $8 was, in the same way that they were unable to evaluate
what 0.003% THD was. Further, in Bazerman et al. (1992), this explanation was explicitly
ruled out. Prior to assessment of any outcomes, participants were given explicit information
about the range of possible values that absolute payoffs could take, and the reversals still
occurred. Thus, the evaluability hypothesis appears to explain only a subset of results.
Similarly, Ritov and Kahneman’s (1997) norm theory has been credited only with the
reversals it explains cleanly. It is possible to extend the theory and argue, for instance, that
Lowenthal’s (1993) selection of political candidates can be accounted for by the category
of alternatives to which candidates were compared. That is, the Tax-evading Jobs Candidate
is mentally compared within a category of referent candidates who are not tax evaders, and
is therefore rated lower in separate evaluation. The Tax-evading Jobs Candidate fares better
when compared against the Clean Candidate who only brings 5000 jobs to the district.
However, to apply norm theory in this way, one must be able to articulate ex ante which
dimensions will evoke within category comparisons and which will evoke across category
Finally, while we believe that the want/should proposition is descriptively useful, we do
not believe that it describes all the data equally well either. For example, to use want/should
to account for Hsee’s (1998) ice cream choice, one would have to make the awkward
argument that somehow participants wanted overﬁlled containers but felt they should take
the container with more ice cream. That explanation is clearly not ﬂuid. Thus, as with the
other accounts, the want/should proposition offers insight, but not complete explanation.
In this section, we develop a framework which integrates the three theoretical accounts
discussed above. We begin by reviewing how individuals cognitively process outcomes
when assessing preferences in decision making, and show how that process differs across
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 53
Experimental ﬁndings and the theoretical explanations that account for them. Theories: (1) Norm theory, (2)
Want/should proposition, (3) Evaluability hypothesis. The alternative listed ﬁrst is always the one preferred in
Authors Stimuli 1 2 3
Bazerman et al. (1992) Dispute with neighbor. Resolution: X
Money Option: $600 for self, $800 for neighbor
Fair Option: $500 for self, $500 for neighbor
Irwin et al. (1993) Better Environment: Improve the air quality in Denver X X
Better Possessions: Get a new VCR
Lowenthal (1993) Choosing between political candidates: X
Jobs Candidate: Brings 10,000 jobs to district, dogged by rumors of
Clean Candidate: Brings 5,000 jobs to district, no rumors of misconduct
Bazerman et al. (1994) MBA job offers: X
Money Job: $85K for self, $95K for other MBA hires
Fair Job: $75K for self, $75K for other MBA hires
Bazerman (1994) MBA job offers: X
Money Job: $75K salary, low procedural justice
Fair Job: $60K salary, high procedural justice
Kahneman and Ritov (1994) Measured willingness to pay for modern problems: X X
Human Issue: Skin cancer among farm workers
Attractive Issue: Extinction threatens Australian mammals
Hsee (1994) Salary options for a 4-year job: X
Year 1 Year 2 Year 3 Year 4
Money Option: $27K $26K $25K $24K
Increasing Pay Option: $23K $24K $25K $26K
Blount and Bazerman (1996) Recruiting participants for an experiment: X
Money Option: $8 for self, $10 for other subjects
Fair Option: $7 for self, $7 for other subjects
Hsee (1996) Selecting music dictionaries: X X
Big Dictionary: 20,000 entries, torn cover
Torn Dictionary: 10,000 entries, intact cover
Hsee (1996) Selecting candidates for a computer programming job: X
Experienced Candidate: 70 programs written, GPA of 3.0
Smart Candidate: 10 programs written, GPA of 4.9
Nowlis and Simonson (1997) Price Brand
Cheap phone: $49 Cobra X
Quality phone: $69 Sony
Hsee (1997) Choosing between servings of ice cream: X
Bigger Serving: More ice cream in large container
Overﬁlled Serving: Less ice cream overﬁlled in small container
Hsee (1997) Picking sets of dinnerware: X X
Large Set: 31 intact pieces, 9 broken pieces
Unbroken Set: 24 intact pieces
54 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
jointversus separateevaluation. Next,we integratethe three theoreticalaccounts into amore
detailed study of how individuals perform separate evaluation, suggesting that it is incon-
sistencies in this process, rather than joint evaluation, that drives preference shifts across
joint versus separate evaluation. Finally, we discuss the implications of recent empirical
ﬁndings and this model for economic research.
4.1. Cognitive processes in decision making
The decision making process is typically modeled in three stages (e.g. Goldstein and
Einhorn, 1987). In the ﬁrst stage, the individual sees and encodes the relevant stimuli for
each potential outcome, including all salient attributes and their values. In the second stage,
theindividual performsa combinatorialprocess on theencoded data toassess utility.Finally,
the individual transforms that perceived utility into a behavioral expression of utility, such
as accepting/rejecting or stating a price.
In separate evaluation, one alternative is encoded, then evaluated. Preferences are ex-
pressed in absolute terms. In joint evaluation, the individual encodes and evaluates two
alternatives simultaneously, and preferences are expressed in relative terms. Comparatively,
separate evaluation is the more difﬁcult cognitive task. The individual must both generate
a decision model and calculate an absolute assessment of preference. The individual may
struggle to make sense of this task, asking ‘How do I know if outcome A is a good one?’
Joint evaluation is more straightforward. The presence of a second alternative frames and
anchors the entire decision process; the decision maker simply enacts it. ‘Which do I like
better—A or B?’
4.2. Biases in separate evaluation
Here,we suggest thatin separate evaluation,there area number ofpotential biases thatthe
decision maker may become susceptible to while performing this more difﬁcult task. These
emanate from both cognitive and motivational inﬂuences. Integrating the explanations from
above, we describe three below. These include (1) generating inconsistent referents, and
overweighting attributes that are (2) easy-to-evaluate or (3) appeal to our more visceral
4.3. Generating inconsistent referents
One way to perform separate evaluation is to evoke an internal norm or referent for
comparison. To do this, this individual generates a counterfactual. ‘Let’s imagine that there
was another option, similar to A, except that it differed on one or two attributes. Then
how would I feel about A?’ Then a comparative process, similar to joint evaluation, can be
used for assessment. Consistent with Ritov and Kahneman (1997), we suggest that while
this is a reasonable approach, it may be prone to error. This is because individuals may
be inconsistent when selecting and evaluating referents in this type of processing. These
inconsistencies may lead to intransitive preferences across alternatives.
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 55
To make this concrete, imagine two outcomes A and B, which are differentiated on at-
tributes 1 and 2. Here, A1<B1 and A2 >B2. In joint evaluation, Attribute 1 is weighted
more heavily, and B is preferred. Now imagine that in separate evaluation of A, the deci-
sion maker generates referent A0, differentiated on attributes 1 and 3, where A1=A01 and
A3>A03. The individual determines that A is better than A0, in light of Attribute 3. He or
she expresses a utility value of xfor A based on that comparison. Now imagine that the same
decision maker later assesses outcome B and generates referent B0for comparison, which
is differentiated along Attributes 2 and 4. Here, B2=B02, but B4 << B04. Here, in light
of Attribute 4, B0is strongly preferred. The decision maker then generates a utility value
of yfor B, where y<x, because he anchors his assessment of Outcome B on its relatively
impoverished value on Attribute 4. Thus, A is preferred to A0leading to a utility assessment
of x, and B is perceived as disadvantaged relative to B0, suggesting an inferior utility value
of y; when A and B are compared directly, B may still be preferred to A. This is because
different referents were selected for comparison in separate evaluation of A and B, and
these different referents biased how utilities were assigned under separate evaluation.
Regarding how different norms or referents might be generated across outcomes in sepa-
rate evaluation, certain features of the decision environment, such as category membership,
may be less salient. For example, in separate evaluation, the cross-category attribute of
species; that is, furry mammals versus humans, may not be salient to the decision maker.
However, in joint evaluation involving outcomes which cross species, this cross-category
attribute may dominate the decision. Other factors such as evaluability might also affect
whichattributes are salient whengenerating referents.It iscognitively easierto self-generate
a referent along dimensions which are familiar than on dimensions that are less familiar
and/or more difﬁcult to evaluate.
4.4. Cognitive short cuts
As we mentioned above, the task of separate evaluation is complex. The decision maker
must not only struggle to make sense of an option in isolation, then he must translate this
assessment into an absolute measure of utility, such as willingness-to-pay. The difﬁculty
occurs when weights across attributes must be assigned spontaneously without the beneﬁt
of another concrete option for comparison. For example, the decision maker may have to
reconcile how important fairness is compared to money, or how important a torn dictio-
nary cover is compared to number of word entries. Rather than generate a full range of
counterfactuals, decision makers may be tempted to simplify this awkward task in several
predictable ways. First, drawing from Hsee (1996), they may tend to ignore attribute data
that is hard to evaluate and compensate by overweighting easy-to-evaluate data. However,
since the evaluability of an attribute may not coincide with its importance to the decision
maker, this method could lead to undesirable inconsistencies.
Second, decision makers may rely on intuitive ‘gut-level’ responses that avoid reﬂection
or analysis for single options. By taking this short cut the individual can avoid some of the
workrequired to contrast andreconcile what one wantson a visceral leveland with whatone
feels that one should do on a normative level (Thaler and Shefrin, 1981; Schelling, 1984;
Bazerman et al., 1998). However, this process is likely to result in the overweighting of
56 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
‘want’ attributes (Bazerman et al., 1998). Avoiding the immediate discomfort of resolving
internal conﬂicts does not always coincide with the longer term consequences of outcomes
for the decision maker. Thus, this tendency, too, could lead to undesirable inconsistencies
across decisions. The choices made by an alcoholic or food addict to indulge are examples
of this phenomenon.
5. Implications for an economic model of behavior
We have reviewed a broad body of evidence which suggests that the temporal structure
of elicitation can inconsistently affect preferences. We suggest that these effects occur be-
cause of differences in processing that occur across separate versus joint evaluation. Some
effects can be explained within a ‘rational’ or informational framework, but others appear
to emanate from more complex psychological processes. Central to these differences is how
the individual makes sense of options in isolation, which we argue may be biased by both
cognitive and motivational factors. For example, we suggest that individuals may be incon-
sistent when self-generating internal referents for comparison. In constructing referents,
they will tend to anchor on more available and evaluable attributes. The relative weight-
ing of the focal outcome on these attributes will, in turn, anchor how utility is assessed
and expressed. Further, we suggest that individuals are not well-calibrated or consistent in
assessing inter-attribute trade-offs. Sometimes the evaluability or ‘want’ appeal of certain
attributes will be overweighted, relative to the true value of those attributes to the decision
maker. These responses can lead to cross-decision inconsistencies.
Together theseﬁndings suggestthat howrelated decisions aretemporally combined ornot
may have large effects on preferences. Thus, it is not only the options themselves and how
they are verbally presented (e.g. framed as gains versus losses), but also whether they are
sequentially or simultaneously presented, that determines preferences. This effect appears
to be driven by inconsistencies in inter-attribute trade-offs across time. This occurs not nec-
essarilybecause preferences change, butbecause people displaypredictable patterns in their
cognitive processing across time. In joint evaluation, the decision maker temporally groups
similar decisions within a single context and time period, resulting in greater consistency.
For economics, these ﬁndings suggest that in both individual and game contexts, the
sequencing of decision nodes is integral to how utility is assigned. Thus, to adequately
model what people actually do, economists must not only consider the full array of alter-
natives, but also a full array of how those alternatives might be sequenced, and the effects
of alternative sequencing on payoff matrices. Speciﬁcally, these results suggest that strate-
gically inconsequential transformations of extensive form games can have real behavioral
The psychological study of how preferences are constructed provides a critical bridge
between psychology and economics. This paper overviews relatively new ﬁndings in psy-
chology on how temporal grouping of alternatives can affect preferences. Like most new
phenomena, multiple theoretical contenders have been generated to account for these
effects. Here, we have developed a framework through which the contributions from each
can be usefully integrated and communicated to an economics audience. Future research
is clearly needed to specify when different biases are likely to operate, in particular, the
M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58 57
conditions under which decision makers generate internal referents in separate evaluation.
Ourgoal in this paper wasto offer an integration of variousconceptual explanations for joint
versus separate evaluation. We hope that our theoretical contribution can help pave the way
for future thought regarding the practical and theoretical implications of these experimental
This paper was supported by National Science Foundation Grant no. SBR-9511977 and
the Dispute Resolution Research Center at the Kellogg Graduate School of Management
at Northwestern University. The authors beneﬁted from many discussions with George
Loewenstein and Chris Hsee, in addition to critical comments by Richard Day, Keith
Murnighan, Tom Ross, and three anonymous reviewers.
Bazerman, M.H., Loewenstein, G.F., Blount White, S., 1992. Reversals of preference in allocation decisions:
Judging an alternative vs. choosing among alternatives. Administrative Science Quarterly 37, 220–240.
Bazerman, M.H., Schroth, H., Shah, P.P., Diekmann, K.A., Tenbrunsel, A.E., 1994. The inconsistent role of
comparison others and procedural justice in reactions to hypothetical job descriptions: Implications for job
acceptance decisions. Organizational Behavior and Human Decision Processes 60, 326–352.
Bazerman, M.H., Tenbrunsel, A.E., Wade-Benzoni, K.A., 1998. Negotiating with yourself and losing:
Understanding and managing competing internal preferences. Academy of Management Review 23, 225–241.
Blount, S., Bazerman, M.H., 1996. Inconsistencies in valuing comparative payoffs: Implications for bargaining
and trade. Journal of Economic Behavior and Organization 891, 1–14.
Goldstein, W.M., Einhorn, H.J., 1987. Expression theory and the preference reversal phenomena. Psychological
Review 94, 236–254.
Hershey, J., Kunreuther, H., Schoemaker, P., 1982. Sources of bias in assessment procedures for utility functions.
Management Science 28, 936–954.
Hsee, C.K., 1994. When trends of monetary outcome matter: Individual vs. joint presentation and judgment of
feelings vs. choice, unpublished manuscript.
Hsee, C.K., 1996. The evaluability hypothesis: An explanation for preference reversals between joint and separate
evaluations of alternatives. Organizational Behavior and Human Decision Processes 67, 247–257.
Hsee, C.K., 1998. Less is better: When low-value options are valued more highly than high-value options. Journal
of Behavioral Decision Making 11, 107–121.
Hsee, C.K., Blount, S., Loewenstein, G.F., Bazerman, M.H., 1998. Preference reversals between joint and separate
evaluations of options: A review and theoretical analysis, in press.
Irwin, J.R., Slovic, P., Lichtenstein, S., McClelland, G.H., 1993. Preference reversals and the measurement of
environmental values. Journal of Risk and Uncertainty 6, 5–18.
Kahneman, D., Miller, D.T., 1986. Norm theory: Comparing reality to its alternatives. Psychological Review 93,
Kahneman, D., Ritov, I., 1994. Determinants of stated willingness to pay for public goods: A study in the headline
method. Journal of Risk and Uncertainty 9, 5–38.
Kahneman,D., Tversky,A., 1979. Prospect theory:An analysis of decision underrisk. Econometrica 47, 263–291.
Loewenstein, G.F., 1996. Out of control: Visceral inﬂuences on behavior. Organizational Behavior and Human
Decision Processes 65, 272–292.
Loewenstein, G.F., Prelec, D., 1993. Preferences for sequences of outcomes. Psychological Review 100, 91–108.
Lowenthal, D.J., 1993. What voters care about: How electoral context inﬂuences issue salience in campaigns?
unpublished dissertation, Carnegie-Mellon University, Pittsburgh, PA.
58 M.H. Bazerman et al. /J. of Economic Behavior & Org. 39 (1999) 41–58
Nowlis, S.M., Simonson, I., 1997. Attribute-task compatibility as a determinant of consumer preference reversals.
Journal of Marketing Research 34, 205–218.
O’Connor, K., De Dreu, C., Schroth, H., Barry, B., Lituchy, T., Bazerman, M.H., 1998. Intrapersonal conﬂict
across time, unpublished manuscript.
Ritov, I., Kahneman, D., 1997. How people value the environment? In: Bazerman, M.H., Messick, D.M.,
Tenbrunsel, A.E., Wade-Benzoni, K.A. (Eds.), Environment, Ethics, and Behavior, New Lexington Press,
San Francisco, CA
Schelling, T.C., 1984. Choice and Consequence: Perspectives of an Errant Economist, Harvard University Press,
Slovic, P., Macphillamy, D., 1974. Dimensional commensurability and cue utilization in comparative judgment.
Organizational Behavior and Human Performance 11, 172–194.
Slovic, P., Grifﬁn, D., Tversky, A., 1990. Compatibility effects in judgment and choice. In: Hogarth, R.M. (Ed.),
Insights in Decision Making: Theory and Applications, University of Chicago Press, Chicago, IL.
Tenbrunsel,A.E.,Wade-Benzoni,K.A.,Bazerman,M.H., 1998. Empirical support for the want/should explanation
for separate versus joint preference reversals, unpublished manuscript.
Thaler, R., Shefrin, H.M., 1981. An economic theory of self control. Journal of Political Economy 89, 392–406.
Tversky, A., Sattath, S., Slovie, P., 1988. Contingent weighting in judgment and choice. Psychological Review
Tversky, A., Slovic, P., Kahneman, D., 1990. The causes of preference reversal. American Economic Review 80,
von Neumann, J., Morganstern, O., 1947. Theory of Games and Economic behavior, Princeton University Press,