Download Parametrizing the Evaluability Hypothesis of Preference Reversals

Parametrizing the Evaluability Hypothesis of Preference Reversals
Kimihiko YAMAGISHI (kimihiko@ky.hum.titech.ac.jp)
Graduate School for Decision Science & Technology
Tokyo Institute of Technology
Tokyo, JAPAN
Yoshiro KUNITAKE (kunitake@ky.hum.titech.ac.jp)
Abstract
the goodness of the Easy attribute, regardless of SE or JE,
because of its familiarity. However, due to the lack of such
familiarity, people face difficulty in evaluating the goodness
of the Hard attribute in SE. It is in JE that people may grasp
the goodness of the particular value of the Hard attribute
because JE allows people to compare the relative strengths
of the two candidates. Moreover, for the programming job,
the Experience should be more directly relevant than GPA.
Consequently, "the hard-to-evaluate attribute has a lesser
impact in separate evaluation than in joint evaluation, and
the easy-to-evaluate attribute has a greater impact" (p.250,
underline added).
Our current investigation aims at detecting the
evaluability changes between JE and SE by parameter
estimation of psychological weighting onto the Easy and
Hard attributes. We scrutinized how well the evaluability
estimates follow the prescription of EH.
Nord Institute for Society and Environment
Tokyo, Japan
We investigated cognitive processes under preference
reversals between "Separate Evaluation (SE)," wherein each
multiattribute choice alternative is evaluated in isolation, and
"Joint Evaluation (JE)," wherein choice alternatives are
evaluated simultaneously.
Previous account for this
phenomenon hypothesized that decision makers pay attention
to specific features of choice alternatives depending on the
nature of assessment. We propose a quantitative model of
such cognitive processes to represent the strengths of attention
paid to particular attribute in choice alternatives. Our model
detected attentional shifts between SE and JE. The change
differed from the hypothesized pattern in the original theory.
Consequently we suggest that the simpler cognitive processes
than the accepted theory explain such preference reversals.
Since its discovery in early 70s, preference reversal has
remained one of central research agenda in behavioral
decision making. Here, we focus on preference reversals
An Evaluative Weighting Model
between "Separate Evaluation (SE)" and "Joint Evaluation
and Empirical Questions
(JE)."
In SE, decision makers evaluate several
multiattribute alternatives simultaneously. In JE, each
Formula (1) shows our Evaluative Weighting Model for
choice alternative is evaluated in isolation. Hsee (1996)
the ith choice alternative with two attributes, x1 and x2.
offered the "Evaluability Hypothesis" (hereafter EH) to
(1)
WTPiJE / SE = "1JE / SE • x1i + " 2JE / SE • x 2i
account for preference reversals between JE and SE. In gist,
The superscripts are either JE or SE. The " • parameters
EH is a theory of change in psychological weighting on
denote the psychological weights associated to each
different attributes of choice alternatives. Here, we propose
attribute x.. In the model, the " • parameters represent the
an "Evaluative Weighting Model" to elaborate the ! evaluability of the corresponding attributes. We remark
!
explanation by EH: The Evaluative Weighting Model allows
here that we regard psychological weighting
of a particular
attribute as the evaluability of the attribute.
us to estimate parameters of psychological weighting and
!
By the model's terms,
we express Hsee's preference
examine the validity of EH 1.
reversal
of
programmer
salaries as follows:
As an example, take the "programmer salary assessment"
(2)
(Hsee, 1996, Study 2). Hsee asked his participants to assess
WTPASE > WTPBSE & WTPAJE < WTPBJE
suitable salary for job candidates of programming in an
The following Formulae (3) express the pattern of
imaginary language. In a [GPA, Experience] format, one
evaluability, as hypothesized by EH (in the previous
underlined quotation), to produce the relationship in
can express Candidate A as [4.9, 20 Programs] and
Candidate B as [3.0, 70 Programs]. Notice that the attribute !Formulae (2):
JE
SE
JE
SE
(3)
values trade off between the candidates. In SE, the
" GPA
< " GPA
& " Exp
> " Exp
participants rated A as deserving a higher payment, whereas
The current empirical question asks if, upon replicating
in JE, the rated salary for B exceeded A.
the WTP preference reversal in Formulae (2), parameter
Hsee explained the phenomenon as follows: The objects
estimates from the same WTP data would follow the EH
!
of assessment (candidates) consisted of a familiar "Easy"
prediction
in Formulae (3).
attribute (GPA) and an unfamiliar "Hard" attribute
(Experience). People reach WTP (willingness to pay: Salary
Experiment
in this case) assessments by evaluating each attribute and
combining the evaluation into WTP. People readily assess
Participants
Two hundred fifty-three Japanese undergraduates, then
enrolled in an introductory psychology course, participated
as a part of course requirement.
1
We used the term "parametrizing" to emphasize our modeling
strategy to use the parameters as evaluability indice. Hence this
term should not be confused with parameterization in geometry.
217
values (R2 = .275, n.s.). As standardized regression
SE
SE
coefficients, we estimated " DEV
and " ARAB
. Likewise, we
JE
JE
estimated " DEV
and " ARAB
as standardized coefficients of
regressing mean WTPs in JE on the attribute values (R2
= .982, p < .001).
JE
SE
! EH !
For ARAB (Hard),
predicts " ARAB
> " ARAB
. The
JE
!
! of " ARAb
significance
(t(6) = 17.70, p<.001) and the
SE
insignificance of " ARAB
(t(6) = -0.14, n.s.) jointly confirm
the prediction.
Moreover, a meta-analysis test (H0:
JE
SE
JE
SE
) rejected the! null in! favor of " ARAB
" ARAB
# " ARAB
> " ARAB
! p<.0001). Hence, the WTPs in JE supported the
(z=8.102,
!
EH prediction
in Formula (4).
SE
However, regarding DEV (Easy), " DEV
failed to reach
JE
significance (t(6) = 1.871, p = .110) !
whereas " DEV
did (t(6)
JE
SE
= 5.597, p < .001). Although EH predicts " DEV
, a
< " DEV
meta-analysis of these t-test results did not indicate a
JE ! SE
reliable difference in " DEV
# " DEV (z = 1.133, n.s.), thereby
!
failed to confirm Formula (5).
!
Decision Making Task
We modified the aforementioned programmer salary
assessment into the following "Manager Candidate Task."
Our participants read the following instruction. "Assume
yourself in charge of selecting a manager of your company's
new branch office in Lebanon. Your task is to assess the
suitable monthly salary for each candidate." The Easy and
Hard attributes for each candidate were "
(Hensachi,
or DEV)" of her/his Alma Meter and score of "Test of
Arabic as a Second Language (ARAB)," respectively. DEV
stands for "Deviation Score," and is a linear transformation
of z-score of each university's competitiveness rating
(mean=50; standard deviation=10). DEV was the most
!
well-known evaluation of Japanese universities, hence an
Easy attribute. ARAB is a fictitious test, and therefore, a
Hard attribute.
Design, Procedure, and Prediction
For DEV, we prepared the following three levels; 46, 55,
and 64. ARAB levels were 109, 326, and 634. By a
factorial design, we prepared 3 * 3 = 9 hypothetical
candidates.
The participants rated each candidate's
deserving monthly salary. The SE group rated each
candidate in isolation. The JE group examined a random list
of the 9 candidates, and assessed a proper salary for each
candidate.
Formulae (4) and (5) denote our empirical predictions:
JE
SE
(4)
" ARAB
> " ARAB
JE
SE
(5)
" DEV < " DEV
Discussion
!
Our Evaluative Weighting model revealed psychological
weighting of multiattribute choice alternatives in JE-SE
preference reversals. Although the weighting parameters
partly followed the prescription of the EH, we noted
inconsistencies with EH, such that the evaluability of the
Easy attribute stayed intact between JE and SE. Therefore,
we speculate that shifting weights on the Hard attribute
ALONE, while weighting the Easy attribute constantly, may
suffice in producing JE-SE preference reversals.
In the poster, we shall present analogous analyses of the
cases of no preference reversals (such as choosing between
choice alternatives consisting of exclusively Easy or
exclusively Hard attributes). Hsee (1996, p. 252) discussed
that, "the relative impact of the two attributes will not
change between two evaluation modes, and there will be no"
(preference reversals). Our analyses revealed that, the
Hard-Hard stimuli invited no preference reversals,
contingent upon enhanced evaluability of ALL attributes in
JE. In contrast, under no preference reversals with EasyEasy choice alternatives, the evaluability parameters stayed
intact across SE and JE.
In conclusion, we remark that the Evaluability Hypothesis,
although mostly in the right direction, failed to capture
details of change in cognitive processes between SE and JE.
Results
Mean Monthly Salary
!
!
Preference Reversals
We selected pairs of candidates wherein the DEV and
ARAB values traded off to compare the mean WTPs in a JESE preference reversal manner. Fig. 1 shows an example.
We use a [DEV, ARAB] notation to express each candidate.
The man salary (WTP) in Fig. 1 is shown in a hundredthousand yen unit.
35
D [64, 326]
30
C [46, 634]
25
Acknowledgments
20
This work was supported by Grant-in-Aid for Scientific
Research (C), No. 16500160, from Japan Society for the
Promotion of Science, to KY.
SE
JE
Rating Mode
Fig.1. An Example of JE-SE Preference Reversal.
The mean ratings in SE showed WTPCSE < WTPDSE (t(28) =
2.842, p < .01). JE showed WTPCJE > WTPDJE (t(115) = 4.830,
p < .001). Thusly, we replicated JE-SE preference reversals.
Estimation of Evaluability Parameters
For all candidates, we calculated 9 mean WTPs for the SE
salary rating to regress the mean WTPs on the attribute
218
References
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.