135
N O N E X P E C T E D - U T I L I T Y T H E O R Y
136
S T A R M E R
A third widely observed finding arguably nudges the decision weighting models
into the lead:
behavior on the interior of the probability triangle tends to conform
more closely to the implications of EU than behavior at the borders
. Although
significant off-border violations are observed in at least some experiments (see for
example Wu and Gonzalez 1996), several studies, including those of Conlisk
(1989); Camerer (1992); David Harless (1992); and Gigliotti and Sopher (1993)
suggest that violations of EU are
concentrated
in comparisons between options in-
volving prospects on or near the borders of triangles. It is important to note that
this observation is unlikely to rescue EU for practical purposes. A natural interpre-
tation of the “border effect” is that individuals are particularly sensitive to changes
in the likelihood of outcomes with “extreme” probabilities (i.e., moving off the
border of the triangle, we introduce a low probability event; in the vicinity of each
corner, some outcome is near certain). It is very easy to think of important choice-
scenarios involving real prospects with “extreme” probabilities; for example, indi-
vidual decisions about participation in national or state lotteries or collective
decisions about nuclear power generation involve high-magnitude outcomes (win-
ning the lottery, suffering the effects of a radiation leak) occurring with very small
probabilities. Consequently, there are good reasons to model sensitivity to “ex-
treme” probabilities. One obvious way to do it is via decision weights.
24
In summary, if one is looking to organize the data from the large number of tri-
angle experiments, then the decision-weighting models are probably the best bet.
Moreover, there is a striking degree of convergence across studies regarding the
functional form to use; for best predictions the key ingredient seems to be an in-
verted s-shaped weighting function. Empirical support for this specification
comes from a wide range of studies including Lattimore, Baker, and Witte (1992);
Tversky and Kahneman (1992); Camerer and Ho (1994); Abdellaoui (1998); and
Gonzalez and Wu (1999), all of which fit the decision-weighting model to experi-
mental data. Collectively, these studies show that models with s-shaped probability
transformations offer significant predictive improvement over EU and outperform
other rivals. Most of the studies in this vein, at least those conducted in recent
times, employ the rank-dependent transformation method, though different math-
ematical forms have been used for the probability-weighting function. Lattimore,
Baker, and Witte use a probability weighting function of the form
(9)
for
i
,
k
5
1, 2,
. . .
,
n
,
k
•
i
and •, •
.
0 (
n
is the number of outcomes as usual). This
captures a number of other proposed forms (e.g., those of Uday Karmarkar 1978
and Quiggin 1982) as special cases. With •
5
•
5
1, • (
p
i
)
5
p
i
, hence we get EU.
More generally, the parameter • controls the inflection point and •
,
1 generates
•
•
•
•
=
+
•
•
=
•
( )
/
p
p
p
p
i
i
i
k
n
k
1
24
Another theoretical possibility suggested by Neilson (1992) is to allow the utility function defined
over outcomes to depend on the number of outcomes: this generates different behavior on and off the
border, but experimental tests of the model (see Stephen Humphrey 1998) have not been supportive.
Do'stlaringiz bilan baham: |