Adv Behav Econ pdf

134
S T A R M E R

---

5
. Evaluating Alternatives to Expected-Utility Theory
5.1. The Recent Experimental Evidence
Starting in the mid-1980s, a number of researchers turned their attention toward
testing nonexpected-utility theories. The majority of this work involved experi-
mental testing, some of it designed to compare the predictive abilities of compet-
ing theories; some designed to test novel implications of particular theories; and
some designed to test the descriptive validity of particular axioms. A very large
volume of work has emerged in this arena, providing a much richer evidential
base against which theories can be judged.
As we have seen, conventional theories all imply the existence of indifference
curves in the probability triangle, and certain of their key properties can be ex-
pressed in terms of characteristics of the indifference maps they generate. For in-
stance, Machina’s theory implies generalized fanning-out, while other theories
imply a mixture of fanning-in and fanning-out. A large number of experimental
studies have explicitly examined individual behavior in choices among prospects
in probability triangles. The data generated from these “triangle experiments”
provides a vantage point from which we can ask the following question: suppose
one were attempting to construct a conventional theory now, with the aim of ac-
counting for the evidence currently available, are there any obvious properties
one should seek to build in?
Although the evidence is both rich and complex, a number of stylized facts ap-
ply across a range of studies. In my view, three observations seem particularly ro-
bust. First, if you want a theory consistent with the available data 
don’t impose
generalized fanning-out
. Evidence from a wide range of studies reveals behavior
inconsistent with linear parallel indifference curves, but the patterns actually ob-
served are more complex than generalized fanning-out. For example, while nu-
merous studies reproduce behavior consistent with Allais paradox violations of
EU in choice pairs moving left to right along the bottom edge of the probability
triangle, another finding replicated across a range of studies—including Camerer
(1989), Chew and Waller (1986), Battalio, Kagel, and Jiranyakul (1990), and
Starmer (1992)—is a tendency for behavior to become less risk-averse moving up
along the left-hand edge of probability triangles. Such behavior would be consistent
with a tendency for indifference curves to fan in. These facts mitigate in favor of
theories like disappointment-aversion, implicit utility, quadratic utility, and models
with decision weights, all of which allow a mixture of fanning-in and fanning-out.
A second general lesson in the data seems to be 
don’t impose betweenness
.
There is considerable evidence—a good part of it is reviewed in Camerer and
Teck-Hua Ho (1994)—that choices are inconsistent with the assumption of linear
indifference curves. Together these two requirements narrow the field consider-
ably: if we want a theory of mixed fanning with nonlinear indifference curves, of
the theories reviewed above the only contenders are quadratic utility, lottery-
dependent utility, and models with decision weights.