130
S T A R M E R
Hence a choice between prospects
q
9 5
(
x
,
p
;
q
, 1
2
p
) and
r
9 5
(
x
,
p
;
r
, 1
2
p
)
may
be evaluated as a choice between
q
and
r
. Although cancellation is effectively
an application of the independence axiom of EU, the editing phase does not imply
that choices will generally satisfy independence, since whether a particular rule is
applied depends upon whether or not it is salient. Although they have no formal
theory of salience they do present evidence that editing is context dependent. One
example shows that cancellation is used in some cases where it is salient and not
in others (see their discussion of the “isolation effect,” p. 271).
One further rule—I will call it the
dominance heuristic
—has the effect of elim-
inating stochastically dominated options from the choice set prior to evaluation.
The addition of the dominance heuristic does not, however, remove all possibility
of monotonicity violation. Kahneman and Tversky assume that individuals scan
the set of options and delete dominated prospects
if they are detected
. This en-
sures the deletion of “transparently” dominated options but leaves open the possi-
bility that some dominated options survive application of the routine. Since the
preference function is not generally monotonic, such options may ultimately be
chosen.
This strategy for imposing monotonicity has the further,
perhaps surprising,
implication that choices may be nontransitive. If • (
?
) is nonlinear, then prospect
theory implies that there will be some
q
and
r
where
q
stochastically dominates
r
such that
V
(
r
)
.
V
(
q
).
18
So long as this dominance is transparent, the dominance
heuristic ensures that there will be no
direct
violation of monotonicity and
r
will
not be chosen over
q
. In general, however, it should be possible to find some other
prospect
s
, such that
V
(
r
)
.
V
(
s
)
.
V
(
q
). If there is no relation of dominance be-
tween
s
and either of
q
or
r
, then pairwise choice among these three gambles will
generate a systematic cycle of choice in which
q
s
c
r
and
r
s
c
s
and
s
s
c
q
where
s
c
is the relation “is chosen over.” Quiggin (1982, p. 327) calls this an “undesir-
able result.”
Quiggin’s reaction would not be untypical of economists more generally, most
of whom have taken both transitivity and monotonicity to be fundamental princi-
ples that any satisfactory theory should embody. On the other hand, several econ-
omists, Quiggin included, have thought aspects of prospect theory appealing and
have sought to build the relevant features into models more in keeping with con-
ventional theoretical desiderata. For example, part of Quiggin’s motivation in de-
veloping rank-dependent expected-utility theory was
to establish that a central
feature of prospect theory—nonlinear decision weights—can be built into a prefer-
ence function without sacrificing monotonicity. By constructing decision weights
cumulatively, we obtain a (transitive) preference function that is monotonic without
the need for an additional editing routine. Papers by Starmer and Sugden (1989),
18
To see how nonlinearity of
p
(
?
) can generate violations of monotonicity, consider a simple case
where
q
5
(
x
, 1) and
r
5
(
x
2
«
,
p
;
x
, 1
2
p
). Suppose
«
.
0 hence
q
dominates
r
: If
p
(
?
) is concave,
probabilities are overweighted,
and the dominated option
r
is
preferred for some
«
.
Now suppose
«
,
0, hence
r
dominates
q
: if
p
(
?
) is convex, probabilities are underweighted, and the dominated op-
tion
q
is preferred for some
«
.