DECISION-MAKING THEORY AND RESEARCH$50 with probability .8 D. Winning $30 with proba-
bility 1.0
Perhaps you preferred A and D, which many
people do. However, according to EU, those choices
are inconsistent, because D does not have a higher
EU than C (for D, EU = $30 = (1.0 $30); for C, EU
= $40 = (.8 $50)). Recognize that A and B differ
from C and D by a .3 increase in probability, and
EU prescribes selecting the option with the highest
EU. The certainty effect may also be seen in the
following pair of options. E. Winning $1,000,000
with probability 1.0 F. Winning $2,000,000 with
probability .5
According to EU, people should be indiffer-
ent between E and F, because they both have the
same EU ($1,000,000 1.0 = $2,000,000 .5).
However, people tend to prefer E to F. As the
cliche goes, a bird in the hand is worth two in the
bush. These results (choosing D and E) suggest
that people are risk averse, because those are the
certain options, and choosing them avoids risk or
uncertainty. But risk aversion does not completely
capture the issue. Consider this pair of options: G.
Losing $50 with probability .8 H. Losing $30 with
probability 1.0
If people were risk averse, then most would
choose H, which has no risk; $30 will be lost for
sure. However, most people choose G, because
they want to avoid a certain loss, even though it
means risking a greater loss. In this case, people
are risk seeking.
The tendency to treat certain probabilities
differently from uncertain probabilities led to the
development of decision-making theories that fo-
cused on explaining how people make choices,
rather than how they should make choices.
Prospect Theory and Rank-Dependent Theo-
ries. Changing from EV to EU acknowledged that
people do not simply assess the worth of alterna-
tives on the basis of money. The certainty effect
illustrates that people do not simply assess the
likelihood of alternatives, so decision theories must
take that into account. The first theory to do so was
prospect theory (Kahneman and Tversky 1979).
Prospect theory proposes that people choose
among prospects (alternatives) by assigning each
prospect a subjective value and a decision weight
(a value between 0.0 and 1.0), which may be func-
tionally equal to monetary value and probability,
respectively, but need not be actually equal to
them. The prospect with the highest value as calcu-
lated by multiplying the subjective value and the
decision weight is chosen. Prospect theory as-
sumes that losses have greater weight than gains,
which explains why people tend to be risk seeking
for losses but not for gains. Also, prospect theory
assumes that people make judgments from a sub-
jective reference point rather than an objective
position of gaining or losing.Prospect theory is similar to EU in that the
decision weight is independent of the context.
However, recent decision theories suggest weights
are created within the context of the available
alternatives based on a ranking of the alternatives
(see Birnbaum, Coffey, Mellers, and Weiss 1992;
Luce and Fishburn 1991; Tversky and Kahneman
1992). The need for a rank-dependent mechanism
within decision theories is generally accepted
(Mellers, Schwartz, and Cooke 1998), but the spe-
cifics of the mechanism are still debated (see
Birnbaum and McIntosh 1996).Improper Linear Models. Distinguishing be-
tween alternatives based on some factor (e.g.,
value, importance, etc.) and weighting the alterna-
tives based on those distinctions has been suggest-
ed as a method for decision making (Dawes 1979).
The idea is to create a linear model for the decision
situation. Linear models are statistically derived
weighted averages of the relevant predictors. For
example: L(lung cancer) = w1*age + w2*smoking +
w3*family history, where L(lung cancer) is the
likelihood of getting lung cancer, and wx is the
weight for each factor. Any number of factors
could be included in the model, although only
factors that are relevant to the decision should be
included. Optimally, the weights for each factor
should be constructed from examining relevant
data for the decision.However, Dawes (1979) has demonstrated that
linear models using equal weighting are almost as
good as models with optimal weights, although
they require less work, because no weight calcula-
tions need be made; factors that make the event
more likely are weighted +1, and those that make it
less likely are weighted −1.Furthermore, linear models are often better
than a person’s intuition, even when the person is
an expert. Several studies of clinical judgment