n identical experiments, an Allais paradox occurs when the addition of an independent event influences choice behavior. Consider the choices in the following table (Kahneman and Tversky 1979).
lottery |
1 to 33 |
34 |
35 to 100 |
preference |
 |
 |
0 |
 |
18% |
 |
 |
 |
 |
82% |
 |
 |
0 |
0 |
83% |
 |
 |
 |
0 |
17% |
In Experiment 1, a choice of
and
was given, and most participants picked
. In Experiment 2, a choice of
and
was given, and most participants picked
.
This observed pattern violates the independence axiom, since in both experiments, the payoff is identical if a
ball is picked, while if the
event is disregarded, the two experiments are identical.
To see it another way, consider the
event to be a black box that is always received if the random ball value is
. Knowing or not knowing the contents of the black box should not influence behavior.