with indicative versions of the selection task, it would seem that we could
explain the diVerence very easily.
However, the hypothesis of a cheater-detection module suggests a more
speciWc prediction than that deontic tasks will be easier than indicative
ones. It predicts that, other things being equal, deontic tasks involving a
trade-oVbetween costs and beneWts will be easier than both indicative
tasksand other deontic tasks. For example, a social rule of the form ‘If you
are F, then you may do X’, which permits something if a particular
qualifying condition is satisWed, will be associated with improved perfor-
mance on a selection task only if subjects take the doing of X to constitute
abeneWt(that is, not just agood, but an item in a cost-beneWt social
exchange). Cosmides and Tooby have devised a series of tests which
conWrm this novel and otherwise unexpected prediction (1989, 1992; Cos-
mides, 1989). For example, the very same statement of a rule – ‘If a student
is to be assigned to Grover High School, then that student must live in
Grover City’ – elicited better selection-test performance when it was
explained that Grover High School was a superior school which citizens of
Grover City had to pay for in terms of higher taxes (hence raising issues of
fairness), even though in the control condition the importance of following
the rule (in order to allow the right number of teachers to be put in a
particular school) was given heavy emphasis.
We would add that it is also possible to detect this phenomenon retro-
spectively in tests carried out before the cheater-detection hypothesis was
advanced, such as the postage stamp version of the selection task
(Johnson-Lairdet al., 1972). In this test subjects had to decide whether the
rule, ‘If a letter is sealed, then it has a 5d stamp on it’, was being observed
in an array which showed: (a) the back of a sealed envelope; (b) the back
of an unsealed envelope; (c) the front of an envelope with a 5d stamp on it;
and (d) the front of an envelope with a 4d stamp on it. The experimenters
described this conditional as ‘meaningful’, in contrast to an ‘arbitrary’
conditional which they used as a control. (In the control condition sub-
jects were asked to decide whether the rule ‘If a letter has a D on one side,
then it has a 5 on the other’ was being followed in an appropriate four-
envelope array.) Of twenty-four subjects tested, twenty-one made the right
selection in the case of the ‘meaningful’ rule, whereas only two got it right
in the case of the ‘arbitrary’ rule. It was striking (and, we think, strikingly
modular)that in spite of the formal similarity between the two conditions
there was no transfer eVect: having got the right answer on the ‘mean-
ingful’ version did not help subjects toWnd the right answer in the control
condition.
This was the sort of result which was interpreted as showing that
reasoning is dependent on familiarity of content, with past experience
122 Reasoning and irrationality