Cognirion.
45 (1992)
141-162
Wishful thinking impairs belief-desire reasoning: A case of decoupling failure in adults? Nigel Harvey Departmem Received
of Psychology. June
University
5. 1991. final revision
London. Cower Srreef, London
College accepted
April
WC1 E 6BT.
UK
10, 1992
Abstract Harvey, N.. 1992. Wishful thinking adults? Cognition, 45: 141-162.
impairs
belief-desire
reasoning:
A case of decoupling
failure
in
Subjects were presented with a scenario that described how a certain type of opinion poll can be manipulated by respondents to put one particular political party (the threatened party) at a disadvantage. In a first experiment, people supporting this party but pretending to oppose it were found to be as likely to say that they would manipulate the poll as people who actually opposed it. In a second experiment, the threat embodied in the scenario was made more direct. It was also more salient because the study was carried out at a time of heightened political awareness when supporters of the threatened party were genuinely concerned about its future. People supporting the threatened party but pretending to oppose it were now about half as likely to say that they would manipulate the poll as those who actually opposed it. Two explanations for this breakdown subserving pretence are considered.
in the belief-desire
reasoning
Introduction Belief-desire
reasoning
desire to interpret, psychologists want
refers
Correspondence fo: Nigel Harvey, Street. London WClE 6BT, UK.
OOlO-0277/92/$05.00
to an individual’s
use of constructs
of belief
and
predict or role-play the behaviour of others. Developmental to know when it develops (e.g., Astington, Harris, & Olson,
0
Department
1992 - Elsevier
of Psychology,
Science
Publishers
University
B.V. All
College
London.
rights reserved.
Cower
142
N. Hanjet
1988; Frye.& to develop Frith, 1989).
Moore, can
1991; Perner,
account
1991; Wellman,
for the symptoms
1990) and whether
of autism
(Baron-Cohen.
its failure Leslie,
&
198.5; Baron-Cohen, Tager-Flusberg, Cohen, & Volkmar. 1991; Frith, Comparative psychologists want to know whether it can be employed by
non-human animals (Byrne & Whiten, 1988; Premack & Woodruff, Whiten, 1991; Woodruff & Premack, 1979). It is generally assumed operates efficiently in adult normal human beings (Astington Stich, 1983; Wellman, 1990). Is this so or are there occasions
1978; that it
& Gopnik. 1991; when it operates
poorly even in this group? Developmental psychologists have used variants of Wimmer and Perner’s (1983) “false belief” task to study belief-desire reasoning. A child is shown a scenario with two doll protagonists (Sally and Anne). Sally places a marble in her box and leaves the scene. Anne moves the marble from this box into her own basket. Sally returns. The child is asked where Sally will look for her marble. Young children and those who are autistic say that she will look in the basket and thereby betray their inability to appreciate the fact that Sally should falsely believe that the marble is still in the box. What is important is not that the children cannot appreciate the effects of a false belief per se but that they cannot appreciate the effects of a belief that is in conflict with their own (Astington & Gopnik. 1991, p. 11). (If the experimenter secretly moves the marble back from the basket to the box before Sally returns, Sally’s belief would be the true one and that of the child would be the false one. Children saying that Sally would look in the box would not be demonstrating an inability to appreciate the effects of a false belief. They would be demonstrating an inability to appreciate the effects of a belief different from their own.) Can adults always appreciate effects of a belief different from their own or are there occasions when their own beliefs (and desires) intrude on their assessments? There has been a dispute about whether or not the sort of belief-desire reasoning that allows successful prediction of other people’s behaviour can be described as theoretical or not: that is, based on a “theory of mind” (Astington & Gopnik, 1991; Harris, 1991; Hobson, 1990, 1991; Leslie & Frith, 1990). Up to now, much of the debate has been philosophically interesting but scientifically unrewarding. It has not generated many testable predictions that would enable the competing positions to be distinguished on empirical grounds. However, Leslie’s (1987) decoupling model of pretence provides a reasonably well-specified account of theory-based belief-desire reasoning in the “false belief” situation. He argues that the child must use its own knowledge of what Sally could and could not see of the episode to form a decoupled representation of what Sally believes. In other words, when a child role-plays, pretends to be or empathises with another person, that person’s beliefs and desires are represented separately from the child’s own beliefs and desires. These separate representations are expressed and manipulated within a special cognitive module. the decoupler. This quaran-
Wishful thinking
tine
ensures
that
they
are
kept
apart
from
and
are
not
influenced
143
by the
pretender’s own beliefs and desires. According to this model, a protagonist playing a game predicts an opponent’s behaviour on the basis of a decoupled representation of the opponent’s beliefs and desires. Because of the decoupling, the protagonist’s own beliefs and desires do not influence
the prediction.
Belief-desire
reasoning
should
not be affected
by
wishful thinking except in cases where the decoupler has not developed (e.g.. young children and autistic people) or where its processing is impaired (e.g.. after some sort of brain damage). Political allegiances provide a useful tool for examining belief-desire reasoning in adults. Support for one political party over the alternatives can usually be taken to indicate a belief that its programme is the best available and a desire for that programme to be implemented. Moreover, supporters of each party know that supporters of other parties believe that they are right and desire power just as much as they themselves do. In each of the experiments reported here. people had to decide whether or not they would act in a way that would threaten the general election chances of one of the main British political parties. The action involved providing misleading responses in an opinion poll and could have been construed as immoral. Some of the people presented with the dilemma actually supported one of the political parties that would benefit from the poll manipulation. Others supported the party that was threatened by it but had to make their decision while pretending that they supported one of those parties that would benefit. If belief-desire reasoning is affected by wishful thinking, people in this second group should be less likely than those in the first one to say that they would manipulate the opinion poll. The rationale underlying the experiments relied on the fact that politicians’ actions are influenced by the results of opinion polls. For example, in Britain, those in power will take opinion poll results into account when deciding on the date for a general election. Politicians of any political party will also take opinion poll information about their leader’s popularity into account when deciding whether or not to push for a leadership election and whether or not to change leader when such an election is under way. (In Britain, the people qualified to vote in a leadership election vary across parties. However, feelings of Members of Parliament (MPs) are always paramount in deciding whether such an election should be called. In the Conservative party, MPs are also the only people allowed to vote in it.) To estimate relative popularity of major figures in the various parties, opinion pollsters often ask respondents whether they would be more likely to vote for a particular party if the current leader were to remain in power or if another named individual prominent in the party were to take over the leadership. A minor variant of this form of question is to ask respondents whether they feel that voters in general (as opposed to the respondents themselves) would be more likely to
144
vote
N. Harvey
for the party
under
one individual
rather
than
another.
Questions
such as
these allow respondents to send messages to politicians who may then act on the basis of their content. However, this means that it is also possible for respondents supporting other parties to send misleading messages to politicians in the hope that they will act on them and thereby
EXPERIMENT For
the
purposes
impair
their chances
in a general
election.
1
of this experiment,
subjects
of all political
persuasions
were
asked to assume that they supported the Conservative party. Whatever their true feelings, they were asked to pretend that they felt that the Labour party would win the next general election only if Neil Kinnock remained unchallenged as its leader. The Conservative party would have a reasonable chance of winning only if he were to be challenged as leader by John Smith, the Shadow Chancellor of the Exchequer. This is because the leadership contest would create deep divisions in the Labour party, thus reducing their public support during the crucial period just prior to the general election. They were asked what they would do if an opinion pollster inquired whether they and the country would be more likely to vote Labour under the leadership of Neil Kinnock or John Smith. If they were truthful, they would answer “Neil Kinnock”. On the other hand, by answering “John Smith”, they could increase the likelihood of the Labour party calling a leadership election and thereby impair its general election chances. Let us suppose that 80% of people who actually support the Conservative party say that they would take this opportunity of manipulating the opinion poll. If people actually supporting the Labour party are able to use their belief-desire reasoning to pretend that they are Conservative supporters, then 80% of them should also say that they would manipulate the poll. If, on the other hand, their true beliefs and desires leak through their pretence, then fewer than 80% of them should say that they would manipulate the poll. There is one potential artefact that should be excluded. Suppose that fewer supporters of the threatened party than of other parties feel that they would have realised that there was scope for poll manipulation. Supporters of the threatened party may just be less likely to say that they would manipulate the poll because (and only because) they felt that they were less likely to recognise that such manipulation was possible in the first place. To check for this possibility, people were asked whether they felt that they would have recognised that poll manipulation was possible. As long as no fewer Labour supporters than other subjects answer this question affirmatively, any evidence in favour of wishful thinking is unlikely to be due to the potential artefact just outlined.
Wishful thinking
Of course, unlikely
after
being
told
that
to find it easy to say whether
had they not been their Machiavellian specific
purposes
poll
manipulation
is possible,
they would have realised
people
145
are
that it was possible
told that it was. Some people may underestimate the extent of insight and others may overestimate it. However, for the of checking
for the artefact
just outlined,
this need
not matter.
The realisation judgment does not have to be made with high fidelity. However, I do need to assume (a) that there is some positive relation between what people judge they would have realised and what they actually would have realised (b) that this relation itself does not depend on political allegiance. Finally, worth emphasising again that results from relevant to present concerns if the decision depend on voting intention.
the realisation to manipulate
and it is
j**?gment are only the poll is found to
Method Subjects Eighty-seven students aged between 18 and 35 years acted as subjects. not paid for their participation in the experiment.
They were
Design and materials Subjects answered three questions by marking boxes on a response form. The questions concerned whether they would manipulate an opinion poll, whether they would have realised that such manipulation was possible and how they intended to vote in the next general election. Questions were presented in this same order for all subjects. Procedure Subjects were run in two groups of approximately equal size. The opinion poll was first described to them orally and the possibility of manipulating it was outlined. They were asked if they understood this. When the experimenter was convinced that they had done so, they were handed the form and assured that the responses that they made would remain anonymous. The form was headed “Political Manipulation of Opinion Polls-An Anonymous
Survey”.
Irrespective
The first question
of whether
you do.
You
believe
Kinnock
remains
summarised
or not you vote Conservative, strongly
unchallenged
that
the scenario
assume for the purposes of this question that
the Conservatives
as leader
as follows:
of the Labour
will only lose the next general party.
election
if
146
N. Harvev
This is because you feel that any Labour leadership contest and that this would reduce their public support at a crucial as leader).
would create deep divisions in the party pre-election time (whoever they vote in
It seems clear to you that the Conservatives can only win if Labour MPs can be misled into thinking that it is worth risking such a contest. An opinion pollster asks you whether you and the country would be more likely to vote Labour if John Smith were to replace Neil Kinnock as Labour leader. You know whether to into risking in the next
that many Labour MPs will take the results of the poll into account when deciding force a leadership contest. Would you take this opportunity of helping to mislead them one by falsely claiming that you and the country would be more likely to vote Labour general election if John Smith replaced Neil Kinnock?
they went on to the Once subjects had answered “yes” or “no” to this question, next question. It was worded as follows. “Do you think that you would have realised that the opinion poll could be manipulated for political ends in this way?” When they had answered “yes” or “no” to this question they turned to the third one. It asked them to specify how they thought that they would vote in the next general election. They had to mark one of four boxes. These were marked Conservative, Labour. Liberal Democrats and Other. (The Liberal Democrats are a relatively small third party that occupies the centre of the British political spectrum. The category “Other” included people who intended not to vote, people who had no idea how they would vote, people who intended to vote for a party that was not explicitly specified and people who were ineligible to vote.)
Results Only four subjects said that they thought that they would vote for the Liberal Democratic party. Hence their data were pooled with those who had marked the “Other” box in answer to the voting intention question. This was appropriate because both these types of subject had to pretend that they were something that they were not (i.e., Conservative voters) in order to decide whether they would manipulate
the poll to impair
the election
chances
of a party
(i.e.,
the Labour
party) that they did not support. Each of the 87 subjects was classified according to whether he or she (a) would manipulate the poll (i.e., manipulation decision), (b) would have realised that this manipulation was possible (i.e., realisation judgment) and (c) would vote for the Labour party, Conservative party or neither of these parties in the general election (i.e.. voting intention). These raw data are presented in Table 1. Percentage of subjects who said that they would manipulate the poll is shown in Figure 1 as a function of realisation judgment and voting intention. A log-linear model assuming independence between all three factors was fitted to the data. Residual deviance was low (x2(7) = 3.898; NS), demonstrating that
Wishful
Table
1.
Experiment
1: Subjects
categorised
according
thinking
147
to poll manipulation
decision, realisation judgment and voting intention. Frequencies predicted on the basis of independence between these three factors are shown in parentheses Voting intention
Would realise manipulation
Would manipulate
possible‘?
poll?
Yes
Yes
Yes
No
No
Labour
Conservative
(:x6.08) ,::.3*,
(E.48, 4
(i.47)
(97.40)
(kg)
(4.45)
(Z.75)
(f.57)
(?79)
(1.51)
(6.23)
4
Yes
No
Neither
No
2
the model describes the data well. Values predicted on the basis of the model are shown in parentheses in Table 1. Of course, independence between all three factors would be obtained if subjects randomly selected response boxes for each question. To ascertain that this did not account for the independence, data were compared with those that
Reclisatian m IX
Labaur
Conservative
Judgment Affirmative Negative
Neither
Voting Intention Figure
1.
Experiment function
1: Percentage
of realisation
of subjects who said that they would
judgment
and voting intention.
manipulate
the poll as a
148
N. Harvey
would
have been expected
had subjects
marked
each response
box in the first two
questions with a probability of one half and each response with a probability of one quarter. High residual deviance
box in the last question showed that this model
did not fit the data (~‘(11) = 67.41; p < .OOl). Instead, to total deviance were made by the effects of non-random
significant contributions responding to the first
(x2( 1) = 34.658; p < .OOl), second 13.103, p < .Ol) questions.
(x2( 1) = 16.248; p < .OOl) and third
(x2(2)
=
Discussion Data are fully consistent with well-nigh perfect belief-desire reasoning in adults. Eighty per cent of subjects who did not support the Labour party said that they would manipulate the opinion poll to impair that party’s chances in a general election. Virtually the same proportion (82%) of Labour party supporters pretending to be Conservative party supporters said that they would manipulate the poll for the same ends. Clearly, people who actually supported the Labour party were able to keep their own true beliefs and desires quite separate from those that they were pretending to have. There was no evidence that wishful thinking contaminated their belief-desire reasoning. The data do not allow us to determine whether each and every person who did not support the Labour party would have manipulated the poll with an 80% probability or whether 80% of them would have definitely manipulated it and 20% of them would have definitely not done so. However, for the purpose of the present argument, this is not important. Ahthat matters is that Labour party supporters pretending to be Conservative party supporters could predict how real Conservative party supporters would behave. They may have accomplished this by thinking that there was an 80% probability that each and every one of their opponents would conspire against them and by then making their decision to manipulate the poll with this same probability. Alternatively, they may have considered that four out of five of their opponents would definitely conspire against them and one out of five would definitely not do so. Each Labour party supporter may then have role-played an opposition conspirator with a probability of 80% and an opposition non-conspirator with a probability of 20%. Again, for present purposes, it does not matter which of these alternative belief-desire reasoning schemes was used. It is the effectiveness of whatever scheme was used that is impressive. An alternative explanation of the results should be considered. It may be that 80% of people would answer affirmatively if asked whether they would be untruthful in order to increase their chances of achieving any desired goal. If all subjects recoded the political manipulation question into this more abstract form before answering it, then 80% of them (irrespective of political allegiances) would
Wishful
be expected
to provide
affirmative
answers.
This recoding
rhinking
explanation
149
suggests
that the proportion of people saying that they would be untruthful should not depend on the particular scenario employed in the experiment. For example, if the potential threat included within the scenario were to be made more realistic or salient in some way, this proportion should still remain constant at 80% for all subjects. In contrast, if people answer such questions directly without recoding them into a common abstract form, the particular characteristics of the scenario may well affect the proportion
of them willing
to act mendaciously
to implement
the threat.
EXPERIMENT
2
Experiment 1 failed to produce any evidence of wishful thinking. People were able to pretend that they were people who supported a political party that was in competition with the one that they actually supported. However, the threat suggested in the scenario was a fairly distant one. A chain of four events had to turn out in favour of the Conservative party for it to be effective. First, the poll manipulation had to provoke a Labour party leadership election. Second, this election had to cause deep divisions within the party. Third, these divisions had to reduce support for the party in a general election. Fourth, the resulting swing towards the Conservative party had to be sufficient to enable them to win this election. Labour party supporters may have felt that the probability of all four events turning out this way was so low that the threat to them was not particularly salient or realistic. Belief-desire reasoning failure may only appear when the threat to someone’s own beliefs and desires is more direct. Experiment 2 was designed to address two questions. First, would a scenario that suggests a more direct threat induce a failure in belief-desire reasoning? Second, if it does not do so, would the proportion of people saying that they would manipulate the poll still be approximately 80%. as the recoding interpretation of Experiment l’s results suggests? The opportunity for carrying out this experiment arose in November 1990 when the Conservative party held its leadership election. The first round was between Mrs Thatcher, the incumbent, and Mr Heseltine. After Mrs Thatcher failed to secure a decisive victory, a second round of the contest was arranged. Prior to balloting in this round, Mrs Thatcher withdrew from the contest and Mr Hurd and Mr Major entered it. Results of various opinion polls were then published. The pollsters had asked respondents whether they would or whether they felt other voters would be more likely to vote for the Conservatives under Mr Hurd, Mr Heseltine or Mr Major. Experiment 2 was carried out at this time (before the results of the contest were known) and was based on this sort of poll.
150
N. Harvq
Subjects the Labour
of all political party.
Whatever
persuasions their
were asked
true feelings,
to assume
that they supported
they were asked
to pretend
that
they felt that the Conservative party would only win the next general election if Heseltine were to be elected as its leader. The Labour party would only have a reasonable chance of winning if the Conservatives could be misled into believing that they were most likely to win a general election with Major as their leader. They were asked what they would do if an opinion pollster asked them whether they and the country would be more likely to vote for the Conservative party if it were led by Major, Heseltine or Hurd. If they were truthful. they would answer “Heseltine”. On the other hand. by answering “Major”, they could impair the Conservative party’s chances in the general election. As before, the question is whether true supporters of the threatened party will be able to role-play their opponents. A certain proportion of true Labour party supporters will say that they would manipulate the poll to favour their party. Will the same proportion of Conservative party supporters who are pretending to be Labour party supporters say that they would manipulate the poll to favour the Labour party? In this experiment, the threat embodied in the scenario is more direct than the one used in Experiment 1. There are three rather than four events that have to turn out in favour of the Labour party for it to be effective. First, the poll manipulation has to persuade Conservative MPs to elect Major as their leader. Second, Major’s leadership has to reduce support for the party in a general election. Third. the resulting swing towards the Labour party has to be sufficient for them to win. The threat is more salient than it was in Experiment 1. The experiment was carried out at a time when Conservative supporters were genuinely very concerned about the future of their party. As before. subjects were asked whether they would have realised that the poll manipulation was possible. The reason for this was the same as in Experiment 1: to exclude the possibility that any evidence in favour of wishful thinking is an artefact of a relative inability to recognise the possibility of poll manipulation among supporters of the threatened party.
Method Subjects
One hundred and two students They were not paid for their Experiment 1.
aged between participation.
18 and 32 years acted as subjects. None of them had taken part in
Wishful thinking
151
Design and materials Design of the experiment and materials used were comparable to those employed in Experiment 1. Only the scenario presented in the first question was different. The exact wording used to describe it is given below.
Procedure As before, subjects were run in two groups that were approximately equal in size. The procedure was identical to that used in Experiment 1. However, the scenario summarised in the first question was as follows:’ Irrespective of whether or not you vote Labour. assume for the purposes of this question do. You believe strongly that Labour will only lose the next general election if Heseltine of the Conservative party.
that you is leader
You also realise Conservative MPs are only likely to elect for their leader the man who they think is most likely to win that election. It seems clear to you that Labour can only win if Tory MPs can be misled into thinking that someone other than Heseltine is most likely to win the election for them. An opinion pollster asks you whether you and the country would be most likely to vote for the Conservatives under Major, Heseltine or Hurd. Knowing that many Tory MPs will take the results of the poll into account when deciding how to vote. would you take this opportunity of helping to mislead them by falsely claiming that you and the country would be most likely to vote Conservative under Major?
Results As before. data from supporters of the Liberal Democratic party were pooled with those from subjects who had marked the “Other” box in answer to the third question. Each of the 102 subjects was then classified according to the same three factors as those that were used in Experiment 1. Table 2 shows these data and Figure 2 shows the percentage of subjects who said that they would manipulate the poll as a function of realisation judgment and voting intention. First, a log-linear model assuming independence between all three factors was fitted to the data. The high residual deviance (x’(7) = 21.38; p < .Ol) revealed that it was inadequate. This is not surprising given that inspection reveals clear differences between the observed data points and those predicted on the basis of the independence model (shown in parentheses in Table 2). An interaction between voting intention and manipulation decision was added ‘“Tory”
is a synonym
for “Conservative”.
152
N. Harvey
Table
2.
Experiment decision,
2: Subjects realisation
dicted
on the basis
shown
in parentheses
categorised
judgment
and
according voting
of independence
to poll
intention.
between
these
manipulation
manipulate
Yes Yes No No
poll?
possible?
Yes No
Frequencies
pre-
three factors
are
Voting intention
Would realise Would
manipulation
Labour
Conservative
8
1s (10.08) 8
(9.15)
(9.68)
(8.79)
(k3)
(6. IS)
4
Neither
&28) 10 (11.42)
8
Yes 9
No
$99)
10
(6.51)
(5.91)
(;.68)
into the independence model. This accounted for a significant amount of deviance (x’(2) = 6.94; p < .05). However, examination of the residual deviance showed that this new model still failed to provide an adequate fit to the data (~~(5) = 14.44; p < .02).
Realisation m m
Labour
Conservative Voting
Figure
2.
Judgment Affirmative Negative
Neither
Intention
Experiment 2: Percentage of subjects who said thar they would manipulate the poll as a function of realisation judgment and voting intention.
Wishful thinking
An then
interaction added
between
manipulation
decision
in as well. This also accounted
and realisation
for a significant
judgment
amount
153
was
of deviance
residual deviance was now low (x2(4) = (x’(1) = 10.38; p < .Ol). Furthermore, 4.06; NS), thus indicating that an adequate fit had been achieved. (Additional analysis showed that the remaining first-order interaction between voting intention and realisation judgment contributed very little to this residual (x’(2) = 1.71; NS).) Adding just
interactions
described
changed
into the independence absolute
levels
model
of deviance
in orders
other
attributable
than the one
to each of them
slightly but did not affect model selection. Table 3 shows predictions from the model for the second-order margins corresponding to the two significant first-order interactions. They have been standardised by averaging fitted values across levels of the remaining factor. It is clear from Table 3(a) that subjects who did not support the Conservative party were more likely to decide in favour of manipulating the poll. In contrast, Conservative party supporters who were pretending to be Labour party supporters were more likely to decide against manipulating the poll. This appears to be strong evidence of a failure to pretend in this context. Table 3(b) shows that people who decided to manipulate the poll were more likely than not to think that they would have realised that such manipulation was possible. In contrast, those who decided against manipulating the poll were more likely to think that they would not have realised that they could have manipulated it. It is important to recognise that this effect was not related to the one shown in
Table
3.
Experiment 2: Predictions from the fitted model for the second-order margins corresponding to the two significant first-order interactions (a) Voring intention x Manipulation decision Voting intention Would manipulate
poll?
Yes No (b)
Conservative
Lahour
Neither
6.0 9.0
11.5 5.0
13.0 6.5
Manipulation decision x Realisation judgment Would realise
Would manipulate Yes No
poll?
manipulation
possible?
Yes
No
13.00 4.33
7.33 9.33
1.54
N. Hurvey
Table
3(a).
judgment
Neither
the
interaction
nor the interaction
reached significance. Conservative party manipulation recognised was possible,
between
voting
intention
and
all three
factors
(included
in the residual)
between
realisation
Thus the belief-desire reasoning failure was not artefactual. supporters were not less likely to decide in favour of poll
because
they
were
that such manipulation
less
likely
was possible.
to think Whether
they were less likely than other subjects
that
they
would
have
or not they realised
it
to say that they would do it.
Results of this experiment are not consistent with the alternative (abstract recoding) interpretation of the data obtained from Experiment 1. That interpretation suggested that the proportion of people saying that they would manipulate the poll should remain approximately constant. However, in the present experiment, this proportion was shown to vary systematically. Moreover, it was always less than the proportion obtained in Experiment 1. It appears. therefore. that the best interpretation of the results from the two experiments is that people succeeded in the pretence required in Experiment 1 but found the pretence required in Experiment 2 more difficult.
Discussion Seventy per cent of Labour party supporters said that they would manipulate the opinion poll to impair the general election chances of the Conservative party. Approximately the same proportion (67%) of people who supported neither main party but who were pretending to be Labour party supporters also said that they would manipulate the poll for these ends. It is clear. then, that there was nothing difficult about pretending per se in this context. Failure of belief-desire reasoning was restricted to Conservative party supporters who were required to pretend to be threatening their own true beliefs. The proportion of these subjects who said that they would manipulate the poll was just 40%. a figure virtually mid-way between what one would expect if they had been acting according to their own true beliefs (i.e.. 0%) and what one would expect if they had succeeded in pretending to be Labour party supporters (i.e., 70%). This suggests that their own true beliefs and desires were contaminating the beliefdesire reasoning that they were using to role-play other people’s behaviour. The effect of this contamination was to produce what a lay person would characterise as wishful thinking. The threat implied in the scenario used in this experiment was more direct than that used in the previous one and more salient because it was presented at a time of heightened political awareness. These factors appear to be responsible for the inability to pretend. However, they also appear to have lowered the proportion of people willing to carry out the threat when it would have benefited the party that
they
truly
supported.’
This
proportion
dropped
from
80%
in Experiment
1 to
70% in Experiment 2. People may be less willing to say they will act immorally when they believe that their action is more likely to be effective -perhaps because they think that they will be more likely to feel guilty or responsible for an unjust outcome. also be
People who decide against manipulating the poll for this reason may more reluctant to admit that they realised that the poll could be
manipulated
or to say that they recognised
fact, they did not do so. This would in Table
GENERAL
account
that it could be manipulated for the appearance
when.
in
of the interaction
3(b).
DISCUSSION
The results from the two experiments need to be considered together. What they show is that adult belief-desire reasoning ability is labile. In some circumstances, adults can emulate the behaviour of other people on the basis of a knowledge of their beliefs and desires. However, there are other circumstances in which performance breaks down-people are prone to think that other people will behave as they themselves would behave on the basis of their own beliefs and desires. This is a cognitive bias that, in lay terms, might be referred to as wishful thinking. Experiment 2 shows that it can be produced in a laboratory setting. It is obtained when effective pretending entails responding in a way that directly conflicts with how subjects would respond on the basis of beliefs and desires that are currently highly salient outside the laboratory. From certain perspectives, the results obtained here can hardly be regarded as surprising. In the literature on decision-making, “optimism” biases are well known. The sort of wishful thinking uncovered here can be considered as just another example of one. Langer’s (1975) “illusion of control” can be classified as an optimisni bias. She had subjects participate in actions that could only result in random outcomes. However, they thought that they had a higher than chance probability of being successful. Furthermore, the “illusion” was greater in subjects who expressed confidence in their own actions being successful than in those who expressed confidence in an experimenter’s action being successful. Harvey and Ayton (1990) showed that this illusion was not restricted to predicting random outcomes. Even when people have a higher than chance probability of controlling a system, they overestimate that probability more than non-participant observers do. Other work suggests that optimism biases are not restricted to situations in which people act on a system in an attempt to control it. Subjects ‘It has not escaped me that the difference between the two experiments could be fatuously explained in terms of a greater tendency towards autism among Conservative party supporters and a greater tendency towards honesty among Labour party ones. However. as there appears to be no support for either of these notions in the literature. I will not explore them further here.
156
have
N. Harvey
also been
found
to be optimistic
about
systems over which they have no control & Ayton, 1986). From other Developmental
their
(Fischhoff
forecasts
for the behaviour
& MacGregor.
of
1982; Wright
perspectives, the results appear to be rather more unexpected. and comparative work on belief-desire reasoning has proceeded
on the basis of an (often implicit) assumption that normal adult human beings exercise this cognitive skill with ease and efficiency. In fact. adults’ reasoning about better
the deceptive behaviour of others in salient political contexts may be little than that exhibited by young children and autistic people in more concrete
settings (Astington & Gopnik, 1991; Baron-Cohen et al.. 1985; Sodian, 1991). So can the present findings inform the developmental debate in any way? Certainly they suggest that it would be unwise to regard possession of the psychological substrate of belief-desire reasoning (e.g., a “theory of mind”) as sufficient for adequate performance in tasks requiring that ability. Conversely, inability to perform those tasks in some particular setting should not be taken as evidence that those substrates do not exist. But these matters concern problems attendant on drawing conclusions about competence from evidence about performance. Developmental psychologists are already well aware of them. More interesting, perhaps, is what the data tell us about the mechanisms underlying pretence. Adults can predict the behaviour of others from a knowledge of beliefs and desires that are different from their own. Clearly they are in possession of whatever psychological mechanism or “theory of mind” is required to do this. It is just that, under certain circumstances, performance is likely to break down. I will outline two crude models that can account for why breakdown occurs. Within the context of Leslie’s (1987) model, the first of these could properly be considered as a failure of the decoupler system. The second one is better regarded as an instance of the output of that system being overridden by the output of other cognitive processes.
Stress-induced
processing
impairment or pertinence-based
response capture?
Mueller (1979) and Eysenck (1982) have argued that their work shows that anxiety leads to reduced or restricted processing of information. Under stress, encoding is less elaborated. Processing is not so deep as it would otherwise have been. In Leslie’s (1987) model, normal input passes through a perceptual system to central cognitive systems and thence to action systems. Pretence involves information passing through the perceptual and central cognitive subsystems to a decoupler system and then back to the central cognitive subsystems before being transmitted to the action systems. In other words, the processing involved in
Wishful
pretence
is deeper
than
that involved
in operating
on information
way. It seems likely that this is true not just for Leslie’s
thinking
157
in the normal
(1987) model
but also for
those that are less well specified than his. Asking people to pretend to act in a way that threatens their own beliefs at a time when they are concerned about the future of the institutions that safeguard those beliefs could reasonably be assumed to provoke anxiety. If Mueller (1979) and Eysenck (1982) are correct, this anxiety would tend to inhibit the deeper processing that subserves pretence. Mechanisms responsible for effective beliefdesire
reasoning
would
be bypassed
or rendered
temporarily
inoperative.
As a
result, people would behave as if they were in the state that Baron-Cohen et al. (1985) argue characterises autistic children. Breakdown in belief-desire reasoning occurs because they have temporarily lost their “theory of mind” (or whatever other structures subserve this ability). There is an alternative to stress-induced processing impairment. The normal processes subserving belief-desire reasoning may have still been operative but the conditions ambient in Experiment 2 may have led to their output being overridden in some way. When people decide to act on the basis of their own beliefs and desires, some potential response is selected. Let us suppose that, given a particular scenario, this selection is normally automatic (Schneider & Shiffrin, 1977). It cannot be inhibited and people have no conscious insight into the processes that underlie it. These processes need not be explicit: selection may be no more than the emergence of one particular response that possesses greater activation than any of its competitors. A selected response need not be executed but would be available to consciousness: people would specify it when describing what they would do on the basis of their own beliefs and desires. Let us suppose that an individual pretending to be some other target person carries out the requisite belief-desire reasoning in a way akin to that described by Leslie (1987). Representations of the target person’s beliefs and desires are copied into a cognitive subsystem (the decoupler) where they are manipulated to produce output that is passed back to the central cognitive systems. In particular, Leslie’s (1987) model specifies that output of the decoupler passes to central cognitive systems that include “systems for planning action”. The decoupler itself does not plan action. Response selection is part of the final common pathway shared by the primary processes involved in acting on one’s own beliefs/desires and by the decoupler that is responsible for emulating the actions of others by manipulating representations of their beliefs/desires. If response selection on the basis of true beliefs/desires is normally automatic and if the decoupler is being used to role-play someone who is in opposition to those beliefs, inputs to the action planning systems will be in conflict. This could be resolved by filtering out the inputs based on true beliefs/desires in much the same way that Broadbent (1958) suggested that non-attended inputs are filtered
158
out
N. Harvey
in dichotic
action
planning
listening
tasks.
systems
would
To produce be locked
successful into
belief-desire
the “decoupler
input from true beliefs/desires would be totally ignored. According very young and autistic people fail to pretend and deceive because yet developed
the executive
control
over their cognitive
processes
reasoning,
channel”
and all
to this model, they have not required
for the
gating of internal attention. Consequently, their action planning systems dominated by processing related to their true beliefs or to the physical world
are (cf.
Russell, Mauthner, Sharpe, & Tidswell, 1991; Wimmer & Perner, 1983). This is because information from these sources is always more salient than that based on representations of other people’s beliefs and desires. To account for the present results, this model could be modified in one of two ways. The notion that internal attention is switched either into a relevant or into an irrelevant source of information could be retained. If it is. then one has to argue that control over the switching mechanism can fail even in adults when salience of the information that has to be ignored is especially high. Alternatively, one could query whether this all-or-none switching model of internal attention is the most appropriate one. Just as important messages (e.g., one’s own name) in an unattended channel need to be able to capture external attention (Moray, 1959), so important information based on one’s own beliefs and desires may need to be able to capture the action planning systems from the decoupler. This would be so whenever the act of pretending becomes likely to interfere with acquisition of primary goals. So instead of totally ignoring input from true beliefs and desires, it would be better to weight it according to its importance or pertinence. The decoupler’s access to action planning systems would only be interrupted when this weighteil input exceeded that arising from the decoupler (cf. Deutsch & Deutsch, 1963; Norman. 1968). This would ensure that people could pretend to act like someone else but do not do so when it is more appropriate to act on the basis of their own beliefs. However. just as one can be inappropriately distracted by one’s own name in an unattended conversation, so the importance or pertinence of one’s own beliefs and desires may inappropriately interrupt or block the decoupler’s access to action planning systems. If pretence is likely to interfere with acquisition of primary goals based on true beliefs and desires, why indulge in it in the first place? One obvious answer to this is that there may seem to be little chance of interference when the pretence starts. The need for primary cognitive processes to recapture control of output from the decoupler may only become apparent later. At some social function, you may be amusing your colleagues with your hyperbolical role-playing of the idiosyncratic personal mannerisms of your head of department. However, when she walks through the door, it is politic to inhibit the pretence. If it is clear from the outset that pretence would interfere with acquisition of primary goals, then it should not be initiated in the first place. However,
Wishful thinking
predictions
of someone
else’s
behaviour
may
well
prove
useful
even
159
when
role-playing that behaviour is detrimental to one’s goals. Predictions must be allowed even when pretence is prohibited. However, this does not mean that the belief-desire reasoning underlying prediction of someone else’s behaviour has to be any different from that involved in role-playing their behaviour. The activity of the decoupler can be the same in the two cases. The distinction between them can be allowed to depend on the responses they activate in the action planning system. Prediction of a target person’s behaviour activates “Target will do X” whereas role-playing the target activates “Do X”. Only one of these response types is primed by the situation or experimental instructions. Furthermore, only the pretence response can conflict with one arising from true beliefs and desires. For example, “Do X” conflicts with “Do not do X” but not with “Target will not do X”. Thus, if a pretence response is required and if an opposite response associated with true beliefs and desires is weighted as highly important or pertinent then role-playing will be likely to fail. On the other hand. if the response associated with true beliefs and desires is not weighted so high or does not conflict with the pretence, then role-playing will succeed. Similarly, a prediction response cannot conflict with one associated with true beliefs and desires and so it will succeed. All this suggests that there are structural constraints that make it difficult for people to pretend to act in a way that conflicts strongly with their own true beliefs and desires. In Experiment 2, there was a failure of belief-desire reasoning among Conservative supporters because (a) pretence rather than prediction was required, (b) this pretence required them to act in a way diametrically opposite to how they would act on the basis of their own beiiefs and desires and (c) the political situation at the time and the directness of the potential threat to their politics ensured that the scenario was highly pertinent to those beliefs and desires. There was no similar failure among Labour party supporters in Experiment 1 because condition (c) did not hold. There was no such failure among people who supported neither main party in either experiment because condition (b) did not hold for them. This account leaves a number of questions open. The notion of importance or pertinence has not been rendered any more precise than it was originally when employed by Deutsch and Deutsch (1963) and Norman (1968). However, it seems reasonable to assume that pertinence weightings are activated by context (e.g., presentation of an experimental scenario) but that their size depends on both long-term and short-term considerations. In the experiments reported here, the long-term factors are likely to derive from outside the laboratory, to increase the weighting of responses associated with true beliefs and desires, and to do so more in Experiment 2 than Experiment 1. The short-term factors are likely to derive from the knowledge that the laboratory task was artificial and could not have any implications for events outside. The experimenter was not asking people
160
from
N. Harvey
threatened
parties
to pretend
that they were their opponents
when
answer-
ing real polls but when answering artificial ones. These short-term factors should have decreased the weighting of responses associated with true beliefs and desires. Stress-induced processing impairment and pertinence-based response capture are alternative ways of accounting for labile belief-desire reasoning ability. They can explain why normally efficient belief-desire certain circumstances. Put simply, the processing
reasoning becomes deficient in that underlies it must either be
temporarily de-activated or else remain active but have its output overwhelmed by that of competing processes. There is no a priori reason to suppose that belief-desire reasoning failure in infants or in autistic people is caused by either of these factors. However, evidence that belief-desire reasoning ability is labile (e.g., Sodian, 1991, Experiment 1) does suggest that one of them is responsible when it fails. They are worth considering even in the absence of such evidence.
Coda:
Methodological
issues
The experiments appear to provide support for the view that political allegiances are useful as tools for studying belief-desire reasoning in adults. However, there is one possible objection to the methodology adopted here: an exact replication of the results is impossible. The Conservative party is no longer deciding on a new leader. In one sense, this objection must be sustained. It is indeed impossible to perform a pure replication of my experiments starting from exactly the same initial conditions. There is no possibility of rejecting my results as reported in Tables 1 and 2. For example, it is not possible to show that the interaction of interest in Experiment 2 arose because data were contaminated by some uncontrolled factor of which I was unaware. I regard this as a problem but not a serious one. Although my results themselves cannot be rejected, the conclusions that I have drawn from them can be. The scenarios that I have used in the experiments should be thought of as tokens exemplifying certain types of situation that are of theoretical interest. For instance, the type of situation exemplified by the Experiment 2 scenario is one in which people are asked to pretend to act in a way that would pose a direct salient threat to their own true beliefs and desires. Other tokens of this same type of situation could be studied to test the conclusions that I came to. It is not necessary to wait for another British Conservative party leadership election. The democratic world abounds in opportunities for studying tasks that are similar to or are interesting variants of those used here. One example must suffice. At the time of writing, primaries for the American presidential election are imminent. I understand that, prior to a party’s state and national conventions, pollsters ask people whether they would be more likely to vote for that party’s candidate in the presidential election if person A or person B were to be chosen
Wishful thinking
161
by convention representatives, to fill that role. Opponents of the party can use this type of poll to send a misleading message to the representatives. They can say that they would be more likely to vote for the candidate who they think would be less likely to win the presidential election. They would hope that this action, in concert with similar ones by other opponents of the party, would lead the representatives to select the weaker individual as their candidate. This situation is very similar to the one used for the scenario in Experiment 2. Thus American primaries (and various other elections) provide opportunities for validating and extending my conclusions. For example, it should be possible to test the claim based on the pertinence-based response capture model that was outlined in the previous section: that predicting the behaviour of an opponent should be easier than pretending to act like one. Inability to carry out exact replications does not prevent progress in other sciences and in other areas of psychology. Astronomical experiments are designed to examine events that will not occur again but the phenomena of interest can still be studied by examining other events of the same type. Archaeological excavations cannot be repeated but sites of the same type can be studied to support the original conclusions. Within cognitive neuropsychology, pure replication of single case studies is not possible but many would argue that progress is (e.g., Caramazza, 1986). Psychologists working on “flashbulb memories” are interested in how people remember events of national or international significance (e.g., John F. Kennedy’s assassination, the space-shuttle disaster).” These events cannot be repeated but the way in which they are remembered is taken to be characteristic of other events of the same type (Brewer, in press; Conway, 1990). Examples such as these suggest that the possibility of performing exact replications, though desirable, is not essential to scientific progress.
References Astington, J.W., & Gopnik, A. (1991). Theoretical explanations of children’s understanding of the mind. British Journal of Developmental Psychology, 9, 7-31. Astington, J.W., Harris, P.L., & Olson, D.R. (1988). Developing theories of mind. New York: Cambridge University Press. Baron-Cohen, S., Leslie, A.M., & Frith, U. (1985). Does the autistic child have a “theory of mind”? Cognition, 21, 37-46. Baron-Cohen, S., Tager-Flusberg, H., Cohen, D., & Volkmar. F. (Eds.) (1991). Understanding other minds: Perspectives from autism. Oxford: Oxford University Press. Brewer, W.F. (in press). An analysis of theoretical and empirical studies of flashbulb memory. In E. Winograd and U. Neisser (Eds.), Affecr and accuracy in recall: Studies of ‘flashbulb memories”. New York: Cambridge University Press. Broadbent, D.E. (1958). Perception and communication. London: Pergamon. Byrne, R.W., & Whiten, A. (Eds.) (1988). Machiavellian intelligence: Social expertise and the evolution of intellect in monkeys, apes and humans. Oxford: Oxford University Press.
3A large-scale Conway, personal
study of flashbulb communication).
memories
of Thatcher’s
fall from
power
is under
way
(M.
162
N. Harvey
Caramazza. A, (1986). On drawing inferences about the structure of normal cognitive systems from the analysis of patterns of impaired performance: The case for single-patient studies. Bruin and Cognition. 5. 41-66. Conway, M. (1990). Autobiographical memory: An infroducrion. Open University Press: Milton Keynes. Deutsch. J.A.. & Deutsch. D. (lY63). Attention: Some theoretical considerations. Psychological Review. 70, 80-90. Eysenck, M.W. (1982). Atrenrion and urousul: Cognirion and performance. Berlin: Springer-Vcrlag. Fischhoff. B.. & MacGregor, D. (1982). Subjective confidence in forecasts. Journal of Forecasring. I. 155-172. Frith. U. (1989). Autism: Explaining Ihe enigma. Oxford: Blackwell. Frye. D.. & Moore. C. (1991). Children’s theories of mind. Hillsdale. NJ: Erlbaum. Harris. P.L. (1991). The work of the imagination. In A. Whiten (Ed.). Nunox theories of mind: Evolurion. development and simulation of elzeryday mindreading. Oxford: Blackwell. pp. 283-304. Harvey. N.. & Ayton. P. (1YYO). Actor-observer differences in judgmental probability forecasting of control response efficacy. Paper given at 31st Annual Meeting of the Psychonomic Society, New Orleans. Abstract in BuLletin of the P.sychonomic Sociery. 2X. 523. Hobson. R.P. (1990). On acquiring knowledge about people and the capactty to pretend: Response to Leslie (1987). Psychological Review,. 97, 114-121. Hobson. R.P. (1991). Against the theory of “Theory of Mind”. British Journal of Developmental Psychology, 9, 33-5 1. Langer. E.J. (1975). The illusion of control. Journal of Personality und Social Psychology. 32. 311-328. Leslie. A.M. (1987). Pretence and representation: The origins of “theory of mind”. Psychological Review, 94. 412-426. Leslie. A.M., & Frith, U. (1990). Prospects for a cognitive neuropsychology of autism: Hobson’s choice. Psychological Review. 97. 122-131. Moray. N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology. 9. 56-60. Mueller. J.H. (1979). Anxiety and encoding processes in memory. Personality and Social Psychology Bulletin, 5, 288-294. Norman. D.A. (1968). Toward a theory of memory and attention. Psychological Reviebl,. 75, 522-536. Perner. J. (1991). Understanding rhe represenrurional mind. Cambridge, MA: MIT Press. Premack. D.. & Woodruff. G. (1978). Does the chimpanzee have a theory of mind‘? Behuviorul and Bruin Sciences, 4, 515-526. Russell. J.. Mauthner. N.. Sharpe. S. 8; Tidswell. T. (1991). The “window task” as a measure of strategic deception in preschoolers and autistic subjects. Brirish Journal of Developmenful Psychology. 9. 331-349. Schneider, W.. & Shiffrin. R.M. (lY77). Controlled and automatic human information processing: I. Detection. search and attention. Psychological Review. X4. 1-66. Sodian. B. (1991). The development of deception in young children. Cognirion. 9. 173-188. Stich. S.P. (1983). From folk psychology IO cognirive science: The case against belief. Cambridge, MA: Bradford Books. Wellman. H. (1990). The child’s rheory of mind. Cambridge, MA: Bradford Books. Whiten, A. (Ed.) (1991). Natural theories of mind: Evolutron, development and simulurion of everyday mindreading. Oxford: Blackwell. Wimmer. H.. & Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognirion. 13. 103-128. Woodruff. G., & Premack. D. (lY7Y). Intentional communication in the chimpanzee: The development of deception. Cognifion, 7. 333-362. Wright, G., & Ayton, P. (1986). Subjective confidence in forecasts: A response to Fischhoff and MacGregor. Journal of Forecusring. 5. 117-123.