Dilemmas and bargains: Autism, theory-of-mind, cooperation and fairness.

Elisabeth Hill1* and David Sally2

1 Institute of Cognitive Neuroscience, University College London, UK.

2 Cornell University, USA.

* Corresponding author.

Institute of Cognitive Neuroscience

University College London

17 Queen Square

London

WC1N 3AR. UK.

Tel: +44 (0)20 7679 1177

Fax: +44 (0)20 7813 2835

Email: e.hill@ucl.ac.uk

KEYWORDS: autism, bargaining, cooperation, dilemmas, theory-of-mind

ABSTRACT

Mentalising is assumed to be involved in decision-making that is necessary to social interaction.

We investigated the relationship between mentalising and two types of strategic games - those

involving the choice to cooperate with another for joint gain or compete for own gain and those

involving bargaining and division of a surplus - in children and adults with and without autistic

spectrum disorders. The results suggest that strategic responses in the first type of game, the well-

known prisoner’s dilemma, are associated with mentalising ability. In contrast, generosity in

bargaining tasks did not depend upon mentalising skills, but initial strategically unequal offers

did. These two essential social games appeared to be differentially compensated for in high-

functioning individuals with autistic spectrum disorders.

2

INTRODUCTION

As crawling gives way to toddling and then striding, a child may move more steadily through the

physical world. So too, improvements in her ability to mentalise—that is, attribute, understand

and manipulate mental states such as beliefs, feelings, thoughts, intentions and deceptions—allow

her to navigate away from the home harbour and into the swift currents and crosswinds of the

broader social world. For individuals with autistic spectrum disorders there exists a fundamental

difficulty in mentalising and social life is a series of strong headwinds, uncertain tacks, buffeting

waves and treacherous eddies. Specifically, individuals with autism fail to understand not only

that others have minds, but also that other minds have different thoughts, and that behaviour is

determined by mental states. Thus, individuals with autistic spectrum disorders are considered to

lack a ‘theory-of-mind’ (Baron-Cohen, Leslie & Frith, 1985, Baron-Cohen, Tager-Flusberg &

Cohen, 1993; 2000).

While difficulties with theory-of-mind are widespread in individuals with autism, certain

of these individuals appear able to acquire some degree of theory-of-mind understanding.

Accumulating evidence suggests that this apparent understanding arises out of a compensatory,

rule-based ability that can reproduce the sympathetic insights of more direct and natural mind-

reading. In a meta-analysis, Happé (1995) reported that children with autism required a higher

level of verbal ability, as measured by the British Picture Vocabulary Scale, in order to pass

simple theory-of-mind tasks (the Sally-Anne and Smarties false belief tasks), than did normally

developing three- and four-year-olds or children with learning disabilities but without autism. It

may be that the high verbal ability of these individuals allowed them to ‘hack out’ a solution to

the tasks (Happé, 1995). Consequently, depending on a situation’s degree of social complexity,

the responses of such individuals with autism may be adequate and yet odd, or simply inapt and

inept.

3

The limitations of a postulated compensatory mechanism have been demonstrated in a

number of recent experiments. Bowler (1992) reported that adults with Asperger syndrome –

considered to represent high-functioning autism – were able to pass tests of false belief, although

they did so using a slow, cumbersome route, as evidenced by the explanations that they gave in

justification of their answers, which lacked higher-order, mentalising bases. Castelli, Frith, Happé

and Frith (2002) asked participants to provide verbal interpretations of animations of two moving

triangles. Each animation was scripted to show random, goal-directed or mentalising movements.

Compared to their normal peers, individuals with high-functioning autism or Asperger syndrome

made fewer and less accurate interpretations only of the animations that evoked mentalising, for

example when two triangles bounced up and down together in glee. A much more natural

interaction was viewed by participants in Klin, Jones, Schultz, Volkmar and Cohen’s (2002)

study. While their eye movements were tracked, these participants watched dramatic scenes from

a famous Hollywood movie. Normal individuals focused mostly on the eyes of the actors;

individuals with autism fixated mainly on the mouths. In contrast, when the scene showed a

character reaching for a gun, the eyes of the autistic individuals moved directly to the object

while those of the matched controls lingered on the actor’s face, presumably to gather clues about

his intentions. Thus, there is striking evidence that individuals with high-functioning autism read

minds differently from their normal peers. While their performance on standard laboratory tasks

of theory-of-mind (generally false belief) can be good, they are severely developmentally delayed

in acquiring such ability, produce unusual explanations of their theory-of-mind understanding,

perform poorly on advanced tests of theory-of-mind and do not show the same spontaneous

reactions to naturalistic task demands as their normal peers.

All the studies above placed the participants in the role of an observer of another person

or of a social interaction. Clearly, individuals with autism experience difficulties with mentalising

in this setting, but little has been documented empirically about the implication of this difficulty

4

in their daily interactions with others. What happens when the person with autism is not spectator

but participant, not outside but inside the interaction? The aim of the current study was to pursue

this question by investigating the performance of individuals with autistic spectrum disorders

using a series of well-known tasks drawn from the field of 'game theory'.

Since its formalization by von Neumann and Morgenstern (1944), game theory has

developed into one of the most important theoretical and empirical tools in the social and

biological sciences. It is at one and the same time a theory of tic-tac-toe, poker and draughts, and

of predator-prey confrontations, arms races and industrial conspiracies, a continuum resting on an

established, adaptable definition of “game.” A game is simply a rule-bound, multiple-agent,

interdependent decision-making problem (Gibbons, 1992). Formally, a game is fully described by

a set of players, which may include Nature or Chance, the strategies available to each player (i.e.,

sequences of moves allowed within the rules), and payoffs to the players that arise from the

particular combination of selected strategies. Games can be analysed to determine their

equilibria, sets of players’ strategies that are stable in the face of hypothetical or real changes in

tactics or the game’s rules. Formal equilibria serve as a predictive benchmark in a game: players

with adequate levels of mutual knowledge, self-interest and coherent decision-making should

arrive at an equilibrium. For example, tic-tac-toe (noughts and crosses) has two players, a set of

weakly dominant strategies involving an initial mark in the centre box, the payoffs representative

of pure conflict—either the players tie, or one wins and the other loses—, and equilibria entailing

only drawn games.

Tic-tac-toe, as any ten-year-old can testify, is a fairly trivial game, in part because the

optimal strategy is so easily learned and is largely impervious to the behaviour or characteristics

of one’s opponent. However, when the game becomes more complex and its strategy space

grows, optimal strategies are harder to calculate, errors are more readily made, and signals are

more easily sent. In these games, the ability to mentalise may help a player anticipate where the

5

other’s memory, information, and calculation are limited and what strategy she is likely to

employ. A well-drawn mental model may also distinguish when the counterpart’s surprising

move is a mistake or a bluff or a trap.

The participants in the complex games of poker and combat testify to the value of

mentalising. A high-stakes card player, whose winnings depend on his abilities to deceive his

competitors and to dispel others’ bluffs, claims (with a bit of grandiosity), “A man’s character is

stripped bare at the poker table. If the other players read him better than he does, he has only

himself to blame. Unless he is both able and prepared to see himself as others do, flaws and all,

he will be a loser in cards, as in life” (Holden, 1990). Similarly, leaders on the battlefield rely on

mental models of their counterparts to construct strategies and decipher tactical movements (see,

for example, the memoirs of Rommel (1953) and Grant (1999[1885]).

Noughts and crosses, poker, and combat are zero-sum games, in which a gain for one side

represents an equal loss for the other. Quite naturally, there exists a category of nonzero-sum

games in which the stakes are not fixed and players’ actions may raise and lower the total

available payoffs. Because they may, together or independently, create and destroy common

value in these settings, players have more shared interests in nonzero-sum games than, for

example, in poker. Nonzero-sum games can be further divided into coordination games in which

the interests of the participants are identical, and mixed-motive games in which their interests are

only partially aligned.

It is principally through mixed-motive games that researchers have analyzed social and

strategic interactions among humans, among other primates, birds and other vertebrates, insects,

plants, genes, corporations, countries, political parties, classes, voters, etc. In the laboratory, three

mixed-motive games in particular, the prisoner’s dilemma, the ultimatum game, and the dictator

game (each of which we will describe in detail below), have been employed to study varying

levels of cooperation and competition, concern for fairness, self-interest and altruism among

6

experimental participants. Thousands of studies have been conducted of these three critical

games, and participants have usually manifested a cooperativeness greater than that predicted by

the models of strict, rational self-interest historically prominent in both biology (natural selection

at the individual level) and economics (homo economicus).

It is not necessary to mentalise in order to play a mixed-motive game or cooperate within

such a game. For example, viruses competing to infect and reproduce in the same set of host cells

are “playing” (insensibly and non-mentally) a prisoner’s dilemma game in which an inability to

cooperate lowers the fitness of each phage (Turner & Chao, 1999). Furthermore, biologists have

discovered relatively sophisticated and accommodating strategies adopted by animals without the

cognitive ability to form mental models—rotating responsibilities for approaching possible

predators among stickleback fish (Milinski, 1987), reciprocal food sharing by vampire bats

(Wilkinson, 1984), and the coordinated capture of grasshoppers and moths by certain species of

spiders (Pasquet & Krafft, 1992). Evolutionary game theorists assume that the tactics of these

creatures are encoded in their genes, that con-specifics are paired with a certain frequency in

strategic interactions, and that fitness and offspring accrue to the genes and the individuals with

the more robust strategy. So, vampire bats share food reciprocally because repeated interactions

among colony-mates make that strategy more “profitable” than one based on always taking food

and never giving. These bats are not analysing, rationalising, mentalising, or improvising: they

are simply following an established evolutionary rule in a familiar situation.

Nevertheless, among humans mixed-motive games do seem to require that players signal

and interpret intentions and develop some theory of the other’s mind (Schelling, 1960). In the

parlance of economics, then, mentalising and rule-following are substitutes in the same way that

butter and margarine, aluminium and tin, work and the national lottery, cars and public

transportation, and wisdom and information are. Few substitutes are perfect for all applications.

A chef may be indifferent about which spread is applied to her morning toast, but would refuse to

7

use oleo in even the simplest pastry, as butter makes a distinctly better crust. Similarly, the key

research question we raise is, how easily do mentalising and rule-following substitute for each

other in mixed-motive games? Does a well-functioning theory-of-mind promote cooperation,

generosity, and fairness, or does it foster treachery and selfishness? Are there natural rules that

mimic the effects of forming a mental model of the counterpart?

In the current study these questions were addressed through a unique, cross-disciplinary

approach—testing the basic games of the social and biological sciences on adults and children

with autistic spectrum disorders to focus on whether: (i) mentalising is necessary for basic

strategic rationality, (ii) mentalising promotes cooperation in the prisoner’s dilemma, (iii)

mentalising allows similarities and differences among related games to be more accurately

perceived, and (iv) rule-following yields the same results as mentalising in bargaining games

when the problem is dividing a given endowment either unilaterally or collaboratively.

Experiment One – prisoner’s dilemma

Without question, the prisoner’s dilemma (PD) is the most thoroughly studied game in the

social and biological sciences because it captures the essence of a frequent social quandary,

namely, that what is good for the group may differ from what is good for each individual. This

game is central to such social phenomena as foraging, over-harvesting of resources, inadequate

investment in public goods, free-riding, retaliation, sacrifice, altruism, etc. The PD involves two

players acting simultaneously, two moves characterized as “defection” and “cooperation,” and

payoffs such that the total return to mutual cooperation is greater than that to mutual defection,

while an individual’s personal payoff is always maximized by defection. A generic PD is shown

in Figure 1, with the payoffs to combinations of moves contained in the appropriate cell of the

matrix. If the other side cooperates, a player is enticed by the temptation to defect

8

(a-c); if the other side defects, cooperation is penalized by the sucker’s loss (b-d). In both

instances, defection guarantees a larger payoff, and so, the only equilibrium is (defection,

defection).

[Figure 1 about here]

Despite the gloomy equilibrium of mutual defection and the predicted triumph of narrow

self-interest over altruism, humans and other creatures often “solve” the PD and achieve mutual

cooperation. Theorists have distinguished two principal means which aid mutual cooperation:

kinship and reciprocity. It makes evolutionary sense for close kin to sacrifice on each other’s

behalf in order that their shared genes may prosper, and so brothers and sisters, rather than third

cousins, will cooperate in the PD (Hamilton, 1964). Kin recognition depends on cognitive

mechanisms able to read and react to perceptible clues of relatedness (Krebs, 1987), and vestiges

of these prehistoric mechanisms translate into our modern predilection for people who are

proximate, similar and familiar (Sally, 2000). Hence, in laboratory experiments, cooperation is

more frequently seen when participants are near each other, can see each other, have information

that they share tastes and opinions, like each other, and have interacted previously (Sally, 1995;

2000). Finally, if the PD is repeated, a population of reciprocal altruists whose strategy is based

on doing this round what the opponent did the previous round, can multiply and survive by

reacting harshly to full-time defectors and nicely to full-time cooperators and to each other.

Axelrod (1984) conducted a famous computerized tournament of repeated PDs in which the

winner was this so-called tit-for-tat strategy.

Tit-for-tat (TFT) is a simple rule that requires absolutely no mentalising. One unilaterally

cooperates in the initial round and then observes what move the counterpart actually makes so

that it can be reciprocated in the next round. Intentions play no role here. Moreover, this type of

player plays TFT with his brother and with a stranger, with a fellow club member and with an

enemy, with a counterpart who promises to cooperate and with one who vows to defect. Such a

9

rule-follower is relatively unaffected by changes in the social context of the game and is more

likely to miss other changes in a game that occur below the surface. By contrast, kinship-based

altruism is more closely linked to mentalising, both because the sharing of mental states is a sign

of similarity and relation, and because the presence of those who are close, similar, and familiar

is more likely to trigger an individual’s theory-of-mind (Sally 2001). Hence, the identity of the

opponent makes a significant difference in the likelihood that a player chooses to cooperate. If

the opponent’s identity is changed from human to machine, even though the pre-programmed

strategy is unaltered, experimental participants are far more likely to cooperate with the former

rather than the latter: 58% versus 41% in Abric and Kahan (1972), and 59% versus 31% in

Kiesler, Sproull and Waters (1996). These two results suggest that the more “human” a

counterpart is to a player, the more likely that player is to cooperate in the PD.

The degree of “humanness” seen in the opponent is a function not only of the opponent’s

identity but also of the identity and cognitive abilities of the perceiver. Chief among these

abilities is mentalising. Insofar as theory-of-mind develops throughout childhood (Astington,

1994), one would expect, then, that older children would cooperate more than younger ones.

Indeed, Fan (2000) found that nine- and eleven-year-old children cooperated more frequently in a

ten round repeated PD than did seven-year-olds. This study was less concerned with accounting

for the choices of children than with the effects of moral suasion and instruction on the promotion

of cooperation in the PD (see also Matsumoto, Haan, Yabrove, Theodorou & Cooke Carney,

1986). The only other study of young children (aged 3-10 years) and the PD that we found

showed that older children were more able and willing to pay attention to their opponent's

interests than younger children, although this was only true if doing so would help improve their

own outcome (Perner, 1979). Other developmental theories and experiments have shown that the

frequency of prosocial behaviour increases throughout childhood (Eisenberg & Fabes, 1998). In

the PD, cooperation, as opposed to defection, is clearly the prosocial move.

10

The experiments adopted in the current study allowed a comparison of patterns of choices

among younger participants and adults playing the same games. Specifically, the performance of

normally developing adults and children (aged 6-, 8- and 10 years) as well as high-functioning

children and adults with autistic spectrum disorders was investigated on three versions of a PD

game. In the PD games the nature of the opponent against whom a participant played was

manipulated - each participant played the game against a human and a computer opponent.

Furthermore a third manipulation of the PD was included in which participants played against a

human opponent where the instructions of the game encouraged participants to cooperate, rather

than compete with their opponent (encouraged cooperation PD).

Accordingly, based on the current understanding of theory-of-mind within developmental

psychology, a number of hypotheses about the decisions of our participant groups were

generated. (It is important to note that these very same hypotheses arise from the theories and

empirical results most frequently cited within behavioural economics.) With respect to the normal

sample a greater degree of cooperation in the PD among the older participants was expected.

Given the difficulties of individuals with autistic spectrum disorders in the area of mentalising,

and the suggestion that mentalising is involved in a participant's choice on the PD, it was

predicted that individuals with autistic spectrum disorders would cooperate less with their

opponent than their normal counterparts when playing against the human opponent. Since

individuals with autism have been shown to have difficulty only representing the “mind” of a

human and not a machine (Leekam & Perner, 1991; Leslie & Thaiss, 1992), a smaller difference

in the cooperation rates between the normal and autistic individuals when playing the PD against

a computer opponent was expected. With respect to the encouraged cooperation game, the

specific instructions negate the need to predict the intentions of the counterpart, and so,

individuals with and without autism should manifest similar, elevated rates of cooperation.

11

METHOD

Participants. A total of 99 children and adults took part in the study, comprising individuals with

and without autism and falling into the following groups: normally developing 6-, 8- and 10-

year-olds, normal adults, children with autism, adults with autism. Participants were tested

individually by an experimenter and - for the experimental tasks – a confederate either in their

school, college or at the Institute of Cognitive Neuroscience, UK. All were native English

speakers. Participant details are shown in Table 1. The mean chronological age of the adults with

autism versus those without did not differ; this was true as well for the 10-year-olds versus the

children with autism.

All participants with autism had been diagnosed formally with either autism or Asperger

syndrome prior to the study. In addition, a checklist was completed by the experimenter and the

confederate for all adult participants and approximately one third of the child participants. This

checklist was based on observation and related to the key characteristics of the disorder and was

used as an aid to confirm an individual’s diagnosis and therefore the results will not be reported

here.

Ethics approval for the study was granted by the National Autistic Society (UK) and by

University College London (UK). Parental consent was required for child participation in the

study and the informed consent of all participants (adults and children) was sought.

General ability. General ability levels were assessed using the British Picture Vocabulary Scale

(BPVS-II, Dunn, Dunn, Whetton & Burley, 1997) for the children and the Wechsler Adult

Intelligence Scale (WAIS-III-R; Wechsler, 1998) for the adults. Mean ability levels fell within

the normal range in all groups (see Table 1).

[Table 1 about here]

12

Theory-of-mind. First-order false belief understanding was assessed in all participants using the

Sally-Anne task (Baron-Cohen et al., 1985; Wimmer & Perner, 1983). Second-order false belief

understanding was assessed using the Coat story (Bowler, 1992) for the adult participants and the

Birthday Puppy story (Sullivan, Zaitchik & Tager-Flusberg, 1994) for the children. Wider aspects

of theory-of-mind were assessed in the adults using Happé’s (1994) eight mentalising stories.

These were scored for accuracy by two independent raters who resolved disagreements by

discussion. The total time taken to read and respond to each story was also recorded. Theory-of-

mind performance is shown in Table 2.

As expected, children with autism were impaired in both first- and second-order false

belief tests in comparison to their normally developing peers [first-order false belief, χ2 (1) =

18.62, p < .001; second-order false belief, χ2 (1) = 9.87, p < .01]. The adults with autism were

impaired in second-order false belief understanding [χ2 (1) = 20.0, p < .001] as well as on the

wider aspects of mentalising as assessed by the Happé stories [accuracy, t (28) = -4.35, p < .001;

time taken when accurate response given, t (19) = 2.88, p < .01]. The lack of a deficit in first-

order false belief understanding in the adults with autism in comparison to the normal adults [χ2

(1) = 2.14, p > .1] is not unexpected given the age and high-functioning status of the adults with

autism in our sample. Compensatory learning in these individuals allowed them to pass the most

simple false belief task, a consistent finding in the published literature (Bowler, 1992; Happé,

1995). Of greater significance is the large number of these adults who failed the second-order

false belief task, as well as their poor performance (both in accuracy and slowed responding) on

the advanced mentalising task. While the performance of the children with autism appeared

superior to that of the adults with autism on the second-order false belief, this was likely to be a

consequence of the differing content of the two stories used to assess this ability in the two age

groups (Bowler, 1992).

13

[Table 2 about here]

Game theory tasks.

The baseline and experimental game theory tasks were presented on a laptop computer, easily

visible to both the participant and the confederate who sat side-by-side (the participant to the

right). Participant responses were recorded on-line for later analysis. The instructions for each

task were presented on the computer, although these were verbalised by the experimenter to

ensure that participants (especially children) understood each task. In all tasks, players were told

that they must try to win as many points as possible and that the total points won on all games

would be exchanged at the end of the testing session for stickers (children) or chocolates (adults).

The greater the number of points won, the greater the reward at the end of the test session.

Baseline tasks. Two baseline tasks were administered in the order described below, before the

experimental tasks, thereby ensuring that the principle of the experimental tasks and the response

methods were familiar to participants.

Show me a colour. A blue and yellow square were shown on the computer screen. Each player

(participant, confederate) independently chose either colour having been told that if they both

chose blue they would both win three points, if they both chose yellow they would both win one

point, and if they each chose different colours they would both win one point. This information

was also presented on the tabletop in front of the players throughout the game, thus ensuring that

players had no need to memorise the points allocation. A partition divided the keyboard down the

middle, with each player responding by pressing one of two keys on one side of the partition. In

this way neither player was able to see their counterpart's response. After players had made their

choices by pressing the appropriate key on the keyboard, the computer presented the choices

made and the number of points won. The confederate always made the rational choice – blue – in

14

this game. Only those participants who responded rationally on this task would be allowed to

continue with the full test battery.

Coin flip. The confederate was not a player in this game, rather the participant ‘played’ against

a coin. A circle and a triangle were shown on the computer screen. The participant chose either

shape by pointing and clicking a mouse, following which the coin was flipped. Points were then

awarded according to the combination of shape chosen and the fall of the coin such that if the

participant had chosen triangle and the coin landed on heads they were awarded three points, or

tails one point. If the participant had chosen circle and the coin landed on heads they were

awarded four points, or tails two points. The equilibrium choice for the participant in this game

against chance is circle, as the payoff is larger in both instances. The intention here was to have

participants play a somewhat “mechanical,” one-sided version of the prisoner’s dilemma.

The Prisoner’s dilemma. Three versions of a prisoner’s dilemma task were completed by

participants, with the response mode being the same (i.e., keyboard) as that in the ‘show me a

colour’ baseline task. Sixteen trials (or rounds) were completed in each version of the task

although participants were not told explicitly that this would be the case. This allowed

comparison of participants' strategy choice on the first round of the game as well as over all

sixteen rounds of each game.

Human opponent. A circle and triangle were shown on the computer screen. Each player

(participant, confederate) independently chose either shape having been told how the points

would be awarded (see Table 3). This information was outlined to the players verbally along the

lines described in the 'show me a colour' baseline task and was also presented on the tabletop in

front of the players throughout the game. This ensured that players had no need to memorise the

points allocation. After players had made their choice by pressing the appropriate key on the

keyboard, the computer presented the choices made and the number of points won. The

15

confederate followed a tit-for-tat (TFT) strategy, always making the cooperative choice – triangle

– on the first round and then copying the participant’s strategy on the previous round for rounds

2-16. The participant was not told that there would be more than one round of the game.

Computer opponent. This version of the task was designed to investigate whether there was a

difference in spontaneous strategy used when playing against a human or computer opponent.

The task was identical to that described above with the computer replacing the human opponent.

The computer was programmed to respond using the same strategy as the human opponent

(cooperation on the first round followed by TFT), although this information was not give to

participants. When playing the human and computer opponent PD games, defection (i.e.,

competition) is the equilibrium choice. Empirically, a wealth of studies and broad reviews of the

literature (Dawes, 1980; Sally, 1995) suggest an expected cooperation rate of approximately 20%

in the first round and then approximately 40% overall as the repeated nature of the tasks is

implicitly recognized. In both games a greater degree of competition than cooperation would be

expected, especially when the opponent is a computer.

Encouraged cooperation. This task was identical to the prisoner’s dilemma with human

opponent except that the points won on each round of the game were combined for the two

players and divided equally at the end of the game. As before, the participant and confederate

made their choice independently by pressing the appropriate key on the keyboard, the computer

then presented the choices made and in this game the total points won by the two participants

combined were also shown. The confederate continued to follow a TFT strategy, always making

the cooperative choice of triangle on the first round of the game and then copying the

participant's strategy on the previous round for rounds 2-16. When both players chose triangle,

six points were won collectively; circle, four points were won collectively; and in cases where

each player chose a different shape, five points were won collectively. Note that with this payoff

structure, cooperation becomes the dominant move: irrespective of what the other chooses, a

16

player gains more points by cooperating. However, this dominance is somewhat obscured by the

visible payoff matrix and the confederate’s continued retaliatory response to the participant’s

choice of circle, and hence, the game is akin to a bluff—it is a PD on the surface and “show me a

colour” or “coin flip” in reality. Mentalising players should shift to a purely cooperative strategy,

but rule-following players may perseverate and not react to the different payoff calculation.

[Table 3 about here]

The order of play against the human and computer opponents was counterbalanced across

the participants, with the encouraged cooperation task being completed last in all cases.

Following completion of all three versions of the task, a semi-structured interview was conducted

to elicit information about a participant’s strategy in each game, how players distinguished

between the human and computer opponents and whether they had identified their opponent’s

strategy. A detailed analysis of the content of the semi-structured interviews is presented in Hill,

Sally and Frith (in preparation).

RESULTS & DISCUSSION

Baseline tasks.

Show me a colour. The rational choice in this game was blue. A small number of

participants, coming from all groups, selected yellow (9.1% of participants with autism, 19.7% of

normal participants), and they were asked why they had made this choice (always saying yellow

was their favourite colour). They were then asked which colour choice they would make if

thinking only of the points to be won. All participants responded blue to this question. They were

therefore considered to have responded rationally and were all included in the full test battery.

Coin flip. The equilibrium choice in this game was circle. Surprisingly, there was a

significant group difference in the choice of shape made on this task, as 24.2% of participants

with autism made the dominated choice of triangle, compared to only 9.1% of the normal

17

participants [F (1, 98) = 4.25, p < .05]. Choosing the triangle is clearly a mistake in this game and

must be driven by a rather deep-seated confusion in the face of the uncertainty of the unflipped

coin. (Note that through the verbalized instructions the experimenter had insured that the

participant had a working understanding of the payoff matrix.)

While this was an unexpected finding it could concur with one other set of findings in the

literature. Imaging studies have revealed that the quandary of guessing and predicting in an

uncertain situation triggers neural activity in the prefrontal cortex (Elliott, Rees & Dolan, 1999;

Paulus et al., 2001). Moreover, damage in this region leads to disadvantageous and erroneous

decisions (Bechara, Damasio, Tranel & Damasio, 1997). Imaging studies of the brains of those

with autism have shown abnormalities in the prefrontal cortex (Abell et al., 1999; Stone, 2000),

and these malformations may underlie a similar pattern of overt knowledge of the correct strategy

(circle, in this instance) but flawed choice (triangle).

Much future work is needed to examine the relationship between autism and uncertain

choice and to determine if the significant difference reported here is anomalous. Whether this

confusion in the face of uncertainty is related to either cooperative or generous behaviour in the

current study will be considered below.

The Prisoner's Dilemma.

Given the lack of PD studies comparing the strategy of normal children to that of adults,

the data for the normally developing groups (adult, 6-, 8- and 10-year-olds) will be presented

first, followed by comparisons of the responses of the normal participants and those with autism.

The mean number of cooperative responses across the 16 rounds of each PD task and the

percentage of each group cooperating on the first round of each task is shown for the normal

adults, 6-, 8- and 10-year-olds in Figures 2a and 2b respectively. A repeated measures ANOVA

with one between factor (participant group) and one within factor (PD game type) was applied to

18

the data for cooperative responses shown in Figure 2a. For performance across the 16 rounds of

each game, there was a significant effect of game type [F (1,61) = 81.25, p < .001], indicating

reliably higher levels of cooperation in the encouraged cooperation version of the task in

comparison to both the human- and computer-opponent versions. This finding indicates that the

premise of each task manipulation was supported. There was no significant difference between

the level of cooperation seen across the groups [F (3,61) = 1.12, p > .1]. As indicated in Figure

2a, there was a significant interaction between group and game [F (3,61) = 4.7, p < .01]. This

interaction shows that the performance of the normal adult group corresponded to the predicted

behaviour far more distinctively than that of the child groups, thereby indicating a developmental

progression towards levels of cooperation being distinguished according to the social situation.

Thus the task manipulations to encourage cooperation were less effective in this sample of 6- to

10-year-olds than in adults, and the majority of the children failed to adjust adequately their

strategy to account for the dominance of cooperation in the third game.

[Figure 2 about here]

Fan (2000) found, in his sample of Taiwanese schoolchildren, that nine- and eleven-year-

olds cooperated more frequently in a ten-round PD than did seven-year-olds. In the tests reported

here (Figure 2a), the ten-year-olds cooperated more with the human opponent than did the six-

and eight-year-olds, although the difference between the normal children was not statistically

significant.

The same overall cooperation rate could disguise a number of very different patterns. For

example, cooperation could start off very high initially and then decay, or distrust could be

prevalent at first with mutual cooperation emerging as TFT strategies are identified and

synchronized. Moreover, given that the participants were never told explicitly about the number

of rounds, cooperation in the first round was less likely to be motivated by reputation or

reciprocity. The first round decisions of participants in each task were examined to determine if

19

age was correlated with a distinct opening move in a given game. In the case of the 6- and 8-

year-old children, there were no significant differences in first moves between any pairing of the

PD tasks, significantly more 10-year-olds cooperated in the encouraged cooperation task than

when playing against the computer opponent [Z = -2.0, p < .05], but not when playing against the

human opponent. The current study, therefore, supports Fan’s (2000) conclusion that young

children are more likely to compete as a default strategy irrespective of task manipulations and

that with age, levels of competition can be reduced. Of interest will be whether there is a gradual

transition from competition to cooperation (when it is appropriate) with age or whether there is a

sudden switch from a default strategy of competition to one that is moulded around the demands

of the situation. The data reported here are suggestive of the former view.

If theory-of-mind promoted only cooperation in these games, one would expect adults to be

even more accommodating than children. If, however, strategic defection can result from

mindreading, and if such perfidy is strongly countered by the overtly prosocial norms of

childhood and less opposed by the more pragmatic norms of adulthood, then adults may be more

competitive than children. The tests reported above revealed that the adults were not uniformly

more or less cooperative than the children. When facing a human counterpart in a competitive

situation, all ages were equally competitive [t (63) = .21, p > .1]. However, this was not the case

when playing the computer opponent and in the encouraged cooperation game. Adults competed

significantly more than the children (6-10 years combined) in the former game [t (63) = 2.4, p <

.02] and cooperated significantly more in the latter game [t (63) = -4.67, p < .001].

This pattern among the average rates of cooperation is based on significant differences in

the underlying distributions: out of fifteen adults, six did not cooperate in a single round with

either the human or computer counterpart, while among the fifty normal children, only five were

uniformly uncooperative. We analysed this difference by categorising individual participants as

either ‘reliably competitive’ (0-1 trials of cooperation), ‘reliably cooperative’ (15-16 trials of

20

cooperation) or ‘variable’ (2-14 trials of cooperation) on each Prisoner's dilemma game. The

results of this allocation can be seen in Figure 3, in terms of the percentage of each group who

fell into each category. Group performance (adults versus all children together) was compared

using a series of chi-square tests for each game separately. This analysis revealed significant

differences between the degree of cooperation between the normal adults and normal children in

all three versions of the Prisoner’s dilemma game. This difference was particularly striking on the

encouraged cooperation game [human opponent, χ2 (2) = 6.84, p < .05; computer opponent, χ2

(2) = 7.49, p < .05; encouraged cooperation, χ2 (2) = 24.46, p < .001]. These differences reflect

the fact that the normal adults were reliably competitive on the human and computer opponent

games and reliably cooperative on the encouraged cooperation game. In contrast, children were

predominantly variable in their strategy on each game.

[Figure 3 about here]

Finally, while there was no statistically significant difference among the normal

participants in terms of the number of people cooperating on the first round of either the human

and computer opponent games of the PD [χ2 (3) = 4.29, p > .1 and χ2 (3) = 5.61, p > .1

respectively], there was a significant difference on the encouraged cooperation PD game [χ2 (3) =

11.08, p < .01]. This difference arose because significantly more of the adults cooperated than

each of the child groups (6-, 8-, and 10-years), who did not differ from each other on this

measure. Adults initiated cooperation significantly more frequently in the encouraged

cooperation task than in both the human and computer opponent versions [Z = -3.21, p < .001 and

Z = -3.21, p < .001, respectively]. None of the children adjusted fully initially or in later rounds

to the ersatz PD—they did not recognize that defection was dominated by cooperation.

By comparing the choice made on the last round of the second PD game played (either

against the human or computer opponent) with the choice made on the first round of the

21

encouraged cooperation version of the PD (rounds 32 and 33 respectively), the extent to which

individuals recognised the switch in the rational choice to be made according to the new task

demands was considered. The data were investigated for each group separately using a series of

Wilcoxon tests to compare the level of cooperation on each pairing of the PD games (see Table

4). There was no significant difference between the choice of shape made in rounds 32 and 33

when the 6-, 8- and 10-year-olds were considered separately [6 years, Z = -1.63, p > .1; 8 years, Z

= -1.73, p > .1; 10 years, Z = -1.0, p > .1], although when combined the children showed a

significant difference between the choice of shape in the two rounds [Z = -2.5, p < .01]. By

contrast, there was a large and significant difference between the choice of shape made in rounds

32 versus 33 in the adults [Z = -3.16, p < .002].

[Table 4 about here]

Having established the existence of a developmental pathway to patterns of cooperation and

competition, the autism samples were then added to the analysis and comparisons made between

the two adult and child groups (normal age groups collapsed) separately for the degree of

cooperation across all 16 rounds of each PD game as well as on the first round of each game only

(see Figure 2). For performance across the 16 rounds of each game, there was a significant effect

of PD game in the comparisons of both the adult and child groups [Adult, F (1,28) = 47.2, p <

.001; Child, F (1,66) = 31.24, p < .001], indicating significantly higher levels of cooperation in

the encouraged cooperation version of the game in comparison to both the human and computer

opponent versions irrespective of participant group. This finding indicates that the premise of

each task manipulation was supported not only in the control, but also in the autistic sample.

There was no significant difference between the level of cooperation seen across the groups

[adults, F (1,28) = .26, p > .1; children, F (1,66) = .09, p > .1], nor a significant interaction

between group and game [adults, F (1,28) = 2.9, p > .1; children, F (1,66) = 1.56, p > .1].

22

Investigating first round choice for the individuals with autism was achieved through the

use of a series of Wilcoxon tests comparing each combination of the Prisoner’s dilemma games

for the adults and children with autism separately. In the case of the adults with autism,

performance reflected the pattern seen in the 10-year-olds in the current sample, with

significantly higher numbers of participants cooperating on the first round of the encouraged

cooperation version of the task in comparison to the computer opponent version only [Z = -2.11,

p < .05]. The level of cooperation between the human and computer opponent and human

opponent and encouraged cooperation games did not differ [human versus computer opponent, Z

= -1.41 p > .1; human opponent versus encouraged cooperation, Z = -1.67, p > .1]. There were no

differences in the degree of cooperation elicited by the children with autism in any pairing of

games, as seen in the profile of the 6- and 8-year-olds in the current sample. These findings

suggest that some degree of appropriate cooperation is seen in adults with autism but is not yet

present in our sample of children with autism, whose chronological age was similar to that of the

10-year-old normally developing children. Thus at the very least the children with autism must be

described as showing a delay – or immaturity – in their development of appropriate levels of

cooperation.

Were there distinctions in the profiles of play over all sixteen rounds of the game? These

data are displayed in Figure 3. Unlike normal adults, many more of whom were reliably

competitive against the human and computer opponents than were the normal children, the

profiles of the adults with autism were less distinguishable from those of the children with autism

across all three games. Group performance (autism adults versus normal adults) was investigated

using a series of chi-square tests for each game separately. This analysis revealed significant

differences in the degree of cooperation between the two adult groups in the human opponent [χ2

(1) = 6.38, p < .05] and encouraged cooperation [χ2 (1) = 6.93, p < .05] versions of the PD game,

23

but not when playing the computer opponent [χ2 (1) = 1.26, p > .1]. The difference between the

groups was of a smaller magnitude on the encouraged cooperation game than seen in the normal

sample, who also differed on the computer opponent version of the PD game.

What of the performance of the individuals with autism when required to switch from the

likely choice of defection (circle) to that of cooperation (triangle) between the last round of the

second PD game (human, computer opponent) and the first round of the encouraged cooperation

PD? Of interest was whether the individuals with autistic spectrum disorders would be able to

make this switch easily and entirely, as did the normal adults, or fitfully and partially, as did the

normally developing children. According to our prediction that individuals with autism would

cooperate less than their normal peers, we would expect that individuals with autistic spectrum

disorders would be less able to switch from a competitive to a cooperative strategy between the

last round of the competitive versions of the PD and the first round of the encouraged cooperation

version. We investigated this through the use of a series of Wilcoxon tests to compare the level of

cooperation on the two critical rounds of the relevant PD games (see Table 4). For the children

with autism, there was no significant difference between the choice of shape in the two rounds

under consideration [Z = .0, p > .1]. For the adults with autistic spectrum disorders, the choice

approached significance [Z = -1.9, p = .058]. It is conceivable that this difficulty in switching

strategies and responding to the new payoff matrix is related to a general tendency in autism

toward perseveration.

Comparison of cross-group performance, and specifically whether the individuals with

autism (at least the adults) cooperated significantly less when playing against the human

opponent than their normal peers was considered. There was no significant difference between

the number of cooperative responses across the sixteen rounds of the prisoner's dilemma played

against the human opponent in the two adult groups [F (2,49) = 1.94, p > .1]. Thus, contrary to

predictions, the behavioural data suggest that the adults with autism did not cooperate less than

24

their normal peers. In fact, they showed a tendency to cooperate more when playing both against

the human and the computer opponent, although the difference between the two adult participant

groups was less in the computer opponent version. While this was not a significant difference, it

accords well with previous studies suggesting that individuals with autism have less difficulty

representing the "minds" of machines such as cameras than they do the minds of other human

beings (Leekham & Perner, 1991; Leslie & Thaiss, 1992).

The percentage of each group cooperating on each round of each Prisoner's dilemma game

is shown in Figure 4. A repeated measures ANOVA with one between factor (group) and two

within factors (Prisoner’s dilemma game; round) was applied to the data for the number of

cooperative moves, first for the normal groups (adults versus all children together). There were

significant effects of group [F (1,62) = 4.08, p < .05], game [F (1,62) = 99.64, p < .001], and

significant interactions of group by game [F (1,62) = 16.19, p < .001] and game by round [F

(1,62) = 4.58, p < .05]. The first three significant effects have been described previously. The

game by round interaction reflected a more similar profile of cooperation in the encouraged

cooperation and computer opponent games than in the human opponent game. In the latter game,

normal children competed progressively more over the course of the sixteen rounds while normal

adults sustained a more even rate of cooperation.

[Figure 4 about here]

The trial-by-trial decisions of the individuals with autism were compared to those of their

normally developing peers for the adult and child groups separately. For the children, there was a

significant effect of game [F (1,65) = 30.01, p < .001], described previously. The normally

developing children appeared to distinguish more between playing the human opponent than the

computer opponent or encouraged cooperation games, highlighted by the less cooperative

strategy that they adopted. This performance profile was not evident in the children with autism.

There was a significant interaction between game and round [F (1,65) = 11.5, p < .001]. Once

25

again, this interaction reflected a more similar profile of cooperation in the encouraged

cooperation and computer opponent games than in the human opponent game.

For the adults, there was a significant effect of game only [F (1,28) = 54.15, p < .001],

described previously and indicating that there were no significant differences between the adults

with autism and the normal adults in terms of cooperative choices across the 16 rounds of each

Prisoner’s dilemma game. Figure 4 shows, however, that the adults with autism were consistently

less cooperative on the encouraged cooperation game and generally less competitive when

playing the other two games, particularly against the human opponent. Thus they were less

influenced by the task demands than their normal peers and this neutrality may reflect the

application of a particular, pre-set decision rule or pattern across all the games.

Furthermore, the responses of the adults with autism in the semi-structured interview

reflected the special status of the human opponent PD, indicating that where the adults with

autism appeared to produce similar patterns of cooperation as the normal adults, this arose out of

their knowledge that they needed to predict the workings of their opponent’s mind, were

generally unable to do this spontaneously and thus needed to rely on rule-based methods. This

was indicated by many of the adults with autism, for example, "I've gotta put a mental state in

Sarah [the confederate], think what she was thinking" (a 32-year-old woman with autism). This

was the nature of the response for many of the adults with autism for the PD tasks, the Happé

stories as well as in their daily lives. The normal adults did not make such statements during their

interviews.

Taken together these results show that although there was no significant difference between

the degree of cooperation elicited in the individuals with autism and controls when all rounds of

each game were taken together, more subtle differences were evident, specifically in the degree

of cooperation between the last round of the human or computer version of the game (whichever

was played second) and the first round of the encouraged cooperation PD, where a lack of

26

switching between a competitive to a cooperative strategy was evident in the individuals with

autism, especially the children. With respect to the human and computer opponents, older

children cooperated initially more than did younger children and normal adults, while neither

group of children nor adults with autism fully reproduced the level and pattern of competition

manifested by normal adults. We can now turn to an explicit analysis of the role of mentalising in

fostering or retarding cooperation in these games.

In order to establish whether there was a relationship between mentalising ability and

choice of strategy when playing each version of the prisoner's dilemma, a comparison was made

between performance on the second-order false belief test and the degree of cooperation

evidenced in each game for the child participant groups. All members of the normal adult group

passed this task, and thus such a comparison was not warranted between the adult groups. An

alternative comparison based on performance on the Happé stories was used to investigate any

relationship between mentalising and degree of cooperation in the adult groups, and is reported

below.

To ascertain whether a relationship between mentalising ability and strategy choice

existed in the children, a repeated measures ANOVA with two between factors (group; second-

order false belief performance) and one within factor (PD game) was applied to the data for

number of cooperative moves (children with autism versus all normally developing children

together). There was a significant effect of game [F (1,62) = 19.72, p < .001], as described

previously. The effect of false belief approached significance [F (1,62) = 3.72, p = .058],

suggesting that second-order false belief passers had a tendency to be more cooperative than

second-order false belief failers, irrespective of whether or not an individual was diagnosed with

autism. A child who failed this false belief test presumably also had a tough time understanding

what the counterpart’s beliefs about the child’s own intentions were in the PD. Generally, a

player wants to be cooperative only if she forsakes narrow self-interest and if she can anticipate

27

that the other will cooperate (Kiyonari, Tanida & Yamagishi, 2000). The target of this latter

anticipation is, of course, a similarly conditional, cooperative expectation, and hence, the

importance of a theory-of-mind.

None of the normal children failed the first-order false belief task, while a third of the

children with autism did so. To determine the relationship between this fundamental mentalising

capacity and cooperation, the performance of the children with autism was analysed with a one

between factor (first-order false belief performance) and one within factor (PD game) repeated

measures ANOVA. There was a significant effect of the game [F (1,16) = 2.13, p < .05], as

described previously. Passing the false belief task significantly decreased cooperation across the

games, mean number of cooperative responses, 4.72 and 7.27 for false belief passers and failers,

respectively [F (1,16) = 4.46, p < .05]. There was no interaction between game and task

performance [F (2,16) = 2.01, p > .1].

This result is the opposite of the one reported for second order false belief: an acutely

malfunctioning theory-of-mind enhanced cooperation. The children who could not decipher the

first-order (Sally-Anne) false belief task cooperated, on average, in approximately half the trials

of each of the PD game versions. One of the ways a player could generate a cooperation rate of

50% is to simply choose randomly on each round without any regard for the decisions of the

counterpart. If this form of decision-making was employed, the player would be equally likely to

choose the circle or the triangle in a round regardless of which shape the opponent chose on the

previous round, in other words, random reciprocation. (The tit-for-tat strategy, by contrast,

assures that circle follows the other’s circle and triangle, triangle with certainty.) For each of the

children with autism, the conditional response rates to the opponent’s cooperating or defecting in

the previous round were calculated. The hypotheses that these conditional response rates were

equal to 50% were tested for those children who had passed or failed the Sally-Anne task. For the

children with first-order false belief troubles the hypothesis of random reciprocation could not be

28

rejected in three of four instances (computer opponent: cooperation after cooperation t (5) = .0, p

> .1, defection after defection t (5) = 1.42, p > .1; human opponent: cooperation after cooperation

t (5) = .44, p > .1, defection after defection t (5) = 4.07, p < .01). In this last case, defection was

reciprocated a little more than 70% of the time. By contrast, those who passed the first-order false

belief test did not respond randomly in any setting (computer opponent: cooperation after

cooperation t (12) = -3.19, p < .01, defection after defection t (12) = 4.83, p < .001; human

opponent: cooperation after cooperation t (12) = -3.71, p < .005, defection after defection t (12) =

6.23, p < .001). This sub-group of test passers was more likely to reciprocate defection and to

exploit cooperation. The evidence, then, strongly suggests that a first-order theory-of-mind or a

rule-based, effective substitute is necessary to reciprocate in a strategic fashion in the prisoner’s

dilemma.

To investigate the relationship between mentalising ability and cooperation in social

dilemmas in the adult groups, each adult participant was allocated a categorical score for their

performance on the Happé stories. Each story had been scored between 0 and 2. A participant's

mean accuracy score was converted into a categorical score of 0, 1 or 2. The data for the adult

groups were then analysed using a repeated measures ANOVA with two between factors (group;

theory-of-mind categorical score) and one within factor (PD game) on the number of cooperative

moves. There was a significant effect of game [F (1,26) = 30.72, p < .001], described previously,

and a significant interaction between game and theory-of-mind category [F (2,26) = 3.85, p <

.05]. This interaction indicated that those participants with superior theory-of-mind performance

adhered to the expected performance profile of more overall competitive play in the human and

computer opponent versions of the game and more overall cooperative play in the encouraged

cooperation version of the game, irrespective of whether or not an individual was diagnosed with

autism. The greater adherence to the expected task performance - as inherent in the rules of each

29

game - suggests that mentalising ability is involved in conditionally responding to the

counterpart.

One possible explanation of this result is that it is driven simply by a disparate recognition

of the changed rules in the initial round of the enforced cooperation game and does not involve an

ongoing divergence in strategy. However, the theory-of-mind category score did not correlate

with the likelihood of switching from a competitive response on the 32nd round of the game

(either playing the human or computer opponent) to a cooperative response on the first round of

the encouraged cooperation PD in either adult group [normal adults, r2 (13) = -.12, p > .1; autism

adults, r2 (13) = -.33, p > .1]. Thus, mentalising did not appear to be involved in responding to

changes in the setting and rules of these PD games, but rather to the maintenance of a consistent

strategy across sixteen rounds.

This flexibility across games and consistency across rounds of the same game may be

attributed to 'real' mentalising skill in the normal adult group, and to a compensatory mechanism

in the adults with autism. The small number of the latter individuals who performed relatively

successfully on the Happé stories reported doing so in a rule-based way ("It's politeness, that's

what my mother taught me. I've never really understood why", a 46-year-old woman with

Asperger syndrome; "I should say the opposite of what I think", a 32-year-old woman with

autism), and many of the adults with autism reported approaching the PD tasks similarly ("The

problem I've got with autism … other people who don't have it, they have slightly different

strategies", a 24-year-old man with Asperger syndrome). Taken together, these comments suggest

that the individuals with autism may be performing rationally by drawing on their logical

reasoning skills. Overall this speculation is supported not only by participant self-report, but also

by their slowed responses to the Happé questions.

Did players distinguish between the human and computer opponent? The inclusion of the

human versus computer opponent version of the Prisoner's dilemma task was a manipulation

30

designed to allow investigation of this issue, in particular relating to a section of the semi-

structured interview where questions were asked concerning any difference in the feeling or

playing of the participants in these two games. On the whole the normal adults felt a difference in

these two games and ascribed this difference to the nature of their opponent, using mental state

terms and/or a sense of empathy to describe this: "There's no way to judge the computer's actions

but I could obviously do so with Sarah [the confederate]." (a 30-year-old normally developing

man); "I felt she could be intuitive to what I was doing whereas I don't perceive a computer as

being intuitive. Its just mechanical." (a 33-year-old normally developing man). While a

proportion of the adults with autism described feeling different in these two tasks, and identified

the locus of this difference as lying in the nature of the opponent, responses included fewer

mental state terms, no empathic sense and had a sense of learned difference rather than the

intuitive human sense seen in the normal adults: "It was more complicated to play with Sarah …

Because it [sic] was a person." (a 37-year-old man with Asperger syndrome). "Playing the

machine was easier because the machine is predictable … and I felt more comfortable ... It’s

easier to read a machine or anticipate a machine … it’s far easier. A person is unpredictable … a

machine is easier because you're not up against emotion. It’s safer playing a machine. I don't fear

playing a machine but I fear the woman." (a 41-year-old man with Asperger syndrome). A final

example that particularly highlights the contrast between the two groups is illustrated by the

opposing sense of prediction seen in the following representative comments of an adult with and

one without autism: "I actually felt there was something about the way the computer was

programmed that I might eventually be able to work out where as I didn't feel that way about

Sarah." (a 46-year-old woman with Asperger syndrome); "I felt I could predict Sarah's responses

more than I could predict the computer's" (a 21-year-old normally developing woman). Although

asked to identify the strategy of their opponent, very few players in either group felt that they

could do so concretely (see Hill, Sally & Frith, in preparation).

31

Given the many comparable results of the individuals with and without autism in these PD

tasks, there was no prominent quantitative manifestation of internal discomfort with the human

opponent. Overall the decisions of the individuals with autism, whether derived through the

generation of a relevant rule or the application of a weakened theory-of-mind, largely reproduced

those of the matched normal participants. It may be that the discomfort of the autistic individuals

matched the distrust of the normal participants, who played the real PDs in a very competitive

manner.

There was one sub-group, however, who might be more prone to react to the perceived

“unpredictability” of the human opponent or take comfort in the assumed “dependability” of the

computer opponent, namely, those individuals who failed the coin flip task. These participants

made a mistake in the face of the uncertain coin, and their decisions might be similarly affected

by an opponent with equal or greater levels of capriciousness. Much evidence suggests that

children in general reflect the thoughts voiced by our participants with autism in that they are

more reluctant than adults to predict stability in the behaviour of other people (Miller & Aloise,

1989), but are no more likely to find machines or physical objects befuddling (Kalish, 2002).

Accordingly, the relationship between levels of cooperation with the human and the

computer opponents and the coin flip decision was investigated. A two between factor ANOVA

(diagnostic group - normal, autism; shape choice on coin flip task - circle, triangle) was applied

to the number of cooperative responses when playing the human and computer opponents

separately. The result of interest in these analyses is that of any influence of coin flip choice on

cooperation in the PD. There was no significant difference in the cooperation rates of coin flip

“passers” and “failers” when playing the human opponent [F (1,94) = 1.58, p > .1]. A significant

difference emerged on this measure when comparing performance on the computer opponent PD

[F (1,94) = 7.03, p < .01]. In this case those making the nondominant choice on the coin flip were

32

more likely to cooperate on the PD (mean cooperation on PD computer opponent for

nondominant versus dominant coin flip choice, 6.21 versus 3.61 respectively).

One potential hypothesis concerning these coin flip results is that these individuals have an

overwhelming preference for triangles over circles. The fact that coin flip “failers” did not pick

triangles (thereby cooperating) more frequently in the human opponent game eliminates this

hypothesis. Rather, it seems that the predictability of the computer made these participants feel

more secure and hence, much more willing to cooperate.

Summary

The performance of normally developing children was found to be less strategic than that of the

normal adults in the current study, supporting the picture of a developmental trajectory in levels

of spontaneous cooperation beginning with competition and ending with competition or

cooperation, as appropriate to task demands in adulthood. It is clear that this pathway is not fully

matured by the age of ten years. In relation to autism, differences existed in comparison to the

normal groups, particularly in terms of less strategic responses to the first round of each game

and across the individual rounds. In terms of mentalising and its influence on levels of

cooperation, mentalising ability aided strategic behaviour irrespective of the presence or absence

of autism. The skills used by those individuals with autism who appeared to have mentalising

abilities, and which could be considered to reflect quasi-mentalising ability, may have allowed

such individuals to 'get-by' in the Prisoner's dilemma task by using strong reasoning skill. These

individuals appreciated that the task required an understanding of another's mind, they had clear

insight into their difficulties in this regard and that they must draw on other resources to work out

what was expected of them. If this is the case, more subtle experimental manipulations of the set-

up used in the current study should elicit the oft-reported autism difficulties in this domain, and

would be particularly striking in real-life situations.

33

In the Prisoner’s dilemma, the issues of cooperation, generosity, and retribution are

interwoven with uncertainty about the counterpart’s intentions and actions and with a multiplicity

of rationales for a given action in each round. In a second study, an alternative form of strategic

interaction that may clarify these—bargaining—was investigated.

Experiment Two – Bargaining

Nature, inaccurate accounting, gravity, an overstocked shelf, chance, a misplaced wallet,

an academic researcher, or some other dea ex machina might endow an individual not only with a

prize or a lump of value but also with a companion and with a decision about division. In the

dictator game, the choice is how much of the prize to grant to the other party who is bound to

accept the grant. In the ultimatum game, the choice is how much of the value to offer to the other

party who then may accept the offer or refuse it. Acceptance achieves the suggested division;

refusal results in the whole prize being withdrawn and both parties receiving nothing. For

example, an experimenter might confer upon a participant ten candies (Murnighan & Saxon,

1998), ten marks or some other unit of currency (Güth & Tietz, 1990), or ten points or tokens

(Harbaugh, Krause & Liday, 2000). This endowed person would, as a dictator, decide how many

of the ten units, if any, to give to another person, and as the proposer in the ultimatum game, how

many to offer the other party knowing that a rebuff would wipe out the grant entirely.

The ultimatum game represents, among other social situations, the possibility in a

negotiation of one bargainer making a final offer to the counterpart and walking away from the

table leaving the other to sign the deal or not. Experience and introspection tell us that in this

setting such a dramatic proposal has a good chance of failing. However, in orthodox economics,

such an ultimatum should work: the equilibrium, corresponding to the prediction based on

rational self-interest, is that the responder should accept any offer greater than zero, and

therefore, the proposer should offer the smallest possible positive amount. In fact, this

34

equilibrium is rarely realized in any of the numerous assays that have been conducted in

laboratories and field sites around the world during the last twenty years. The regular findings,

rather, are the following: (i) the modal offer is between 40% and 50% of the whole prize, (ii) tiny

offers are almost always rejected, and (iii) a majority of responders will refuse offers below a

third of the total (see Güth & Tietz, 1990; Camerer & Thaler, 1995; Oosterbeek, Sloof & van de

Kuilen, 2001 for reviews).

The literature has focused on understanding why offers are so robust, and why responders

are so rancorous. One possible explanation for the generosity of the proposers is that they have a

certain taste for fairness and a preference for sharing some part, or even half, of their dowry.

Forsythe, Horowitz, Savin & Sefton (1994) asked participants to play an ultimatum game and a

dictator game to test this hypothesis, since fairness considerations would dictate that proposers

give the same amount in both games. However, these authors found that the proportion of equal

split offers declined from 75% in a $10 ultimatum game to 21% in a $10 dictator game. The

mean offer in a standard 10 unit dictator game is between 20% and 25% (Rigdon, 2002), and in

the ultimatum game, 40% and 45%. Roughly, then, half of the typical proposer’s generosity is

driven by a taste for fairness and half by strategic considerations of the possible spite of the

responder.

Subsequent dictator experiments have shown that the taste for fairness can be heightened

or slaked by the context of the game. The degree of selfishness among dictators is raised by

allowing them complete anonymity, even from the experimenter (Hoffman, McCabe & Smith,

1996) and by placing them in a business setting of buying and selling (Hoffman, McCabe, Shacat

& Smith, 1994); the frequency of altruistic grants is raised if the recipient is a charity (Eckel &

Grossman, 1996) and if more personal characteristics of the recipient are identified (Bohnet &

Frey, 1999; Charness & Gneezy, 2000). Subsequent ultimatum experiments have shown that the

strategic and fairness concerns of both parties move in predictable directions: when the proposed

35

split of $10 is generated by a roulette wheel as opposed to another person, the mean minimal

offer acceptable to responders is much lower—$1.20 rather than $2.91 (Blount, 1995); when a

written note saying “I know you’d like more, but that’s the way it goes” is attached to a small

offer, a rebuff from the responders is more likely (Kravitz & Gunto, 1992); when the endowment

is worth more to one side than the other (e.g., 50¢ versus 10¢ per chip) and the other side is

ignorant of this fact, advantaged proposers are more likely to suggest an even split of the

counting units rather than total value, and advantaged responders are more likely to reject fair

splits of the underlying units to induce a more equitable split of surplus value (Croson, 1996;

Kagel, Kim & Moser, 1996).

The variants just described could be reflective of mentalising ability or social rule-

following. For instance, a person might reject a small offer because she imagines that the

proposer thinks she is unworthy, gullible, or dim-witted, or because she recognizes this game as a

sharing situation which demands that greedy people be punished. Similarly, a munificent dictator

might utilize her theory-of-mind to anticipate the disappointment of an unfunded recipient, or she

might recognize the applicability of a sharing norm. A norm may substitute for mentalising: as

the other driver in a narrow lane approaches, you need not read his eyes, thoughts, or intentions,

you need only remember the locale and move to the left in the UK and to the right in the US.

Mentalising may become necessary only if the interaction does not proceed as expected: when

the other driver fails to move to the proper side, then you need to notice the direction of his gaze,

the tenseness of his hands, and the expression on his face.

The evidence on the relative utilization of mentalising versus norm-following in the

standard, anonymous ultimatum game is mixed. On the one hand, Henrich et al. (2001) found

that much of the variance in mean offers among fifteen small-scale societies (ranging from 26%

among the Machiguenga in Peru to 58% among the Lamelara in Indonesia) could be explained by

two factors—the importance of cooperation in the society’s economic production and the reliance

36

on market exchange in daily life. These factors would seem to develop and mould norm

formation rather than mentalising abilities. The authors themselves suggest that their participants

applied the analogous (and varying) norms found in their societies:

[W]hen faced with a novel situation (the experiment), they looked for analogues in their daily experience, asking “What familiar situation is this game like?” and then acted in a way appropriate for the analogous situation. For instance, the hyper-fair…offers (greater than 50 percent) and the frequent rejections of these offers among the Au and Gnau reflect the culture of gift-giving…, accepting gifts, even unsolicited ones, commits one to reciprocate at some future time to be determined by the giver (Henrich et al., 2001, p. 76). A second experiment that both documented the under-utilization of mentalising and its

potential impact is that of Hoffman, McCabe and Smith (2000). Here, in a $10 ultimatum game

with a business context, an additional line was added to the instructions encouraging the proposer

(i.e., seller) to strategize and read the mind of the opponent: “Before making your choice,

consider what choice you expect the buyer to make. Also consider what you think the buyer

expects you to choose.” This encouragement caused the mean offer to rise to $4.17 from $3.71 in

a control condition where no explicit mindreading prompt was given, and this increase suggests

that proposers in the control condition were solving the game without fully employing their

theory-of-mind.

On the other hand, studies in which an asymmetry in information between the offerer and

responder is strategically exploited support the relative prominence of mentalising. Respondents

in one experiment received a set sequence of twelve offers for either $1 or $2 out of a total

surplus of $20 (Pillutla & Murnighan, 1996). There was a complicated information structure

overlaid on the set of games: (i) during the first half of the sequence, none of the responders knew

that the total prize was $20, making it difficult to deem an offer unfair; (ii) half of the responders

knew that the (fictitious) offerers knew that the low offers were unfair and therefore could more

easily attribute greedy intentions to them. (Note that this second manipulation depends upon a

theory-if-mind both to understand the different contents of the other’s mind and to respond

37

emotionally.) After making each decision, participants reported how they reacted to the offer and

how they felt. These verbal responses were coded for the degrees of unfairness and anger

expressed. In relation to the current study there were two important findings: first, the

manipulation relying on theory-of-mind was successful—participants who knew both that an

offer was low and that the other knew it was low were far angrier. Second, this anger resulted in a

greater frequency of rejection, and was more predictive of the likelihood of rejection than was the

expressed degree of unfairness alone.

As any overtaxed parent can testify, children reject various ultimatums all day long and

readily employ multiple notions of “fairness.” There is a vast literature on prosocial behaviour

among children, some of it emphasizing social rules and some, perspective taking. Numerous

donation studies (similar to the dictator game) have found that children are more generous when

they have seen a model being generous (e.g., Harris, 1970; Wilson, Piazza & Nagle, 1990).

Within social learning theory (Bandura, 1986), a model affects the observer by directly

representing the presence or application of a rule rather than by triggering a mediating internal

process, suggesting that mentalising is less important in giving. On the other hand, a child’s

abilities to take the perspective of another person visually, emotionally, and cognitively are

positively related to prosocial behaviour in most studies (Underwood & Moore, 1982), and in one

specific study, two factors relying on a child’s theory-of-mind, affective reasoning and sympathy,

caused a large increase in donations to a needy person (Knight, Johnson, Carlo & Eisenberg,

1994). Both sources of prosociality should become stronger as children grow up, and indeed, a

meta-analysis by Eisenberg and Fabes (1998) found that sharing and donating became more

prevalent from preschool through adolescence.

Two specific ultimatum studies, however, showed a less clear developmental trend.

Harbaugh, Krause and Liday (2000) found that fourth and fifth graders made larger grants as

dictators than did second graders or ninth graders. While these authors found that ultimatum

38

proposals on average increased with grade level, Murnighan and Saxon (1998) found a non-

monotonic pattern with kindergarteners offering more candies than third graders, who, in a game

with a dollar at stake, offered fewer cents than did sixth graders, who were more generous than

ninth graders. Finally, younger children in both studies were more likely to accept small offers.

This mixture of results demonstrates that the offering and responding behaviour of children may

be affected by the specific details of game presentation, and may reflect a general inconsistency

in inference about social interaction. Kalish (2002) has shown that while children and adults will

equally predict consistency in repeated physical events such as pumice floating in water, children

will much more often predict that a person would behave differently in the future than he or she

did in the past, for instance, preferring Bert tomorrow despite preferring Ernie today. If the

reaction of the other party is inconsistent or unreliable, then it makes sense to accept whatever the

current offer is, and to not be too strategic in formulating an offer.

By investigating the giving and receiving behaviour of children and adults with and

without autism, the importance of mentalising for both generosity and consistency can be

determined.

METHOD

The participants were the same as those included in the first experiment (social dilemmas), with

the addition of one eight-year-old.

Materials and Methods. The testing set-up remained the same as that reported for the first

experiment with participants being assessed in a quiet room, sitting at a laptop to the right of the

confederate. Responses were recorded on-line for later analysis. Task instructions were presented

on the computer and the experimenter verbalised them to ensure that participants understood each

task. Players were told that they must try to win as many points as possible and that the total

points won on the games would be exchanged at the end of the testing session for stickers

39

(children) or chocolates (adults). The greater the number of points won, the greater the reward at

the end of the test session. Two versions of a bargaining task - dictator game and ultimatum game

- were completed by participants, with the dictator game always played first

Dictator game. The dictator (participant) was given ten points and asked how much s/he

wanted to give to the opponent, knowing that s/he would keep the remainder. Eleven cards were

presented on the computer screen, outlining all possible permutations by which the points could

be split, ranging from the dictator keeping all ten points for her/himself to giving all ten points to

the opponent. The dictator made his or her choice and indicated this by selecting the relevant card

on the computer screen. The choice that the dictator had made and the points allocated to both

players were displayed. This process was repeated 16 times throughout the course of the game,

with the participant acting as the dictator for rounds 1-4 and 9-12, and the confederate taking the

part of the dictator for the remaining rounds. The participant was unaware that there would be

more than one round of the dictator game and that the confederate would also take a turn as the

dictator. In the latter case the confederate allocated approximately the same amount of points to

the participant as the participant had to her.

Ultimatum game. This game was essentially the same as the dictator game but the opponent

had the choice to accept or reject the offer made to them by the proposer in each round of the

game. The game started as the dictator game. Once the proposer had made an offer, the opponent

indicated whether s/he accepted or rejected that offer. If the opponent accepted the proposer's

offer, the points were divided as proposed (exactly as in the dictator game). If the opponent

rejected the proposer's offer, neither player received any points. The choices made by each

player, as well as the points won were displayed after each round of the game. The set-up of the

ultimatum game was identical to that of the dictator game, with the participant acting as the

proposer and the confederate as the opponent for rounds 1-4 and 9-12, and with the roles reversed

for the remaining rounds. The participant was not told explicitly that there would be more than

40

one round of the game and that the confederate would also take a turn as the proposer. In the

latter case the confederate allocated approximately the same amount of points to the participant

as the participant had to her and always rejected offers of less than an equal split (i.e. four or

fewer points being allocated to the confederate).

RESULTS & DISCUSSION

Once again, there is a lack of studies which track the strategy of normal children and adults on

our version of the dictator and ultimatum games. This comparison will be presented first,

followed by the performance of the individuals with autism. Comparisons of the full dataset will

then be reported.

Offers made.

The offers made by the participant to the confederate were expected to be lower in the dictator

game than in the ultimatum game. The mean points offered to the confederate by the participant

across the first four rounds of each game, and on the first round only, are shown for each group in

Figures 5a and 5b respectively. A repeated measures ANOVA with one between factor (group)

and one within factor (game) was applied to the data for mean points offered. For performance

across the first four rounds of each game, there was a significant effect of group [F (3,62) = 3.17,

p < .05]. A series of Tukey tests revealed this difference to arise from higher offers being made

by the six- in comparison to the eight-year-old children [p < .05]. There was a significant effect

of game [F (1,62) = 77.87, p < .001], reflecting higher offers being made to the confederate in the

ultimatum game, and a significant interaction between group and game [F (3,62) = 3.5, p < .05].

The interaction reflected a greater difference between the offers made across the two games by

the adult group. Thus, while both adults and children responded in the predicted manner, the

adults did so more strikingly, making a greater distinction between the size of their offers in the

41

two games, with a greater number of points being kept for self in the dictator game versus the

ultimatum game. This pattern of performance was also seen in the analysis of the mean offers

made to the confederate on the first round of each game, with the exception of the difference

between the groups [group, F (3,62) = .62, p > .1; game, F (1,62) = 59.03, p < .001; group by

game, F (3,62) = 3.07, p < .05]. These results correspond to the equivocal results of previous

ultimatum studies in that there was no clear trend in offer amounts across age groups. More work

is needed to explain this finding within the findings of the prosocial literature which shows that

sharing and donating become more prevalent as children grow up (Eisenberg & Fabes, 1989). A

tentative explanation for the differences may be that this task evoked exchange norms which are

already fairly firmly implanted by the age of six.

[Figure 5 about here]

Having established the pattern of performance across the two bargaining tasks used in the

current study, the autism samples were added to the analysis and comparisons made between the

two adult and child groups separately for the mean number of points offered by the participants to

the confederate across the first four rounds of each game as well as on the first round only (see

Figure 5). In all cases (adults; children; four rounds; first round) there was a significant effect of

game [adults, F (1,28) = 34.19, p < .001; children, F (1,67) = 27.75, p < .001], with more points

being offered to the confederate in the ultimatum game, but no effect of group and a mildly

significant interaction between group and game [adults: group, F (1,28) = .02, p > .1; group by

game, F (1,28) = 3.27, p > .05; children: group, F (1,67) = .7, p > .1; group by game, F (1,67) =

2.95, p > .05. On average, it appears that the individuals with autism approached the bargaining

tasks in a way that was comparable to their normally developing peers, suggesting that some

individuals with autism may have an intact mechanism that deals with fairness. There is, also,

some indication that individuals with autism did not react as vigorously to the strategic

dimensions of the ultimatum game.

42

More evidence for the relative effects of autism on fairness and tactical giving appear

when looking at the underlying distributions of offers. In particular, Figure 6 displays the

dispersal of first round ultimatum offers by adults and children with and without autism.

Visually, there appears to be a strong divergence. The powerful Epps and Singleton (1986) test

was used to determine if these two samples of discrete data were likely to be from identical

populations. This CF statistic, based on the empirical characteristic function, is asymptotically

distributed as chi-square with four degrees of freedom and can be corrected for small samples.

For the first round ultimatum of autistic and normal adults displayed in Figure 6a, the null

hypothesis of similar distributions is strongly rejected [CF = 11.83, p < .02]. One interpretation of

this result is that adults with autism applied one of two rules: split the amount fairly and squarely,

or take everything that you can. A far greater proportion of the normal adults tried to strategically

shade their offers by taking a point or two extra for themselves.

[Figure 6 about here]

We can contrast the distributions of first round ultimatum offers of children with and

without autistic spectrum disorder in Figure 6b. Here, again, there was evidence of distinct

patterns of first offers [CF = 9.96, p < .05]. There were no significant differences in offer

distributions between the adults and the children who shared the same condition [normal, CF =

6.59, p > .1; autism, CF = 2.54, p > .1]. Moreover, when comparing first round dictator offers

there were no significant differences in the underlying distributions across any groups. Since the

initial dictator grants were the same, it can be concluded that core generosity did not vary by age

or with autism. However, once an element of strategic anticipation was added by empowering the

responder in the ultimatum game, those with autism seemed to employ one of two salient rules:

cut the total in half, or keep it all. In contrast, the mentalising abilities of the normal participants

were utilized to generate mildly unequal, slightly shaded offers.

43

The direct effects of mentalising on first round offers among the children can be seen in

Figure 7. The striking visual difference in the distribution of offers is strongly confirmed by an

Epps-Singleton test (CF = 9.64, p < .05). Here, the majority of those who failed the second-order

false belief test offered the responder nothing or only one point, while the majority of those who

passed the test offered an even split. (These distributions of initial offers have means whose

distinctiveness approaches significance [F (1,63) = 3.73, p = .058].) A theory-of-mind that was

effective due to intuition or synthetic construction was very helpful to the child making a

reasonable offer in this game.

[Figure 7 about here]

The extent of learning across rounds of the ultimatum game was investigated, in light of

the apparently different strategies of the individuals with and without autism. The confederate

consistently rejected offers of four points or fewer, so participants had the chance to learn, adjust

and converge on the optimal offer of an even split. A comparison of the groups on a round by

round basis was made on the dictator and ultimatum games for all eight rounds (see Figure 8). A

repeated measures ANOVA with one between factor (group) and two within factors (bargaining

game, round) was applied to the data for the mean offer made by each group on all eight rounds

of each game. For the comparison of the normal adults and children, there was a significant effect

of game [F (1,64) = 118.66, p < .001], and a significant interaction between group and game [F

(1,64) = 7.93, p < .01]. These significant effects have been described previously. When

comparing the mean offers of adults with and without autism, there was a significant effect of

game only [F (1,28) = 38.19, p < .001]. Unlike the first round, however, this similarity in the

means across the groups for rounds two through four and nine through twelve is also reflected in

comparable underlying distributions of individual offers (all CFs < 7.0, all p’s > .1). Moreover,

mentalising, as measured by performance on the Happé stories, was insignificant in an ANOVA

of the average offer over multiple rounds. Once autistic adults had a single experience with the

44

ultimatum game, their offers were largely indistinguishable from those of the control group.

Hence, the combination of a scaffolded theory-of-mind, choice of a relevant norm and one

reaction from another person allowed the adults with autism to mirror the behavior of their

normal counterparts.

With respect to the average offers of the children there was a significant effect of game [F

(1,67) = 62.38, p < .001], described previously and a significant interaction between group,

bargaining game and round [F (1,67) = 4.21, p < .01]. This interaction highlighted divergence

between the amounts offered to the confederate in the two games on round one in the normally

developing children only. Most of the children seem to learn the game as quickly as the adults: by

the second round the offers of autistic and normal children as a group are dispersed in statistically

similar ways. This concurrence is confirmed by the insignificance of second-order theory-of-

mind in an ANOVA of average offer over all eight rounds. The one group that remains distinct

are the six-year olds: even after eleven rounds of the game, in their final turns as proposers, their

offers are scattered over all the possibilities and are distributed differently from those of the

eight- and ten-year olds [CF = 11.62, p < .02] and those of the autistic children [CF = 8.29, p <

.1].

[Figure 8 about here]

Offers rejected.

The mean points rejected over the eight rounds of the ultimatum game by each participant group

are shown in Figure 9. A one factor ANOVA was used to compare the mean points rejected by

each participant group. There was a significant difference between offers rejected by the normal

groups [F (1,65) = 5.09, p < .05], reflecting a developmental trend from childhood to adulthood,

with lower offers being acceptable to the children [mean (SD) points rejected: adults, 3.23 (2.71);

children, 2.34 (1.22)]. No significant differences existed between acceptable offers to either

45

adults with or without autism [F (1,29) = .44, p > .1], children with or without autism [F (1,68) =

.08, > .1], or autistic adults or children [F (1,32) = .62, p > .1],. Thus in terms of offers made by

the confederate that were deemed acceptable by the participant, a developmental progression

emerged: Both those with and without autism were tolerant of another's gain, accepting offers of

less than 50% for themselves. There was, however, a limit in all groups with adults rejecting

offers to self of less than 32% and children being more tolerant, rejecting offers of less than 23%

of the share for self. This conclusion is in accord with the most robust finding in the literature on

children and bargaining: children, younger children in particular, are less likely to reject smaller

offers.

[Figure 9 about here]

Lastly, mentalising appeared to play no role in the rejection of offers. Performance on the

relevant theory-of-mind task did not explain any of the variance in average offer rejected by

normal adults and children or by autistic adults and children. Hence, for our participants the

ability to decipher the confederate’s intentions and envisage her disdain did not affect the

likelihood of a small offer being rejected. It is as though the offence generated internally within

the recipient of an undersized offer is sufficiently motivating.

Summary.

The youngest normally developing children (aged 6 years) were found to be more prone to

sharing than the 8-year-olds in this sample, and the normally developing children as a whole

more tolerant of their opponent's gain (as measured by the offers rejected by participants).

Normal children demonstrated more pure generosity in the dictator game and less strategic

munificence in the ultimatum game than did adults. There were hints of a similar pattern among

our autistic participants. Lastly, for each group with the exception of the 6-year-olds, repeating

the ultimatum game caused differences to dissipate. The older children learned and adapted so

46

that by the second round their offers were indistinct from those of the adults; meanwhile, even by

the twelfth round, the youngest children were still somewhat at a loss. Although the evidence is

not overwhelming, there is a suggestion of a developmental trajectory in levels of bargaining.

Autistic children and adults were distinguished from their peers in the initial round. The

adults with autism spectrum disorder were more likely to choose either an even split or a tender

of zero. Autistic children showed a similar predilection for proposals of nil or five in contrast to

normal children who were most likely to just divide the prize in half. An ineffective theory-of-

mind was apt to result in an initial offer of no more than one point, while deciphering the second-

order false belief story correctly tended to lead to an equitable offer. Autistic adults proposed a

little less than four points on average, and normal adults a little more, but this resemblance masks

bimodal roots of the former and the deliberately strategic nature of the latter. The development of

theory-of-mind skills may help the child first to recognize and act upon relevant norms of

behaviour such as fairness, and later, to stretch those norms and improvise away from them when

the situation calls for it.

A single experience as proposer was sufficient to counteract any deficits in mentalising in

the succeeding rounds of the this ultimatum game. Without question, the static nature of the

payoffs and counterpart in this bargaining game is essential to the observed decay in the effects

of mentalising. A more dynamic and realistic negotiation would probably continue to demand the

application of an active theory-of-mind.

In contrast, rejection of offers did not vary with the efficacy of the participant’s

mentalising capability. The ability to sense another’s unfairness or contempt was not necessary to

reject offers of a third or less. So, the generation of an initial appropriate offer depends on theory-

of-mind, but the denial of an inapt or inequitable offer does not.

47

GENERAL DISCUSSION

In the experiments reported here the approach to two-way, reciprocal social interaction was

investigated using three mixed-motive games. The study had two broad aims, first to track the

development of decision-making on these games, and second to investigate the relationship

between mentalising and strategy choice.

Normally developing children did not, on the whole, perform in the same way as the

normal adults in situations involving a social dilemma (the prisoner's dilemma). This was the

case both for the games that encouraged competition between players and for the game that

encouraged cooperation. While normal adults showed a clear pattern of competition and

cooperation according to the task requirements, normally developing children aged 6-10 years

mixed both behaviours with their opponent in all situations, irrespective of whether such an

approach was overtly beneficial to them. There was some evidence of a developmental trend as

10-year-olds were more likely to cooperate in the initial round of the game playing against a

human opponent and pursued a level of cooperation that was more responsive to the identity of

the counterpart and the rules of the game than did the younger children. Normal children as a

group were more cooperative with the computer opponent than were normal adults, and they

became relatively more competitive with their human opponent over the course of the sixteen

rounds.

In contrast, many differences in the bargaining games were resolved after one or two

repetitions. As with the prisoner’s dilemma, adults were more sensitive to the strategic

possibilities offered by the structure of the games, in this case, the dictator versus the ultimatum

game. Adults, on average, changed their offers more in the latter game, an acknowledgement of

the responder’s potential veto power. There were no clear age trends with respect to the average

generosity of the offers of the young participants, but the very youngest were significantly more

likely to accept a share of the endowment that was less than 30%. In addition, many of these 6-

48

year-olds seemed to remain baffled by the ultimatum game after repeated plays, as their offers in

the twelfth round were scattered relative to the adults and other children. This excess dispersion

may be in keeping with the finding that younger children are less likely to predict consistency in

the actions of other people, in this case, the responder (Kalish, 2002).

The adult players with autism appeared, on the surface, to perform similarly to their

normal adult counterparts on both kinds of game. However, strategy differences were identified

on both the prisoner's dilemma and ultimatum game, although on the former to a greater extent

than the latter. Specifically the adults with autism showed less extreme behavioural choices of

competition and cooperation on the one hand (prisoner's dilemma) and different distributions of

opening offers on the other (ultimatum game). Contra our hypotheses, autistic adults were not

less (more) likely to cooperate with the human (computer) opponent, and they failed to adjust

fully for the dominant choice in the encouraged cooperation version. Moreover, by the second

round of the ultimatum game the discrepancy in offer distributions had disappeared. The overall

pattern was similar when comparing the two groups of children. The autistic children did not

adapt in any noticeable way to the encouraged cooperation PD, a strategic failure that may reflect

the perseverative continuation of earlier choice patterns. Normal and autistic children gave the

same unilateral gifts in the dictator game. Finally, with the just noted exception of the 6-year-

olds, the large difference in initial ultimatum offer distributions among the children was erased by

the second round.

We also investigated mentalising performance directly. A child’s ability to pass the

second-order false belief test, arising from either a functional theory-of-mind or rational, rule-

based deduction, was found to be positively related to the likelihood of cooperation in the three

PD games and to perfectly fair ultimatum offers rather than very small proposals. Those autistic

children who failed the simplest theory-of-mind test cooperated more than did those who passed

the test, but this difference seemed to result largely from the much more random responses of the

49

test failers to the moves of the counterpart. Adults with more effective mentalising abilities

employed them both to compete against the human and computer opponents and to cooperate in

the encouraged cooperation game. However, with this group, performance in the mentalising test

(interpretation of Happé stories) did not explain the variance in individual ultimatum offers.

All told there are numerous differences in the decisions and underlying strategies of the

participant groups. Roughly speaking, the development of theory-of-mind does seem to be

correlated with increasing generosity through childhood and then more strategic behaviour in

adulthood. However, when compared to the dramatic distinctions anticipated by our hypotheses

or the markedly varying introspections of the adults on strategy and choices, these findings are

weaker than expected. Arguably, the rather more surprising findings are those of the initial

similarity in performance profile, in particular in terms of the mean number of cooperative moves

across all rounds of the PD games and the offers made/rejected in the bargaining games. Such

similarity does not mirror the behavioural differences observed in autistic and nonautistic

individuals in similar real-life situations. How could this similarity on the experimental tasks be

explained in light of the differences in strategy-choice adopted by all players, the striking

mentalising deficits evidenced by the autistic participants and their relevant difficulties in daily

life?

A relationship between strong mentalising ability and increased social cooperation in

mixed-motive games has intuitive, and some experimental support. On the basis of the findings

of the current study it might seem that mentalising is differentially involved in the prisoner's

dilemma versus the bargaining games, being more critical in the former and less in the latter,

especially when supplemented by direct experience. However, the underlying differences in

strategy identified through statistical analysis, and corroborated in introspective reports as well as

in the comments of autistic adults about their interactions in daily life suggest, rather, that some

form of compensatory mechanism is driving the choices of the autistic individuals. One candidate

50

for compensation would be the substitute of intuitive understanding with a laboured, rule-based

approach in autism whereby the 'rules' of social interaction are gradually learned. These rules

must then be applied in on-line situations. This would likely be a laborious process. By such an

account it would not be surprising that behaviour could appear odd since social interaction does

not, in reality, operate on the strict application of rules alone. If a rule-based mechanism was

acting as a substitute for intuitive, easy interactions in the mixed-motive games reported here, it

is conceivable that the greater group differences highlighted in choices on the prisoner's dilemma

suggest that these social games were compensated for differentially in high-functioning

individuals with autistic spectrum disorders. This could arise because the 'rules' recruited for

bargaining can be more easily learned and/or applied than those for dealing with a dilemma. If

this were the case one might predict that an offer of exactly half would be made to the opponent

in the ultimatum game, a logical rule which cannot fail to be acceptable. In the first round, this

was exactly the profile identified in 60% of the autistic participants but in less than a third of the

normal participants. Moreover, more than a quarter of the autistic adults and children wished to

keep all ten points for themselves in the initial trial. Perhaps these participants had not developed

a rule for fairness, or identified the ultimatum game as a “finders keepers” game, i.e., if you’re

given something, it is yours alone. In contrast the 'rules' of the prisoner's dilemma are more

opaque and therefore harder to identify and use as successfully. Though relatively simple, a rule

such as ‘tit-for-tat’ may not suggest itself as easily in the PD as the fairness rules do in the

dictator and ultimatum games and so, the need to mentalise remains prominent throughout the

PD.

While it is not within the scope of the current study to pursue the issue of neural

activation when performing mixed-motive social games, published findings might shed some

light on the compensatory mechanism potentially adopted by the autistic individuals. Neuro-

imaging studies of mentalising in normal individuals have identified a network of brain regions

51

that is consistently active during mentalising over and above the other task demands. This

network involves the medial prefrontal cortex (especially anterior paracingulate cortex), the

temporal-parietal junction and the temporal poles (Fletcher et al., 1995; Brunet, Sarfate, Hardy-

Bayle & Decety, 2000; Castelli, Happé., Frith & Frith, 2000; Gallagher et al., 2000; Vogeley et

al., 2001). These areas appear to be activated less in the brains of autistic adults when performing

mentalising tasks (Happé et al., 1996; Baron-Cohen et al., 1999; Castelli, Frith, Happé., & Frith,

2002).

With respect to mixed-motive games, a recent functional magnetic resonance imaging

(fMRI) study in which playing a two-way reciprocal trust game against both a human and

computer opponent was contrasted showed increased activation in non-specified areas of the

prefrontal cortex in those players who consistently attempted cooperation with a human

opponent, than when the same players played against a computer following a fixed, known

probabilistic strategy. Those players who did not cooperate consistently did not show this pattern

of increased prefrontal cortex activation (McCabe, Houser, Ryan, Smith & Trouard., 2001). This

finding may accord with the notion of a social brain network described above, in which the

medial prefrontal cortex plays an important role in verbal and non-verbal mentalising tasks, as

well as in monitoring your own inner states. In a second fMRI study, Rilling et al. (2002)

required normal individuals to play an iterated prisoner's dilemma against both a human and a

computer opponent. In this study, mutual cooperation was associated with consistent activation in

brain areas that have been linked with reward processing, in particular the nucleus accumbens,

caudate nucleus, ventromedial frontal/orbitofrontal cortex and rostral anterior cingulate cortex.

Taken together these two studies suggest the potential involvement of mentalising and/or

reward/punishment systems in the brain. While the current study can not draw specific

comparisons to brain imaging data, it would be of interest to establish whether the brain

activations of autistic individuals would reflect those of normal individuals when making

52

behavioural choices that are - in outcome - identical across the two groups. Given the differing

levels and patterns of neural activation of this population for mentalising tasks, and the differing

strategies adopted by the two groups in the current study, it would seem most likely that the

neural activations of autistic and non-autistic individuals would differ. Information as to the

nature of the approach taken by the individuals with autism could be revealed by such imaging

studies.

The tests reported in this paper identified an unexpected profile of similarities as well as

differences in the performance of individuals with and without autism. Mentalising per se may

not be necessary for basic strategic rationality, but a quasi-mentalising compensatory mechanism

may be needed. Better mentalising (or greater success of a compensatory strategy) increased

levels of cooperation in some settings and of strategic behaviour in others. Thus on the surface a

putative compensatory mechanism in the autistic participants yielded similar behaviour to that

seen in the normal individuals. This contrasts with the striking difficulties in social interaction

experienced by individuals with autism in social interactions in the real-world. Perhaps it is the

on-line aspects of mentalising and mental flexibility which cause the greatest difficulty for high-

functioning autistic individuals in dilemma and bargaining situations in the real-world where far

more distractions and far fewer cues to guide behavioural choices exist than do in the almost

sterile laboratory where tasks can be seen to be abstract. We are currently investigating this

possibility.

53

ACKNOWLEDGEMENTS

This research was supported by funding from the UK's Medical Research Council (grant number

G9716841). We gratefully acknowledge the willing participation of all individuals in this study.

We are indebted to Sarah Griffiths, Zoë Fortune and Sakina Adam-Saib for substantial help with

data collection and especially to Professor Uta Frith for invaluable support for, and discussion of

the project.

54

REFERENCES Abell, F., Krams, M., Ashburner, J., Passingham, R., Friston, K., Frakowiak, R., Happé, F., Frith,

C., & Frith, U. (1999). The neuroanatomy of autism: a voxel-based whole brain analysis of structural scans. Neuroreport, 10, 1647-1651.

Abric, J.C., & Kahan, J.P. (1972). The effects of representations and behavior in experimental

games. European Journal of Social Psychology, 2, 129-144. Astington, J.W. (1994). The Child's Discovery of the Mind. London: Fontana. Axelrod, R. (1984). The Evolution of Cooperation. New York: Basic Books. Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory.

Englewood Cliffs, NJ: Prentice-Hall. Baron-Cohen, S., Leslie, A., & Frith, U. (1985). Does the autistic child have a “theory of mind”?

Cognition, 21, 37-46. Baron-Cohen, S. Tager-Flusberg, H., & Cohen, D.J. (1993). Understanding Other Minds.

Perspectives from Autism. Oxford: Oxford University Press. Baron-Cohen, S., Tager-Flusberg, H., & Cohen, D.J. (2000). Understanding Other Minds.

Perspectives from Developmental Cognitive Neuroscience. Oxford: Oxford University Press. Baron-Cohen, S., Ring, H., Williams, S., Wheelwright, S., Bullmore, E., Brammer, M., & Andrew

C. (1999). Social intelligence in the normal and autistic brain: an fMRI study. European Journal of Psychiatry, 11, 1891-1898.

Bechara, A., Damasio, A., Tranel, D., & Damasio, A.R. (1997). Deciding advantageously before

knowing the advantageous strategy. Science, 275, 1293-1295. Blount, S. (1995). When social outcomes aren’t fair: the effect of causal attribution on preferences.

Organizational Behavior and Human Decision Processes, 63, 131-144. Bohnet, I., & Frey, B.S. (1999). Social distance and other-regarding behavior in Dictator Games:

Comment.” American Economic Review, 89, 335-339. Bowler, D.M. (1992). “Theory-of-mind” in Asperger’s syndrome. Journal of Child Psychology &

Psychiatry, 33, 877-893. Brunet E., Sarfate, Y., Hardy-Bayle, M.C., & Decety, J. (2000). A PET investigation of the

attribution of intentions with a nonverbal task. Neuroimage, 11, 157-166. Camerer, C., & Thaler, R.H. (1995). Anomalies: ultimatums, dictators and manners. Journal of

Economic Perspectives, 9, 209-219. Castelli, F., Happé, F., Frith, U., & Frith, C.D. (2000). Movement and mind: a functional imaging

study of perception and interpretation of complex intentional movement patterns. Neuroimage, 12, 314-325.

55

Castelli, F., Frith, C., Happé, F., & Frith, U. (2002). Autism, Asperger syndrome and brain mechanisms for the attribution of mental states to animated shapes. Brain, 125, 1839-1849.

Charness, G., & Gneezy, U. (2000). What’s in a name? Anonymity and social distance in dictator

and ultimatum games. Manuscript, University of California at Santa Barbara. Croson, R.T.A. (1996). Information in ultimatum games: an experimental study. Journal of

Economic Behavior and Organization, 30, 197-212. Dawes, R. (1980). Social dilemmas. Annual Review of Psychology, 31, 169-193. Dunn, L.M., Dunn, L.M., Whetton, C., & Burley, J. (1997). British Picture Vocabulary Scale (2nd ed.). Windsor: NFER Nelson. Eckel, C.C., & Grossman, P.J. (1996). Altruism in anonymous dictator games. Games & Economic

Behavior, 16, 181-191. Eisenberg, N., & Fabes, R.A. (1998). Prosocial development. In W. Damon (ed.), Handbook of

Child Psychology (5th ed., vol. 3), pp. 701-778 New York, NY: Wiley & Sons. Elliott, R., Rees, G., & Dolan, R.J. (1999). Ventromedial prefrontal cortex mediates guessing.

Neuropsychologia, 37, 403-411. Epps, T.W., & Singleton, K.J. (1986). An omnibus test for the two-sample problem using the

empirical characteristic function. Journal of Statistical Computation & Simulation, 26, 177-203.

Fan, C-P. (2000). Teaching children cooperation—an application of experimental game theory.

Journal of Economic Behavior & Organization, 41, 191-209. Fletcher, P.C., Happé, F., Frith, U., Baker, S.C., Dolan, R.J., Frackowiak, R.S.J., & Frith, C.D.

(1995). Other minds in the brain: a functional imaging study of “theory of mind” in story comprehension. Cognition, 57, 109-128.

Forsythe, R., Horowitz, J.L., Savin, N.E., & Sefton, M. (1994). Fairness in simple bargaining

experiments. Games & Economic Behavior, 6, 347-369. Gallagher, H., Happé, F., Brunswick, N., Fletcher, P.C., Frith, U., & Frith, C.D. (2000). Reading

the mind in cartoons and stories: an fMRI study of “theory of mind” in verbal and non-verbal tasks. Neuropsychologia, 38, 11-21.

Gibbons, R. (1992). Game Theory for Applied Economists. Princeton: Princeton University Press. Grant, U.S. (1999/1885). Personal Memoirs. New York: The Modern Library. Güth, W., & Tietz, R. (1990). Ultimatum bargaining behavior. Journal of Economic Psychology,

11, 417-449. Hamilton, W.D. (1964). The genetical evolution of social behavior: I and II. Journal of Theoretical

Biology, 7, 1-52.

56

Happé, F.G.E. (1994). An advanced test of theory of mind: understanding of story characters’ thoughts and feelings by able autistic, mentally handicapped and normal children and adults. Journal of Autism & Developmental Disorders, 24, 129-154.

Happé, F.G.E. (1995). The role of age and verbal ability in the theory of mind task performance of

subjects with autism. Child Development, 66, 843-855. Happé, F., Ehlers, S., Fletcher, S., Frith, U., Johannsson, M., Gillberg, C., Dolan, R., Frackowiak,

R., & Frith, C. (1996). ”Theory of mind” in the brain. Evidence from a PET scan study of Asperger syndrome. Neuroreport, 8, 197-201.

Harbaugh, W.T., Krause, K., & Liday, S. (2000). Children’s bargaining behavior: differences by

age, gender, and height. Manuscript, University of Oregon. Harris, M.B. (1970). Reciprocity and generosity: some determinants of children. Child

Development, 41, 313-328. Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., & McElreath, R. (2001). In

search of homo economicus: behavioral experiments in 15 small-scale societies. American Economic Review, 91, 73-78.

Hill, E.L., Sally, D., & Frith, U. (in preparation). Investigating the influence of mentalising in the

Prisoner’s dilemma: introspective evidence from a study of individuals with autism. Hoffman, E., McCabe, K., & Smith, V.L. (1996). Social distance and other-regarding behavior in

dictator games. American Economic Review, 86, 653-660. Hoffman, E., McCabe, K., & Smith, V.L. (2000). The impact of exchange context on the activation

of equity in ultimatum games. Experimental Economics, 3, 5-9. Hoffman, E., McCabe, K., Shacat, K., & Smith, V. (1994). Preferences, property rights, and

anonymity in bargaining games. Games & Economic Behavior, 7, 346-380. Holden, A. (1990). Big deal: a year as a professional poker player. New York: Viking. Kagel, J.H., Kim, C., & Moser, D. (1996). Fairness in ultimatum games with asymmetric

information and asymmetric payoffs. Games & Economic Behavior, 13, 100-110. Kalish, C.W. (2002). Children’s prediction of consistency in people’s actions. Cognition, 84, 237-

265. Kiesler, S., Sproull, L., & Waters, K. (1996). A prisoner’s dilemma experiment on cooperation with

people and human-like computers. Journal of Personality & Social Psychology, 70, 47-65. Kiyonari, T., Tanida, S., & Yamagishi, T. (2000). Social exchange and reciprocity: confusion or a

heuristic? Evolution & Human Behavior, 21, 411-427. Klin, A., Jones, W., Schultz, R., Volkmar, F., & Cohen, D. (2002). Defining and quantifying the

social phenotype in autism. American Journal of Psychiatry, 159, 895-908.

57

Knight, G.P., Johnson, L.G., Carlo, G., & Eisenberg, N. (1994). A multiplicative model of the dispositional antecedents of a prosocial behavior: predicting more of the people more of the time. Journal of Personality & Social Psychology, 66, 178-183.

Kravitz, D., & Gunto, S. (1992). Decisions and recipients in ultimatum bargaining games. Journal

of Socio-Economics, 21, 65-84. Krebs, D. (1987). The challenge of altruism in biology and psychology. In C. Crawford, M. Smith,

& D. Krebs (eds.), Sociobiology and Psychology: Ideas, Issues, and Applications. pp. 81-118. Hillsdale, NJ: Lawrence Erlbaum.

Leekam, S., & Perner, J. (1991). Does the autistic child have a metarepresentational deficit?

Cognition, 40, 203-218. Leslie, A.M., & Thaiss, L. (1992). Domain specificity in conceptual development: evidence from

autism. Cognition, 43, 225-251. Matsumoto, D., Haan, M., Yabrove, G., Theodorou, P., & Cooke Carney, C. (1986). Preschoolers’

moral actions and emotions in prisoner’s dilemma. Developmental Psychology, 22, 663-670. McCabe, K., Houser, D., Ryan, L., Smith, V., & Trouard, T. (2001). A functional imaging study of

cooperation in two-person reciprocal exchange. Proceedings of the National Academy of Sciences, 98, 11832-11835.

Milinski, M. (1987). Tit for tat in sticklebacks and the evolution of cooperation. Nature, 325, 434-

435. Miller, P.H., & Aloise, P.A. (1989). Young children’s understanding of the psychological causes of

behavior: a review. Child Development, 60, 257-285. Murnighan, J.K. & Saxon, M.S. (1998). Ultimatum bargaining by children and adults. Journal of

Economic Psychology, 19, 415-445. Oosterbeek, H., Sloof, R., & van de Kuilen, G. (2001). Cultural differences in ultimatum game

experiments: evidence from a meta-analysis. Manuscript, University of Amsterdam. Pasquet, A., & Krafft, B. (1992). Cooperation and prey capture efficiency in a social spider,

anelosimus-eximius. Ethology, 90, 121-133. Paulus, M.P., Hozack, N., Zauscher, B., McDowell, J.E., Frank, L., Brown, G.G., & Braff, D.L.

(2001). Prefrontal, parietal, and temporal cortex networks underlie decision-making in the presence of uncertainty. NeuroImage, 13, 91-100.

Perner, J. (1979). Young children’s preoccupation with their own payoffs in strategic analysis of 2 x

2 games. Developmental Psychology, 15, 204-213. Pillutla, M.M., & Murnighan, J.K. (1996). Unfairness, anger, and spite: emotional rejections of

ultimatum offers. Organizational Behavior & Human Decision Processes, 68, 208-224. Rigdon, M.L. (2002). Dictator games. Encyclopedia of Cognitive Science, in press.

58

Rilling, J.K., Gutman, D.A., Zeh, T.R., Pagnoni, G., Berns, G.S., & Kilts, C.D. (2002). A neural basis for social cooperation. Neuron, 35, 395-405.

Rommel, E. (1953). The Rommel Papers. B.H. Liddell Hart & P. Findlay (eds.). New York:

Harcourt, Brace, & Jovanovich. Sally, D. (1995). Conversation and cooperation in social dilemmas: a meta-analysis of experiments

from 1958 to 1992. Rationality & Society, 7, 58-92. Sally, D. (2000). A general theory of sympathy, mind-reading, and social interaction, with an

application to the prisoners’ dilemma. Social Science Information, 39, 567-634. Sally, D. (2001). Into the looking glass: discerning the social mind through the mindblind.

Advances in Group Processes, 18, 99-128. Schelling, T.C. (1960). The Strategy of Conflict. Cambridge: Harvard University Press. Stone, V.E. (2000). The role of the frontal lobes and the amygdala in theory of mind. In S. Baron-

Cohen, H. Tager-Flusberg, & D.J. Cohen (eds.), Understanding Other Minds. Perspectives from Developmental Cognitive Neuroscience (2nd ed.), pp. 253-273. Oxford, England: Oxford University Press.

Sullivan, K., Zaitchik, D., & Tager-Flusberg, H. (1994). Preschoolers can attribute second-order

beliefs. Developmental Psychology, 30, 395-402. Turner, P.E., & Chao, L. (1999). Prisoner’s dilemma in an RNA virus. Nature, 398, 441-443. Underwood, B., & Moore, B. (1982). Perspective-taking and altruism. Psychological Bulletin, 91,

143-173. Vogeley, K., Bussfeld, P., Newen, A., Herrmann, S., Happé, F., Falkai, P., Maier, W., Shah, N.J.,

Fink, G.R., & Zilles, K. (2001). Mind reading: neural mechanisms of theory of mind and self-perspective. Neuroimage, 14, 170-181.

von Neumann, J., & Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton:

Princeton University Press. Wechsler, D. (1998). Wechsler Adult Intelligence Scale Revised (3rd ed.). New York: The

Psychological Corporation. Wilkinson, G.S. (1984). Reciprocal Food Sharing in the Vampire Bat. Nature, 308, 181-184. Wilson, C.C., Piazza, C.C., & Nagle, R.J. (1990). Investigation of the effect of consistent and

inconsistent behavioral example upon children’s donation behaviors. Journal of Genetic Psychology, 151, 361-376.

Wimmer, H., & Perner, J. (1983). Beliefs about beliefs: representation and the constraining function

of wrong beliefs in young children’s understanding of deception. Cognition, 13, 103-128.

59

FIGURE LEGENDS

Figure 1: A generic prisoner's dilemma

Figure 2: Mean number of cooperative responses across all 16 rounds of each Prisoner's

dilemma game (Figures 2a and 2c) and percentage of each participant group

cooperating on the first round of each Prisoner's dilemma game (Figures 2b and 2d)

Figure 3: Percentage of each group (6-10 year olds combined) being 'reliably cooperative',

'reliably competitive' and 'variable' when playing the human opponent (Figure 3a),

computer opponent (Figure 3b) and encouraged cooperation (Figure 2c) versions of

the Prisoner's dilemma game

Figure 4: Percentage of each group (6-10 year olds combined) cooperating on each of the

sixteen rounds of each Prisoner's dilemma game

Figure 5: Mean points offered (max=10) by the participant groups to the confederate across

four rounds of the dictator and ultimatum games (Figures 5a and 5c) and on first

round only (Figures 5b and 5d). Error bars show standard deviation

Figure 6: Dispersal of first round ultimatum offers by adults and children with and without

autism (Figures 6a and 6b, respectively)

Figure 7: First round ultimatum offers based on second order false belief test result

60

Figure 8: Mean points offered (max=10) by the participant to the confederate for each

participant group (6-10 year olds combined) on the eight rounds of the dictator and

ultimatum games

Figure 9: Mean number of points (max=10) rejected by participants (6-10-year-olds

combined). Error bars show standard deviation

61

Table 1.

Participant details

Normal participants Autism participants

6 years 8 years 10 years Adult Children Adult

N.

14

19

18

15

18

15

Male (female) 8 (6) 10 (9) 10 (8) 7 (8) 16 (2) 12 (3)

CA

Mean

SD

Range

(yr.mth)

6.7

.2

6.04-6.11

8.5

.3

8.00-8.11

10.6

.3

10.03-10.11

34.0

12.3

21-62

10.6

3.1

6-15

34.4

11.0

18-49

VIQ*

Mean

SD

Range

109.22

10.87

98.4-133.3

103.22

16.35

75.1-136.5

109.3

11.92

89.8-129.6

117.13

11.01

95-130

96.29

33.7

63.2-211.9

92.29

21.21

60-128

* calculated from verbal mental age for children and WAIS for adults

62

Table 2a.

Percent of each group passing first- and second-order false belief tasks

Normal participants Autism participants

6 years 8 years 10 years Adult Children Adult

1st order

2nd order

100

71.43

100

94.12

100

100

100

100

66.67

55.56

86.67

13.33

Table 2b.

Mean (SD) score (max=16) of each adult group on Happé’s mentalising stories (left panels of table) and mean (SD) time (sec) taken to

complete only those stories where a mentalising response (i.e., response gaining the maximum score) was given (right panels of table)

Normal adults Autism adults Normal adults Autism adults

Accuracy:

Mean

Range

13.8 (1.61)

11-16

8.87 (4.09)

1-15

Time (sec):*

Mean

Range

23.43 (5.75)

15.5-36.24

53.13 (33.82)

22.09-137.12

* based on 88 and 47 mentalising responses in the normal adult and autism adult groups respectively.

63

Table 3.

Points awarded to the participant and confederate in the Prisoner's dilemma games

PARTICIPANT

Cooperate

(triangle)

Compete

(circle)

Cooperate

(triangle)

3, 3

1, 4

CO

NFE

DER

ATE

Compete

(circle)

4, 1

2, 2

Note. The players did not see the terms cooperate and compete. The terms confederate and

participant were replaced with the appropriate names of the two players in each case.

64

Table 4.

% of each participant group making the cooperative (triangle) choice on the last round of the

human or computer opponent (round 32) and the first round (round 33) of the encouraged

cooperation PD

Last round against human

or computer opponent

First round of encouraged

cooperation PD

Sig. level

Normal adults (n=15) 26.67 93.33 < .002

6 years (n=14) 7.14 35.71 n.s.

8 years (n=18) 16.67 50 n.s.

10 years (n=18) 38.89 55.56 n.s.

6-10 years combined (n=50) 22 48 < .01

Autism adults (n=15) 33.33 73.33 .058

Autism children (n=18) 38.89 38.89 n.s.

65

FIGURE 1.

a > c > d > b and c+c > a+b

Player 1

Player 2

Coop.

Defect

Defect Cooperate

b, a

d, d a, b

c, c

66

FIGURE 2A FIGURE 2B

0

2

4

6

8

10

12

14

16

Human opponent Computer opponent Encouragedcooperation

010

2030

4050

6070

8090

100

Human opponent Computer opponent Encouragedcooperation

FIGURE 2C FIGURE 2D

0

2

4

6

8

10

12

14

16

Human opponent Computer opponent Encouragedcooperation

0

1020

3040

50

6070

8090

100

Human opponent Computer opponent Encouragedcooperation

67

Figure 3a: Human opponent Figure 3b: Computer opponent Figure 3c: Encouraged cooperation

��������������������������������������������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������

���������������������������������������������������

Autism adults

�����������������������������������������������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������

��������������������������������

Normal adults

����������������������������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

Autism children ����������������������������������������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������������������������������������������������������������������

Normal children

������������������������������������������

�����������������������������������������������������������������������������������������������������������������������������

����������������������������

Autism children

���������������������������������������������������������������������������������������������������������������������������������������������������������

Autism adults ���������������������������������������������������������������������������������������������������������������������������������������

��������������������������������������������������������������������������������

Normal adults

���������������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������

����������������������������

Normal children

��������������������������

������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������

Autism adults

���������������������������������������������

����������������������������������������������������������������������������������������������������������������������������������

Normal adults

������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������

����������������������������

Autism children

����������������������������

�������������������������������������������������������������������������������������������������������������������

������������������������������������������

Normal children

���reliably-competitive

���variable

���reliably-cooperative

68

FIGURE 4.

69

Normal adults

0102030405060708090

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16Round

Autism adults

0102030405060708090

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Round

Normal children

0102030405060708090

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16Round

Autism children

0102030405060708090

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16Round

Human opponentComputer opponentEncouraged cooperation

FIGURE 5A FIGURE 5B

FIGURE 5C FIGURE 5D

������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������

���������������������������������������������������������������������������������������������������������

������������������������������������������

���������������������������������������������������������������������������������������������������������������������������0

2

4

6

8

10

Dictator game Ultimatum game

Mea

n po

ints

off

ered

������������������������������

���������������������������������������������������������������������������������������������������������������������������������������������

���������������������������������������������������������������

���������������������������������������������������������������������������������������������������������������������

������������������������������������������������

������������������������������������������������������������������������������������������������������������������0

2

4

6

8

10

Dictator game Ultimatum game

Mea

n po

ints

off

ered

������ 6 years8 years������ 10 years������ Adults

��������������������������������������������������������������������������������

��������������������������������������������������������������������������������������������������������������������������������������������0

2

4

6

8

10

Dictator game Ultimatum game

Mea

n po

ints

off

ered

��������������������������������������������������������������������������������

��������������������������������������������������������������������������������������������������������������������������������0

2

4

6

8

10

Dictator game Ultimatum game

Mea

n po

ints

off

ered

������ AdultsChildren

70

FIGURE 6A

������������������������������������������������

������������������

��������������������������������������������������������������������������������

������������������

0

0.1

0.2

0.3

0.4

0.5

0.6

0 1 2 3 4 5 6 7 8 9 10

Offer to Counterpart

Prop

ortio

n of

Sam

ple

��� AutisticNormal

FIGURE 6B

������������������������������������������������������

������������������������

������������

������������������������������������������������������������

������������������������

������������

����������

0

0.1

0.2

0.3

0.4

0.5

0.6

0 1 2 3 4 5 6 7 8 9 10

Offer to Counterpart

Prop

ortio

n of

Sam

ple

���

Normal

Autistic

71

FIGURE 7.

������������������������������������������������

����������������������������������������������������������������������

������������������

������������������������������

���������������������

���������������������

������������������

0

0.1

0.2

0.3

0.4

0.5

0.6

0 1 2 3 4 5 6 7 8 9 10

Offer to Counterpart

Prop

ortio

n of

Sam

ple

����

Passed

Failed

72

FIGURE 8.

73

Normal adults

0

2

4

6

8

10

1 2 3 4 9 10 11 12

Round

Autism adults

0

2

4

6

8

10

1 2 3 4 9 10 11 12

Round

Autism children

0

2

4

6

8

10

1 2 3 4 9 10 11 12

Round

Normal children

0

2

4

6

8

10

1 2 3 4 9 10 11 12

Round

Dictator gameUltimatum game

FIGURE 9.

0

2

4

6

8

10

Normal adults Autism adults Normal children Autism children

Mea

n po

ints

reje

cted

74