The crucial issue is why most children with autism fail on the ‘belief’ question in this experiment, while children in the two control groups mostly pass. One explanation is that since children with autism have difficulties with pretend play, the participants misunderstood the pretend ‘drama’ that the experimenter enacted with the dolls. But this was ruled out by later experiments that replicated the task using real people to enact the scenario.
Baron-Cohen et al. argued that the children with autism tend to fail the belief question because, instead of 'putting themselves in Sally's shoes', they assume that her belief about where the marble is hidden is the same as their own knowledge of where the marble really is. In short, the study appears to show that children with autism lack the capacity for understanding another person's mental state (which in this case is different from their own). This basic finding has been replicated (repeated with the same outcome) many times, employing numerous variants of the task. Note, however, that the number of children on the autism spectrum who fail Sally-Anne varies from one study to another.
The following reminiscence, from an able person with Asperger syndrome, illustrates the difficulties that arise:
"One of the most recurrent problems throughout middle childhood was my constant failure to distinguish between my knowledge and that of others. Very often my parents would miss deadlines or appointments because I failed to tell them of these matters."
(Sarah in Sainsbury, 2000, p.60)
Besides the main differences in test performance, the video clip also illustrates differences in how the children with autism and the typically developing child engage with the task. The children with autism tend to take the task fairly seriously, while the typically developing child clearly sees it as a playful scenario or a game.
It is important to note that the Sally-Anne task is not a way of diagnosing autism. Some people who are not on the autism spectrum may also find the Sally-Anne task difficult. Moreover, critics of the ToM approach point out that false belief tasks only test one kind of theory of mind skill - and one that is not very close to everyday 'mindreading'.