Because autism is a pervasive disorder and can affect so many different aspects of cognition, it's often very difficult to pin down the underlying cause of any particular deficit or weakness. If someone with autism struggles on a test they are given, it could potentially be for any number of reasons - memory difficulties, poor concentration, stress brought on by the test situation, failure to understand what it was they were supposed to be doing. And so on. All of these factors might be expected to lead to poor performance. But when people with autism perform exceptionally well on a task, we can rule out these possible alternative explanations. If we can understand the origin of these strengths, then we might also gain some insight into some of the difficulties faced by people with autism.
One of the most intriguing findings from autism research in recent years is that (on average at least) people with autism perform much better than expected on visual search tasks. In a typical visual search task, participants view a computer display containing multiple objects (usually small shapes or letters) and simply have to determine whether a particular object is present or absent in the screen. For example, in the display below, the task is to detect an orange square. The task can be made more or less difficult by changing the number of distracters (non-target objects) in the display or by altering the similarity between the target and distracters.
In general, it seems that the autistic advantage increases as the task difficulty increases. The reason for this, however, remains a matter of conjecture.
To try and answer this question, Joseph et al. (2009) tested 17 high-functioning children with autism and a control group of typically developing children matched on age and nonverbal intelligence. The kids had to search for a letter T on the screen and press one button if it was present and another if they couldn't find it (on half the trials it genuinely wasn't there). Their eye-movements were monitored while they were doing the task to determine whether there were any differences in search strategy.
Overall, the kids with autism were on average quicker to respond than the controls. Importantly, they were also slightly (but not significantly) more accurate. Often in these kinds of experiments there's a trade-off between speed and accuracy. Some people respond quickly and make more errors. Others make sure they get the response correct all the time but as a result are a bit slower. The fact that kids with autism were quicker and at least as accurate means that they genuinely were better at the task.
So why were the kids with autism so good?
One possibility is that people with autism are somehow better at filtering out distracters and guiding their search towards the target. However, this is unlikely to explain good performance on this particular task. Joseph et al asked their participants to search for a letter 'T' in a display that was otherwise full of letter 'L' distracters. To make things even more difficult, the target could be in any orientation (ie right-way-up, up-side down, on one side, or on the other side), and the 'L's on each trial were a mixture of the four different orientation. This meant that there were no simple features that could be used to filter out some of the distracters. For example, ignoring anything with a horizontal line in it would have been a useless strategy because the target contains a horizontal line too. If kids with autism were good at visual search because they were better able to filter out distracters, they shouldn't have had an advantage here. But they still did.
Another possibility is that people with autism have better memory for the locations they've already searched. To test this possibility, in half of the trials, the objects on the screen kept moving around (twice every second). Again, this should have removed the 'autism advantage' but didn't, suggesting that the answer lies elsewhere.
The eyetracking data also revealed surprisingly little in the way of group differences. Joseph et al looked at a number of different factors:
- Did kids tend to look more in some quadrants of the screen than in others?
- Did they look more at objects in the centre or in the periphery?
- Did their eyes move from object to neighbouring object or zip around all over the screen?
- How many different fixations did they make per trial?
- How long were they fixating each object?
Finally, the authors looked to see whether any measures derived from the visual search task correlated with measures of autism symptom severity. The only statistically significant correlation was between the Social scale of the Autism Diagnostic Observation Schedule and 'search intercept' in the static condition. On the face of it, the correlation is intriguing as it suggests that there may be a link between whatever low-level perceptual or attentional mechanisms are affecting visual search and the brain mechanisms that lead to social impairment.
However, before getting too excited, its important to understand what search intercept is. Participants are generally slower to respond the more items there are on the screen. If we plot reaction time against display size, we can typically fit a straight line through all the points. It's usually argued that the slope of the line relates to the actual search process (ie locating the target) because the more objects there are on the screen the more searching you have to do. If the search is difficult and slow then the slope will be steep because adding more items has a big effect on search times.
The intercept is derived by extrapolating the line fitted to the data and seeing where it crosses the y-axis. In other words, it provides an estimate of what the reaction time would have been if there hadn't been any items in the display. Of course, this is hypothetical because if there weren't any items in the display then there wouldn't be a visual search to do! However, the intercept is often treated as an estimate of the time to complete all the processes that contribute to reaction times but aren't involved in the actual searching process (eg the time to press the button having made a decision).
Unfortunately, slopes and intercepts are very unreliable. Low reliability means that, if we conducted the same measure again with the same person, we might well get a very different outcome. Slopes are unreliable because they are calculated from multiple measures and all the small measurement errors made along the way can add up. It's even worse for the intercept because small errors get magnified the further you extrapolate beyond the data. This concern with reliability is born out when we look at the intercepts of each individual (plotted in Joseph et al's Figure 10). Some of the kids with autism have negative intercepts. If we take the intercepts at face value as a measure of non-search related process time, then we have to conclude that these kids are able to do these things in negative time (ie they are magic).
Two, the authors conducted 24 correlational analyses between various measures relating to visual search performance and autism symptoms. This was the only one that was statistically significant. Statistical significance means that it's quite unlikely to have achieved this correlation by chance if you were just looking at the one correlation. But if you run 24 analyses then there's a fairly high probability that at least one of them will be significant. It's like buying a million lottery tickets. Even though each one is unlikely to win, you've got a pretty good chance of at least one having the winning combination. Strictly speaking, the authors should have corrected for multiple comparisons, which raises the barrier for statistical significance if you're doing lots of tests. If they'd done that, then none of the correlations would have been significant.
So the correlation between visual search slope and social impairment doesn't really stand up to closer scrutiny. It may well have arisen by chance and in any case, the intercept scores really aren't to be trusted. But this doesn't take away from the fact that the main results are pretty intriguing.
To sum up, Joseph et al's paper is entitled "Why is visual search superior in autism spectrum disorder?" The authors are able to rule out some potential explanations - superior memory for locations, superior filtering of distracters. However, the study really poses more questions than it answers. In particular, previous visual search studies have used static displays and found that people with autism are generally faster to respond than controls. Joseph et al found that kids with autism were faster to respond to dynamic displays but not to static displays. All very confusing.
Clearly, there is something very interesting going on. For some reason, people with autism are good at visual search tasks, and the harder we make the tasks it seems, the more they excel. Joseph et al's eye-tracking data suggests that there isn't anything obviously unusual about the way they are completing the task. They're just very good at it.
Understanding why they are good at visual search may well tell us something about how the autistic brain is wired up. But we're still a long from understanding visual search - let alone anything else.
Reference:
Robert M. Joseph