Excellent comments and important questions, thanks. Can only scratch the surface, but here's a belated start, I hope some of this is helpful.

About Coding, low processing speed seems unlikely; motor atypicalities seem more likely, possibly with other factors (task instructions, atypical experiences) thrown in. Asperger individuals might do well on the non-scored part of Coding. But these are all only guesses.

Coding does show the problem of intelligence tests requiring specific abilities which are important, but don’t necessarily involve intelligence. If I suddenly lose my dominant hand or my vision, and my Coding score plummets, am I less intelligent now?

Of course autism issues are complicated, and the point isn’t that Raven is the only approach (e.g., see John Connolly’s work on PPVT performance via N400). Still, there are plausible though as yet mostly speculative reasons why Raven can be a good way to assess autistic intelligence. This is keeping in mind that, as Deary et al. (2010; doi:10.1038/nrn2793 p.209) wrote, there is a lot of room for “differences in how individuals use their brains for intelligent performance.” And as we point out in our discussion, the interpretation (which we have encountered) that Raven does not really assess intelligence in autistics is problematic.

The well-known problems with Raven norms (we used RSPM for all participants) are very serious and extremely difficult to address. Possibly this helps explain why a range of Raven findings in the autism literature, going back to the 1970s, remained uninterpreted and ignored, leading to the claim that autistics must have a major deficit in fluid intelligence.

When we first presented an autism Raven vs Wechsler scatterplot at IMFAR 2004, the universal response was that these striking data were merely an artifact of bad Raven norms. While we recognized the difficulties and pitfalls (especially if you take the major Raven manual seriously, and we did), we thought this should be tested rather than assumed.

It has been fascinating to see the work of other groups who have, since 2007, looked into this important but very diffcult problem. Some groups have access to independent Raven norms which allow more conventional analysis but raise other questions. Overall, very different approaches have been attempted which, like ours, have a range of strengths and shortcomings. Also, the literature has offered a few incidental glimpses into Raven data from large databases.

Possibly at some point there should be a warts-and-all review. Ideally, the norms problem would be resolved, and data from very large autistic samples would be reported in a more than incidental way. If it turns out (as our findings here suggest) that some or many autistics perform well in the newish Wechsler Matrix Reasoning subtest, I hope this won’t automatically be dismissed as just another aberrant islet of ability.