Doesn't appear to have held up to replication. [1]
>However, unlike Tsukuhara et al., we could establish only small to nonexistent associations between cognitive ability and pupil size (r’s -.01 to .06, r with g specifically = .05)
In fact, a quick scan down the Scholar results shows other replications have found the opposite effect [2]
>The results found that high-Gf adolescents showed smaller tonic pupil size in both tasks
If this is a replicable result, they are probably measuring subjects' interest in being in the lab and doing tasks. Some people care, and pay attention and do well. Others are just there for class credit and blow it all off.
But their paper argues that the null results were due to A)using tasks that do not have the strongest association (i.e. they did not measure fluid intelligence), and B)setting baseline lighting levels too brightly so that pupils began at the minimum possible size.
https://www.sciencedirect.com/science/article/abs/pii/S00100...