[zfell]

Could there be a much simpler explanation?

The most recent study the article sites (Morgan et. al, 2018) states that out of 468k people in the population "1870 children developed schizophrenia (0.4%) while 9120 developed a psychotic illness (1.9%). None of the 66 children with cortical blindness developed schizophrenia or psychotic illness."

If we don't assume there is a relationship between schizophrenia and cortical blindness, it's not surprising that none of the 66 people who had cortical blindness developed schizophrenia. Simple binomial approximation will yield a 77% (0.996^66) probability. Am I missing something?

Also I find the difference in prevalence rates of schizophrenia of 0.4% in the Morgan et al. paper vs 0.72% in McGrath et al. 2008 odd.

Morgan et al. 2018: https://www.sciencedirect.com/science/article/abs/pii/S09209...!

[zfell]

Found a link to the full paper here: https://www.gwern.net/docs/psychology/2018-morgan.pdf

It explains the difference in prevalence rates: "While median lifetime morbid risk for schizophrenia is estimated to be 0.72% (Saha et al., 2005), many of our young cohort had still not passed through the window for schizophrenia onset."

However, their claim for a relationship is indeed very weak: "In our data, this dropped to 0.2% for congenital or early peripheral blindness and was zero for congenital or early cortical blindness [0 cases out of 66]. Our data further suggest that the protection offered by cortical blindness may extend to a broader range of psychotic disorders and that risk of psychosis may effectively be reduced to zero."

It's ironic that they start the discussion stating "The results from this whole-population cohort, although possibly underpowered, lend confidence to findings from smaller case studies". Yes, it is underpowered, and I'm surprised it was published.

[Fomite]

I mean, if it is genuinely a whole population cohort, there's a legitimate question to what "power" even means in that context, because there's no statistical uncertainty. It simply is.

Whether or not hypothesis testing in registry data is justifiable was once the topic of a legendary fight in my graduate department during someone's defense.

[throwawayhhakdl]

No that’s not correct.

I mean if we’re evaluating the claim that no blind people got schizophrenia then sure. But the claim is that blind people don’t get schizophrenia, which is different. The population for this claim is not “all the humans alive” but rather “all potential humans” (which is not possible to sample).