[gwern]

If you look at the abstract, all of the correlations actually show that having a cat does predict mental illness at all ages considered. But the effect sizes are relatively low (albeit consistent with the more direct studies looking at people actually infected with toxoplasma rather than people who might be infected), so with a few hundred or thousand people, it technically doesn't satisfy p<0.05. Nevertheless, this is further evidence for toxoplasma having bad effects.

This seems to be a trend with toxo papers - take a too-small sample, do no power or Bayesian analysis, find all bad correlations, but declare evidence of safety anyway.

[smallnamespace]

Yes, frequentist inference only allows you to reject the null hypothesis (cats do not cause mental illness), but not to accept it if you fail to reject [1].

The abstract as written is simply wrong from a statistical point of view:

our study strongly indicates that cat ownership in pregnancy or early childhood does not confer an increased risk of later adolescent PEs [2]

If you instead adopt Bayesian reasoning, then the direction of the measured effect is weak evidence that cats cause illness later in life. Either way you cut it, the study author's interpretation fails to coincide with the actual data they collected.

[1] https://liesandstats.wordpress.com/2008/09/08/accept-the-nul...

[2] https://www.cambridge.org/core/journals/psychological-medici...

[abandonliberty]

That's how you get published! To get in Nature, you need to demonstrate a shocking, novel result. Small sample size really helps.

Edit: Source: Neuroscientist, multi-term dean, honorary doctor, and I would keep going but it would end up personally identifiable. Let's just say I have this on extremely good authority. There's a reason why "Most science research findings are false. " https://qz.com/530064/most-science-research-findings-are-fal...

[jessriedel]

The jump from "we could not detect a difference" to "there is no difference" is so frustrating. (And this is a mistake the authors make in this case, not just the journalists.)

[beojan]

They did say "no evidence of a link" in the title and abstract, though the conclusion at the end of the abstract unfortunately says the study strongly suggests there is no link. Given the quoted odds ratios are > 1, this is a strange statement to make.

[norea-armozel]

I think for most people in the world, pork products are more likely to be a source of the parasite than cats though. And there's quite a few people who eat pork even outside the US.

[dkhenry]

I don't think its just toxo papers, people use bad math in all kinds of scientific papers.