With the usual #statistician's caveat that #causality is really hard to sort out in data like this even with good #controls ... yes, I believe this. And the #mechanisms aren't hard to find, either. 😐
🇳🇱
@medigoth Causality is not as hard as people think. The problem is you cant use correlation to get there in any reliable way (since most variables are unknowns)... But that isnt to say you cant effectively determine causality, you just lean into other tests like causality tests rather than relying on correlation alone.
In this case its simple, look at something like a granger causality test. In populations where there was a sudden rise in christianity (or racism) then look to see if the other one spiked with some time delay and if so which of the two preceded the other.
This approach isnt perfect of course, but its going to get you far closer to an accurate analysis than trying to look at correlation.
@medigoth Good, then we mostly agree.
That said correlation only shows correlation I think we all agree it **never** is evidence in and of itself of causation, for that wee either 1) need to control all variables in a lab setting (usually not possible) or 2) use other methods that can demonstrate (or at least suggest) causation... By the sound of it we agree here as well.
@freemo Sure, causal inference, largely from time series data and/or Mendelian randomization, is a big part of my work. The methods as described in the article don't seem to be Granger or SEM, though. Of course one should never rely on popular science reporting for a thorough understanding of methods. 😀 I'd have to read the paper to be sure.
Because I loathe the blithe use of "correlation is not causation" to dismiss legitimate causal inference results, I want to see researchers being really careful when making causal claims from observational data. _If_ they met that standard here, good for them.