Read on Twitter
@talyarkoni This. We fully acknowledge methods and interpretability limitations (loading particularly heavily on the causality part) and I firmly believe we'd have agreed on most of the raised concerns. Just wish I had to respond to p2p criticism instead of insults [thread] https://twitter.com/talyarkoni/status/1314293522082787328
1/ Whilst we agree that FTSE100 may represent a latent variable, which is not only reflective of economic performance of a country, prvs studies've indicated that (despite small ESs) SM indexes are assocd w population's wellbeing, suicide rates, and even number of car accidents
2/ Our conclusions rely on random sampling assumptions and even despite the fact that our results survived all of the undertaken adjustments for potential analytical biases we cannot completely rule-out a possibility that these assumptions are not violated.
3/ To put it simply,SM moves may influence enrolment likelihood of individuals with certain psychological and biological traits and therefore the collected individual measures inherently represent the consequences of the (economic) state and not rand "snapshots" of the population
4/ one of the major concerns was potential scanner drifts (see e.g. https://www.biorxiv.org/content/10.1101/646695v2). This is why we also adjusted for seasonal effects, tried de-trending the TS and looked into different frequency bands.
5/ one of the first questions we asked ourselves was "yeah, you’d expect some correlation with so much data". This is where the comparison to other markets is important. And that the lagged correlations go in a specific direction.
6/ whilst we agree that the identified associations may be spurious and be a subject for many interpretations, this does not mean that they are not worth investigating, provided the limitations are acknowledged and caution is advised.
7/the causality part was also extensively discussed within the group.Note, it is not claimed as the main (prereg'ed) analysis and was exploratory. We're aware of the autocorrelations and non-stationarity in the data, explicitly reporting and trying to address them in the analyses
8/ This is also one of the reasons why we are abstaining from commenting on duration of the estimated effects.
