Excited to be hanging out at the Tinbergen Institute @erasmusuni today and tomorrow to hear Guido Imbens speak!! And on Friday I'll be talking about (what else) Bayesian evidence aggregation! Say hi if you're around!
Being invited to the econometric institute at Erasmus has been a dream of mine so I am extremely chuffed to be here!
Ok we are here and I was late because I got lost (smart!) but fortunately I didn't miss Guido's comments on Angus Deaton's claim that RCTs are no more internally valid than observational studies: "he's wrong".
(Yeah yeah he's already on the record on this, sure)
Guido has also just noted that the troubles people have with understanding p values arise because the p value is not the right object to care about when you are facing a decision theoretic problem. 🌟💥⚡️
Whoohoo it's time for SUTVA!
Timely reminder from Guido Imbens that clustering standard errors changes the estimand and hypothetical causal unit. If you cluster at e.g. the state level, you are now estimating the uncertainty you'd have about the effect in a hypothetical new state you have not seen yet.
