RT @jondr44: #EconTwitter I have created a new DiD Resources page on my website. It has links to all of my DiD teaching slides, coding exer…

@jakekohlhepp Thanks! Thanks these counterexamples are helpful! Now thinking of the special case where the signal is normally distributed with mean equal to the true state (Then the claim holds if for example the prior is normal)

@jakekohlhepp I'm interested in the posterior mean conditional on the signal being equal to some realization (so in that sense that signal is no longer random). I'm not interested in thinking about the posterior mean as being random due to the sampling variance of the signal

@jakekohlhepp To clarify: this holds when the prior is normal with mean zero and the sampling distribution is normal (because then the posterior mean is just the signal pulled toward zero by some positive shrinkage factor)

RT @katy_milkman: If you’re in or adjacent to the field of #BehavioralScience & aren’t talking about the new evidence of data fraud by ⁦@Ha…

@c_r_walt @nberpubs Excellent. Thanks!

@Andrew___Baker At least in some languages merge/join calls check that the keys are as expected. In SQL it’s the Wild West, with 1:1, m:1, and 1:m joins executing without error even if join is truly m:m

@Andrew___Baker @AndreyFradkin

@tyleransom You could use alternative math delimiters \( \)

@FairweatherPhD Travis Barker? Sure. But Paramore? This seems like a much bolder bet

Can anyone point me to papers on how best to formalize tradeoffs between type I and type II error? Hoping to formalize tradeoff via some objective function to make clear whether on net decrease in one error but increase in other error is "good"

@paulgp Think carefully about the economic content underlying identifying assumptions

@palapid @jenniferdoleac Makes sense! (But IMO is confusing Econ jargon that could be stated more clearly!)

@adamjnafa They’re often used as a diagnostic for AB tests. If we aim to allocate 50 percent of units to A and 50 percent of units to B but we can reject the null that the expectation of the observed shares is 50:50 then something is wrong. See for example

@jt_kerwin @GioCirco Just make sure you interact your treatment indicators with your included covariates. Otherwise precision gains (relative to no covariates) aren’t guaranteed!

RT @florianederer: Local governments are borderline criminal when auctioning off claims to homes with overdue property taxes. 1) Average t…