John Horton points to Sims‘s comment on Angrist and Pischke:
Top of page 8—he criticizes economists for using clustered standard errors—suggests using multilevel models instead.
Awesome! So now there are at least two Nobel prize winners in economics who’ve expressed skepticism about controlled experiments. (I wonder if Sims is such a danger in a parking lot.)
P.S. I’m still miffed that this journal didn’t invite me to comment on that article!
Nicely written paper especially like the last phrase
“regrettable because increased computing power and the new methods of inference
that are arising to take advantage of this power make such narrow, simple
approaches [sandwich estimation] to data analysis increasingly obsolete”
Maybe is I copied it and changes the econometerics references to epidemiological it would
make a really nice paper for me to submit some where…
I would just have to add the right spelling and other errors ;-)
Is Heckman the other skeptic of experiments you’re referring to?
Let’s just say that all the clues are right there in my post.
You guys do know he gave lecture literally called, “Bayesian Methods in Applied Econometrics, or, Why Econometrics Should Always and Everywhere Be Bayesian.”
Here is the free link: http://sims.princeton.edu/yftp/EmetSoc607/AppliedBayes.pdf
Nice paper. Thanks for posting it.
I guess it’s Heckman. Gelman already linked to Heckman’s remarks about natural experiments in this blog…