Comments for Mostly Harmless Econometrics

Comment on 42 clusters references swap by josh

josh — Sun, 29 Jan 2012 23:59:44 +0000

try this one Ilyssa: Journal of Econometrics 141 (2007) 597–620 (C Hansen, ‘Asymptotic Properties . . .”
Hansen doesn’t mention Canada but he does report Monte Carlos for a scenario with N=10 in the cross-sectional dimension.

A caution, as we noted in MHE 8.2.3 on fewer than 42 clusters: Alas, the bottom line here is not entirely clear.

Also, we don’t know too much about the finite sample behavior of two-way clustering procedures. It’s not clear you wouldn’t be better off just to fatten up the cluster in one dimension and ignore the second. I’m sure someone is working on that (or at least they should be!)

Comment on 42 clusters references swap by Ilyssa

Ilyssa — Sun, 29 Jan 2012 16:27:36 +0000

I don’t see the Canada reference in either Hansen paper. I’m not lucky enough to be studying the 50 U.S. states but rather the 10 Canadian provinces. Would the t distribution of G-k be such that k is at least 2? That is, k = 2 if there is exactly one regressor that induces that Moulton problem.

Comment on Imbens and Angrist Discover LATE by Jonathan

Jonathan — Sun, 27 Nov 2011 04:48:29 +0000

Nothing better than two geniuses at work!

Comment on High Fashion at the Spring Meeting of Young Economists by bruna zolin

bruna zolin — Mon, 11 Jul 2011 16:17:12 +0000

bellissimi

Comment on Regression what?! by josh

josh — Mon, 11 Jul 2011 01:29:54 +0000

sorry Dean . . . I mean weighted by the variance of treatment given X (now corrected). In other words the weights are P[D=1|X](1-P[D=1|X])

why would I want to weight in this manner? One reason is that these are the effecient weights under constant effects (the weights are maximized in cells with P=1/2, i.e., equal numbers of treatment and controls). So that seems like a good benchmark. Somewhat more causally, where there’s more variance in treatment there’s more information about treatment. But you’re right that regression does not estimate the effect of treatment on the treated. MHE chapter 3 compares the two (based on my 1998 paper on the effects of military service.)

Comment on Regression what?! by Matt Bogard

Matt Bogard — Mon, 11 Jul 2011 01:02:54 +0000

Thank you for the followup on this post. I also tried to contact Dr. Gelman via email to get more clarification from his perspective. I did not realize he also followed up with comments on his blog . I did not intend to start a debate, I just wanted an answer and I wasn’t sure if I would get one. (being more than just a curiousity, I’m in the throws of trying to apply some of these techniques in my own work). My intentions were ‘mostly harmless.’ You have assured me that my initial reading/interpretation of your book so far has been on track. Now on to chapter 5!

Comment on Regression what?! by Dean Eckles

Dean Eckles — Sun, 10 Jul 2011 22:41:11 +0000

Why do we want to weight by the variance of X? When there is essential heterogeneity in treatment effects, don’t we want an average treatment effect for a particular goal (say, the treated).

Comment on Principals of principal stratification by josh

josh — Tue, 15 Feb 2011 03:26:41 +0000

Thanks for your question Mike. I have to admit that I don’t know anything about PS beyond what I read in the papers . . . but since the strata you mention are unobservable, how could we use them for conditioning?

Can you treat a BC as endogenous and instrument it? Yes (in principal, at least, if you’ve got extra instruments lyin’ around.) Griliches and Mason (1972) is a classic example. See also Chamberlain (1978).

Comment on Multiple endogenous variables – now what?! by josh

josh — Thu, 13 Jan 2011 04:11:31 +0000

Good eye Lulu! This is a case of the exception proving the rule perhaps. In Angrist (2006), two treatments are randomly assigned with less than full compliance. Two instruments for compliance are available in the form of the original assignment. Pretty clean set-up! As I wrote earlier, “models with multiple endogenous variables are indeed hard to identify and the results can be hard to interpret”. But not this one: its a randomized trial, and IV is the appropriate adjustment for non-compliance. JA

Comment on Multiple endogenous variables – now what?! by lulu millcreek

lulu millcreek — Wed, 12 Jan 2011 04:20:45 +0000

What about “Instrumental variables methods in experimental criminological research: what why and how?” by Angrist (which describes multiple endogenous regressors).
Now I am confused…