Amongst others, MSc students who took my Research Design class this year and tore their hair out over the mid-term assignment - here and here - might be gratified to know that the Gordian knot I asked you to struggle with is discussed by Gary King at a recent Harvard symposium as one of the 12 'Hard Problems in the Social Sciences'. There is a nice video (requires Realplayer) and some useful slides. Whatever your complaint about Oxford sociology, you can't say that we don't take you to the 'cutting edge'.
Posts
Keith O’Rourke’s final published paper: “Statistics as a social activity:
Attitudes toward amalgamating evidence”
-
Keith O’Rourke passed away two years ago. Here’s his obituary, which was
sent to me by Bart Harvey: Lover of earth, wood, water and fire, Keith left
us aft...
5 hours ago
1 comment:
Andrew Gelman and Jennifer Hill have a brief and elegant discussion of the same problem in their "Data Analysis Using Regression and Multilevel/Hierarchical Models" (p.188-194).
But I think King is talking about a slightly different thing here. In many cases avoiding post-treatment bias is straightforward. We simply should not control for "post-treatment" variables (although many people are still readily doing this). But the story gets more complicated once we have two-way causality (i.e., health affects income, income affects health). In this case it is simply impossible to determine what is "post-treatment". I think this is why King says that currently there is no solution to this problem.
Also, the video of his presentation is available on Facebook in the format that does not require RealPlayer - http://www.facebook.com/#!/video/video.php?v=1416961137774 .
Alexey
P.S. Thanks for maintaining this blog, I really enjoy reading it. I hope it's public and my comment is appropriate - my apologies if it isn't.
Post a Comment