Comments on: Never Too LATE, Part 2

By: LATE with Multiple Instrumental Variables | Marc F. Bellemare

LATE with Multiple Instrumental Variables | Marc F. Bellemare — Tue, 18 Sep 2012 09:01:22 +0000

[...] up on a three-part series published over the past few weeks. Click the following links for parts 1, 2, and [...]

By: Marc F. Bellemare

Marc F. Bellemare — Sat, 08 Sep 2012 18:01:33 +0000

Feedback like yours is not helpful without specific examples. Oh, and you might want to be more creative in your spamming.

By: BT Women Sandals

BT Women Sandals — Sat, 08 Sep 2012 17:43:27 +0000

I was looking around your site and noticed broken links. You might want to change it.

By: Marc F. Bellemare

Marc F. Bellemare — Thu, 06 Sep 2012 17:17:49 +0000

Thank you for your comment and question, Chuhang! You are right: for every different IV, you will (in principle) get a different LATE. Different instruments don’t change the effective treatment groups in an experimental setting (the experimenter chooses who is assigned to each of control and treatment groups). In an observational (i.e., nonexperimental) context, however, the IV does change who takes up a treatment or not, and the LATE measures the ATE on compliers (i.e., those who change their treatment-taking behavior because of the IV).

By: Chuhang Yin

Chuhang Yin — Thu, 06 Sep 2012 15:22:08 +0000

Hi Prof. Bellemare, I just want to check my understanding about LATE: LATE is just an estimate of ATE based on the choice of instrument, right? Because people respond to different instruments differently, we don’t really know what the true effect of the endogeneous variable is. So are we assuming instrument variables can change the effective treatment groups?