Dear All --
Our speaker tomorrow at Applied
Stats<http://projects.iq.harvard.edu/applied_stats>
will be Teppei Yamamoto from MIT PoliSci. Professor Yamamoto's talk is titled
"Identification and Estimation of Causal Mediation Effects with Treatment
Noncompliance." As per usual, the talk will be held in CIGS
K354<http://map.harvard.edu/?bld=04471&level=9> at 12 noon and lunch will be
served. The abstract for Professor Yamamoto's talk can be found below and you can
download the associated article from our website at:
http://projects.iq.harvard.edu/applied_stats/event/teppei-yamamoto-mit.
Abstract:
Treatment noncompliance is a common problem in program evaluation. The problem is
particularly severe when the analyst is interested in causal mediation effects. This is
because, somewhat counterintuitively, the mediated portion of an intention-to-treat (ITT)
effect cannot be nonparametrically identified even when treatment assignment is randomized
and the ignorability of the observed mediator is assumed. This paper shows that, once the
standard instrumental variables assumptions are made, the mediated ITT effects and the
local average causal mediation effects (LACME) for compliers can in fact be identified
under a local sequential ignorability assumption. The commonly-used naïve estimators for
the mediated ITT effect and LACME are shown to be biased in unknown directions. The
proposed estimators are illustrated via a simulation study and applied to data from a job
training experiment. The proposed method, implemented in an open-source R package, enables
researchers to investigate causal mechanisms by which the treatment affects the outcome of
interest even when treatment noncompliance exists.
Tess
-----------------
Tess Wise
PhD Candidate
Harvard Department of Government
http://tesswise.com