Re: st: ivreg2 and LATE

[Austin Nichols <austinnichols@gmail.com>]

Francesco Burchi <fburchi@uniroma3.it> :

Since you don't have a dummy endog var, you don't have defiers or
compliers, per se.  What you want to know is, what is the impact on
the density of the endog variable, or the difference in CDFs; see
http://www.nber.org/~kling/kling_iv.pdf and the published version:
Journal of Business and Economic Statistics 19:3 (July 2001), 358-364.
Angrist and Pischke also discuss this same point:
http://books.google.com/books?id=bUZzyXeV1JUC&printsec=frontcover&dq=angrist+pischke&cd=1#v=snippet&q=%22CDF%20difference%22&f=false

On Wed, Mar 10, 2010 at 1:49 PM, Francesco Burchi <fburchi@uniroma3.it> wrote:
> Dear Statalisters,
>
> I have a question which is partly related to Stata and partly to general
> econometric analysis, I hope you can give me some suggestions.
>
> I am running a 2SLS estimation using -ivreg2- because one of the
> explanatory variable is endogenous. This endogenous variable is
> continuous and standardized, thus it varies between 0 and 1. I have used
> two dummy variables as instrument. I have run several tests that show
> that the instruments are valid (exogenous, not weak, and not redundant)
> and not likely to be endogenous; moreover the Wu-Hausman test confirms
> that the instrumental variable approach is the right one.
>
> Now, since the coefficient of the endogenous variable is about 6 times
> larger in the 2SLS estimates as compared to the OLS estimates (while the
> significance is only slightly different), and since it is the only
> coefficient substantially changing depending on the estimator, I was
> asked to give some "size" interpretation. In particular, I was asked
> whether the 2SLS estimates are LATE (Local average treatment effect) and
> whether I am sure that there are no "defiers". To my knowledge, mainly
> based on the works of Imbens and Angrist, LATE assume that some
> properties such as "monotonicity" are satisfied, which means that there
> are no defiers. My questions are: 1) Is there a way to check whether
> there are "defiers"? (I believe the answer is "no"); 2) How can I
> interpret the size of the coefficient and not just its significance? 3)
> Is there any relevant aspect to consider when we look closely at the
> relationship between the instrumental variables and the endogenous
> variable, which might be of great relevance to derive policy
> recommendations?
>
> I should add that I am using Stata 10.
>
> Thank you in advance for your precious help,
>
> Best,
> Francesco
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/