Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: IV regression with panel data: random effects, clustered s.e.'s, postestimation tests

From   Austin Nichols <>
Subject   Re: st: IV regression with panel data: random effects, clustered s.e.'s, postestimation tests
Date   Fri, 27 Apr 2012 10:48:35 -0400

Jana von Stein <>:
The manual entry for -xtreg- provides a very good discussion, and the
References section provides much additional reading.

On Fri, Apr 27, 2012 at 10:44 AM, Jana von Stein <> wrote:
> Dear Austin,
> Ah, I see. That makes good sense. Is there a good stats source you might
> point me to on this? This is sort of new territory for me and I want to make
> sure I understand the statistics/theory behind what choices I'm making.
> (I.e., on the u_i in particular and that [a] not okay to to xt R.E. in the
> presence of non-orthogonal u_i; and [b] that it is kosher(ish) to ignore
> panel structure and use R.E. ivreg?
> Does the same apply to between effects?
> Thanks,
> Jana
> On Apr 27, 2012, at 10:25 AM, Austin Nichols wrote:
>> Jana von Stein <>:
>> For RE to be consistent, the effects u_i must be orthogonal to all
>> predictors, so you can ignore the u_i and estimate a pooled model.
>> I.e. use -ivreg2- (ignoring panel structure except for clustering on
>> country and/or year to account for correlated errors) and sacrifice a
>> little efficiency in favor of robustness against implausible
>> assumptions.
>> On Fri, Apr 27, 2012 at 9:21 AM, Jana von Stein <> wrote:
>>> Hello,
>>> I have time-series panel (country) data and an endogenous independent
>>> variable. My sense is that the best way to handle this is to instrument,
>>> use
>>> a random effects model, and cluster on country. Does anyone know what
>>> command(s) I can use that will do (1) clustered s.e.'s and (2)
>>> postestimation tests (underidentification, Sargan, etc.) for instrument
>>> validity and identification?
>>> Xtivreg2 is great, but -- unless I'm missing some extra code someone's
>>> programmed -- isn't available for random or between effects. Fixed
>>> effects
>>> (available in xtivreg2) are out of the question, as some of my
>>> instruments
>>> don't vary over time within country.
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index