Statalist The Stata Listserver

[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Too few clusters in ivreg2 - again

From   "Austin Nichols" <>
To, "Ramani Gunatilaka" <>
Subject   Re: st: Too few clusters in ivreg2 - again
Date   Wed, 8 Nov 2006 14:07:54 -0500

Ramani Gunatilaka --
I will not reply to your private correspondence with Mark, who is
eminently suited to comment, but I do want to point out that you have
a number of problems:

1. *weak instruments*
as evidenced by the diagnostics reported after the first-stage output,
excerpted below:

Test of excluded instruments:
F(  2,    11) =     3.91
Prob > F      =   0.0522
NB: first-stage F-stat cluster-robust
Weak identification statistics:
Cragg-Donald (N-L)*minEval/L2 F-stat       4.24
NB: identification statistics not robust

See -findit condivreg- for tests (-condtest-) robust to that problem.

2. *insufficient clusters*
Even if you resolve your "estimated covariance matrix of moment
conditions not of full rank" problem, you only have 12 clusters, which
is not really approaching infinity IMHO.  Recall that the
cluster-robust estimator is consistent as m (the number of clusters)
gets large, and "large enough" seems to often be 50 or more.

3. *model spec may not be appropriate for dependent variable*
I am speculating here, but I am guessing that "happy" is a qualitative
dependent variable, and you may prefer -ivprobit- or somesuch.

. which ivprobit
*! version 1.1.9  11oct2006

On 11/8/06, Ramani Gunatilaka <> wrote:
Hi Mark,
Sorry to bother you again.
I updated my version of ivreg2.
I am sorry if these seem basic questions, but I have not been able to
figure this out.
I did use fwl and specified the model as follows:
But I got an error.
*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index