Statalist The Stata Listserver

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

Re: st: treatreg (rho=1) and ivreg ( negative model SS)

Subject   Re: st: treatreg (rho=1) and ivreg ( negative model SS)
Date   Fri, 02 Mar 2007 14:29:05 +0100

Did you have a look at the FAQ before posting? explains the reasons why you can have missing or negative R-squared with -ivreg- and -reg3-, and I think this may answer to (2). 
I add that -condivreg- (SJ6-3: st0033_2; SJ5-2: st0033_1; SJ3-1: 0033) fits IV regression with potentially weak instruments and one endogenous variable (no robust options, however).
Sorry, I guess that (1) maybe somewhat related to the problem of rho outside the [-1, 1] interval in -heckman-, but I will leave the question open since I am a novice in these kinds of model. Maybe you can wait some days and then repost with an informative title like "problems with rho in treatreg, twostep" (remember the Statalist FAQ: A rewrite or even one repost of the original is tolerable, but more than one repost is not.)


At 02.33 02/03/2007 -0500, Richard Lo wrote:
>Hello everyone,
>I have two questions which had been posted previously
>but not answered yet. I believed they are related.
>(1) I ran a treatment effect model using ML and
>two-step procedure. The estimated coefficient for rho 
>is quite different (0.5385326 by ML and 1.00000 by
>two-step) and the lamda no longer equals to rho 
>times sigma in the two-step results.
>(2) I also ran a iverg (ivreg2) with robust option for
>my dada. The results show that the model SS is 
>negative in ivreg (negative centered and uncentered
>R-square in ivreg2 ) Related analysis indicate that 
>the first stage F statistic is low (2.6 for a single
>endogenous variable with 5 excluding instruments), 
>indicating there may be weak instrument problem); The
>over-identifying restrictions test however shows 
>that it can not be rejected.  
>I just do know whether they are attributed to program
>bugs or model mis-specifications. If it is the 
>latter case then what might go wrong during the
>calculations (or what kind of model mis-specification 
>might have contributed to this) ?  
>Thank you for your kind help in advance.

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index