From: "Nick Cox" <email@example.com>
Subject: st: RE: IVTOBIT, missing values, panel
Date: Mon, 15 Sep 2003 12:25:11 +0100
> I have three questions. First is a follow up to a question
> that was posted
> last year (31 Jul 2002, to be precise) by Kee-Lee Lee. The
> problem is one
> of getting the error message ---matrix has missing
> values--while running
> IVTOBIT. There is no missing value in the variables used, though.
> I followed the suggestion of Nick Cox to --set trace on--.
> I attach the
> stata output below. Could anyone figure out the reason?
> My second question is, is anyone aware of any stata module
> that extends
> ivtobit to a panel setting?
> Finally, I would like to know if it is possible to estimate the Log
> Likelihood after estimating ivtobit.
> Many thanks in advance,
> Joe Jacob
> ****Stata Output******
> ivtobit ird, endog(expsal) iv(wager ltlnou size2) exog(outs forg
> - matrix alpha[1,`K'+1]=tmp[1,`K'+`nendog'+1]
> matrix has missing values
> end of do-file
-ivtobit- is a user-written program. You
could try contacting the author to see whether
he is willing to provide support.
It seems as if this is a program not being maintained,
In the thread in July 2002, I highlighted what
looked like a bug to me. I don't recollect either
refutation (i.e. it's not a bug really) or fix
(as the program on SSC appears earlier).
That may or may not be behind what is here. I
recommend copying the program and inserting
matrix li alpha
matrix li tmp
just before the statement which causes
-ivtobit- to crash.
Perhaps Stata is dropping some variable
as collinear. No idea otherwise.
If you ever get this working, -ereturn li-
will show what is left behind in memory.
* For searches and help try: