[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: Flagging estimation errors in -ivreg2-
Hi all. This didn't go through the first time ... hopefully it's not too late to be useful to Mike.
> -----Original Message-----
> From: Schaffer, Mark E
> Sent: 14 February 2007 08:39
> To: email@example.com
> Subject: AW: Flagging estimation errors in -ivreg2-
> If -ivreg2- finds collinearities, it saves two versions of
> assorted varlists (included exogenous, excluded exogenous,
> etc etc) in the list of e() macros. One version is the
> original varlist specified by the user, with every variable
> there. The other version (same name but ending with a 1) is
> the varlist after the collinear variables have been dropped.
> If -ivreg2- doesn't find any collinearities, then only the
> first version of the e() macros is there.
> So... for your case (1) you should be able to flag whether or
> collinearities were found by whether or not the e() macros
> ending in 1 are present. I think these are all listed in the
> -ivreg2- help file, along with a description of what they are.
> For case (2), this could be flagged by whether or not the
> e(j) statistic is missing, since, as the error message
> indicates, when you have the problem, the overid stat isn't
> reported. But this works only if you are estimating
> overidentified models.
> -----Ursprüngliche Nachricht-----
> Von: firstname.lastname@example.org im Auftrag von
> Michael S. Hanson
> Gesendet: So 2/11/2007 7:40
> An: email@example.com
> Betreff: st: Flagging estimation errors in -ivreg2-
> I have a do file that uses a lengthy set of nested
> loops to automate estimation of a large set of regressions
> using -ivreg2-, and iteratively build up a summary report.
> The -ivreg2- commands are
> executed -quietly- and -capture-d to facilitate building the
> Spot-checking a few individual estimations, I discovered that
> two problems that have silently occurred:
> 1. Some regressors are dropped to perfect collinearity.
> (This colinearity is not trivial, such as two identical
> regressors or a set of dummies reproducing the constant;
> hence such regressions are very difficult to identify ex ante.)
> 2. The VCV matrix fails a rank condition; the error
> message displayed on the screen when estimated -noisily- is
> below. (In my application, the "covariance matrix uses too
> many lags" may be the statistical source of the problem
> (still investigating), but the specified equations are
> theoretically sound.)
> "Error: estimated covariance matrix of moment conditions not
> of full rank;
> overidentification statistic not reported, and
> standard errors and
> model tests should be interpreted with caution.
> "Possible causes:
> covariance matrix of moment conditions not positive definite
> covariance matrix uses too many lags
> singleton dummy variable (dummy with one 1 and N-1 0s
> or vice versa)"
> Note that -capture- still returns _rc = 0 in either of these
> cases -- after all, some kind of estimates are reported by
> -ivreg2- -- so that does not appear to be a feasible way to
> "flag" these potentially problematic estimation results.
> What I am looking for is some other way to determine that one
> (or both) of these errors has arisen with - ivreg2-, and
> thereby "flag" the output in the summary report so that I can
> go back and examine the small number (out of hundreds or
> thousands) of cases that may suffer from either issue.
> Any suggestions would be appreciated.
> -- Mike
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
* For searches and help try: