Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

RE: st: RE: condivreg/Multicollinearity


From   "Vassilopoulos Achilleas" <[email protected]>
To   <[email protected]>
Subject   RE: st: RE: condivreg/Multicollinearity
Date   Fri, 10 Sep 2010 01:43:59 +0300

Then I guess that your collinearity doesn't arise only from the first step
but also from the next steps of the -ivregress-. 
-ivregress- uses three tests for collinearity, one after the other, before
the estimation. What you can do is to replicate these tests by hand and save
the non-collinear variables of every step so you can use them later. See
below :

_rmdcoll ENDOGENOUS INCLUDED-EXOGENOUS, forcedrop      //checks for
collinearity between endogenous vars and included exogenous//

global first `r(varlist)'    //This global contains the set of non-collinear
included vars//

_rmcoll dummy1-dummy500     //checks for collinearity among instruments//

global inst_first `r(varlist)'    //this global contains the dummies that
are not collinear between themselves//

_rmcoll2list, alist($first) blist($inst_first)   //checks for collinearity
among the endogenous and all exogenous vars and further drops instruments
until a linearly independent set is obtained //

global inst `r(blist)'  //this global contains the final set of
non-collinear instruments// 

I guess all you need from this point on are the globals $inst and $first


Hope this helps,
_____________ - _______________

Achilleas Vassilopoulos

Agricultural University of Athens,
Dept. of Agricultural Economics and Rural Development,
Lab. of Political Economy and European Integration.
Iera Odos 75, 11855, Athens, Greece

Tel: (+30) 210-5294726
Fax: (+30) 210-5294786
email : [email protected]



-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of richard boylan
Sent: 09 September, 2010 17:12
To: [email protected]
Subject: Re: st: RE: condivreg/Multicollinearity

That does not work, I still get the same error message after "condivreg."

ivreg eliminates many more dummy variable than only estimating the
first stage with reg.

So, ideally, I would want to follow the same procedure you described
using ivreg if I knew how
to access the first stage coefficients.

On Thu, Sep 9, 2010 at 2:53 AM, Vassilopoulos Achilleas
<[email protected]> wrote:
> If your dummies are collinear, they will be dropped no matter which model
> you employ.
> As a result, a simple way to get what you need (not very elegant though)
is
> to run a -reg- or -probit- or -logit- with all your dummies as
independents
> and generate a global or local with the names in the coefficient vector
i.e.
> :
>
> reg DEPENDENT dummy1 dummy2 ..... dummy500
>
> global list : colnames e(b)  or
>
> local list : colnames e(b)
>
> Then you can use this $list or `list' in your subsequent estimations...
>
> Hope this helps,
> _____________ - _______________
>
> Achilleas Vassilopoulos
>
> Agricultural University of Athens,
> Dept. of Agricultural Economics and Rural Development,
> Lab. of Political Economy and European Integration.
> Iera Odos 75, 11855, Athens, Greece
>
> Tel: (+30) 210-5294726
> Fax: (+30) 2105294786
> e-mail : [email protected]
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of richard boylan
> Sent: Wednesday, September 08, 2010 10:14 PM
> To: [email protected]
> Subject: st: condivreg/Multicollinearity
>
> I am estimating a model with a very large number of interacting dummy
> variables in the
> first stage (> 500); so it is difficult a priori to determine which
> dummy variables are going
> to be collinear.
>
> When I estimate the model with ivreg, STATA eliminates a variety of
> collinear variables
> and produces estimates.  However, when I use condivreg I get an error
> message
>
> r(498);
>
> and a message "Multicollinearity!"
>
> It seems to me that one way of trying to solve this problem is to
> obtain from ivreg the list
> of variables that were not dropped in the first stage, and then
> estimate condivreg using
> the same variables.
>
> If this seems like a reasonable way of going about it, can anyone
> point me about how
> I would go about doing that? I.e., putting in a local variable the
> list of variables that
> were used in the first stage?
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index