[R] Dealing with large nominal predictor in sem package
John Fox
jfox at mcmaster.ca
Tue Apr 10 03:51:17 CEST 2007
Dear adschai,
> -----Original Message-----
> From: adschai at optonline.net [mailto:adschai at optonline.net]
> Sent: Monday, April 09, 2007 7:33 PM
> To: John Fox
> Cc: r-help at stat.math.ethz.ch
> Subject: Re: RE: [R] Dealing with large nominal predictor in
> sem package
>
> Hi John,
>
> Thank you. I think (2) from your explanation hits the right
> point. The reason is that when I made my own dummy variables
> and my original nominal variable has 10 possible values, it
> makes each each observed exogeneous variable vector of mine
> has 9 zeros and 1 one value. And I have about 400000
> observations. So it will make the matrix almost zero.
>
I'm afraid that I don't follow that, unless you're saying that some of the
levels of the factor have very few observations in them.
> One more question. If I have a nominal response, I guess the
> tsls would no longer work. How can I go around with this?
If the response is ordinal, then you can use sem() with
polyserial/polychoric correlations. Otherwise, the sem package won't handle
it.
> Says, I have 3 equations in my structure model whose
> responses are continuous whereas another one has multinominal
> response. Thank you so much.
>
As I said, neither tsls() nor sem() will handle an unordered response.
John
> - adschai
>
> ----- Original Message -----
> From: John Fox
> Date: Monday, April 9, 2007 8:04 am
> Subject: RE: [R] Dealing with large nominal predictor in sem package
> To: adschai at optonline.net
> Cc: r-help at stat.math.ethz.ch
>
> > Dear adschai,
> >
> > It's not possible to know from your description exactly what you're
> > doing, but perhaps the following will help:
> >
> > (1) I presume that your nominal variable is exogenous,
> since otherwise
> > it wouldn't be sensible to use 2SLS.
> >
> > (2) You don't have to make your own dummy regressors for a nominal
> > variable; just represent it in the model as a factor as you would,
> > e.g., in lm().
> >
> > (3) Do you have at least as many instrumental variables
> (including the
> > dummy
> > regressors) as there are structural coefficients to
> estimate? If not,
> > the structural equation is underidentified, which will produce the
> > error that you've encountered.
> >
> > I hope this helps,
> > John
> >
> > --------------------------------
> > John Fox
> > Department of Sociology
> > McMaster University
> > Hamilton, Ontario
> > Canada L8S 4M4
> > 905-525-9140x23604
> > http://socserv.mcmaster.ca/jfox
> > --------------------------------
> >
> > > -----Original Message-----
> > > From: r-help-bounces at stat.math.ethz.ch
> > > [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of
> > > adschai at optonline.net
> > > Sent: Sunday, April 08, 2007 11:07 PM
> > > To: r-help at stat.math.ethz.ch
> > > Subject: [R] Dealing with large nominal predictor in sem package
> > >
> > > Hi,
> > >
> > > I am using tsls function from sem package to estimate a
> model which
> > > includes large number of data. Among its predictors, it has a
> > > nominal data which has about 10 possible values. So I expand this
> > > parameter into 9-binary-value predictors with the coefficient of
> > > base value equals 0. I also have another continuous predictor.
> > >
> > > The problem is that, whenever I run the tsls, I will get
> 'System is
> > > computationally singular' error all the time. I'm
> wondering if there
> > > is anyway that I can overcome this problem? Please kindly
> suggest.
> > > Thank you so much in advance.
> > >
> > > - adschai
> > >
> > > [[alternative HTML version deleted]]
> > >
> > > ______________________________________________
> > > R-help at stat.math.ethz.ch mailing list
> > > https://stat.ethz.ch/mailman/listinfo/r-help
> > > PLEASE do read the posting guide
> > > http://www.R-project.org/posting-guide.html
> > > and provide commented, minimal, self-contained, reproducible code.
> > >
> >
> >
> >
>
More information about the R-help
mailing list