Statalist


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

Re: AW: AW: st: AW: redprob: could not calculate numerical derivatives... r(430)


From   Doris Oberdabernig <[email protected]>
To   [email protected]
Subject   Re: AW: AW: st: AW: redprob: could not calculate numerical derivatives... r(430)
Date   Mon, 16 Nov 2009 11:27:51 +0100

If I run a pooled probit model all the results are robust. they stay the same
using the -probit- command vs. -xtprobit-. I already checked this.

In my -probit- specification I use quite a lot of dummies, this should not be a
problem for -redprob- though, right? (the problem remains if I estimate
-redprob- with continuous variables)

Cheers,
Doris


Zitat von Martin Weiss <[email protected]>:

>
> <>
>
>
> I suspect that the -from()- issue that we have discussed thus far is a side
> issue to some more fundamental problem with your data. What happens when you
> run a pooled -probit- model? Do you get sensible results?
>
>
>
> HTH
> Martin
>
> -----Ursprüngliche Nachricht-----
> Von: [email protected]
> [mailto:[email protected]] Im Auftrag von Doris
> Oberdabernig
> Gesendet: Montag, 16. November 2009 10:00
> An: [email protected]
> Betreff: Re: AW: st: AW: redprob: could not calculate numerical
> derivatives... r(430)
>
> Hello Martin,
>
> I already found the examples in
> http://www2.warwick.ac.uk/fac/soc/economics/staff/academic/stewart/stata/red
> probnote.pdf
> and http://www.stata-journal.com/sjpdf.html?articlenum=st0106,
> and read what was written at
> http://www.stata.com/statalist/archive/2006-04/msg00379.html and
> http://www.stata.com/statalist/archive/2006-04/msg00369.html, but i could
> not
> find a solution for the problem.
>
> I still get the error message:
>
> could not calculate numerical derivatives
> flat or discontinuous region encountered
> r(430)
>
> also when I proceed like in the examples above with my data. I already made
> sure
> that the panel is balanced and tried out different specifications.
> Unfortunately I could not find a response to the post of Georgios who seemed
> to
> have the same problem neither.
>
> I suppose the problem has to be somewhere else, I just cannot find out
> where. I
> would be grateful for any ideas.
>
> Cheers,
> Doris
>
>
>
>
> Zitat von Martin Weiss <[email protected]>:
>
> >
> > <>
> >
> > Well, using the example contained in
> >
> http://www2.warwick.ac.uk/fac/soc/economics/staff/academic/stewart/stata/red
> > probnote.pdf, section 2.3, here you can see how the point estimates
> > emanating from a pooled -probit- can be used subsequently for -redprob-.
> > Weirdly enough, the example in the pdf does not run for me w/o adding the
> > -bstart1- -matrix- myself. Omitting the call to -from(bstart1)- from the
> > command yields the result in the pdf, though.
> >
> >
> >
> > *************
> > u http://www.stata-press.com/data/r9/union.dta, clear
> > drop if year<78 | year==83
> > bys idcode: keep if _N==6
> >
> > sort idcode year
> > by idcode: gen byte tper = _n
> > by idcode: gen Lunion = union[_n-1]
> >
> > probit union Lunion age /*
> > */ grade south if tper!=1
> >
> > mat bstart1=e(b)
> >
> > redprob union Lunion age /*
> > */ grade south (age grade south not_smsa), /*
> > */ i(idcode) t(tper) quadrat(24)  /*
> > */ from(bstart1)
> > *************
> >
> > The issue has not come up too often on the list (but see
> > http://www.stata.com/statalist/archive/2006-04/msg00369.html), presumably
> > because the -program- is not accessible via -ssc- or -findit-...
> >
> >
> >
> > HTH
> > Martin
> >
> >
> > -----Ursprüngliche Nachricht-----
> > Von: [email protected]
> > [mailto:[email protected]] Im Auftrag von Doris
> > Oberdabernig
> > Gesendet: Sonntag, 15. November 2009 17:55
> > An: [email protected]
> > Betreff: Re: st: AW: redprob: could not calculate numerical derivatives...
> > r(430)
> >
> > Hi Martin,
> >
> > thanks for your help. I tried out to define the matrix like you described
> it
> > (-mat A=e(b)- after estimation, and supply it to -redprob- via
> -from(A)-...)
> > but unfortunately this does not solve my problem. I still get the error:
> >
> > could not calculate numerical derivatives
> > flat or discontinuous region encountered
> > r(430)
> >
> > do you have any other suggestions?
> >
> > Thanks,
> > Doris
> >
> >
> > Zitat von Martin Weiss <[email protected]>:
> >
> > >
> > > <>
> > >
> > > So -redprob- is not accessible via -findit-, but through
> > >
> http://www2.warwick.ac.uk/fac/soc/economics/staff/academic/stewart/stata/
> > >
> > > As discussed recently on the list, your -e(b)- is an -ereturn- matrix,
> but
> > > not what -redprob- obviously wants, a full-fledged Stata matrix.
> > >
> > > Try -mat A=e(b)- after estimation, and supply it to -redprob- via
> > > -from(A)-...
> > >
> > >
> > >
> > > HTH
> > > Martin
> > >
> > > -----Ursprüngliche Nachricht-----
> > > Von: [email protected]
> > > [mailto:[email protected]] Im Auftrag von Doris
> > > Oberdabernig
> > > Gesendet: Sonntag, 15. November 2009 16:55
> > > An: [email protected]
> > > Betreff: st: redprob: could not calculate numerical derivatives...
> r(430)
> > >
> > > Hello everybody,
> > >
> > > I am trying to estimate a dynamic probit model with a lagged dependent
> > > binary
> > > variable. I am using the -redprob- command (from Prof. Mark Steward,
> > > University
> > > of Warwick, 2006) to take care of the initial conditions problem (using
> > > Stata
> > > 10.1).
> > >
> > > The command looks like:
> > > redprob depvar varlist (varsinit) [if exp] [in range] [, i(varname)
> > > t(varname)
> > > quadrat(#) from(matname) ]
> > >
> > > where varlist contains the lagged dependent (binary) variable,
> > > varsinit contains determinants of the outcome in the first year,
> > > i(varname) is the cross-section identifier,
> > > t(varname) the time-identifier,
> > > quadrat(#) specifies the number of of Gaussian--Hermite quadrature
> points
> > > for
> > > the evaluation of the required integral and
> > > from(matname) specifies a matrix containing starting values for the
> > > parameters
> > > of the model.
> > > Redprob estimates a random effects dynamic probit model by Maximum
> > > Likelihood.
> > >
> > > The default for the "from" matrix uses a pooled probit for t>1 and a
> > > separate probit for the intial period.
> > > If I specify from(e(b)), i get the error:
> > >
> > > could not calculate numerical derivatives
> > > flat or discontinuous region encountered
> > > r(430)
> > >
> > > If I do not specify any from() option, I get the error:
> > > Probit Model for t=1
> > > no observations
> > > r(2000)
> > >
> > > It would be great if somebody could help me on this problem.
> > > Cheers,
> > > Doris
> > >
> > > *
> > > *   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/
> >
>
>
>
> *
> *   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 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index