[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Heckman or Heckman-twostep? (was: Things to consider when regressions don't converge)

 From "de la Garza, Adrian" To Subject st: RE: Heckman or Heckman-twostep? (was: Things to consider when regressions don't converge) Date Tue, 18 May 2004 16:10:44 -0400

```Thank you, David.

My rho is far from 1, it is 0.18 if I run the model using the two-step
procedure. But if I run it without it, it doesn't converge! Two
questions:

1/ Is 0.18 "far" from 1, so that you would recommend that I don't use
Heckman's two-step procedure?

2/ Do you know why the method using the ML function would not converge?

I read somewhere that Heckman's two-step method yielded more efficient
estimators... so I was more inclined to use the two-step procedure until
you told me this. Do you know of a book or other source that might talk
about this (ML vs two-step)?

Thanks a lot.
Adrian

> -----Original Message-----
> From: David Greenberg [mailto:dg4@nyu.edu]
> Sent: Tuesday, May 18, 2004 3:49 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: Heckman or Heckman-twostep? (was: Things to
> consider when regressions don't converge)
>
>
> When the estimated rho is close to 1, the two-step procedure
> may handle the estimation better. Otherwise, I think it is
> now considered preferable not to use it. David Greenberg,
> Sociology Department, New York University
>
> ----- Original Message -----
> From: "de la Garza, Adrian" <ADelagarza@imf.org>
> Date: Tuesday, May 18, 2004 2:24 pm
> Subject: st: Heckman or Heckman-twostep? (was: Things to
> consider when regressions don't converge)
>
> > Since neither I nor anybody else have found an answer to my problem,
> > here is another rather technical question:
> >
> > In which cases would I prefer to use the Heckman model without the
> > twostep option and in which cases would I rather run it with the
> > twostepoption?
> >
> > Thank you, all.
> >
> > Adrian
> >
> > > -----Original Message-----
> > > From: de la Garza, Adrian
> > > Sent: Tuesday, May 18, 2004 11:38 AM
> > > To: statalist@hsphsun2.harvard.edu
> > > Subject: st: Things to consider when regressions don't converge
> > >
> > >
> > > Jean, thanks a lot. But perhaps I should have explained myself
> > better.> It is my -R- that I want, that is, the number of bonds
> > issued by a
> > > particular borrower SO FAR at each different point in time. Your
> > -R-
> > > would give me the FINAL total of bonds that a particular
> > > borrowed issue
> > > throughout the whole sample and that number would be repeated
> > for each
> > > observation within the same borrower. This is not what I want.
> > >
> > > And yes, I tried the twostep option and it does run my
> > regression more
> > > smoothly but I am not sure why it doesn't work without the twostep
> > > option.
> > >
> > > Thank you.
> > > Adrian
> > >
> > > > -----Original Message-----
> > > > From: jean ries [ries@ires.ucl.ac.be]
> > > > Sent: Tuesday, May 18, 2004 11:26 AM
> > > > To: statalist@hsphsun2.harvard.edu
> > > > Subject: Re: Things to consider when regressions don't converge
> > > >
> > > >
> > > > At 16:29 18/05/2004, you wrote:
> > > > >I am running a Heckman selection model (shown below) and
> > after this
> > > > >non-converging story I started playing around with my
> > equation and
> > > > >noticed that my -lR- variable might be the one that's giving
> > > > me trouble.
> > > > >-lR- is ln(R), and -R- is generated as follows:
> > > > >
> > > > >sort borrower indic signdate mtydate amount
> > > > >by borrower: g R = _n
> > > > >
> > > > >so -R- is the number of bonds issued by a particular
> > > borrower at each
> > > > >different point in time.
> > > >
> > > > I don't think that R represents what you expect it to
> > > > represent. The way
> > > > you define it, R contains the current observation number for
> > > > each borrower.
> > > > Try the following to obtain the number of bonds issued by a
> > > > particular
> > > > borrower :
> > > >
> > > > bysort borrower: g R = _N
> > > >
> > > > and:
> > > >
> > > > help _variables
> > > >
> > > > In any case, have a look at the Stata reference manual. It
> > > > contains a nice
> > > > discussion on problems related to Heckman selection models.
> > > > As suggested
> > > > there, you should try to fit your model using the
> two-step method.
> > > >
> > > > Hope this helps,
> > > >
> > > > jean
> > > >
> > > > *
> > > > *   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:
> > > *   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:
> > *   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:
> *   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:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```

 © Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index