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

From |
"Erik Ø. Sørensen" <sameos@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: RE: simulated maximum likelihood estimation |

Date |
Tue, 10 May 2005 10:03:44 +0200 |

On 9. mai. 2005, at 20.13, Jun Xu wrote:

It seems to me that you draw new random numbers for each evaluation.qui gen double `ed`i'' = -ln(uniform())*`b'

I don't think you should draw new random numbers each time. If you do so, your likelihood function will never be smooth and you cannot hope to use gradient based methods to maximise it.

You should instead first draw a number of random variates, and pass these on to the evaluator as if they were data. Alternatively, I guess you could set the seed to be a constant everytime my_ll is called.

Best regards,

Erik.

qui replace `sp`i''= `sp`j''*(invlogit( `xb'+`ed`i'')) if $ML_y1 == 1--

qui replace `sp`i''= `sp`j''*(invlogit(-(`xb'+`ed`i''))) if $ML_y1 == 0

}

qui replace `lnf' = ln(`sp100')

end

use binlfp2.dta, clear

ml model lf my_ll (lfp = k5 k618 age), technique(nr bhhh dfp bfgs)

ml search

ml maximize

**************************************************

Jun Xu

Ph.D. Candidate

Department of Sociology

Indiana University at Bloomington

http://mypage.iu.edu/~junxu/home

_________________________________________________________________From: "Nick Cox" <n.j.cox@durham.ac.uk> Reply-To: statalist@hsphsun2.harvard.edu To: <statalist@hsphsun2.harvard.edu> Subject: st: RE: simulated maximum likelihood estimation Date: Mon, 9 May 2005 01:11:33 +0100 I am not sure exactly what you are trying to do here. It is usually better to set up a log likelihood function directly. What you are logging here does not have the usual form of a likelihood function. Nick n.j.cox@durham.ac.uk Jun Xu > I am not sure if some expert could give me some hint about > where to go. Here > I am trying to estimate a logit with an added expotential > distributed random > error (with mean of 5). I am using simualted mle with 100 > replications, but > cannot get it converged. Not sure if I had a correct setup. > > > set trace off > set more off > cap program drop my_ll > program my_ll > version 8.2 > args lnf xb > tempname b > sca `b' = 5 > > forval j = 1/100 { > tempname ed`j' > } > > qui replace `lnf' = 0 > forval i = 1/100 { > qui gen double `ed`i'' = -ln(uniform())*`b' > qui replace `lnf'= `lnf' + invlogit( `xb'+`ed`i'') > if $ML_y1 == > 1 > qui replace `lnf'= `lnf' + invlogit(-(`xb'+`ed`i'')) > if $ML_y1 == > 0 > } > > qui replace `lnf' = ln(`lnf') > end > > use binlfp2.dta, clear > ml model lf my_ll (lfp = k5 k618 age), technique(nr bhhh dfp bfgs) > ml maximize > > > initial: log likelihood = 2948.488 > alternative: log likelihood = 2952.4599 > rescale: log likelihood = 2952.4599 > (setting optimization to Newton-Raphson) > numerical derivatives are approximate > flat or discontinuous region encountered > Iteration 0: log likelihood = 2949.0411 (not concave) > numerical derivatives are approximate > flat or discontinuous region encountered > Iteration 1: log likelihood = 2950.7797 (not concave) > numerical derivatives are approximate > flat or discontinuous region encountered > no observations > r(2000); > * * 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/

Don’t just search. Find. Check out the new MSN Search! http://search.msn.click-url.com/go/onm00200636ave/direct/01/

*

* 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/

Erik Ø. Sørensen, dept of Econ., Norwegian School of Economics

phone: (+47) 55959985 [office], <http://homepage.mac.com/sameos/research/>

--

Erik Ø. Sørensen, dept of Econ., Norwegian School of Economics

phone: (+47) 55959985 [office], <http://homepage.mac.com/sameos/research/>

*

* 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/

**References**:**RE: st: RE: simulated maximum likelihood estimation***From:*"Jun Xu" <mystata@hotmail.com>

- Prev by Date:
**Re: st: Dropping variables with all observations missing** - Next by Date:
**st: Mata and memory** - Previous by thread:
**RE: st: RE: simulated maximum likelihood estimation** - Next by thread:
**RE: st: RE: simulated maximum likelihood estimation** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |