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: During simulations, what to do if the regression-based model does not converge?
From
"Tiago V. Pereira" <[email protected]>
To
"Tiago V. Pereira" <[email protected]>
Subject
Re: st: During simulations, what to do if the regression-based model does not converge?
Date
Wed, 20 Jun 2012 12:12:40 -0300 (BRT)
Thank you very much, Brendan! It worked like a charm!
All the best,
Tiago
--------------
On Tue, Jun 19 2012, Tiago V. Pereira wrote:
> The most intuitive algorithm would be
>
> if [mvmeta b V, reml takes too much time to run] {
>
> break
>
> and continue as
>
> mvmeta b V, mm
> }
Try:
. mvmeta b V, reml iterate(100)
. if e(converged)==0 {
. di in red "NO CONVERGENCE, fitting method of moments model"
. mvmeta b V, mm
. }
This assumes mvmeta takes the standard iterate() option and returns
e(converged), which is the normal Stata practice.
Brendan
--
Brendan Halpin, Department of Sociology, University of Limerick,
Ireland
Tel: w +353-61-213147 f +353-61-202569 h +353-61-338562; Room F1-009 x
3147
------------
Dear statalisters,
I am running some simulations and need to run -mvmeta-. In some cases,
-mvmeta- fails to converge, and the simulation stucks at a certain point
and doesn't finish until I manually cancel it.
Example (data given below):
[simulate data]
mvmeta b V, reml
Note: using method reml
Note: using variables b1 b2
Note: 22 observations on 2 variables
initial: log likelihood = -6.0965061
rescale: log likelihood = -6.0965061
rescale eq: log likelihood = 1.5713054
Iteration 0: log likelihood = 1.5713054
Iteration 1: log likelihood = 1.5735159 (not concave)
Iteration 2: log likelihood = 1.5735286 (not concave)
Iteration 3: log likelihood = 1.5735293 (not concave)
Iteration 4: log likelihood = 1.5735293 (not concave)
Iteration 5: log likelihood = 1.5735293 (not concave)
Iteration 6: log likelihood = 1.5735293 (not concave)
Iteration 7: log likelihood = 1.5735293 (not concave)
.
.
.
.
[neverending]
If I run instead
. mvmeta b V, mm
I get the method-of-moments approach, which would be the best estimate if
the reml option doest not work.
The most intuitive algorithm would be
if [mvmeta b V, reml takes too much time to run] {
break
and continue as
mvmeta b V, mm
}
I have failed to find a solution to that Can you tell me if it is possible?
All the best,
Tiago
*/ ------------ example data ---------------------------------
input b1 b2 V11 V22 V12
.233144 .1990245 .13194766 .00934892 .00169556
-.28375212 .19690602 .21868578 .01441359 .00277669
.00327333 .01059937 .22549555 .01792554 .00327333
.50068392 .1364699 .16447968 .01013253 .00197968
.17374169 .24701389 .24077677 .02068396 .00466566
.50303291 .06834434 .20411383 .01143459 .00219075
-.45014787 .26331932 .14863599 .01185804 .00217135
.65685769 .33265554 .31337754 .01162247 .00226642
1.1248355 .06663273 .44921284 .02497867 .0047684
1.2422897 .2806992 .32947018 .01348576 .00254711
-.48948849 .24811046 .16104793 .01191982 .0022244
.260668 .3046892 .20409032 .01089658 .00207011
.31184814 .11819502 .19728779 .01629585 .00284334
-.34414868 .31684138 .28063149 .01408479 .00285371
-.01790121 .48026602 .21454526 .01923754 .00343415
.11029605 .06801034 .1932147 .01195094 .0023056
.04299536 .2128944 .28981034 .02116782 .00409606
-.30519299 .31617451 .24653605 .02153186 .00367891
.31049675 .10128629 .19748667 .01613425 .00304222
-.22549042 .15365738 .12865038 .00878746 .00166625
-.10485559 .15266178 .13572105 .00932605 .00179248
.62737876 .18297067 .1572774 .00949862 .00172184
end
*/ ------------ example data ---------------------------------
*
* 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/