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: RE: triprobit convergence problem
From
Maarten buis <[email protected]>
To
[email protected]
Subject
Re: st: RE: triprobit convergence problem
Date
Mon, 22 Mar 2010 09:21:46 -0700 (PDT)
These type of results scream: simplify the model. Remember
that an observation contains only very indirect information
that can be used to estimate this model. In particular this
model estimates a corelation matrix between latent variables,
i.e. variables we did not observe directly.
The most obvious way forward is to stack your two datasets.
This will constrain the effects to be the same across
datasets, which means you have more information you can
use to estimate your model. If you are uncomfortable with
this assumption, add some dummies or interactions to
relax some of these constraints.
-- Maarten
--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://www.maartenbuis.nl
--------------------------
--- On Mon, 22/3/10, Roy, Manan <[email protected]> wrote:
> From: Roy, Manan <[email protected]>
> Subject: st: RE: triprobit convergence problem
> To: "[email protected]" <[email protected]>
> Date: Monday, 22 March, 2010, 15:54
> Hi Stephen,
>
> 1) I used -triprobit- on ssc.
>
> 2) The data is on adults 25-59 years old with at least 1
> child between 5-18 years old (N=1600), and adults between
> 25-59 with at least 1 child between 5-10 years old (N=1027).
> The model is trying to identify the effect of school meal
> program participation on different time use categories.
>
> 3) TIME4_0, NSLP, SMEAL, male, WNonHisp South are all dummy
> variables
>
> 4) This exact same trivariate model with TIME4_0, however,
> converges with the data set with N=1027, without any
> options.
>
> 5) The following output is for the data set with
> N=1600
>
> (a)
> triprobit (TIME4_0 = NSLP SMEAL male
> teage agesq WNonHisp South) (NSLP =
> male teage agesq WNonHisp South) (SMEAL = male teage agesq
> WNonHisp South) [w=eufinlwgt], difficult
>
> (analytic weights assumed)
>
> trivariate probit, GHK simulator, 25 draws
>
> Comparison log likelihood = -2995.3698
>
> initial: log likelihood =
> -2995.3698
> rescale: log likelihood =
> -2995.3698
> rescale eq: log likelihood = -2995.3698
> Iteration 0: log likelihood =
> -2995.3698
> Iteration 1: log likelihood =
> -2810.0233 (not concave)
> Iteration 2: log likelihood =
> -2798.2733 (not concave)
> Iteration 3: log likelihood =
> -2790.7672 (not concave)
> Iteration 4: log likelihood =
> -2790.6208 (not concave)
> Iteration 5: log likelihood =
> -2790.2056 (not concave)
> Iteration 6: log likelihood =
> -2789.4005 (not concave)
> Iteration 7: log likelihood =
> -2789.3472 (not concave)
> Iteration 8: log likelihood =
> -2789.1896 (not concave)
> Iteration 9: log likelihood =
> -2788.9758 (not concave)
> Iteration 10: log likelihood = -2788.3554 (not
> concave)
> Iteration 11: log likelihood = -2788.204
> (not concave)
> Iteration 12: log likelihood = -2787.9404 (not
> concave)
> Iteration 13: log likelihood = -2787.9003 (not
> concave)
> Iteration 14: log likelihood = -2787.6958 (not
> concave)
> Iteration 15: log likelihood = -2787.3079 (not
> concave)
> Iteration 16: log likelihood = -2787.1486 (not
> concave)
> Iteration 17: log likelihood = -2786.8424 (not
> concave)
> Iteration 18: log likelihood = -2786.8261 (not
> concave)
> Iteration 19: log likelihood = -2786.7436 (not
> concave)
> Iteration 20: log likelihood = -2786.6794 (not
> concave)
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> Iteration 21: log likelihood = -2786.6747 (not
> concave)
> Iteration 22: log likelihood = -2786.6708 (not
> concave)
> could not calculate numerical derivatives
> missing values encountered
> r(430);
>
> (b)
> triprobit (TIME4_0 = NSLP SMEAL male
> teage agesq WNonHisp South) (NSLP =
> male teage agesq WNonHisp South) (SMEAL = male teage agesq
> WNonHisp South) [w=eufinlwgt], difficult draws(45)
> (analytic weights assumed)
>
> trivariate probit, GHK simulator, 45 draws
>
> Comparison log likelihood = -2995.3698
>
> initial: log likelihood =
> -2995.3698
> rescale: log likelihood =
> -2995.3698
> rescale eq: log likelihood = -2995.3698
> Iteration 0: log likelihood =
> -2995.3698
> Iteration 1: log likelihood =
> -2876.4216
> Iteration 2: log likelihood =
> -2801.6882
> Iteration 3: log likelihood =
> -2795.6717
> Iteration 4: log likelihood =
> -2794.0842 (backed up)
> Iteration 5: log likelihood =
> -2791.9006
> Iteration 6: log likelihood =
> -2790.852 (not concave)
> Iteration 7: log likelihood =
> -2790.7997 (not concave)
> Iteration 8: log likelihood =
> -2790.7166 (not concave)
> Iteration 9: log likelihood =
> -2790.6944 (not concave)
> Iteration 10: log likelihood = -2790.5223 (not
> concave)
> Iteration 11: log likelihood = -2790.4437 (not
> concave)
> Iteration 12: log likelihood = -2790.4041 (not
> concave)
> Iteration 13: log likelihood = -2790.375
> (not concave)
> Iteration 14: log likelihood = -2790.3516 (not
> concave)
> Iteration 15: log likelihood = -2790.3329 (not
> concave)
> Iteration 16: log likelihood = -2790.3146 (not
> concave)
> Iteration 17: log likelihood = -2790.0194 (not
> concave)
> Iteration 18: log likelihood = -2789.7409 (not
> concave)
> Iteration 19: log likelihood = -2789.5673 (not
> concave)
> Iteration 20: log likelihood = -2789.1887 (not
> concave)
> Iteration 21: log likelihood = -2789.1792 (not
> concave)
> Iteration 22: log likelihood = -2789.0136 (not
> concave)
> Iteration 23: log likelihood = -2788.9393 (not
> concave)
> Iteration 24: log likelihood = -2788.8466 (not
> concave)
> Iteration 25: log likelihood = -2788.8227 (not
> concave)
> Iteration 26: log likelihood = -2788.7829 (not
> concave)
> Iteration 27: log likelihood = -2788.7466 (not
> concave)
> Iteration 28: log likelihood = -2788.6929 (not
> concave)
> Iteration 29: log likelihood = -2788.5878 (not
> concave)
> Iteration 30: log likelihood = -2788.5645 (not
> concave)
> Iteration 31: log likelihood = -2788.5494 (not
> concave)
> Iteration 32: log likelihood = -2788.5383 (not
> concave)
> Iteration 33: log likelihood = -2788.5355 (not
> concave)
> Iteration 34: log likelihood = -2788.5213 (not
> concave)
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> Iteration 35: log likelihood = -2788.5166 (not
> concave)
> Iteration 36: log likelihood = -2788.5163 (not
> concave)
> could not calculate numerical derivatives
> missing values encountered
> r(430);
>
> 6) Will try the other options.
>
> Thanks!
>
> Manan
> ________________________________________
> From: [email protected]
> [[email protected]]
> On Behalf Of Stephen P. Jenkins [[email protected]]
> Sent: Monday, March 22, 2010 7:59 AM
> To: [email protected]
> Subject: st: triprobit convergence problem
>
> ------------------------------
>
> Date: Sun, 21 Mar 2010 17:27:20 -0500
> From: "Roy, Manan" <[email protected]>
> Subject: st: triprobit convergence problem
>
> Hi,
>
> I am trying to estimate triprobit models with different
> time
> categories (as dummies) and 2 binary program participation
> variables.
>
> I have 2 almost identical data sets, one with N=1600 and
> the
> other with N=1000.
> There are 2 time categories for which the models are not
> converging. Let's call them TIME1 and TIME2.=20
> TIME1 converges in N=1600 data while it doesn't in N=1000
> data.
> The exactly opposite case holds for TIME2.
>
> I have tried using the technique option. However, I get the
> error
> that this option's not allowed with triprobit.
>
> I have also tried the difficult option, and specified
> different
> number of draws.
>
> Any suggestions on how I can make it work will be greatly
> appreciated.
>
> Thanks,
>
> Manan
> >>>>>>>>>>>>>>>>>>>
>
> You should state the source of the user-written program
> -triprobit- (it is on SSC, I believe)
>
> You do not provide, as the Statalist FAQ asks, the precise
> Stata
> commands that you typed and the output that was produced.
>
> And, sorry, the nature of your trivariate probit
> specification is
> unclear from what you write, in any case.
>
> There are at least 3 other ways to estimate trivariate
> probit
> models, and you could try them (they also allow -maximize-
> options like -difficult- and -technique(...)-:
>
> * -mvprobit- on SSC
>
> * Generic code using a plugin (and so fast): see Cappellari
> &
> Jenkins, Stata Journal 6(2), 2006 [article downloadable
> from
> Stata Journal website]
>
> * -cmp- on SSC
>
>
> Stephen
> -------------------------------------
> Professor Stephen P. Jenkins <[email protected]>
> Institute for Social and Economic Research (ISER)
> University of Essex, Colchester CO4 3SQ, UK
> Tel: +44(0)1206 873374. Fax: +44(0)1206 873151
> http://www.iser.essex.ac.uk
> Survival Analysis using Stata:
> http://www.iser.essex.ac.uk/survival-analysis
> Downloadable papers and software:
> http://ideas.repec.org/e/pje7.html
>
>
> *
> * 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/