MCMC error

Welcome to the forum for runmlwin users. Feel free to post your question about runmlwin here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Go to runmlwin: Running MLwiN from within Stata >> http://www.bristol.ac.uk/cmm/software/runmlwin/
Forum rules
Post Reply
luisrvaz
Posts: 5
Joined: Mon Aug 18, 2014 10:22 am

MCMC error

Post by luisrvaz »

Hi,

I am carrying out a bivariate response model using -runmlwin-. The code is as follows:

Code: Select all

xi: runmlwin (mean_reach cons soc_net type_121 have_current mental_health, eq(1)) ///
		(effectiveness cons type_121, eq(2)), /// 
	level2(SHA: (cons, eq(1)) (cons, eq(2))) ///
	level1(service: (cons, eq(1)) (cons, eq(2))) nopause
	
xi: runmlwin (mean_reach cons soc_net type_121 have_current mental_health, eq(1)) ///
		(effectiveness cons type_121, eq(2)), /// 
	level2(SHA: (cons, eq(1)) (cons, eq(2))) ///
	level1(service: (cons, eq(1)) (cons, eq(2))) ///
	mcmc(on) initsprevious
However, when running the second MCMC part of the code, an error message returns reading
error while obeying batch file
C:\Users\mcxlv1\AppData\Local\Temp\ST_000000cp.tmp at line
number 259:
MCMC 0 500 1 5.8 50 10 C1499] C1500]111114
Any help understanding what might have gone wrong would be appreciated.

Many thanks,
Luis
Top
GeorgeLeckie
Site Admin
Posts: 432
Joined: Fri Apr 01, 2011 2:14 pm

Re: MCMC error

Post by GeorgeLeckie »

Hi Luis,

Not immediately obvious what the problem is. Does the following two-level bivariate response model work for you?

Best wishes

George

Syntax:

Code: Select all

* Load the data
use http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial, clear

* Fit the model by IGLS
runmlwin ///
  (normexam cons girl, eq(1)) ///
  (standlrt cons girl, eq(2)), /// 
  level2(school: (cons, eq(1)) (cons, eq(2))) ///
  level1(student: (cons, eq(1)) (cons, eq(2))) ///
  nopause

* Fit the model by MCMC   
runmlwin ///
  (normexam cons girl, eq(1)) ///
  (standlrt cons girl, eq(2)), /// 
  level2(school: (cons, eq(1)) (cons, eq(2))) ///
  level1(student: (cons, eq(1)) (cons, eq(2))) ///
  mcmc(on) initsprevious ///
  nopause
Output

Code: Select all

. * Load the data
. use http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial, clear

. 
. * Fit the model by IGLS
. runmlwin ///
>   (normexam cons girl, eq(1)) ///
>   (standlrt cons girl, eq(2)), /// 
>   level2(school: (cons, eq(1)) (cons, eq(2))) ///
>   level1(student: (cons, eq(1)) (cons, eq(2))) ///
>   nopause
 
MLwiN 2.31 multilevel model                     Number of obs      =      4059
Multivariate response model
Estimation algorithm: IGLS

-----------------------------------------------------------
                |   No. of       Observations per Group
 Level Variable |   Groups    Minimum    Average    Maximum
----------------+------------------------------------------
         school |       65          2       62.4        198
-----------------------------------------------------------

Run time (seconds)   =       2.10
Number of iterations =          3
Log likelihood       = -10263.911
Deviance             =  20527.822
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
normexam     |
      cons_1 |  -.1612301   .0572912    -2.81   0.005    -.2735188   -.0489415
      girl_1 |    .261202   .0402489     6.49   0.000     .1823155    .3400885
-------------+----------------------------------------------------------------
standlrt     |
      cons_2 |  -.1142065   .0469738    -2.43   0.015    -.2062735   -.0221395
      girl_2 |    .155753   .0401182     3.88   0.000     .0771227    .2343833
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                 var(cons_1) |   .1613726   .0311917      .1002379    .2225073
          cov(cons_1,cons_2) |   .0917669   .0213467      .0499282    .1336056
                 var(cons_2) |   .0913072    .019033      .0540031    .1286112
-----------------------------+------------------------------------------------
Level 1: student             |
                 var(cons_1) |   .8394563   .0187846      .8026391    .8762734
          cov(cons_1,cons_2) |   .4990042   .0158479       .467943    .5300655
                 var(cons_2) |   .8985292   .0201032      .8591278    .9379307
------------------------------------------------------------------------------

. 
. * Fit the model by MCMC   
. runmlwin ///
>   (normexam cons girl, eq(1)) ///
>   (standlrt cons girl, eq(2)), /// 
>   level2(school: (cons, eq(1)) (cons, eq(2))) ///
>   level1(student: (cons, eq(1)) (cons, eq(2))) ///
>   mcmc(on) initsprevious ///
>   nopause
 
MLwiN 2.31 multilevel model                     Number of obs      =      4059
Multivariate response model
Estimation algorithm: MCMC

-----------------------------------------------------------
                |   No. of       Observations per Group
 Level Variable |   Groups    Minimum    Average    Maximum
----------------+------------------------------------------
         school |       65          2       62.4        198
-----------------------------------------------------------

Burnin                     =        500
Chain                      =       5000
Thinning                   =          1
Run time (seconds)         =       23.7
Deviance (dbar)            =   20272.58
Deviance (thetabar)        =   20158.25
Effective no. of pars (pd) =     114.33
Bayesian DIC               =   20386.91
------------------------------------------------------------------------------
             |      Mean    Std. Dev.     ESS     P       [95% Cred. Interval]
-------------+----------------------------------------------------------------
normexam     |
      cons_1 |  -.1643378   .0600507      221   0.003    -.2823821   -.0454367
      girl_1 |   .2621391   .0405411     1943   0.000     .1849588      .34239
-------------+----------------------------------------------------------------
standlrt     |
      cons_2 |  -.1155206   .0480626      385   0.008    -.2112595   -.0225297
      girl_2 |   .1558245   .0403793     2074   0.000     .0790241    .2364237
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |     Mean   Std. Dev.   ESS     [95% Cred. Int]
-----------------------------+------------------------------------------------
Level 2: school              |
                 var(cons_1) |  .1729842  .0344736   3558   .1169745  .2514282
          cov(cons_1,cons_2) |  .0982398   .023249   3267   .0603594  .1506764
                 var(cons_2) |  .0976112  .0206698   2658   .0650132  .1460618
-----------------------------+------------------------------------------------
Level 1: student             |
                 var(cons_1) |  .8398604  .0188513   5116   .8047218  .8778029
          cov(cons_1,cons_2) |  .4995166  .0161975   4812   .4681756  .5326727
                 var(cons_2) |  .8998426   .020322   4673   .8613644  .9406142
------------------------------------------------------------------------------
Top
luisrvaz
Posts: 5
Joined: Mon Aug 18, 2014 10:22 am

Re: MCMC error

Post by luisrvaz »

Hi George,

Yes, this model seems to work fine. Presumably the problem is to do with the data then? The IGLS model works.

Many thanks,
Luis
Top
ChrisCharlton
Posts: 1384
Joined: Mon Oct 19, 2009 10:34 am

Re: MCMC error

Post by ChrisCharlton »

This may be a starting value issue. Do the IGLS results look sensible (i.e. no parameters have a value of zero, etc)?
Top
luisrvaz
Posts: 5
Joined: Mon Aug 18, 2014 10:22 am

Re: MCMC error

Post by luisrvaz »

Hi Chris,

The IGLS output is as follows, and is sensible I think:

Code: Select all

MLwiN 2.30 multilevel model                     Number of obs      =       106
Multivariate response model
Estimation algorithm: IGLS

-----------------------------------------------------------
                |   No. of       Observations per Group
 Level Variable |   Groups    Minimum    Average    Maximum
----------------+------------------------------------------
            SHA |       10          6       10.6         20
-----------------------------------------------------------

Run time (seconds)   =       3.96
Number of iterations =          4
Log likelihood       =  -767.8653
Deviance             =  1535.7306
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
mean_reach   |
      cons_1 |   8.505443   2.035181     4.18   0.000     4.516561    12.49432
   soc_net_1 |   5.722118   2.270977     2.52   0.012     1.271086    10.17315
  type_121_1 |   6.279435   1.768567     3.55   0.000     2.813107    9.745763
have_curre~1 |   4.123179     2.0874     1.98   0.048     .0319506    8.214407
mental_hea~1 |   4.455556   2.073707     2.15   0.032     .3911659    8.519947
-------------+----------------------------------------------------------------
effectiveness|
      cons_2 |   42.13049   1.804684    23.35   0.000     38.59338    45.66761
  type_121_2 |   7.038652   2.074887     3.39   0.001     2.971948    11.10536
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: SHA                 |
                 var(cons_1) |   10.55678   7.644933     -4.427016    25.54057
          cov(cons_1,cons_2) |  -11.41561    7.48183     -26.07973    3.248507
                 var(cons_2) |   12.55333   10.33496     -7.702825    32.80948
-----------------------------+------------------------------------------------
Level 1: service             |
                 var(cons_1) |   64.05847   9.320689      45.79026    82.32669
          cov(cons_1,cons_2) |  -10.17931   8.600607     -27.03619    6.677567
                 var(cons_2) |   107.8221   15.51508      77.41306    138.2311
------------------------------------------------------------------------------
I wondered whether the issue could be to do with the cluster variables, having a small number of level 2 groups i.e. 10 Strategic Health Authorities?

Luis
Top
GeorgeLeckie
Site Admin
Posts: 432
Joined: Fri Apr 01, 2011 2:14 pm

Re: MCMC error

Post by GeorgeLeckie »

Hi Luis,

The starting value for the level-2 covariance implies a level-2 correlation of effectively -1 which is on the boundary on the feasibly parameter space

. display -11.41561/sqrt(10.55678*12.55333)
-.99163972

You could try manually specifying a less extreme starting value.

Yes, 10 clusters would be considered by many as rather low for a multilevel analysis.

Best wishes

George
Top
luisrvaz
Posts: 5
Joined: Mon Aug 18, 2014 10:22 am

Re: MCMC error

Post by luisrvaz »

Hi George,

I tried to specify less extreme starting values than the IGLS model produced, however, after changing the matrix with the stored initial values, the model still produces the same error. Is it possible that it is a similar error obeying batch file to the one listed on the following site:

http://www.bristol.ac.uk/cmm/software/s ... rrors.html

I wondered what might be happening as it specifically mentions lines C1499 and C1500?

Many thanks,
Luis
Top
ChrisCharlton
Posts: 1384
Joined: Mon Oct 19, 2009 10:34 am

Re: MCMC error

Post by ChrisCharlton »

C1499 and C1500 are where the starting values for the residuals and residual s.e. are being stored. You may be able to get a more informative error message if you remove the nopause option from your -runmlwin- call, and then after MLwiN displays the equation click abort macro and then try running the model via the start button.
Top
Post Reply

Return to “runmlwin user forum”