Page 1 of 1
2-level empty ordinal model
Posted: Tue Apr 12, 2016 11:00 am
by vivian1234
Hi,
I am new to runmlwin and I have a problem in running a 2-level empty model for my ordinal outcome variable.
This is my code:
Code: Select all
runmlwin outcome cons, ///
level2(hh: (cons)) ///
level1(id:) ///
discrete(distribution(multinomial) link(ologit) denominator(cons) basecategory(4) pql2) nopause
Code: Select all
runmlwin outcome cons, ///
level2(hh: (cons)) ///
level1(id:) ///
discrete(distribution(multinomial) link(ologit) denominator(cons) basecategory(4)) ///
mcmc(on) initsprevious nopause
However, the result looks strange. rather than giving me the value for var(cons_123), it gives me 6 random effect parameters.
Thank you very much!
Vivian
Re: 2-level empty ordinal model
Posted: Tue Apr 12, 2016 11:55 am
by ChrisCharlton
Is the following more like what you are after?
Code: Select all
use http://www.bristol.ac.uk/cmm/media/runmlwin/alevchem, clear
egen school = group(lea estab)
Code: Select all
runmlwin a_point cons, level2(school: (cons, contrast(1/5))) level1(pupil: ) discrete(distribution(multinomial) link(ologit) denom(cons) base(6) pql2) nopause
MLwiN 2.36 multilevel model Number of obs = 2166
Ordered multinomial logit response model
Estimation algorithm: IGLS, PQL2
-----------------------------------------------------------
| No. of Observations per Group
Level Variable | Groups Minimum Average Maximum
----------------+------------------------------------------
school | 220 1 9.8 94
-----------------------------------------------------------
----------------------------------
Contrast | Log-odds
-------------+--------------------
1 | 1 vs. 2 3 4 5 6
2 | 1 2 vs. 3 4 5 6
3 | 1 2 3 vs. 4 5 6
4 | 1 2 3 4 vs. 5 6
5 | 1 2 3 4 5 vs. 6
----------------------------------
Run time (seconds) = 4.84
Number of iterations = 8
------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Contrast 1 |
cons_1 | -1.372275 .1031697 -13.30 0.000 -1.574484 -1.170066
-------------+----------------------------------------------------------------
Contrast 2 |
cons_2 | -.4809893 .0982812 -4.89 0.000 -.6736169 -.2883616
-------------+----------------------------------------------------------------
Contrast 3 |
cons_3 | .2940195 .0976683 3.01 0.003 .1025931 .4854459
-------------+----------------------------------------------------------------
Contrast 4 |
cons_4 | 1.171295 .1001301 11.70 0.000 .9750433 1.367546
-------------+----------------------------------------------------------------
Contrast 5 |
cons_5 | 2.391542 .1099275 21.76 0.000 2.176088 2.606996
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school |
var(cons_12345) | 1.329894 .1784171 .9802029 1.679585
------------------------------------------------------------------------------
Code: Select all
runmlwin a_point cons, level2(school: (cons, contrast(1/5))) level1(pupil: ) discrete(distribution(multinomial) link(ologit) denom(cons) base(6)) mcmc(on) initsprevious nopause
MLwiN 2.36 multilevel model Number of obs = 2166
Ordered multinomial logit response model
Estimation algorithm: MCMC
-----------------------------------------------------------
| No. of Observations per Group
Level Variable | Groups Minimum Average Maximum
----------------+------------------------------------------
school | 220 1 9.8 94
-----------------------------------------------------------
----------------------------------
Contrast | Log-odds
-------------+--------------------
1 | 1 vs. 2 3 4 5 6
2 | 1 2 vs. 3 4 5 6
3 | 1 2 3 vs. 4 5 6
4 | 1 2 3 4 vs. 5 6
5 | 1 2 3 4 5 vs. 6
----------------------------------
Burnin = 500
Chain = 5000
Thinning = 1
Run time (seconds) = 46.6
Deviance (dbar) = 7046.19
Deviance (thetabar) = 6891.03
Effective no. of pars (pd) = 155.16
Bayesian DIC = 7201.36
------------------------------------------------------------------------------
| Mean Std. Dev. ESS P [95% Cred. Interval]
-------------+----------------------------------------------------------------
Contrast 1 |
cons_1 | -1.324595 .103499 40 0.000 -1.506988 -1.094635
-------------+----------------------------------------------------------------
Contrast 2 |
cons_2 | -.462988 .098712 36 0.000 -.6341004 -.2398282
-------------+----------------------------------------------------------------
Contrast 3 |
cons_3 | .2886934 .1000466 35 0.000 .1082631 .5150148
-------------+----------------------------------------------------------------
Contrast 4 |
cons_4 | 1.142674 .1030695 38 0.000 .9477457 1.380694
-------------+----------------------------------------------------------------
Contrast 5 |
cons_5 | 2.339768 .113265 47 0.000 2.12508 2.583499
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Mean Std. Dev. ESS [95% Cred. Int]
-----------------------------+------------------------------------------------
Level 2: school |
var(cons_12345) | 1.271291 .1928117 469 .9335589 1.688362
------------------------------------------------------------------------------
Re: 2-level empty ordinal model
Posted: Tue Apr 12, 2016 12:07 pm
by vivian1234
So I have to specify contrast(1/3) in the 2nd level random part? Is it because the var(cons.123) is the only random part?
Thank you so much!
Vivian
Re: 2-level empty ordinal model
Posted: Tue Apr 12, 2016 12:15 pm
by ChrisCharlton
The contrast option is used where you want a parameter to be common across more than one of the response categories (as in your case), otherwise the default is to have a separate parameter for each of the response categories.