Characteristic is not defined at correct level of analysis

[KatrijnDelar]

Dear all,

I am trying to run a model that has a nesting structure as shown in the attached figure.
However, when I want to incorporate a country-cohort characteristic, MLwiN does not recognize it as defined at this level (neither in IGLS, nor in MCMC). MLwiN considers it as a cohort characteristic, and subsequently I cannot obtain the correct coefficient.
When I leave the cohort level out of my nesting structure, MLwiN recognizes my country-cohort characteristic correctly and gives a completely different coefficient.

I really hope that someone can give me some advice!

Thanks,
Katrijn

Nesting structure.JPG (19.44 KiB) Viewed 12848 times

[ChrisCharlton]

It looks like your data structure is cross-classified, so you would need to fit the model using MCMC with this option turned on. See chapter 15 of the MCMC manual (http://www.bristol.ac.uk/cmm/media/soft ... mc-web.pdf) for more details.

[KatrijnDelar]

Dear Chris,

Thank your for your quick reply.

However, I actually do use MCMC (with the cross-classification option turned on).
When I add my country-cohort characteristics, the variable has a lm-subscript (which corresponds to the cohort level).
Do you know whether something else could cause this problem?

In any case, many thanks,
Katrijn

[ChrisCharlton]

As described in 15.2 of the MCMC guide cross-classified models are normally written with a single subscript. The multiple subscript notation assumes that the model is nested, so this may be what is causing the confusion. To turn on the single subscript mode you need to click the Notation button on the equations window and untick multiple subscripts. To check that the number of units being used is what you expect you need to go to the Model->Hierarchy Viewer window.

[KatrijnDelar]

Thanks!
Katrijn