www.cmm.bristol.ac.uk/forum

Evening all,

I'm a relatively novice MLwiN user so apologies if my query is easily resolved or if there have been lots of previous posts.

I'm trying to fit a multilevel multinomial model of couple HIV status on three levels, the individual, the couple, and their district, using data from Lesotho.

However, I haven't been able to fit anything in MLwiN yet because every time I try to fit the model, I get the following messages:

"no non linear component to mixed response model - check group 11"

"no non-linear component to mixed response model - check group 11(offsets 1)"

And yet I've gone Data Manipulation - Groups and found nothing in group 11.

What does this error message usually mean, and would anyone be able to hazard a guess as to why I'm receiving it?

Please see attached pictures for further clarification.


Many thanks,
A very stressed student. (my name is Chris)

Unlike a lot of statistical packages MLwiN does not automatically add an intercept to the model, and from the screenshots that you provided it looks like this is missing. To add this create a column of ones the same length of your dataset (the MLwiN convention is to name this cons). Once this is created click the "Add term" button, select cons and choose "Add separate coefficients". You will also need to specify a denominator, which you can do by clicking the red n in the equation and again selecting your cons variable.

If you are unfamiliar with fitting multinomial models in MLwiN I would suggest reading chapter 10 of the user guide (available at http://www.bris.ac.uk/cmm/media/softwar ... al-web.pdf) as well as module 10 of the LEMMA online course (see: http://www.bris.ac.uk/cmm/learning/online-course/).

This is brilliant help, thank you very much! And thanks for getting back to me so quickly also!

Would you mind if I asked a follow-up question?

After having some success with models, I am now stuck with another error. This time it says "columns of unequal length in group (calc g18=g19). I've gone through the user manual and can't find any luck. Do you know what might be happening here?

As the estimation method to fit these type of model involves transforming the variables a copy of the ID, response and explanatory variable columns is made at the start of each iteration using the command you referenced. What the error message is saying is that the length of these columns do not match, so I would suggest checking the length reported in the Data Manipulation->Names window of each of the columns in the model.