www.cmm.bristol.ac.uk/forum

Dear,

I have a dataset (the first in which this has happened) that has a problem when estimating a binomial model (prevalence of pain=30%) with IGLS (MQL1), normal data, sample size n=3135, 2 levels (n=36 clusters), 32 columns used. MLwiN gives the following error message:
------------------------------------------------------
Error while obeying bath file: .... line number 59:
calc F~(H) = expo(H)*(1+expo(H))^(-2)
----------------------------------------------------------

Does anyone knows why this could have happened and how to solve it?

Thanks a lot.
Roger.

This sort of error message is often a sign that the algorithm is having trouble fitting the model. One thing that sometimes helps in this case is to build the model up from simpler models, using the results from the previous model by using the More button instead of Start.