www.cmm.bristol.ac.uk/forum

Hello

The MLwiN in the binary logistic model provide information only about the between-school (level 2) variance. However, there is no information about the level 1 variance. Is there any way to obtain this variance so as to have a better picture of the model? For example in SPSS for logistic regression the Cox and Snell's R2 or the Negelkerke's R2 values inform about the overall fit of the model.

Thanks

Hi Stasis,
Not all single level logistic regression techniques translate directly to the multilevel case. Even in normal models the concept of R^2 is not usually referred to in multilevel situations as you do not have a single variance to reduce! The multilevel logistic also has interesting properties in that in some sense the level 2 variance is on a different scale from the level 1 variance due to the link function and so adding predictor variables can inflate the level 2 variance. I guess the best I could offer in terms of goodness of fit is to consider the predictive power of the model on the data i.e. if you predict from the model on to the actual data and then use some threshold to indicate whether your prediction should be considered a 0 or a 1 you could calculate a statistic for how often the model gives the correct prediction.
Hope this helps,
Best wishes,
Bill.