www.cmm.bristol.ac.uk/forum

I have estimated a 3-level model, with respondent as level 1, interviewer as level 2, and (27 European) countries as level 3. Now that I see the parameters for the respondent variables (beta's), and the interviewer variation (sigma²_v0), I know these apply for the whole 'EU-27' area, and I'm ...

Thank you,

I will definitely take this advice into consideration.

Hi,

I am having trouble choosing the best estimation method, and because different methods lead to different conclusions in my case, I wonder what solution to put more trust in.

I have a 3-level binary random intercepts logit regression, and am under the impression the data is under dispersed. I ...

I actually went to predictions -> predict the lododds as a function only of Beta(age) x Age (no intercept). Then calculated this to probabilities with this formula:

exp(logodds)/(1+exp(logodds))

And then plotted these probabilities with age. It looks OK.

However if anyone sees a problem with ...

I have fitted a two-level random intercept logistic model with 3 predictors. I want to make a graph in MLwiN that plots this:

graph example.png

Although in my case I want to plot the predicted probabilities for increase in age.

I have tried customised predictions -> predict mean probabilities ...

"Unfortunately because of the way that logit model are estimated in MLwiN through quasi-likelihood, it is not possible to use the usual deviance to compare models. One could use the Intervals and Tests procedures to test individual and sets of estimates for significance. But using MCMC methodology ...