www.cmm.bristol.ac.uk/forum

Dear all,

I am seeking to derive the ICCs of a cross-classified multilevel logit model. The data are 5,000,000 individuals nested in Level A (150 units) and Level B (40 units); it is an additive cross-classifed model.

Group sizes vary considerably, however, from 500 to 1,000,000 in the case of ...

Edit: Of course the cross is necessary. Given that the resulting dataset of 200,000 obs * 1,000 iterations becomes unmanageable, would you have any suggestions on how this could still be plotted?

Great, many thanks for these clarifications.

Given n(level1) = 200,000 and n(level2) =40, would you have any recommendation for the chain length?

This leads to a related issue: The sandbox example works great in the relatively small tutorial dataset. I have been unable to run it on my real dataset ...

Dear Chris,

Many thanks, also to George, for having a look at this.

This is very accessible and fantastic.

A few questions to make sure I understand it all:

Why is thinning = 5? I understand that every fifth result is stored but I am not sure I understand why it is done. Should (burnin(#) and ...

Dear Chris,

This is great, many thanks. Allow me a final and perhaps naive question: Once predlo predmd predhi are predicted, how would I then go about plotting them?

Many thanks in advance!

Thank you for your continued help. I have worked through these slides and tutorials a while ago, they are a tremendous resource (and actually also taught me a lot about Stata).
The predicted probabilities in 7.6. are also great.
What is a challenge for me is to get the confidence intervals of ...

Thanks for sharing this link and the example. There is so much yet to be discovered for me! As far as I can tell, the worked example visually represents the different slopes for individuals nested in different schools.

What I would be keen on achieving is to estimate an interaction across levels ...

Dear Chris,

Your solution worked like a charm, many thanks for your guidance!!

May I follow up with one further question? Suppose we expect the effect of educ to decline in d_lit . Without predicted probability and marginal effects plots it is hard to interpret this logit interaction model ...

Dear Chris,

Thank you very much for your very generous offer!

Please find a stylized sandbox example below, it is based on the "bang" data and the example given in the runmlwin helpfile.
Substantively, it of course doesn't make much sense; it assumes d_pray to be related to use only through d_lit ...

PD. To be more specific, when I run a sandbox example (hence only 10 repetitions), and using e-class, all seems to work well until I get a "name conflict"


capture program drop myreg
program myreg, eclass
tempname bb
capture drop Res_X_uhat

reg A B C if pickone_c==1, r
predict Res_X_uhat ...