Margins/Predict & MEOLOGIT

[Korbinian Jellinek]

Hi there,

I am calculating a multilevel mixed effects model with an ordinary outcome. Therfore I use

meologit depvar var1 var2 [..varlist...] || country: var1 , vce(cluster country)

a command that is equivalent to

meglm depvar var1 var2 [..varlist...] || country: var1 , vce(cluster country) family(ordinal) link(logit)

I have tried all integrations methods, meaning "mvaghermite", "mcaghermite", "ghermite". "laplace" is not viable as "initial values not feasible; r(1400);". However, I cannot use the predict or margins command (neither spost13 package) because

could not compute empirical Bayes' means;
missing values were returned by the evaluator
r(322);

In the update 06aug2014 it says at point 5.

predict after gsem and me would incorrectly return zero-valued empirical Bayes' mean estimates when
intmethod(mvaghermite) was not a viable integration method.  In this case, gsem and me will now report a note
below the coefficient table acknowledging that the results were produced using nonadaptive quadrature, and
predict will issue the error message "could not compute the empirical Bayes' means".

But as already said, I tried all integration methods. Additionally I dropped the multilevel part so this part here

 || country: var1

With this configuration of the command, everything works.

What can I do?

Best
korbinian

[Korbinian Jellinek]

nobody?

[Clyde Schechter]

Just grasping at straws here: have you tried re-running your model using -mvaghermite- but giving it the results of your non-adaptive quadrature estimates as starting values? Perhaps from that starting point it will converge, and then you will be able to use -predict- and -margins-. That's all I can think of. If that doesn't work, you might consult Stata technical support.

[Henrik L. Lolle]

Hi Korbinian,

I have the same problem as you. If you have found a solution, I would very much like to hear about it.

Best,
Henrik

[Rich Goldstein]

why is "country" both a random effect and being used for cluster-adjusted SE's? I would start by dropping "vce(cluster country"