Brant test or Wald test for variables across models after meglm (meologit)

[Pierr Exe]

Hello everyone,

I am running a multilevel ordered logit model (5 outcomes) and would like to check the parallel lines assumption. However, the Brant test is not readily available in this context. I was thus thinking of running a series of logit regressions on the various outcomes such that 1 is run against 2, 3, 4; 1 and 2 against 3 and 4; 1, 2 and 3 against 4. After that I would want to see whether the coefficients are significantly different across equations.

The original model is as follows:

meologit y1 x1 x2 x3 || lev3: || lev2:

The subsequent series of logit regressions for the 4 combinations mentioned above is as follows:

logit comb1 x1 x2 x3 || lev3: || lev2:
est sto m1
logit comb2 x1 x2 x3 || lev3: || lev2:
est sto m2
logit comb3 x1 x2 x3 || lev3: || lev2:
est sto m3
logit comb4 x1 x2 x3 || lev3: || lev2:
est sto m4

The coefficients are not too different from one another (even though their statistical significance is).
Ideally, I would like to compare the coefficients as follows:

suest m1 m2 m3 m4
test [m1=m2=m3=m4], common

But the command suest is not compatible with multilevel models. I thought the best would be to find a way to give a name to each equation after each model is stored but I cannot find the macro to do just that. Has anyone any recommendation?

Thank you very much for your help,

**Stata 17** user

[Federico Tedeschi]

Here (https://www.statalist.org/forums/for...-ordered-logit) is a reply to a similar question. I think that, by adapting that code, we could get this:

eq tresh: x1 x2 x3
gllamm y1, i(lev2 lev3) family(binomial) link(ologit) thresh(thresh)
test [_cut11]_b[x1] = [_cut21]_b[x1]
test [_cut11]_b[x2] = [_cut21]_b[x2]
test [_cut11]_b[x3] = [_cut21]_b[x3]

that would perform 3 separate tests for x1, x2 and x3. Maybe this could be a starting point to build a global test for parallel lines/proportional odds.

[Pierr Exe]

Thank you, Federico, that is very helpful!

[Federico Tedeschi]

You're welcome, Pierr.

test [_cut11]_b[x1] = [_cut21]_b[x1]
test [_cut11]_b[x2] = [_cut21]_b[x2]
test [_cut11]_b[x3] = [_cut21]_b[x3]

To make it a global test:

test [_cut11]_b[x1] - [_cut21]_b[x1]=[_cut11]_b[x2] - [_cut21]_b[x2]=[_cut11]_b[x3] - [_cut21]_b[x3]=0

[Pierr Exe]

Dear Federico,

Apologies to answer your last message this late. Thank you very much for that, it is most helpful and I am sure it will also help other fellow stata users! All the best,

Pierr