ordinal logistic regression - problems with option vce ( )

[Soren Nielsen]

I am working on a rather large dataset - ~ 10 000 'exposed' persons and ~ 30 000 'non-exposed' controls , matched on sex, age and place of residence. After the preliminary screening and visual analyses of data, I am proceeding to fitting ordinal logistic regression models. As my dataset is from an observational epidemiological study, and matched, I need to use 'robust' standard errors (SEs), in casu vce(cluster stratum). When fitting a proportional odds model I am able to do this. But, when I want to fit adjacent category models, using the 'adjcatlogit' command, and continuation-ratio models, using the 'ccrlogit' - command, I get the reply 'option vce ( ) not allowed'.
Has anyone encountered this problem?
Is there a way around the problem?
Is there an explanation?
The vce ( ) option is allowed when using the command 'ologit', so I am a bit surprised.

Sincerely
Søren Nielsen

[Richard Williams]

Those are user-written commands that are around 10 years old, apparently written with Stata 13. Here is the Stata Journal article on them:

https://journals.sagepub.com/doi/pdf...867X1401400414


For whatever reason, the author did not or could not support the vce option. Perhaps they thought it was inappropriate. Perhaps they didn't know how to support it, or felt no need to support it, or it just never occurred to them to support it.

You can try directing your Qs to the author. Maybe they will explain why supporting vce is a bad idea. Alternatively, maybe they'll be inspired to release an updated version.

You can also try tweaking the programs yourself. It often doesn't take much to add support for an option.

When an official Stata command doesn't do something I would expect it to do, I figure either (a) Stata is smarter than me and knows that it would be wrong to do this, or (b) nobody has gotten around to programming it. For a user-written command, I think (b) is even more likely than it is with an official Stata command.

[Richard Williams]

You might also check out the user-written gencrm command, which does Generalized continuation-ratio regression models. I don't know if it does what you want, but it does support vce.

[Soren Nielsen]

Thanks for the response. I have contacted one author, and consulted Xing Liu(2016) who recommend using mlogit which allows vce(cluster), so I will try that.
Søren Nielsen