Adjusting for clustering in ordinal regression

[Sophiay Dulal]

Hi,

I was wondering if someone could help me. I have binary outcomes and ordinal outcomes.

I am using the following command for the binary logistic regression adjusting for the cluster.
melogit `v' `BinVar' || cluster: , or nolog

How do I adjust for clustering in ordinal regression?

When I tried this:
.ologit time1stbreast_cat3G i. wealth, ||cluster: or nolog

Stata showed:
option | not allowed

I really appreciate any help you can provide.

[Jeff Wooldridge]

The word "adjust" can mean different things. What you've done with melogit is model the cluster correlation in a specific way and apply joint MLE. But you could also use pooled MLE and cluster your standard errors:

logit `v' `BinVar', vce(cluster clusterid)
ologit time1stbreast_cat3G i. wealth, vce(cluster clusterid)

If you want to model the cluster correlation using ordered logit, you can use the xtologit command (even if you don't have longitudinal data):

xtset clusterid
xtologit time1stbreast_cat3G i. wealth, vce(robust)

By the way, using the vce(robust) option means you're admitted the model may be misspecified in ways that causes inconsistent parameter estimators. But you should do it anyway. Same with melogit, in my opinion.