ologit or melogit or just regress?

[Curtis Wray]

I have a continuous variable 90-day death rate at the hospital level. In the database, there are 1000 individual hospitals. I have subdivided the 90-day death rate into quartiles.

I want to look at factors associated with treatment in poorer performing hospitals.

Is an ordered logit the correct model? More specifically a multilevel ologit since patients are nested within hospitals.

Thanks

[Marcos Almeida]

I guess so. You may use a two-level aproach, or - ologit - having hospitals as clusters. You may check with the multilevel model whether it improved, compared to the one-level model.

[Carlo Lazzaro]

Curtis:
whenever counts or rates are named, my jerk-knee reaction is -poisson-.
I do share your approach of considering (as they actually are) patients nested within hospitals; hence, I would take a look at -mepoisson- entry in Stata .pdf manual.

PS: crossed in the cyberspace with Marcos'interesting reply, who has a different take.