topic Re: Adjusting for clustered data in regression in Discussions

Adjusting for clustered data in regression

jswislar — Fri, 26 Jan 2018 15:00:27 GMT

I am looking to predict the proportion of patients in a hospital with a specific disease, adjusting for hospital characteristics. However, the data are clustered (hospitals clustered within states) and I want to adjust for this, too. I am using JMP Pro 13.

Is the proper approach to use a Generalized Linear Model and select state with the "Nest" button? This seems correct, but when I do that JMP includes each state (n=51) as a variable. This is effectively what I want, but adds a lot of DF and variables to a data set with relatively few observations.

Am I doing this right, or is there a better way?

Thanks

Re: Adjusting for clustered data in regression

cwillden — Fri, 26 Jan 2018 15:26:24 GMT

Normally, I would want to make state and hospital[state] random effects. That would dramatically reduce the impact of the total number of parameters to estimate. However, random effects are not supported for GLMs in JMP.
You could do a cluster analysis and use the clusters in place of states.
If you have access to SAS, you could use Proc GLIMMIX to do the mixed model GLM.

Re: Adjusting for clustered data in regression

jswislar — Fri, 26 Jan 2018 15:30:19 GMT

Thank you for the quick response. Would the new cluster variable be entered as a "Nest" variable in GLM? I already know there are 4 clusters I can group the states into.

Re: Adjusting for clustered data in regression

cwillden — Fri, 26 Jan 2018 15:34:43 GMT

You certainly can do that.