2-level hierarchical analysis of covariance

[Jennifer May]

Good evening,

I have an RCT with an intervention arm and standard care, with a continuous outcome variable ('score'). The initial plan was to conduct an ANCOVA and to control for baseline 'score'. But we now want to cluster by hospital.
Is a multilevel ANCOVA possible? Or do I use the 'mixed' command and conduct a hierarchical linear model?

Any help/comments appreciated.

Thank you

[Rich Goldstein]

if all you want to do is adjust your SE's for the clustering, that can be done by adding "vce(cluster hospital)" as an option to your command (but change "hospital" to whatever the correct variable name is for your data)

note that -mixed- does estimate a multi-level ANCOVA (if we are using the term in the same way); if you want more detail, you need to show, at least, the exact command you fed to Stata for your ANCOVA

[Joseph Coveney]

Is a multilevel ANCOVA possible? Or do I use the 'mixed' command and conduct a hierarchical linear model?

I recommend mixed, and relegating anova for pedagogical use with textbook examples.

[Jennifer May]

Thank you both for your help.

My outcome variable is 'score', I don't want to have change in 'score' as my outcome variable as I want to control for baseline score. If I use the command -mixed- is it correct to have 'score' as my outcome variable and adjust for baseline score?

[Leonardo Guizzetti]

You may call your variables whatever you like. One variable -score- contains all your follow-up values, and a second variable (call it -base- or whatever) is the baseline value for all records for each subject.

[Rich Goldstein]

yes, the example on p. 504 of the manual shows just such a situation

[Jennifer May]

thanks for your help. please would you be able send me the link to the example?

[Leonardo Guizzetti]

In Stata, type -help mixed- and hit enter. In the help window, click the link at the top to view PDF documentation. In the PDF, go page 504 and read the example.