[Hiro farabi]

Hiro,

The residual specification in this model is not related to the partially nested design. It relaxes the testable assumption that a single residual parameter is appropriate for your model. You have partial nesting whether or not you estimate separate residuals by treatment status.

thank you so much Erik, your response is really helpful.

Just one last question: since my dataset is in long format, based on your recommendation, the sid is repeated once for QALYs and once for costs. Am I using t (type) correctly? or it should be repeated over time?

[Hiro farabi]

.

[Erik Ruzek]

Hiro,

When you reshape the data as I suggested in post # 3 for a multivariate model, what residuals(un, t(type)) does is allow for separate correlated residuals by outcome type. I don't know how time is involved in your study and cannot address that. I assumed you had an outcome that was also measured at baseline, which is reflected in my earlier post with the out (final outcome) and out_b (baseline outcome) code.