I'm conducting survival analysis in Stata with data structured as multiple records over time. To illustrate, let's say I'm studying patient mortality. I'd like to check whether two explanatory variables, say weight and blood pressure, are correlated.

I could run a pairwise correlation using pwcorr weight bloodpressure, but this wouldn't take into account the fact that the observations can come from the same patients. How should I test for correlation between weight and bloodpressure while controlling for clustering on patientid?

I've seen the Variance Inflation Factor (VIF) used for testing correlations in survey data, but I'm not sure how to transfer it to survival data.

  • $\begingroup$ What is your ultimate interest here? Correlation between weight and blood pressure won't really hamper your ability to estimate their effect on mortality, unless the correlation is very high. And if your interest is purely in the correlation between these predictors, your problem isn't survival analysis anymore, it's about summarizing correlation longitudinally. $\endgroup$ – JAW Aug 27 '13 at 15:01
  • $\begingroup$ @JAW I'm interested in the correlation because one of the predictors I'm using in my actual analysis is novel, and I want to demonstrate that it's not too related to other predictors. I suppose you're right that it's not really a problem of survival analysis, but I want to make sure that the survival structure of the data doesn't cause problems for my correlation calculation. $\endgroup$ – ASGM Aug 27 '13 at 15:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.