I am performing a longitudinal analysis and I am curious if the predictors we are including in the model will introduce any unexpected effects.

We have subjects with multiple points of follow-up and they are entered into the study at various times and ages. The main hypothesis we want to test is if there is an effect of what age you entered the study and how long you've been in the study (e.g. at a given age, if you've been in the study for 10 years or 20, is there a difference in your measures). It is also possible that different criteria were used to allow subjects to enter the study depending on when they entered.

Thus, we are adjusting for their age at entry, age at measurement, and decade of entry. To test the main hypothesis, we are looking at the age-at-entry * age-at-measurement interaction. I think people usually do age at entry and time since entry, but this should give identical predictions, and time itself is not directly interesting to us, but both age variables are.

My understanding of APC models is that prediction is unaffected. But the problem is in the interpretation of individual effects -- each coefficient for age at entry and age at measurement is confounded with the duration of study. Similarly, I think that age at entry and decade of entry will be confounded with birth year, though we assume birth year has no effect. Are any new problems introduced with when including interactions? Is including all 3 of these variables problematic? I feel like we might be doing something profoundly stupid with this model, yet I'm not quite sure how else to test the hypotheses and control for selection criteria at entry.

  • 1
    $\begingroup$ After taking the measurement the subject will no longer be in the study. Is that correct? $\endgroup$
    – Stat
    Nov 11, 2013 at 21:02
  • $\begingroup$ No, there are sometimes multiple measures per subject. Basically, subjects come in for a procedure, and we measure a continuous outcome at sporadic times afterwards to monitor their health over the years. My intent is to use a mixed model or GEE. $\endgroup$
    – rjweyant
    Nov 11, 2013 at 21:57


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.