We conducted an experiment to determine if type of blood is a main effect or interact with a treatment (2 treatments). The subjects were measured on several hours (5 times). There are 3 subjects by type of blood (these are replicates). Then, we have the effects of TYPE, TREATMENT, TIME.
We are not sure if Time is a covariate, we think Time as a factor because we don't look for a specific behavior on Time, only at 5 Times, to have a replication in Time (then Time can be random factor). However, TYPE and TREATMENT are fixed factors.

This is a repeated measures design or mixed model, or simply ANOVA with 2 fixed and one random factor (within ¿?).

Thank you.

  • $\begingroup$ interesting... you can have two people on her with the exact same handle... $\endgroup$ – John Jul 27 '11 at 20:13

Don't forget the SUBJECT random effect. That's the one that makes it into a repeated measures/mixed model design. You are correct that TYPE and TREATMENT are fixed, but how to treat TIME will depend on what assumptions you want to make about time. The simplest thing to do would just be to leave it out and treat the five measures as independent subsamples on each individual, but more generally, you could treat TIME as a fixed effect/repeated measure and model the correlation between each time point.

The preferred terminology in these models can differ depending on your field; as the same model can often be described several ways, so it's not really a matter of deciding whether it's a repeated measure/mixed model/ANOVA; you could probably use any of those terms to describe the model you end up with. What's more important is to define what terms you want to include in the model and how you want them to be able to vary.

  • $\begingroup$ Thank you for your quickly answer. I wonder if I treat "TIME as a fixed effect/repeated measure and model the correlation between each time point." Which is the meaning of intercept in this case?. And finally, How can do such analysis if I'm using R? $\endgroup$ – John Jul 27 '11 at 20:58
  • $\begingroup$ just use lmer (package lme4). Something like this: lmer(y ~ type + treatment + type*treatment + (1|subject)). This will run a regression with intercept varying by subject (specific effect by subject) $\endgroup$ – Manoel Galdino Jul 28 '11 at 0:42
  • $\begingroup$ If you want to fit the correlation between time points, you'll need the lme function in the nlme package instead of lme4. Also, if you want to assume that the five points are independent subsamples and everything is balanced, it's (almost nearly) equivalent to use the averages for each individual in a simple ANOVA. $\endgroup$ – Aaron left Stack Overflow Jul 28 '11 at 13:55

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.