just a simple question, to make sure this is correct. I watched some videos with similar study questions and I'm following them. I have 4 groups of treatments(Ct+A1+A2+A3), and two different moments, pre(baseline) and post(8 weeks after). To compare the effect of the drugs on my variable on R studio I'm using:

> treatment = factor(as.cathegoric(treatment)
> is.factor(treatment)
> model1 = lm(post~pre+treatment+pre:treatment)

Them after that I get my summary(model1) and anova(model1).

But my question is, My PI told me to use the basal values as the covariate comparing post to pre. I'm assuming that I'm doing that by using this formula. Because lm(post~pre+treatment+pre*treatment) I'm comparing the post to a variable fixed by groups, and along with the interaction of the pre values with the group. I've being seeing a bunch of people attesting this on papers ("..Tested with ANCOVA with basal values as the covariate and “study group” as the fixed factor..."). I just wanted to be sure I did the same before going on with the paper.

Thank you!!


  • $\begingroup$ the page is omitting my asterisks.. the formula is: model1 = lm (post~pre + treatment +pre:treatment). $\endgroup$
    – Alan F
    Sep 30, 2019 at 22:06

1 Answer 1


You are doing what your PI wanted.

This is not necessarily a good idea. If the variable measured in pre and post is measured very accurately (like, for instance, weight) you should be OK. But if it is measured with error (like, for instance, blood pressure) then you will be correlating with error.

It would be much better to have more than two points for each person, but a multilevel model will still be better than this sort of analysis, if the MLM converges.

Pre- post studies with unreliable variables are a big problem.


Your Answer

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

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