Basic regression model question.

Suppose I treated 3 groups of mice, group 1 with water, group 2 with drug A, and group 3 with drug B. The water treatment thus serves as baseline.

Now my question is what is a more appropriate linear model to use in order to assess the effect of each drug relative to the water treatment:

  1. Do 2 pairwise comparisons: A vs. water and B vs. water?

  2. Use a linear model: response ~ treatment, where the levels in treatment are: water, A, and B with dummy coding.


1 Answer 1


I'd sort of use a mix of the two options you have suggested. I'd fit the linear model response ~ treatment where treatment is at three levels. (note that if you are working in R then dummy coding is automated). Since you have a specific hypothesis in mind, I would also set a priori contrasts on the linear model to test whether water = a and water = b. Again, if you are working in R then setting contrasts like this is fairly straightforward.

  • $\begingroup$ So I guess by running a linear model as response ~ treatment in R, it is actually doing all pairwise contrasts vs. the baseline level, correct? $\endgroup$
    – dan
    Commented Mar 25, 2017 at 4:21
  • $\begingroup$ Yep, R uses the constraint of reference level set to zero. As long as water is your reference level then it will test the others against that $\endgroup$ Commented Mar 25, 2017 at 4:33

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.