I'm comparing several vectors of numbers using a two-sided t-test. There are multiple treatment groups: Control & Treatment 1 & Treatment 2, ...
Within each treatment group, we have two sub populations: A and B. Thus we have:
- Treatment 1:A
- Treatment 1:B
- Treatment 2:A
- Treatment 2:B ...
In each case, the B group can be thought of as a sort of baseline.
I initially compared all treatment groups (A/B) to "control:B" and got a few significant differences. However, to rule out the variations in the baseline measure (that is, variations in "B" subgroups between the control, treatment 1, etc), I've taken the difference between the A and B subgroups, and using those differences, I've done a new t-test.
Essentially, instead of doing a t-test between, say, treatment 1:A vs Control:B and treatment 1:B vs Control:B, now it's a t-test between Control:B-Control:A vs. Treatment:B - Treatment:A.
Does this make sense as a good way to account for variations in the "B" (baseline) groups? Also as an FYI, the A and B groups were measured independently and represent independent samples (i.e. B isn't a measurement of doing something funny to A). Thanks!