Let us have variable X (it is skin conductance level) that has a great interindividual variability in "baseline" due to which all between subject effect are masked by this baseline "intercept". Something like:

Within subject effect (Baseline phase / Task phase)
Between subject effect (Task / Task + stressor)

CONTROL
Person 1:::: Baseline X = 10 / Task (no stressor) X = 13
Person 2:::: Baseline X = 4 / Task (no stressor) X = 4,6
Person 3:::: Baseline X = 14 / Task (no stressor)X = 16

EXPERIMENTAL = Task + STESSOR
Person 1:::: Baseline X = 3 / Task (with stressor)X = 6
Person 2:::: Baseline X = 7 / Task (with stressor)X = 11
Person 3:::: Baseline X = 14 / Task (with stressor)X = 20

Baseline phases are not manipulated, but the task is by an additional stressor.

As you can see, the experimental stress (Task with stressor) did create an higher effect on X if you look at the relative change, but not if you compare mere means.

I was thinking about this transformation:

X_relative = (X_TASK - BASAL_X) / BASAL_X

Now we have great difference!

Person1 X (nostress) = 0,3
Person2 X (nostress) = 0,15 MEAN X = 0,2 %
Person3 X (nostress) = 0,14
EFFECT is 0,2 -> 0,66
Person1 X (w stress) = 1,0
Person2 X (w stress) = 0,57 MEAN X = 0,66 %
Person3 X (w stress) = 0,43

So, this "%" way counts for "baseline". But is it ok to ANOVA or Regress this percentage values and say "The experimental manipulation /between factor/ counted for XY% of variance of relative change of X". I think, it is ok, but just to make sure.