[danbruc]

Naively I would expect regression toward the mean to weaken the placebo effect - using a placebo seems a better idea in light cases and even in a controlled study you would probably not be allowed to only use placebos in severe cases so that it seems more likely that the situation worsens which would reduce the observed strength of the placebo effect in the other direction.

[ekianjo]

But placebo only work in the situations considered as light cases, where regression to the mean is still feasible. Once you consider cancer, severe infections or other stuff like that, there is no placebo effect anymore. That's why I said most of the placebo effect is regression to the mean, and once you study diseases where regression to the mean never occur, then the placebo effect is in-existent.