> If it is selection bias then the training has no effect whatsoever.

That doesn't follow. Both factors could have an effect at the same time and both effects could, in principle, go in either direction. The only thing that we can conclude from the present study is that the joint effect of placebo and selection bias is positive.

Yes. I just meant that the entire effect could have been due to selection bias, so we don't know if there was any placebo effect at all.