......there were statistical methods for defining "improvers" in the protocol and in the PACE paper and we can now clearly see the difference using the different methods, with far fewer people improving using the protocol method.
But what about what I'll call "worsers"?
I'm not thinking here of those who suffered some major adverse event, those were recorded, I'm thinking of those whose score worsened with CBT/GET to a comparable degree to how the "improvers" improved!
...
Were they "worsers" or did their score stay static before and after treatment?
IF their subjective scores are normally distributed, I would think that, since the mean score has improved, any sensible definition of improvers and "worsers" will show more improvers than worsers. That is, if we look at the area under the distribution curve of the 52 week results and compare the area above the baseline mean plus some constant with the area below the baseline mean minus the same constant, the first area must be bigger if the curves are normal distributions and the 52 week mean is higher than baseline.
So what is interesting is seeing if the scores are NOT normally distributed, which can be analysed with the data. Ceiling and floor effects will change the picture I outlined above but so also would a double humped curve. Of course, the 95% confidence level error bars probably also rely on a normal distribution.
But, at the end of the day, these scores are subjective measures from an unblinded trial so any conclusions drawn from them don't really mean much.