First the link to this week’s complete list as HTML and as PDF.
***
Interleukin 6 is an inflammation marker that can also rise under stress. The normal value in healthy subjects is about 1 pg/ml but it can rise up to 1000. Creswell et al. report on the success of “mindfulness meditation”
in lowering a heightened IL-6 value in long-term unemployed subjects. Both arms of the randomised study had normal and statistically identical values to begin with – 1.81 ± 2.03 and 1.21 ± .76 pg/ml. The slightly elevated value for the intervention group looks like most participants with normal and one or two with spuriously high values – as always data are not given. Four months after the intervention both groups show exactly identical values at 1.45 ± .78 and 1.41 ± .73 pg/ml. (I have converted the given SE back into the correct SD here.) If anything, all this study shows is the regression to the mean of a spurious difference in initial conditions.
Addendum
I have been strongly criticized for my evaluation above. Apparently I failed to appreciate the distinction between a within-group and a between-group study. This is entirely possible, because I still fail to understand the difference. Maybe I’m just too dense and should desist from presuming to write these criticisms. All I see is a randomized study, where the claimed effect is smaller than the initial difference between the two arms, where the claimed effect is no larger (if in the opposite direction) than the no-effect in the no-intervention arm, – so the most parsimonious explanation is simple regression to the mean – and where the claimed effect stays well inside the standard error of the initial condition, which for me equals the very definition of “no effect”
. I also fail to see what within-group even means here. All we are given are agglomerated data for the group as a whole and these values show no change. If there were to be any more, the authors keep it well hidden from us readers. I can only judge by what I’m given and what I can see.
***
Benson-Amram is another misrepresented non-result. As a comparison of figures 2a and 3a clearly shows the predictive value is zero. At least this is a scientific article and data are given – without them this evaluation would have had to be pure guesswork and the non-result might have gone unnoticed and accepted by the readership.
***
As Prof. em. John Brignell 
, one of the fathers of automated measurement and computer modelling continues to emphasize, all a model can do is build a hypothesis of what might be. It is never and can never be the equivalent of actual data. Bellprat & Doblas-Reyes make another case in point for the most egregious current abuse.