# NO NO NO NO...

From this paper on a Randomized Controlled Trial of acupuncture for allergic rhinitis

We therefore started with noninferiority tests of change in RQLQ score and concluded that real acupuncture was noninferior if the left limit of the covariance-based, 2-sided 95% CI surrounding the between-group difference between real and sham acupuncture was greater than the noninferiority margin of ⫺0.5 point (12). If we showed noninferiority, we repeated the analysis for the RMS outcome using a noninferiority margin of ⫺1.5 points. This threshold was chosen on the basis of a review that we performed of unpublished RQLQ and RMS data suggesting a rough equivalence of scales at a ratio of 1:3, so the RQLQ threshold of 0.5 point translated to an RMS threshold of 1.5 points. If this procedure also showed noninferiority, we tested for superiority and concluded that real acupuncture was superior to sham acupuncture if at least one of the Bonferroni-adjusted, analyses of covariance– based, 2-sided 97.5% CIs surrounding the between-group difference in RQLQ score and RMS was completely greater than zero.

This is how NHST turns people's brains to mush. In this study **both** the acupuncture group and the "medication only" group got acupuncture, it just happened at different times. They have assessment at 4 time points: t=0 initial baseline, t=8 weeks (only acupuncture group has had acupuncture), t=16 weeks (acupuncture group had acupuncture between week 0 and 8, medication group had acupuncture between t=8 and t=16), and again at week 60 (no treatment between t=16 and t=60 weeks). Clearly, what's needed is a time-series analysis for score as a function of what treatments happened when, but instead, we get "noninferiority and superiority testing at a predefined threshold of ..." bullshit.

Unfortunately, their reproducible research statement says "Data set: Certain portions of the analytic data set are available to approved individuals through written agreements with the author or research sponsor." which means getting the data is a lot more complicated than just downloading it off an open public website.