Processing data for graphical display can sometimes hide underlying problems with
the data. In this example, it seems that injecting a dose of performance-enhancing
drugs into one arm of weightlifters significantly increases the number of repetitions
they can perform before the onset of muscle fatigue, compared with the number achieved
by the sham-injected control arm (Fig. 1a; regression analysis coefficient R
2
?=?0.7995, p??0.001). The greater the proportion of the recommended safe dose of drugs injected,
the higher the fold difference in lifting endurance compared with the control. But
these processed data hide a somewhat surprising result, which was not clearly described
in the study: lifting ability actually decreased in control arms with an increasing
dose of the drug in experimental arms (Fig. 1b).
Fig. 1. Number of weight lifts performed by healthy males in a steroid-injected arm compared
with the control. a The ratio of lifts performed by steroid versus control arms across a range of drug
doses. b The absolute number of lifts. R
2
?=?regression analysis coefficient
The control varies between treatments, and so the experimental to control ratio increases
with drug level in the experimental arm largely, though not entirely, because the
function of the control arm is worse. Regardless of whether this represents a real
biological effect, the way the data were processed and displayed in the top graph
was inappropriate — this example highlights the importance of examining raw and not
just processed data.