Tukuk-OSThe relativity trick: “brain tumour risk doubles with X” sounds apocalyptic and may mean 1 in 10,000 becoming 2 in 10,000. Relative risk multiplies; absolute risk adds — and only the absolute version lets you decide anything sane. Whenever a headline gives you a percentage increase, hunt the baseline: the same result honestly stated as “an extra one case per 10,000 people” often reverses the emotional conclusion.
The subgroup trick: slice data enough ways and “significant” findings appear by chance alone. Screen 20 outcomes at the usual p<0.05 bar and statistically you expect about one false positive even when nothing is true; the “works only in men under 40” finding is usually that stray. This is not hypothetical — psychology and medicine have lived a replication crisis: landmark findings repeatedly failing retesting, p-hacking (collecting, slicing and modelling until the threshold trips) now estimated to inflate a substantial share of published positives. The 2010s correction — pre-registration, larger samples, open data — is real but incomplete.
The checklist that survives contact with any study article: how many subjects (a finding from 12 people is anecdote with decimals)? Compared against what (control group, or before-after with time confounds)? Measured or self-reported? Published in a venue that peer-reviewed it, or a press release? Who funded the analysis? And finally, the meta-rule the whole academy teaches: one study is never a verdict — science accumulates in the boring direction, and the trustworthy claim is the one several independent groups have now failed to knock down.