Relative Risk (Risk Ratio)

Relative risk divides the risk in the exposed group by the risk in the unexposed group. It is the natural measure for a cohort study or a randomised trial, because both designs follow defined groups forward and can therefore observe a genuine risk in each. A relative risk of 3 means the exposed group's risk was three times the other's, and the interpretation is as direct as it sounds.

What it hides is scale, and the hiding is not accidental — this is the number a press release quotes. A relative risk of 2 sounds alarming and describes both a rise from 40 % to 80 % and a rise from one in a million to two in a million. The absolute risk difference separates those cases and the ratio cannot. The rule of thumb for reading any risk report: find the baseline risk, or treat the relative figure as decoration.

Two structural cautions on the arithmetic itself. Because a risk cannot exceed 1, a large baseline risk caps how large the relative risk can be — with an unexposed risk of 40 %, no relative risk above 2.5 is arithmetically possible, and the ceiling is not a finding about the exposure. And relative risks are not symmetric under relabelling: an RR of 2 for developing a condition is not an RR of 0.5 for avoiding it, which is one reason odds ratios, which do behave symmetrically, are preferred in regression models.

The last caution is the oldest one in the field. A relative risk quantifies an association and nothing more. Confounding, selection and reverse causation all produce ratios that look exactly like a causal one, and randomisation is what removes them. Bradford Hill's 1965 criteria — strength, consistency, temporality, biological gradient and the rest — are the framework for reasoning past an observational ratio toward causation, and Hill was explicit that they were considerations for judgement rather than a checklist to be scored.