Relative Risk vs Odds Ratio: 2×2 Table Worked Examples

Relative risk (RR), also called the risk ratio, is the probability of the outcome in the exposed group divided by the probability in the unexposed group — a direct "how many times more likely" statement. The odds ratio (OR) is the odds of the outcome given exposure divided by the odds given no exposure, computed as the cross-product ad/bc from the 2×2 table. The key practical difference is design: RR requires that you can estimate actual risks, which you can in cohort studies and randomized trials but NOT in case-control studies, where participants are selected by outcome status. In case-control studies only the OR is valid. The OR is also the native effect measure of logistic regression. When the outcome is rare (under about 10%), the OR closely approximates the RR; when the outcome is common, the OR is pushed further from 1 than the RR and overstates the effect.

Lay the table out consistently: rows are exposure (exposed / unexposed), columns are outcome (disease / no disease). Cell a = exposed with the outcome, b = exposed without it, c = unexposed with the outcome, d = unexposed without it. Then: risk in the exposed = a/(a+b); risk in the unexposed = c/(c+d); odds in the exposed = a/b; odds in the unexposed = c/d. Getting the cells in the right corners is half the battle — a surprising share of wrong answers come from transposing exposure and outcome. Always label the margins before computing anything.

RR = [a/(a+b)] / [c/(c+d)]. An RR of 1 means no association; RR > 1 means exposure raises risk; RR < 1 means exposure is protective. The interpretation is clean and intuitive: an RR of 2.0 means the exposed group is twice as likely to develop the outcome. Because RR uses the denominators (a+b) and (c+d), you must KNOW the size of the exposed and unexposed populations and observe outcomes prospectively — which is exactly why RR works in cohort studies and trials but not case-control designs, where the investigator fixes the number of cases and controls and the "risk" denominators are artifacts of sampling.

OR = (a/b) / (c/d) = ad / bc — the cross-product ratio. An OR of 1 means no association, with the same directional reading as RR. The OR’s mathematical advantage is that it does not depend on the at-risk denominators, so it stays valid even when cases and controls are sampled separately — that is why case-control studies report ORs. It is also symmetric and the quantity logistic regression naturally estimates (the exponentiated coefficient is an OR). The cost is interpretability: an OR is a ratio of odds, not of probabilities, so "2.5 times the odds" is not the same plain-English claim as "2.5 times the risk," and the two diverge as the outcome becomes common.

A cohort follows 200 smokers and 200 non-smokers for a respiratory outcome. Smokers: 40 develop it, 160 do not (a = 40, b = 160). Non-smokers: 10 develop it, 190 do not (c = 10, d = 190). Risk in smokers = 40/200 = 0.20. Risk in non-smokers = 10/200 = 0.05. RR = 0.20 / 0.05 = 4.0 — smokers are 4 times as likely to develop the outcome. Now the OR = ad/bc = (40 × 190) / (160 × 10) = 7600 / 1600 = 4.75. Both are valid because this is a cohort, but notice the OR (4.75) is larger than the RR (4.0): the outcome here (20% in the exposed) is not rare, so the OR overstates the risk ratio. Report the RR as the headline for a cohort with a common outcome.

A case-control study recruits 100 cases (people with a rare cancer) and 100 controls, then asks about a past exposure. Cases: 30 exposed, 70 not (a = 30, b = 70 — but here the "rows" are set by the design). Controls: 10 exposed, 90 not. Because the investigator FIXED 100 cases and 100 controls, the column totals do not reflect true population risk, so RR is meaningless. Compute the OR = ad/bc = (30 × 90) / (70 × 10) = 2700 / 700 = 3.86. Since the cancer is rare in the population, this OR is a good approximation of the RR you could not measure directly — the rare-disease assumption in action. Interpretation: the exposure is associated with roughly 3.9 times the odds of disease, and because the disease is rare, about 3.9 times the risk as well.

Both RR and OR are skewed ratios, so confidence intervals are built on the log scale and then exponentiated. For the OR, the standard error of ln(OR) is √(1/a + 1/b + 1/c + 1/d); the 95% CI is exp[ln(OR) ± 1.96 × SE]. If the interval includes 1, the association is not statistically significant. The biggest interpretation trap is treating an OR as if it were an RR when the outcome is common — media reports doing this routinely exaggerate effects. A second trap is sign confusion with protective effects: an OR of 0.5 means halved odds, which for a rare outcome is roughly halved risk, but for a common outcome the RR would be closer to 1 than 0.5. When the outcome is common and the design allows it, prefer RR; when only an OR is available, say "odds," not "risk."

Snap a photo of a 2×2 table or an epidemiology problem and StatsIQ labels the cells, identifies the study design, computes the valid measure (RR and OR for cohorts/trials, OR only for case-control), and builds the log-scale confidence interval. It flags the common-outcome case where the OR diverges from the RR so your interpretation stays honest. This content is for educational purposes only.