Interpret a Ratio of Arithmetic Means Without Calling It a Geometric Ratio

Specify the two means

A ratio of arithmetic means divides one group’s arithmetic mean by another’s. A ratio of geometric means instead compares averages on a logarithmic scale after back-transformation. They can differ substantially in skewed data, so the phrase mean ratio is not enough to identify the estimand.

Cochrane’s continuous-outcome guidance discusses alternative effect measures and their interpretation. Before using a ratio, establish that the outcome scale and its zero make a relative comparison meaningful. Adding an arbitrary offset can change a ratio even when an absolute difference stays unchanged.

Check scale and population compatibility

Record units, group definitions, analysis counts and whether the reported means are raw or model-adjusted. Verify that the denominator mean is not zero or near zero in a way that makes the ratio unstable. Negative means can make a ratio difficult to interpret even when arithmetic computation is possible.

Do not divide a median from one group by a mean from another. If studies report different summaries, recover compatible information or use a synthesis strategy that acknowledges the mismatch. A familiar-looking ratio should not conceal incompatible numerator and denominator quantities.

Compare a hypothetical scale change

Suppose group A has arithmetic mean 20 units and group B has mean 10 units. Their ratio is 2 and their difference is 10 units. Multiplying both measurements by 100 changes the means to 2,000 and 1,000; the ratio remains 2.

Adding 10 to every measurement changes the means to 30 and 20, so the ratio becomes 1.5 while the difference remains 10. This illustrates why an arbitrary zero, as on some temperature or questionnaire scales, can make a relative mean comparison scientifically misleading. The calculation itself cannot establish that the scale supports a meaningful ratio.

Carry uncertainty through the chosen method

A ratio of point estimates is not accompanied by a valid interval merely by dividing separate confidence limits. Use an appropriate method that accounts for sampling variation and any dependence between the means. Independent groups and repeated measurements require different covariance handling.

Report the two underlying means alongside the ratio, name the mean type and describe the uncertainty method. If a log-transformed model was used, explain what its exponentiated coefficient actually estimates. The final interpretation should answer a relative question that is meaningful on the original scale, with no hidden switch between arithmetic, geometric and adjusted summaries.

Source verification: 6 November 2026. Examples are hypothetical.