Ask Mallard

Generalizability and transportability

Generalizability asks whether the effect estimated in the study population would be the same in the population you want to apply it to.

Two different questions

Internal validity asks whether the estimate is right for the people in the study. Generalizability asks whether it is right for anyone else. A study can be perfectly internally valid and irrelevant to your patients, and the second question cannot be answered by any amount of care about the first.

Why representativeness is the wrong frame

The intuition is that a sample should look like the population. But an effect transports unchanged whenever no effect modifier is distributed differently between study and target, whatever else differs. If the treatment works equally in men and women, a men-only trial generalises to women perfectly. If it works differently by kidney function, a trial that matched the population on age, sex and ethnicity but excluded renal impairment does not generalise to renal patients at all.

When a study and its target differ

An effect that grows with an effect modifier, above two distributions of that modifier: the study sample concentrated at low values and the target population at high values, so the study's average effect does not describe the target.Transport depends on effect modifiers, not on looking representativeEffect modifierSize of the effect grows with the modifierStudy sampleTarget populationWhere the effect varies with this factor and the two populations carry it differently, the study's averageeffect will not describe the target.
Conceptual diagram. It shows the structure of the idea, not data from any study.

How to reason about it

  • Name the target population explicitly, before the study, not afterwards.
  • List the plausible effect modifiers from mechanism and prior evidence.
  • Compare the distribution of those modifiers between the study sample and the target, and report it.
  • Where the distributions differ, standardise or reweight the estimate toward the target rather than asserting that the result applies.
  • Report eligibility criteria in enough detail that a reader can tell who was excluded, and what proportion of real patients that would be.

Transporting an estimate formally

If the modifiers are measured in both the study and a sample of the target population, the study estimate can be reweighted so that the modifier distribution matches the target. This makes the assumptions explicit: that the modifiers are all measured, and that everyone in the target had some chance of appearing in the study population.

What to watch out for

A non-representative sample is not automatically a threat to generalizability, and this is the point most often got backwards. What matters is whether the study population differs on something that MODIFIES the effect, not whether it looks like the population in general. Restricting a study to a narrow group can improve internal validity at no cost to transportability if the effect does not vary on the restricting variable; conversely a demographically representative sample generalises badly if a strong modifier is distributed differently in the target.

Clinical example

A trial in patients under 65 with normal renal function reports a benefit. Whether it applies to your 80-year-old with reduced clearance depends on whether age and renal function modify the effect, which is a pharmacological question, and not on whether the trial sample resembled the national population.

Recent studies that used this
Searching PubMed…
Related Learn pages
Sources & further reading
  1. Why representativeness should be avoided (opens in a new tab) · Int J Epidemiol · doi 10.1093/ije/dys223
  2. Generalizability of randomized trial results to target populations: Design and analysis possibilities (opens in a new tab) · Res Soc Work Pract · doi 10.1177/1049731517720730
Last source review: 2026-09-06
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.