What is the difference between internal and external validity?

Internal validity is whether the estimate is correct for the people actually studied (free of bias); external validity is whether that estimate generalises to other populations or settings.

Does a large sample size guarantee internal validity?

No. A large sample reduces random error and narrows confidence intervals, but it does nothing to remove systematic errors such as confounding, selection bias, or information bias, which determine internal validity.

Which threats does internal validity summarise?

It summarises freedom from the three main systematic errors — confounding, selection bias, and information bias — as distinct from random (sampling) error.

Internal Validity

Internal validity is the degree to which a study's estimate of an exposure-outcome association is correct for the people actually studied — that is, free from systematic error. A study is internally valid to the extent that confounding, selection bias, and information bias have been adequately controlled. It is the prerequisite for asking the separate question of external validity: whether the finding generalises beyond the study population.

Definition

Internal validity is the extent to which the measured association between exposure and outcome in a study reflects the true association in the study population, unbiased by confounding, selection bias, or information bias.

Scope

The entry covers the meaning of internal validity, the three systematic threats it summarises, its relationship to random error and to external validity, and how reporting standards ask studies to address it. It is a methodological reference and offers no clinical guidance.

Core questions

Has confounding been adequately controlled by design or analysis?
Could selection of or attrition from the study sample have distorted the estimate?
Were exposure and outcome measured accurately enough to avoid material misclassification?
Is the estimate's uncertainty (random error) reported alongside these systematic concerns?

Key concepts

Systematic error (bias)
Confounding
Selection bias
Information bias
Random error and precision
External validity (generalisability)
Reporting standards (STROBE)

Mechanisms

Internal validity is best understood as the absence of three systematic errors. Confounding mixes the exposure's effect with that of a common cause; selection bias distorts the association through how subjects enter or remain in the analysis; and information bias distorts it through mismeasurement. A study with strong internal validity has addressed all three so that the remaining departure from the truth is mainly random error, which is quantified by confidence intervals and shrinks with sample size. Internal validity is logically prior to external validity (generalisability): an estimate that is biased for the study population cannot reliably be extrapolated to others. Tools that strengthen internal validity include randomisation, restriction and matching, appropriate adjustment guided by causal reasoning, blinded and standardised measurement, and minimising loss to follow-up.

Clinical relevance

Internal validity is the first thing weighed when judging whether a study's result should be believed, because a finding that is biased for its own participants provides little dependable evidence. The concept describes how the trustworthiness of evidence is assessed; it does not direct any individual's diagnosis or treatment.

Epidemiology

Assessing internal validity is part of every critical appraisal of observational and experimental research, and reporting guidelines such as STROBE require authors to describe potential sources of bias and the limitations bearing on validity. The systematic separation of internal from external validity is a standard organising principle in epidemiologic methods.

Evidence & guidelines

The STROBE statement (von Elm et al., 2007) asks observational studies to report efforts to address bias and confounding and to discuss limitations affecting internal and external validity, making the concept operational in the reporting of research.

History

The distinction between internal and external validity was articulated in mid-twentieth-century research-methods writing and was adopted into epidemiology as the field formalised its treatment of bias and confounding. Late-twentieth-century causal-inference work and, later, reporting guidelines such as STROBE gave the concept of internal validity precise content and a place in the routine appraisal and reporting of studies.

Debates

How should internal and external validity be traded off?: Strengthening internal validity (for example, through restriction or tightly controlled designs) can narrow the population studied and so limit generalisability; commentators differ on how much priority to give each, though internal validity is generally treated as a prerequisite for any useful external inference.

Key figures

Kenneth Rothman
Sander Greenland
David Grimes
Kenneth Schulz

Seminal works

grimes-schulz-2002-bias
vonelm-2007

Frequently asked questions

What is the difference between internal and external validity?: Internal validity is whether the estimate is correct for the people actually studied (free of bias); external validity is whether that estimate generalises to other populations or settings.
Does a large sample size guarantee internal validity?: No. A large sample reduces random error and narrows confidence intervals, but it does nothing to remove systematic errors such as confounding, selection bias, or information bias, which determine internal validity.
Which threats does internal validity summarise?: It summarises freedom from the three main systematic errors — confounding, selection bias, and information bias — as distinct from random (sampling) error.