Internal validity

The property of an empirical study that warrants its causal claim within its own sample and setting — whether the observed intervention-outcome association is genuinely produced by the intervention rather than by confounders, selection, or bias — established by ruling out a closed catalog of named threats.

Version

v4 · 2026-09-19 · History · 1 correction

Domain-specific #

561

Domain group

Formal Sciences

Origin domain

Experimental Design & Statistics

Subdomain

Research Design → Experimental Design & Statistics

Core Idea

Internal validity is the property of an empirical study that warrants its causal conclusion within its own sample and setting — the degree to which the observed intervention-outcome association is genuinely produced by the intervention rather than by confounders, selection, measurement bias, or chance. The question is strictly local and logically prior to external validity, since an internally invalid claim cannot be rescued by wider generalization. The Campbell-Stanley / Cook-Campbell framework supplies a finite catalog of named threats — history, maturation, testing, instrumentation, regression, selection, attrition — ruled out by design or analysis.

How would you explain it like I'm…

Did It Really Work?

Say you water a plant with special juice and it grows taller. Was it really the juice? Maybe it just got more sun that week, or it was going to grow anyway. Internal validity means you've checked those other reasons, so you can trust that the juice did it.

Ruling Out Sneaky Causes

Internal validity is about whether an experiment really proves what it claims. If kids who used a new reading app read better, was it the app — or did they just get older, practice the test, or were they better readers to begin with? Scientists have a list of these sneaky other explanations. A study has good internal validity when it rules out each one, for example by putting kids into groups at random. It only answers whether the app worked for those kids in that study, not for everyone everywhere.

Ruling Out Rival Causes

Internal validity is how well a study supports its claim that the intervention caused the observed outcome, within that study's own participants and setting. It asks whether the effect could instead come from confounders, selection differences, measurement bias, contamination, or chance. It's different from external validity, which asks whether the result generalizes to other people and places — and internal validity comes first, since a result that isn't real can't be generalized. Researchers use a list of named threats, such as history, maturation, testing effects, instrumentation changes, regression to the mean, selection, and attrition. They rule each out with design choices like randomization and blinding, or with analysis showing the threat is implausible.

 

Internal validity is the property of a study that warrants its causal conclusion within the sampled participants and setting — the degree to which an observed intervention-outcome association is genuinely produced by the intervention rather than by confounding, selection artifacts, measurement bias, procedural contamination, or chance. It is strictly local and logically prior to external validity, since generalizing an invalid causal claim cannot rescue it. Campbell and Stanley (1963), elaborated by Cook and Campbell (1979), organized it as a finite catalog of named threats: history, maturation, testing, instrumentation, regression to the mean, selection, attrition, and their interactions. Each threat specifies a mechanism by which an effect could be spurious. Internal validity is established by systematically ruling out threats, either through design (randomization against selection, blinding against observer bias, pre-registration against outcome switching) or through analysis showing a threat is implausible. The catalog turns open-ended skepticism into a structured, checkable audit.

Scope of Application

Internal validity lives across the empirical-research fields that share the substrate of controlled inquiry on quantitatively measured outcomes.

• Experimental psychology and education research — the textbook home of the Cook-Campbell taxonomy.
• Clinical trials — randomization, blinding, and intention-to-treat blocking specific named threats.
• Program and policy evaluation — difference-in-differences, regression discontinuity, synthetic controls.
• Software A/B testing — sample-ratio mismatch, novelty effects, and network interference as the catalog restated.
• Marketing and pricing experimentation — holdout groups, geo-experiments, and synthetic control.

Clarity

Internal validity separates two questions practitioners fuse into "is this study any good?": did the intervention cause the change in this sample, and will the effect appear elsewhere. Naming the first as internal and the second as external makes them independently assessable and reveals they vary in any combination. Because a broken causal warrant cannot be rescued by generalizing wider, internal validity is logically prior. Its second service is converting diffuse skepticism into a structured audit against a finite, named threat list.

Manages Complexity

The objection over every causal claim — "maybe something else produced this" — is literally infinite. Internal validity compresses that unbounded space onto a closed catalog of mechanism-categories, so the question becomes "for each named threat, is it operating here?" The analyst then tracks one three-valued status per threat — blocked by design, checkable post hoc, or residual — and sums across the catalog to read the overall warrant. A priority ordering keeps the compression honest: run the threat checklist first, defer generalization until the local warrant survives.

Abstract Reasoning

The defining move is a catalog-driven audit — walking the closed threat list and inferring from a design feature which threat it activates. Each threat drives a three-valued classification with a security ranking (design-blocked beats analysis-blocked, since the latter rests on measuring the right covariates). A foundational ordering move makes internal validity prior to external, preventing the error of widening the sample to fix a broken warrant. Its interventionist face runs at design time, predicting which threats a contemplated control would neutralize and choosing the design that secures the warrant in advance.

Knowledge Transfer

Within quantitative empirical research internal validity transfers as full apparatus: the internal/external separation and ordering, the closed threat catalog, the three-valued per-threat status, and the design-beats-analysis ranking restage identically across psychology, clinical trials, policy evaluation, A/B testing, and marketing experiments — only the setting changes. Beyond that substrate the named apparatus does not travel; engineering root-cause analysis, intelligence, and history face the same spurious-cause problem with their own threat lists. The portable lesson belongs to the parents Causal Inference, Intervention, and Validation; the Cook-Campbell catalog is a domain-specific archetype-level checklist.

Relationships to Other Abstractions

Current abstraction Internal validity Domain-specific

Parents (3) — outgoing typed relations

• Internal validity presupposes Causal Inference Domain-specific

  Internal Validity presupposes a Causal Inference whose in-setting warrant it audits against confounding, selection, history, attrition, and other threats.

• Internal validity presupposes, typical Experimental Design Prime

  Internal validity typically presupposes experimental design because its strongest threat controls are built into assignment and measurement before analysis.

• Internal validity presupposes, typical Intervention Prime

  Internal validity typically audits whether an externally assigned treatment produced the outcome after its normal incoming causes were controlled.

Children (1) — abstractions linked here by typed relations

• External Validity Domain-specific presupposes Internal validity

  A positive transport warrant presupposes that the in-sample causal effect being transported is locally warranted rather than an artifact.

Hierarchy paths (8) — routes to 7 parentless roots

• Internal validity → presupposes Causal Inference → is a kind of Statistical Inference → is a kind of Inductive Reasoning

Show alternative paths (7)

Neighborhood in Abstraction Space

Internal validity sits in a sparse region of the domain-specific corpus (75^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Warning Fatigue & Threat Response (12 abstractions)

Nearest neighbors

• Bradford Hill criteria — 0.84
• Hazard-Control Decay — 0.83
• Identity Threat — 0.83
• Causal Inference — 0.82
• Latent Condition — 0.82

Computed from structural-signature embeddings · 2026-10-25