Internal validity
Whether a study really shows that the cause it claims produced the effect.
What it means
Internal validity is the degree to which a study establishes that the intervention or independent variable — and not some confounding factor — actually caused the observed effect. It is threatened by a well-catalogued set of alternative explanations identified by Campbell and Stanley, including history (outside events), maturation (natural change over time), selection (pre-existing differences between groups), regression to the mean, testing and instrumentation effects, and differential attrition. The principal defenses are randomization, which on average equates groups on all confounds known and unknown, together with control or comparison groups, blinding, and careful measurement, all of which isolate the treatment as the only systematic difference. A fundamental nuance is that internal validity often trades off against external validity: the tight control that secures clean causal inference can create artificial conditions that limit how far the result generalizes. It matters because causal claims are the currency of behavioral science and policy — if internal validity is weak, an apparent treatment effect may be an artifact, and any decision based on it rests on sand.
Examples
If the group that received a new tutoring program also happened to be more motivated to begin with, you cannot tell whether the program or the prior difference raised their scores — internal validity is compromised.
A city's road-safety campaign runs the same winter that fuel prices spike and driving drops. Crashes fall, but the campaign and the petrol price are both plausible causes.
In a fitness-app trial, the people who quit the app also quit answering the surveys, so the final data comes only from those it suited — and the app looks better than it is.
First described in Campbell & Stanley (1963).