noesisnoesis
How measurement works

What is construct validity?

The hardest question in measurement: whether the thing you are measuring exists as you have defined it, and whether your instrument reaches it rather than something adjacent.

Construct validity asks whether an instrument measures the theoretical construct it claims to, rather than something correlated with it, something narrower than it, or nothing coherent at all. It is the deepest form of validity evidence and the one that takes years rather than a study to establish.

The difficulty is structural. A construct such as resilience or emotional intelligence is not an object; it is a proposed regularity in behaviour. To show that a scale measures it, you have to simultaneously defend the claim that the construct exists as described and that these particular items reach it. The two claims cannot be separated, which is why Cronbach and Meehl, who introduced the idea in 1955, framed it as testing a whole theoretical network rather than one instrument.

How the case is built

Predicted patterns of correlation. The theory says which things the construct should relate to and how strongly. Self-compassion should correlate positively with life satisfaction, negatively with depression, and only moderately with self-esteem — if it correlated 0.9 with self-esteem, it would not be a distinct construct. Every one of those predictions is a testable claim, and failing one is informative.

Expected group differences. If the theory says a trait develops with age, or is elevated in a clinical population, the instrument should show it. Failing to find a difference the theory demands is evidence against the instrument, the theory, or both.

Response to intervention. If a construct is supposed to be trainable, scores should move after training and not move in a control group. Constructs that never respond to anything are hard to distinguish from artefacts.

Internal structure. The factor structure should match the theorised structure — and, importantly, in a fresh sample. Exploratory factor analysis on the development data will find whatever structure was designed in; confirmatory analysis on new data is where the claim is actually tested.

The multitrait-multimethod idea

Campbell and Fiske's 1959 contribution was to insist that construct validity requires two things at once: measures of the same trait by different methods should agree, and measures of different traits by the same method should not.

That second half is where most instruments struggle. If your self-reported empathy correlates 0.7 with your self-reported agreeableness and 0.2 with observer-rated empathy, the strongest thing your questionnaire is measuring is not empathy — it is your general way of describing yourself on questionnaires. Method variance is one of the most persistent problems in self-report research, and it is the reason serious validation work seeks a non-questionnaire criterion wherever one exists.

Why this matters when reading your results

Construct validity is what stands between a score and its interpretation. When a well-validated instrument reports that you score high on a trait, the chain of reasoning behind that statement includes decades of published work establishing that the trait behaves as a coherent quantity, that these items track it, and that the score means something similar across the people who take it.

When a test with no construct validation reports the same thing, the score is a number generated by an algorithm nobody has checked. Both look identical on the results screen. The difference is entirely in the documentation, which is why the documentation is worth looking for.

Put it to the test

Reading about measurement is one thing. Seeing your own score reported with its source, its norm sample and its limits is another. Free to take, no signup required.

Keep reading