What is Discriminant Validity?
Discriminant validity shows that a survey or test measures a distinct concept rather than something else. It tells you that two different scales are not just re‑measuring the same thing.
Discriminant validity is a property of measurement tools (like questionnaires) that confirms different constructs are actually different. For example, a scale for ‘satisfaction with services’ should not simply duplicate a scale for ‘trust in staff.’ Practically, researchers check discriminant validity by showing low-to-moderate correlations between measures of different constructs and by using statistical methods (factor analysis, HTMT ratio, Fornell‑Larcker checks) to confirm items group with the right construct and not with others. Good discriminant validity gives confidence that each survey section is capturing a unique idea.
Usage example
You run a community feedback survey with separate scales for ‘service accessibility’ and ‘perceived fairness.’ If those scales correlate very highly, you might lack discriminant validity — respondents may be answering as if the two concepts are the same. You'd review the wording or item set to make the distinction clearer.
Practical application
Discriminant validity matters because it affects how you interpret results and make decisions. If two measures are not distinct, you might misattribute cause or double-count an issue when planning services or reporting findings. For Hearo users, checking discriminant validity is especially important when running multilingual surveys: translations or cultural differences can blur distinctions between constructs. Verifying discriminant validity (and rewording or re-testing items where needed) helps ensure comparisons across groups and languages are meaningful and defensible.
FAQ
How do I check discriminant validity in my survey?
Common checks include examining correlations (different constructs should not be extremely high), running exploratory or confirmatory factor analysis to see whether items load on the intended factors, and using metrics like the HTMT ratio as a rule of thumb. For most practical purposes, start by looking for unexpectedly high correlations and review item wording where distinctions are unclear.
What counts as ‘too high’ a correlation between different constructs?
There is no hard cutoff, but as a rule of thumb correlations above about 0.7–0.8 suggest substantial overlap and merit review. Modern approaches (like HTMT) often use thresholds around 0.85–0.90. Treat these numbers as guidance — also consider theoretical expectations and item content.
Can a single-question measure have discriminant validity?
Single-item measures can be useful, but they make validity checks harder because you can’t test internal structure. With single items, rely more on clear wording, pilot testing, and external comparisons to other measures to build confidence that the item is capturing a distinct concept.
Could translation affect discriminant validity?
Yes. A translation can change nuance so that two concepts that were distinct in the original language appear similar (or vice versa). When using multilingual surveys, check discriminant validity separately by language where possible and use participant feedback to refine wording so constructs remain distinct across translations.