Longitudinal Discriminant Validity
Also known as: LDV, longitudinal construct distinctiveness, cross-time discriminant validity, temporal discriminant validity
Longitudinal discriminant validity tests whether a psychological construct measured at two or more time points is empirically distinct across occasions — ensuring that the same construct does not collapse into a single undifferentiated mass over time. It is a prerequisite for meaningful change modeling in panel and longitudinal research.
Key highlights
- Provides a rigorous, model-based test of whether temporal change can be meaningfully modeled rather than assumed.
- The chi-square difference test is straightforward to implement in any SEM software given a longitudinal CFA baseline.
- Complements convergent validity and reliability checks to give a full picture of longitudinal measurement quality.
- Applies to any number of time points and any construct type measurable via reflective indicators.
- Catches measurement collapse early, preventing invalid conclusions about stability, change, or causal dynamics.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use longitudinal discriminant validity assessment whenever you model change, growth, or temporal mediation in panel data and the same construct is measured at two or more waves. It is especially critical before running latent growth curve models, cross-lagged panel models, or autoregressive models, because these all assume that the construct at T1 and the construct at T2 are meaningfully distinct. Do not apply this test as a stand-alone procedure without first establishing measurement invariance; if configural or metric invariance fails, the cross-time factor correlation is not yet meaningfully interpreted.
Strengths & limitations
- Provides a rigorous, model-based test of whether temporal change can be meaningfully modeled rather than assumed.
- The chi-square difference test is straightforward to implement in any SEM software given a longitudinal CFA baseline.
- Complements convergent validity and reliability checks to give a full picture of longitudinal measurement quality.
- Applies to any number of time points and any construct type measurable via reflective indicators.
- Catches measurement collapse early, preventing invalid conclusions about stability, change, or causal dynamics.
- Highly dependent on the quality and comparability of measurement across waves; requires metric invariance at minimum.
- In very large samples the chi-square difference test almost always rejects the unit-correlation constraint even when the practical difference is trivial; supplementary effect-size criteria are needed.
- A rejected constraint confirms cross-time distinction but says nothing about whether the observed change is substantively meaningful or due to measurement drift.
- The AVE criterion is computed at the item level and can be misleading when factor loadings vary substantially across time.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Do I need full measurement invariance before testing longitudinal discriminant validity?
Metric invariance (equal factor loadings across waves) is the minimum requirement, as it ensures the latent variable scale is the same at each occasion and makes the cross-time correlation interpretable. Scalar invariance is needed if you also want to compare latent means. Configural invariance alone is insufficient.
What if the chi-square difference test is non-significant?
A non-significant result means you cannot reject the hypothesis that the cross-time factor correlation equals 1.0, which calls into question whether the construct is truly distinct across occasions. Before concluding poor discriminant validity, inspect the actual correlation estimate and its confidence interval; if it is well below 1.0 but the test lacks power (small sample), consider confidence-interval methods.
How is this different from test-retest reliability?
Test-retest reliability (typically an observed-score correlation) conflates true stability with measurement error across occasions. Longitudinal discriminant validity operates at the latent level and tests whether the construct at T1 and the construct at T2 are statistically distinct entities, not merely how consistently individuals are ranked. You want high test-retest reliability AND good discriminant validity.
Can I assess this with more than two time points?
Yes. With three or more waves you test the constraint that each cross-time factor correlation equals 1.0, either pairwise or jointly. A joint test constraining all cross-time correlations to 1.0 is the strongest, but pairwise tests are more informative about which particular intervals show validity problems.
Is there a recommended threshold for the cross-time correlation below which discriminant validity is acceptable?
Common practice follows Fornell and Larcker (1981): the squared cross-time correlation should be less than the AVE for the construct at each wave. An observed cross-time phi below about .85 is sometimes cited as a practical heuristic, but the formal chi-square or CFI-difference test is preferable to any arbitrary cutoff.
Sources
- 1.Cole, D. A. & Maxwell, S. E. (1993). Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology, 112(4), 558–577.
- 2.Vandenberg, R. J. & Lance, C. E. (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Longitudinal Discriminant Validity. ScholarGate. https://scholargate.app/psychometrics/longitudinal-discriminant-validity