Longitudinal Confirmatory Factor Analysis
Also known as: longitudinal CFA, repeated-measures CFA, longitudinal measurement model, panel CFA
Longitudinal confirmatory factor analysis (longitudinal CFA) applies a theoretically specified measurement model to data collected at two or more time points. Its primary purpose is to verify that a scale measures the same latent construct in the same way over time — a prerequisite for drawing valid conclusions about change from repeated-measures data.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+6 more
When to use it
Use longitudinal CFA when you have collected the same scale at two or more time points and intend to compare means, examine change, or model growth. It is the correct method when you need to verify that observed score differences reflect genuine changes in the construct and not artefacts of differential item functioning over time. Longitudinal CFA requires theory-driven factor structure specification — use exploratory factor analysis first if the structure is unknown. Do not use it when the sample is very small (fewer than roughly 100–200 per time point for typical model sizes), when data are missing in a non-random pattern that cannot be handled by FIML, or when items are dichotomous or severely non-normal without applying the appropriate estimator (such as WLSMV or MLR).
Strengths & limitations
- Provides a rigorous, sequential test of measurement invariance that is required before any latent change analysis.
- Separates measurement error from true score variance, allowing change to be assessed at the latent level.
- Can accommodate partial invariance, permitting substantive conclusions even when a minority of items behave differently across time.
- Integrates naturally with broader longitudinal SEM frameworks including latent growth curve models and latent change score models.
- Handles missing data efficiently via Full Information Maximum Likelihood (FIML), preserving all available observations.
- Produces model fit indices and modification indices that help diagnose which specific parameters are non-invariant.
- Requires a pre-specified factor structure; model mis-specification at baseline propagates through all invariance tests.
- Chi-square difference tests for invariance are sensitive to sample size: trivial non-invariance can appear significant in large samples.
- Demands reasonably large samples — rule-of-thumb minimums around 100–200 per time point — to yield stable parameter estimates and adequate power.
- Partial invariance complicates interpretation and requires careful reporting so readers understand the scope of valid comparisons.
- Computational and conceptual complexity is substantially higher than a simple repeated-measures ANOVA comparison of observed means.
Frequently asked
Is longitudinal CFA the same as testing measurement invariance?
Measurement invariance testing is the central application of longitudinal CFA, but the method is broader. Beyond invariance, longitudinal CFA can model correlated latent factors across time, estimate latent means at each occasion, and serve as the measurement part of larger longitudinal SEM models such as latent growth curves.
What if only partial scalar invariance holds — can I still compare latent means?
Partial scalar invariance — where at least two items per factor have equal intercepts — allows latent mean comparisons with caveats. You must constrain the invariant intercepts, free the non-invariant ones, and clearly acknowledge in your report that comparisons are anchored on the subset of invariant items rather than on the full scale.
How do I handle missing data across time points?
Full Information Maximum Likelihood (FIML) is the preferred approach: it uses all available data at each time point without imputation, producing unbiased estimates under the Missing At Random assumption. Listwise deletion of cases with any missing wave is inefficient and can bias results.
Should I correlate residuals of the same item across time points?
Yes, in almost all longitudinal applications. The same item administered repeatedly shares method variance beyond what the latent factor explains, so correlated residuals between the same item at different occasions are substantively justified and should be included in the model from the outset.
Which estimator should I use with Likert-type items?
For ordinal items with five or fewer categories, WLSMV (Weighted Least Squares Mean- and Variance-adjusted) based on polychoric correlations is preferred over normal-theory ML. If ML is used, the robust variant MLR corrects standard errors and provides a scaled chi-square, reducing but not eliminating bias from non-normality.
Sources
- Widaman, K. F. & Reise, S. P. (1997). Exploring the measurement invariance of psychological instruments: Applications in the substance use domain. In K. J. Bryant, M. Windle & S. G. West (Eds.), The science of prevention: Methodological advances from alcohol and substance abuse research (pp. 281–324). American Psychological Association. link ↗
- Millsap, R. E. (2011). Statistical Approaches to Measurement Invariance. Routledge. ISBN: 9780805864786
How to cite this page
ScholarGate. (2026, June 3). Longitudinal Confirmatory Factor Analysis. ScholarGate. https://scholargate.app/en/psychometrics/longitudinal-confirmatory-factor-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Confirmatory factor analysisPsychometrics↔ compare
- Longitudinal EFAPsychometrics↔ compare
- Longitudinal Measurement InvariancePsychometrics↔ compare
- Measurement InvariancePsychometrics↔ compare
- Multilevel CFAPsychometrics↔ compare
- Structural Equation ModelingResearch Statistics↔ compare