Longitudinal Scale Development
Also known as: LSD, longitudinal measurement development, repeated-measures scale construction, scale development with panel data
Longitudinal scale development is the systematic process of constructing and validating a measurement instrument using data collected at multiple time points. It extends classical scale development by additionally testing whether the scale measures the same construct in the same metric across occasions, enabling valid tracking of change over time.
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.
When to use it
Use longitudinal scale development when you are constructing a new instrument intended to track change or growth over time in a panel or cohort study, or when adapting an existing cross-sectional scale for repeated use. It is essential whenever latent mean comparisons across waves are planned. Do not apply the full longitudinal framework when only a single administration is intended — cross-sectional scale development with CFA suffices in that case. Also avoid using this framework when the construct itself is expected to change in meaning over the study period (construct evolution), as invariance constraints would then be theoretically inappropriate.
Strengths & limitations
- Provides rigorous evidence that score changes across time reflect genuine construct change rather than measurement artifact.
- Integrates scale construction and longitudinal SEM within a single coherent framework.
- Identifies which specific items drift in meaning over time, allowing targeted revision.
- Supports both latent growth curve and autoregressive longitudinal models once invariance is established.
- Produces a scale with well-documented temporal psychometric properties, strengthening manuscript credibility.
- Requires a minimum of two, preferably three or more, measurement occasions, substantially increasing data collection costs and attrition risk.
- Testing measurement invariance demands large samples at each wave — small samples yield poorly powered tests that may miss real non-invariance.
- When partial invariance is found, interpretation of mean comparisons becomes complex and requires additional assumptions.
- Attrition across waves can introduce selection bias, undermining the representativeness of later-wave estimates.
Frequently asked
How many time points are needed for longitudinal scale development?
A minimum of two waves is required to test any longitudinal invariance, but three waves are strongly preferred because they allow autoregressive and growth models and provide more stable invariance estimates. With only two waves, configural and metric invariance can be tested, but identifying all scalar constraints may require additional assumptions.
What is the difference between longitudinal scale development and just administering an existing scale repeatedly?
Longitudinal scale development builds measurement invariance testing into the construction process itself, producing an instrument with documented temporal psychometric properties. Administering an off-the-shelf scale repeatedly assumes invariance without testing it — a potentially serious error if the items drift in meaning across occasions or populations.
What should I do if some items are not longitudinally invariant?
First, investigate whether non-invariance is systematic (e.g., item wording becomes less relevant over time) or random. If at least two items per factor remain invariant, a partial invariance model may be estimated, with non-invariant parameters freed across waves. Latent mean comparisons remain possible but must be interpreted cautiously, and all freed parameters must be reported transparently.
Can I use this framework with ordinal Likert-type items?
Yes, but use polychoric correlations and a robust weighted least squares estimator (e.g., WLSMV in lavaan or Mplus) rather than maximum likelihood. Ordinal versions of metric and scalar invariance — testing thresholds instead of intercepts — should then be applied.
How large a sample is needed at each wave?
As a rough guide, aim for at least 200 cases per wave for a simple unidimensional scale and 300–500 for multidimensional scales, accounting for expected attrition. Monte Carlo simulation specific to your model is the most reliable way to determine adequate power for the invariance tests you plan.
Sources
- Millsap, R. E. (2011). Statistical Approaches to Measurement Invariance. Routledge. ISBN: 978-0805864311
- Little, T. D. (2013). Longitudinal Structural Equation Modeling. Guilford Press. ISBN: 978-1462510160
How to cite this page
ScholarGate. (2026, June 3). Longitudinal Scale Development. ScholarGate. https://scholargate.app/en/psychometrics/longitudinal-scale-development
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 CFAPsychometrics↔ compare
- Longitudinal Measurement InvariancePsychometrics↔ compare
- Multilevel Scale DevelopmentPsychometrics↔ compare
- Scale developmentPsychometrics↔ compare
- Test-Retest ReliabilityPsychometrics↔ compare