Latent Growth & Longitudinal SEM Services
When you measure the same people, teams, or firms repeatedly, the interesting questions are about change: how do individuals develop over time, do they follow different trajectories, and how do variables influence one another across waves? Longitudinal structural equation modelling answers these—from latent growth curves to cross-lagged panel models.
Longitudinal SEM models change over time using repeated measures within a structural equation framework. Latent growth curve models estimate each unit’s trajectory (an intercept and slope) and what predicts it; growth mixture models allow distinct trajectory classes; and cross-lagged panel models (including the RI-CLPM) estimate reciprocal, over-time relationships between variables.
What longitudinal SEM does
Cross-sectional data can describe how things stand at one moment; longitudinal data—repeated measurements of the same units—let you study how they change, and how variables shape one another over time. Longitudinal structural equation modelling brings the flexibility of SEM (latent constructs, measurement models, complex path structures) to that repeated-measures setting, and it comes in several forms suited to different questions.
The latent growth curve model (LGCM) treats each unit’s trajectory as defined by latent factors—typically an intercept (starting level) and a slope (rate of change)—then estimates the average trajectory and, crucially, how much individuals vary around it. Predictors can explain that variation (why some grow faster than others), and the growth factors can themselves predict later outcomes. Growth mixture models extend this by allowing distinct latent classes of trajectory—subgroups that develop in qualitatively different ways—rather than assuming everyone varies around a single average. And the cross-lagged panel model (CLPM) family, including the random-intercept CLPM (RI-CLPM), focuses on the reciprocal question: does earlier X predict later Y, does earlier Y predict later X, or both, once each variable’s own stability is accounted for.
Choosing the right longitudinal model
The model should follow the question. Use a latent growth curve model when the interest is the shape and predictors of change—how a construct develops over time and what accounts for individual differences in that development. Use a growth mixture model when theory suggests there may be distinct developmental subgroups rather than one population trajectory, and you want to identify and profile them. Use a cross-lagged model when the question is about reciprocal influence between variables across waves.
Within the cross-lagged family, the choice between the traditional CLPM and the RI-CLPM matters. The traditional CLPM does not separate stable, between-person differences from within-person change, which can distort the cross-lagged estimates; the RI-CLPM adds a random intercept to model those stable differences, so the cross-lagged paths reflect within-person processes. Where the question is genuinely about within-person dynamics, the RI-CLPM and related models are generally more appropriate—and we set out the trade-offs rather than defaulting to one.
Which longitudinal model fits the question
| Model | Core question | Estimates |
|---|---|---|
| Latent growth curve | How does the construct change over time? | Average intercept & slope; individual variation; predictors of change |
| Growth mixture | Are there distinct trajectory subgroups? | Latent trajectory classes and their profiles |
| CLPM | Do X and Y influence each other over time? | Cross-lagged & autoregressive paths |
| RI-CLPM | Within-person reciprocal effects? | Cross-lagged paths net of stable between-person differences |
| Multilevel SEM | Nested / clustered longitudinal data? | Effects across levels (e.g. within & between units) |
Getting longitudinal models right
Longitudinal SEM rests on foundations that must be checked, not assumed. The first is measurement equivalence over time: for change in a construct to be interpretable, the construct must be measured the same way at each wave, which is exactly what longitudinal measurement invariance establishes. Without it, apparent “change” can be an artefact of a shifting instrument. The second is the treatment of missing data, which is near-universal in panel studies through attrition; modern estimation (full-information maximum likelihood, or multiple imputation) handles it under stated assumptions, and how missingness is addressed should be reported.
Each model also carries its own cautions. Growth models require a sensible specification of the functional form of change (linear, or a justified non-linear shape) and enough time points to estimate it. Growth mixture models are powerful but can over-extract classes and are sensitive to specification—the number of classes is decided using several indicators together with interpretability and theory, and a recovered class is not automatically a real subpopulation. In the cross-lagged family, the CLPM-versus-RI-CLPM choice is consequential, as noted above. We make these decisions explicitly and report them.
Change is only interpretable if the measure holds over time. Before reading a growth curve or a cross-lagged path as substantive, longitudinal measurement invariance should be established—otherwise the “change” may be movement in the instrument, not the construct.
Software
We fit longitudinal SEM in established, reproducible tools—R’s lavaan (and tidySEM) and Mplus for growth, mixture, and cross-lagged models—with FIML or multiple imputation for missing data and versioned code and output provided.
How we deliver a longitudinal SEM
This work sits within our wider SEM & Psychometrics practice, and connects to our longitudinal & panel methods—so the model is matched to the question and built on measurement that holds over time.
We start from your repeated-measures data and the change question you want to answer, and select the model that fits it—growth curve, growth mixture, or a cross-lagged specification. Where constructs are latent, we establish longitudinal measurement invariance first, then specify the model, handle missing data appropriately, and estimate it.
Reporting follows current longitudinal-SEM conventions: the invariance evidence, the model and its fit, the growth or cross-lagged parameters, class-enumeration criteria for mixtures, and the missing-data approach, all transparently presented.
You receive the fitted longitudinal model with its parameters and fit—average and individual trajectories, trajectory classes, or cross-lagged and autoregressive paths as the model requires—the invariance and missing-data documentation, a clear interpretation of the change process, and reproducible analytical code and analysis-ready files (where appropriate and permitted).
Longitudinal SEM across Management & Allied Studies
Questions about development, trajectories, and reciprocal influence over time run throughout the social sciences—so longitudinal SEM applies across the disciplines we serve.
Management & Organizational Research
How employee attitudes, performance, or capabilities develop over time, and reciprocal effects such as engagement and performance across waves.
Applied Psychology & HR
A core setting—developmental trajectories of well-being, motivation, and behaviour, and within-person reciprocal dynamics.
Marketing & Consumer Research
How consumer attitudes, loyalty, or usage evolve, and whether distinct customer trajectory segments exist.
Education & Learning Sciences
Growth in achievement, motivation, or engagement over a programme, and reciprocal effects between them.
Strategy & Entrepreneurship
How firm capabilities or venture outcomes develop, and reciprocal relationships among strategic variables over time.
Economics & Public Policy
Trajectories of outcomes following a reform or programme, and reciprocal dynamics among indicators across periods.
Longitudinal SEM: common questions
Studying change over time?
Whether your question is about developmental trajectories, distinct trajectory subgroups, or reciprocal effects across waves, we match the longitudinal SEM to the question—built on measurement that holds over time and reported in full.