Description
Models each person's repeated measures as an individual trajectory around an average curve, estimating mean change, individual differences in change and their predictors.
Growth curve models treat each person's scores over time as a weighted sum of time functions with person-specific coefficients. Fixed effects give the average starting level and rate; random effects give individual differences; their covariance shows whether starting level predicts change. The same model can be fitted as a multilevel (random-coefficient) model or as a structural equation model in which repeated measures load on intercept and slope factors. Forms include no growth, linear (loadings 0, 1, 2, 3), quadratic, latent basis (free loadings capturing any shape), exponential, logistic, sinusoidal and splines; unequal intervals change the loadings (for example 0, 1, 3, 7). Time-invariant covariates predict growth factors; time-varying covariates predict each wave. The approach uses all available data under a missing-at-random assumption and allows unequal timing, but needs at least two more occasions than the order of the highest random trend.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Siloed
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Weeks
Steps
- Plan occasions to match the expected shape: at least three for a linear random slope, more for curves, placed where change is expected to bend.
- Screen the data, plot individual trajectories, and code time meaningfully (for example months since sign-up, with zero at a substantive point).
- Fit a random-intercept model, then add a random slope, and compare nested models by likelihood-ratio or chi-square difference tests.
- Try latent basis or nonlinear forms where theory suggests a plateau or learning curve, and alternative residual covariance structures.
- Handle missing waves with full-information maximum likelihood or multiple imputation; do not bin unequal measurement times as if they were equal.
- Add person-level predictors of intercept and slope (segment, plan, prior experience) to form the conditional model.
- Plot model-predicted individual curves against observed ones to check the model represents individuals, and report how growth factors were scaled.
Sources
- Designing Experiments and Analyzing Data (Maxwell et al., 2018)
- Principles and Practice of SEM (Kline, 2023)
- Handbook of Child Psychology Vol 1 (Lerner et al., 2015)
- Mixed Models: Theory and Applications (Demidenko, 2004)
- Mixed Models: A Flexible Approach to Correlated Data (Seltman, n.d.)
- Nonlinear Regression (Seber & Wild, 1989)
- Applied Bayesian Modelling (Congdon, 2003)
- Intensive Longitudinal Methods (Bolger & Laurenceau, 2013)
- A Beginner's Guide to Structural Equation Modeling (Schumacker & Lomax, 2004)
- The SAGE Handbook of Quantitative Methods in Psychology (Millsap & Maydeu-Olivares, 2009)
- Bayesian Statistical Modelling (Congdon, 2006)
- Generalized Least Squares (Kariya & Kurata, 2004)
- Linear Statistical Inference and Its Applications (Rao, 1973)
Also known as
- Random-coefficient growth curve model
- Latent growth curve modelling
- Latent growth model
- LGM
- Latent trajectory model
- Random intercept and slope model
- Mixed-effects model for longitudinal data
- Growth curve modelling in SEM
- Nonlinear growth curve
- Sigmoid growth model
- Spline growth model
Categories
- Analytics & Behaviour
- Statistics & Modelling