Latent Growth Curve Model (LGC)

Both approaches model individual growth trajectories and are mathematically equivalent under many parameterisations. The LGC model is formulated within the SEM framework, which makes it straightforward to add latent predictors, multiple parallel growth processes, or measurement error correction. Mixed-effects models are typically preferred when the number of time points varies across individuals or when the data structure is highly unbalanced, because they handle such designs more flexibly.

At least three waves are required to estimate both the intercept variance and the slope variance and their covariance. With only two waves the model is equivalent to a difference score and the variance of the slope cannot be separated from measurement error. Four or more waves are strongly preferred when a quadratic or other nonlinear growth form is hypothesised.

Full-information maximum likelihood (FIML) is the recommended approach. It uses all available data from each individual at each wave without imputation or listwise deletion, producing unbiased estimates under the assumption that data are missing at random. Most SEM software implements FIML as a standard option.

Yes — this is one of the key strengths of the LGC framework. Time-invariant covariates (such as sex or experimental condition) are regressed on the intercept and slope latent factors in the structural part of the model. This allows you to test whether a predictor explains why individuals differ in their starting level, their rate of change, or both.