How is This Different from a Trajectory Analysis, Latent Growth Curve, Mixture Model, or Generalized Estimating Equation?
A key difference between RMLCA and other approaches is that it is not fitting a particular function to the data, so the trend over time can take any shape that fits the data, no matter how complex. Unlike variations of the generalized linear model, this approach is not attempting to fit the data to a linear model. RMLCA does not generate slope estimates or characterize variance across subjects in these slopes over time. The emphasis in RMLCA is not on the variable (even time), but is on the person (what are the patterns over time that emerge across persons). The approach is also not subject to the same assumptions as the general linear model (e.g., normality) or the generalized linear model. RMLCA makes some novel assumptions, however, such as the assumption of conditional independence (the idea that latent class fully accounts for item response probabilities within each class). RMLCA also requires that the indicators be categorical.