As an important practical matter, though, mixed effects models are much more computationally intensive than the marginal models and are more likely to have problems with convergence; i.e., mixed effects models require more extensive data. For this reason, marginal models may be seen as having an advantage over mixed effects models in many clinical trials and epidemiologic studies, or whenever data are limited. As mentioned, though, certain types of analysis require mixed effects models. In general, we recommend a biostatistician be involved when implementing and interpreting results from mixed effects models.
