Population averaged models are the models for kind of clustered data in which the marginal expectation of response variable is the main focus of interest. An alternative approach is taken in use to subject-specific models which concentrate on the modelling of the changes in an individual's response.

This is accomplished by the introducing subject-specific random effects into model. The mixed effects model or the multilevel model is an instance. There are two key points which differentiate the two types of model.

* The regression coefficients of the population averaged model define what the average population response looks like. By contrast regression coefficients of a subject-specific model define what the average individual's response curve looks like. In number of the cases and in particular when the model is linear in subject-specific effects, the two interpretations will coincide. In the more usual non-linear setting, though, the two approaches can lead to quite different conclusions.

* A additional distinction lies in specification of the underlying variance-covariance structure. In the population averaged models the marginal expectations are explicitly modelled while choosing the variance-covariance structure which adequately describes the correlation pattern between the repeated measurements. In subject-specific models, yet, individual heterogeneity is modelled using the subject-specific effects and it is these random effects which partially determine variance-covariance structure. c random effects into model. The mixed effects model or the multilevel model is an instance. There are two key points which differentiate the two types of model.