Technically the multivariate analogue of the quasi-likelihood with the same feature that it leads to consistent inferences about the mean responses without needing specific suppositions to be made about second and higher order moments. Most frequently used for the likelihood-based inference on longitudinal data where the response variable cannot be supposed to be normally distributed. Easy models are used for within-subject correlation and a working correlation matrix is introduced into the model specification to accommodate these correlations. The process gives consistent estimates for the mean parameters even if the covariance structure is incorrectly specified.

The technique assumes that the missing data are missing completely at the random; otherwise the resulting parameter estimates are biased. The amended approach, weighted generalized estimating equations, is available which produces the unbiased parameter estimates under the less stringent assumption that the missing data are missing at random.