While this approach is quite feasible for the Gaussian mixed model, the computations are intractable in other cases. Breslow and Clayton (1993) review approaches to approximate these calculations. There are two main approaches and both reduce to reweighted versions of the calculations for the Gaussian linear mixed model. Although these perform quite well in a wide range of circumstances, they are generally quite laborious since they typically involve, at each iteration, inversion of a square matrix of size equal to the total number of /3 and b parameters. When the random effects have a nested, or 'multilevel' structure, inversion partitioning can dramatically reduce the computational burden, but there are many useful models for which this is not possible.