Tags: melff/mclogit
Tags
Better support of starting values and more stable PQL/MQL algorithm `mclogit()` and `mblogit()` now have an infrastructure that allows to provide starting values for models with random effects. Starting values should be given for the coefficents and for the variance parameters. If the PQL method is used, also starting values for the random effects have to be provided. The PQL/MQL algorithm is made more stable by re-using in for the inner iteration the intermediate estimates of coefficients, variance parameters, and random parameters from previous outer iterations.
PreviousNext