One highlight of the new release is improved support for multinomial baseline logit models, i.e. models for categorical dependent variables with unordered categories (a.k.a. nominal-level dependent variables) with help of the function
mblogit(). Such models could already be fitted with
mclogit(), yet this required some additional work in data-management and model construction.
mblogit() function allows to fit models with categorical dependent variables right away. Its output will be numerically identical with the output of
multinom from the package nnet, but the results are reported in more “conventional” way similar to to the results of
glm() and the like.
At present, there is no support for random effects in multinomial baseline logit models. These are going to be added in the next release (0.5), along with random slopes in multinomial conditional logit models.
CHANGES SINCE 0.3
mblogitto fit multinomial baseline logit models.
mclogitobjects, so that
drop1.defaultshould work with these.
mclogit.fitnow allow user-provided starting values.
getSummarymethods now return “contrasts” and “xlevels” components.
Fixed prediction method for
Corrected handling of weights and standard errors of prediction.
Matrices returned by the
vcov()have row and column names.
mclogit.fit.rePQLare exported to enable their use by other packages.