I try to migrate my models that use the "mlogit" package to the "mnlogit" package. The latest seems to be faster as it uses the multi-core capacities of a computer during estimation. It is also sometimes considered as more robust.

The mlogit package offers the possibility to pass the "weights" of the observations as a parameter. These weights can be a column of the data frame that contains the explanatory variables. The data set in in "long" format, and so are the "weights".

If I do this with the mnlogit package, it complains that the "weights" don't have the same length than the data. Knowing that, in the "long" format, the weights (and other data) are replicated (one row for each alternative), I tried to pass the "weights" as they are stored in the "wide" dataset I used to generate the "long" format data set with mlogit.data() method.

The model runs, but I'm not sure that my interpretation is right as my estimators are slightly different from those I obtained with mlogit.