1

I'm having some trouble with an analyses I want to conduct and were wondering if there are any alternatives. Essentially I have a longitudinal dataset (or panel data), and a dependent variable (group membership) that is categorical and time invariant. For example:

ID Time Group Sex Age X1 X2 X3
1 1 A M 23 4 6 2
1 2 A M 23 2 6 1
1 3 A M 23 3 5 4
2 1 B F 31 5 6 3
2 2 B F 31 5 7 2
2 3 B F 31 1 3 1
3 1 C M 27 4 2 4
3 2 C M 27 5 3 4
3 3 C M 27 2 3 3

So we have both time invariant variables (Sex, Age) and time variant variables (X1, X2, X3), with our DV (Group) being constant within-person. Typically, I believe the correct analysis here would be a mixed multinomial logit regression, however I haven't found a way to run a model like that in R after attempting numerous packages.

I was wondering if there were other types of analyses that I could use to model Group membership. I have explored other analyses such as longitudinal discriminant analysis, but they tend to use binary outcomes, rather than multinomial. I have also looked at packages like "multgee" but the outcome variable in those packages is measured at every time point (whereas Group here is static across time).

Is there a package related to mixed multinomial models that I've missed, or another package I can use to account for this data structure/outcome variable? Any help here would be appreciated.

Dhawal Selarka
  • 63
  • 1
  • 1
  • 5

1 Answers1

0

For this type of data, I think that a multinomial logit model would be appropriate. The coefficients in this case would represent the log-odds that an independent variable is associated with a change in an individual's group.

Regarding which packages you might consider, I'll point you to another answer here. Seems like mlogit will be your best bet for general purpose multinomial logit modeling, although nnet might be preferable if you prefer a mixed log-linear model.

Dharman
  • 30,962
  • 25
  • 85
  • 135
Matthew G
  • 91
  • 2