0

I would like to jointly estimate 3 variables. Two of them are categorical and the other one is binary. So I thought about a "multivariate multinomial logit model". I found a lot of theory about it (for Example Agresti 2007 Ch. 9 or Beel and Paap 2014) but I cannot find a package for R. Is there a built-in function or package I can use? I can switch to a bivariate multinomial logit if it is needed.

Thank you very much for your help in this matter!

PrintOnFire
  • 13
  • 3
  • adding some code and more flags would always help more clarity. – narcis dpr Sep 24 '20 at 15:08
  • I am sorry, I do not have any code yet because I will have the data in October. The problem is that I need to know if it is possible do such thing with R and if there is some package to do this. This is likely to determine the background of my research. I only need to know if there is a package in R to do a multivariate multinomial logit (clustered multinomial in Agresti 2007 chap. 9) – PrintOnFire Sep 25 '20 at 16:04

1 Answers1

1

There are several packages that might interest you for a multinomial logit model. They are mlogit, mnlogit, antitrust, and nnet.

mlogit: This is the most direct Multinomial Logit package currently available. It provides sample data, tools to estimate a multinomial logit model, and additional useful functions such as mlogit.optim to optimize specific parameters of multinomial logit functions.

mnlogit: This package is similar to mlogit, but it does not provide as many additional functions. It may be faster for the actual estimation process though.

antitrust: This package can estimate merger effects under logit (or nested logit) demand. This does not directly provide the multinomial logit coefficients, but it is very good at solving for the bottom line HHI and price effects of a merger.

nnet: This is a package for general multinomial log-linear models, and it can also estimate multinomial logit models.

Hope one of these packages helps for your purposes!

Matthew G
  • 91
  • 2