2

I was trying to estimate the mixed-effects logit in R using mlogit package. I would be grateful if someone could point out where is my mistake.

My original data has a panel structure - each individual makes choice in 6 different scenarios. Each time he has 4 options (3 brands and non getting anything). That is, there are 24 rows for each individual in the dataset.

I have few individual specific demographics. There are also few alternative-specific characteristics: one is price (it is different for each alternative in each scenario), others are just dummy variables that are alternative specific. These dummies indicate specific brands of the same product. The coefficient for each dummy varies according to the demographic characteristics.

I have read the Yves Croissant's pdf on the estimation of multinomial logit models in R, and was able to adapt the data to the format suitable for mlogit estimation. 

data.logit = mlogit.data(data, shape = "long", 
                         alt.var = "alt_id",
                         choice = "pick",
                         id.var = "idnum",
                         sep = "")

me.logit = mlogit (pick ~ price | gender + age + educ + income, 
                   data.logit,
                   reflevel = 4,
                   rpar = c("brand1:(intercept)" = "n",
                            "brand2:(intercept)" = "n",
                            "brand3:(intercept)" = "n"),
                   R = 500,
                   halton = NA, 
                   panel = TRUE)
summary(me.logit)

The results I get do not make sense to me. For two brands the alternative specific constants (coefficients for the dummies) are negative, the coefficient for the price is, obviously, negative too. Then the WTP for those brands becomes negative which is definitely wrong.

I think that there may be a mistake in the way I assigned parameters when using mlogit.data.

Fujifilm
  • 21
  • 1
  • 3
  • Looks more like a statistical interpretation difficulty, more appropriate for CV.com. If a coding solution is needed then `data` was not provided. – IRTFM Oct 14 '16 at 04:33
  • @42- Thank you for your response. I have updated my post with the link to the data on pastebin.com – Fujifilm Oct 14 '16 at 05:01
  • In case anyone encounters the error in loading `mlogit` due to the error message that included `Symbol not found: ___addtf3`, it was resolved after updating the required stamod package to version 1.26. I think it is possible to include such requirements in the package DESCRIPTION file and will suggest that the mlogit's maintainer. There was also a discrepancy in column names between the formula and the data. Need to change `income` to `inc`. – IRTFM Oct 14 '16 at 16:17

0 Answers0