Home/ Questions/Q 7055441
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-05-28T03:44:58+00:00

I’m using the mlogit package in R to estimate a mixed logit model with

  • 0

I’m using the mlogit package in R to estimate a mixed logit model with a log-normal parameter. The package runs fine, but is there a way to extact the random coefficients, particularly for non-normally distributed parameters?

Using an example from “Kenneth Train’s exercises using the mlogit package for R”, p 22,

library(mlogit)
data("Electricity", package = "mlogit")
Electr <- mlogit.data(Electricity, id = "id", choice = "choice", 
               varying = 3:26, shape = "wide", sep = "")

Electr$rev.tod <- -1*Electr$tod  # Reverse sign on tod parameter
Elec.mxl <- mlogit(choice ~ pf + cl + loc + wk + rev.tod + seas | 0, data=Electr, 
               rpar = c(cl = "n", loc = "n", wk = "u", rev.tod = "ln", seas = "n"),
               R = 100, halton = NA, print.level = 0, panel = TRUE)
summary(Elec.mxl)

Generates this (truncated) output:

Coefficients :
            Estimate Std. Error t-value  Pr(>|t|)    
pf         -0.866325   0.032452 -26.696 < 2.2e-16 ***
cl         -0.203770   0.013411 -15.194 < 2.2e-16 ***
loc         2.038715   0.079918  25.510 < 2.2e-16 ***
wk          1.481339   0.065181  22.727 < 2.2e-16 ***
rev.tod     2.105324   0.033971  61.973 < 2.2e-16 ***
seas       -8.490331   0.279262 -30.403 < 2.2e-16 ***
sd.cl       0.360140   0.017474  20.610 < 2.2e-16 ***
sd.loc      1.575765   0.089507  17.605 < 2.2e-16 ***
sd.wk       1.600303   0.122982  13.012 < 2.2e-16 ***
sd.rev.tod  0.390088   0.021940  17.780 < 2.2e-16 ***
sd.seas     1.997713   0.106031  18.841 < 2.2e-16 ***

random coefficients
              Min.    1st Qu.     Median       Mean     3rd Qu.     Max.
cl            -Inf -0.4466810 -0.2037701 -0.2037701  0.03914082      Inf
loc           -Inf  0.9758776  2.0387151  2.0387151  3.10155255      Inf
wk      -0.1189636  0.6811879  1.4813394  1.4813394  2.28149087 3.081642
rev.tod  0.0000000  6.3104924  8.2097637  8.8587752 10.68065958      Inf
seas          -Inf -9.8377681 -8.4903311 -8.4903311 -7.14289412      Inf

Is there any way to extract the mean of these random coefficients? This is mainly an issue for the non-normally distributed parameters. As pointed out by Train & Croissant, you can use the following code to calculate the mean coefficient for a log-normally distributed parameter, but I’m wondering if there is a more straightforward (and simpler!) approach.

-exp(coef(Elec.mxl)["rev.tod"]+(0.5*(coef(Elec.mxl)["sd.rev.tod"])**2))
Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    Actually … it generates no output. You must have also executed summary(Elec.mxl). The other problem besides not telling us the source of the output is that the mean of a sequence that includes -Inf is going to be either -Inf or undefined depending on the pleasures of the authors of the system. If you mean the column named “Mean” then (after looking at the structure of the summary object) this should do the trick:

         summary(Elec.mxl)$summary.rpar[ , "Mean"]
#        cl        loc         wk    rev.tod       seas 
#-0.2551692  1.9159597  1.6331273  8.9872786 -8.4093959

    You did say you wanted only the random coefficients, right? (Aren’t you a bit worried about a logit model that is producing coeffocients on the order of 8? at least in regular logit models it generally indicates a data situation that is creating pathological results.)

    • 0