CODEWITHSUNDEEP
rcoefficientsmlogit
Or Zuckerman
Or Zuckerman

Reputation: 31

mlogit in R - coefficients and unknown random parameter

I'm trying to run mlogit on my data. My data is:

> head(df[c(1:3, 33:45)])
      ID Gender Age Option equipment.A equipment.B equipment.C clean.A clean.B clean.C people.A people.B people.C price.A price.B
1 108630      M  56      A        3        3        1      1      3      3         2         3         2     4     3
2 115547      F  34      B        3        3        1      1      3      3         2         3         2     4     3
3 118359      F  51      C        3        3        1      1      3      3         2         3         2     4     3
4 126656      F  40      C        3        3        1      1      3      3         2         3         2     4     3
5 127439      F  26      C        3        3        1      1      3      3         2         3         2     4     3
6 130846      M  69      C        3        3        1      1      3      3         2         3         2     4     3
  price.C
1     4
2     4
3     4
4     4
5     4
6     4

My code is:

Result <- mlogit.data(mydata, shape = "wide", choice = "Option", varying = 34:45)
Result.mxl <- mlogit(Option ~ equipment + clean + people + price | 0, Result)
summary(Result.mxl)

Then, my output is:

Frequencies of alternatives:
     A      B      C 
0.3745 0.3525 0.2730 

nr method
4 iterations, 0h:0m:0s 
g'(-H)^-1g = 3.12E-07 
gradient close to zero 

Coefficients :
         Estimate Std. Error z-value  Pr(>|z|)    
equipment2  0.467333   0.100203  4.6639 3.103e-06 ***
equipment3  0.431747   0.089822  4.8067 1.535e-06 ***
clean2    1.278133   0.094188 13.5701 < 2.2e-16 ***
clean3    1.603504   0.118145 13.5724 < 2.2e-16 ***
people2 0.438747   0.094280  4.6537 3.261e-06 ***
people3 0.292673   0.115142  2.5419   0.01103 *  
price2     0.641272   0.111956  5.7279 1.017e-08 ***
price3     0.786393   0.100451  7.8286 4.885e-15 ***
price4     1.045555   0.151903  6.8830 5.859e-12 ***
price5     1.965825   0.211786  9.2821 < 2.2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Log-Likelihood: -2009.3

So the problem is that I want to get a table with just one observation of 'equipment', 'clean', 'people' and 'price', as well as table of random coefficients. (something like this example: https://cran.r-project.org/web/packages/mlogit/vignettes/e3mxlogit.html)

I tried to add another parameters to my code and run:

Result.mxl <- mlogit(Option ~ equipment + clean + people + price | 0, Result, 
                   rpar = c(equipment = 'n', clean = 'n', people = 'n', price = 'n'), 
                   panel = TRUE, id.var = "id", alt.levels = c("A","B", "C"))

but now I get an error: 

Error in mlogit.start(formula = formula, data = data, mf = mf, start = start,  : 
  unknown random parameter

I would appreciate very much your help!

Upvotes: 1

Views: 937

Answers (1)

Michiel
Michiel

Reputation: 1

If you use factor variables, mlogit apparently first makes new variables for each factor level with new names. It then works with these names and therefore states 'unknown random variable' if you use the original names. You can figure out what the factor names are by running a normal multinomial logit. Then use coef(summary()) to see what the names of the new variables (one per factor level are) and use these to tell mlogit which variables it should use random coefficients for.

Upvotes: 0

Related Questions