0
$\begingroup$

On page 13 of this document, it indicated the estimated log-sum coefficient. How can I obtain the result with R?

The R output I think is relevant for this answer:

##
## Call:
## mlogit(formula = depvar ~ ich + och + icca + occa + inc.room +
## inc.cooling + int.cooling | 0, data = HC, nests = list(cooling = c("gcc",
## "ecc", "erc", "hpc"), other = c("gc", "ec", "er")), un.nest.el = TRUE)
##
## Frequencies of alternatives:
## ec ecc er erc gc gcc hpc
## 0.004 0.016 0.032 0.004 0.096 0.744 0.104
##
## bfgs method
## 11 iterations, 0h:0m:0s
## g'(-H)^-1g = 7.26E-06
## successive function values within tolerance limits
##
## Coefficients :
## Estimate Std. Error z-value Pr(>|z|)
## ich -0.554878 0.144205 -3.8478 0.0001192 ***
## och -0.857886 0.255313 -3.3601 0.0007791 ***
## icca -0.225079 0.144423 -1.5585 0.1191212
## occa -1.089458 1.219821 -0.8931 0.3717882
## inc.room -0.378971 0.099631 -3.8038 0.0001425 ***
## inc.cooling 0.249575 0.059213 4.2149 2.499e-05 ***
## int.cooling -6.000415 5.562423 -1.0787 0.2807030
## iv 0.585922 0.179708 3.2604 0.0011125 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Log-Likelihood: -178.12
$\endgroup$
0
$\begingroup$

It looks like the log-sum coefficient is generated automatically by mlogit and output as 'iv'. If you change un.nest.el to 'FALSE' it doesn't assume unique elasticity and generates separate log-sum coefficients for each nest ('iv:cooling' and 'iv:other').

$\endgroup$
  • $\begingroup$ Why does it say in the document that the log-sum coefficient is -0.59 while the iv coefficient is positive? $\endgroup$ – Aqqqq Oct 17 '18 at 4:26
  • $\begingroup$ Typo. If you enter the code in the paper to run the t-test - (coef(nl)['iv'] - 1) / sqrt(vcov(nl)['iv', 'iv']), which calls the 'iv' variable - it gives the same -2.304171 reported in the paper. $\endgroup$ – cgrafe Oct 17 '18 at 14:21
  • $\begingroup$ So the log-sum coefficient should be 0.59 here? If I obtain a negative log-sum coefficient, how should I calculate the correlation? (It does not make sense to have a correlation bigger than 1, i.e. 1-(negative log-sum coefficient).) $\endgroup$ – Aqqqq Oct 17 '18 at 18:07
  • $\begingroup$ A little farther down it says "The log-sum coefficient is over 1" where the inclusive value ('iv') is positive (1.36201), so the paper is inconsistent. Here's another example of a similar analysis: help.statwizards.com/data-wizard/statistics_programs/r.htm. They just used 'iv'. $\endgroup$ – cgrafe Oct 17 '18 at 19:36
  • $\begingroup$ Your link does not say anything about correlation. Do you know how to calculate correlation from iv correctly? $\endgroup$ – Aqqqq Oct 18 '18 at 5:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.