#===========================================================================================
Mauerer, I. & Tutz, G. (2024). "Replication Data for: Heterogeneity in General Multinomial Choice Models",Harvard Dataverse, V1.  https://doi.org/10.7910/DVN/UL0RHQ

Content: 

- The file Replication.R contains the commands to run the models and create the tables 
- The file GHMNL.R contains the R function to perform the model


#-----------------------------------------------------------------------------------

Related Publication: 

Mauerer, I. & Tutz, G. (2023). "Heterogeneity in General Multinomial Choice Models." Statistical Methods & Applications 32: 129–148. doi:10.1007/s10260-022-00642-5

#===========================================================================================


The R function fits the General Heterogeneous Multinomial Logit Model (GHMNL), which can handle two types of covariates: global (chooser-specific) and category-specific (choice-specific) variables.


#=============================================================================================================
# GHMNL(dat, namesglob, Indglob, namescats, Indcats, nameshet, nameresp, k, penalty)
#
# - dat:        data set in long format (k rows for each observation) 
#               the response is indicated by a 1 in the corresponding row, 0 in all other rows
#
# - namesglob:  names of global variables (chooser-specific attributes)
# - Indglob >0: global variables are included, otherwise ignored
# 
# - namescats:  names of category-specific  variables (choice-specific attributes),
#               if no category-specific variables are in the data set, choose any variable and set Indcats=0
# - Indcats >0: category-specific variables are included, otherwise ignored
#
# - nameshet:   names of variables in the heterogeneity term (can be empty)
#               if nameshet <- c()  no heterogeneity term in the model
#
# - namesresp:  name of response variable (vector of 0-1 dummy variables, based on nominal response)
# - k:          number of categories in the response
#
# - pen:        penalty on estimates, choose small, pen = 0.0000001 or even pen=0 (unpenalized) 
#               number >= 0, if penalty=0 pure maximum likelihood
#               if penalty>0 ridge penalized maximum likelihood 
#==============================================================================================================