README.md

In xf-pan/dicmo: Discrete Choice Modeling

updated at 2019-12-13

What is dicmo?

dicmo is an R package for estimating various discrete choice models. In addition, it provides some useful functions to manipulate data before estimating the model. The current version of dicmo is 0.1.0.

What can dicmo do?

• data manipulation
• convert a date from a wide format to a long format
• generate the data with long format that is directly used in model estimation
• multinomial logit model estimation using maximum likelihood approach
• help to convert the data from a wide format to a long format automatically
• help to code the categorical attributes/variables automatically
• allow the existence of observations that some alternatives are not available in the choice set
• allow the existence of none-option in the choice set
• allow the existence of case that some alternative-specific attributes are not available for certain alternatives
• allow to estimate alternative-specific parameters for alternative-specific attributes, context varialbes ans ASCs
• allow to estimate interaction effect between alternative-specific attributes
• allow to estimate interaction effect of context variables with alternative-specific attributes and ASCs
• allow to compute the AVC matrix using numeric hessian matrix or BHHH algorithm
• allow to customize the initial values of the estimated parameters
• allow to fix the values of certain estimated parameters
• report various goodness of fit, i.e. initial log-likelihood, convergent log-likelihood, AIC, BIC, Rho-squared, and adjusted Rho-squared
• 2-level nested logit model estimation using maximum likelihood approach
• generalized ordered logit/probit model estimation using maximum likelihood approach
• exploded (2-level nested) logit model, including best-worst scaling case 3

How to install dicmo?

• first make sure the devtools package is installed and loaded: install.packages("devtools") library(devtools)
• then install dicmo from GitHub and load it: devtools::install_github("xf-pan/dicmo") library(dicmo)

Attentions

There are some rules have to be followed: The input data should be a tibble with a wide format. The column name of the input data could consist of English letters, numbers and underline - others are not accepted - and start with English letters. The name of alternatives could consist of English letters, numbers and under line, and it also could only have numbers as long as the numbers are treated as characters. Names of columns that indicating alternative-specific attributes (including the alternative-available column) should consiste of the names of alternatives and attributes (or the "avi" argument) - they are combined by a colon, for instance, "attribute:alternative" or "alternative:attribute" The artument "attrs" could have no slice "context" and "asc", that means they are all set as 0 - same function to set them as NULL. The order of the slice in the argument "attrs" does not matters. In the argument "attrs", negative number is not allowed. The elemets in the argument "attr_coding" and "attr_level" do not necessary have one-to-one correspondence in sequence. Highly recommend to always include the alternative-available column, i.e. the "avi" argument. Highly recommend to first estimate the model without setting anything for the argument "interact", "param_start" and "param_fixed" to know the names of estimated parameters, then set these arguments if it is needed. For ordered choice model, the argument "attrs" has no "asc" element. For ordered choice model, the elements in the argument "rate" must be sorted from the worst to the best.

xf-pan/dicmo documentation built on Jan. 25, 2021, 3:04 p.m.