Software for discrete choice model estimation

From DDL Wiki

Jump to: navigation, search

Most commercial statistics software packages offer build-in logit model estimation functions. Some academic researchers in economics and marketing science field provide their source codes for academic use.

Contents

Excel

Excel is the popular spreadsheet software. With a good understanding of choice model structure, the discrete choice results can be transformed into a maximum likelihood problem (log-likelihood), which is solvable by Excel's build-in solver. However, such approach is not suitable for logit models with random coefficients since the construction of random draws for simulation is tedious in spreadsheets.

  • Prof. Jeremy Michalek's website provides his logit regression spreadsheet for the model in his paper "Linking marketing and engineering product design decisions via analytical target cascading," Journal of Product Innovation Management, 2005, v22 p42-62.

SAS/MDC

SAS is a powerful software package for statistical application. The MDC (Multinomial Discrete Choice) module is capable to perform choice model regression for various discrete models, such as conditional logit, heteroscedastic extreme value, mixed logit, nested logit, and multinomial probit models. However, beginners without programming experience will spend more time to become familiar with the interface and be able to create codes.

Stata

Stata is a data analysis and statistical package that provides everything you need for data analysis, data management, and graphics. It is also capable of logit model regression.

LIMDEP/NLOGIT

NLOGIT is an extension program of commercial LIMDEP statistical software. It provides the functions for estimation, model simulation and analysis of multinomial choice data. According to the company website, the latest version of NLOGIT is able to handle heterogeneity in variances of utility functions and mixed logit model.

Sawtooth

Sawtooth is a specialized in marketing research software for discrete choice model and conjoint analysis. It is able to generate conjoint survey questionnaire (on-line or print-out) through a friendly user interface. Sawtooh can measure the perceived values of specific product features, learn demand of a particular product, and forecast market acceptance.

R: bayesm

bayesm is a package written in R language by Peter Rossi and Rob McCulloch to illuminate ideas presented in their textbook, Bayesian Statistics and Marketing. The package can estimate a number of choice models including Multinomial Probit (MNP), Multinomial Logit (MNL) and Hierarchical Multinomial Logit (MNL) with Mixtures of Normals heterogeneity. In addition to choice models, the package also includes functions for estiamating a number of models popular in marketing such as Linear Regression, Seemingly Unrelated Regression (SUR), Multivariate Probit (MVP), Negative Binomial (Poisson) Regression and Multivariate Mixtures of Normals.

The estimation is all done by Bayesian MCMC methods and the functions generally produce a set of draws from the posterior distribution of the parameters. Users must be familar with Bayesian MCMC methods and know the proper methods for analyzing the posterior draws (e.g., convergence diagnostics). The documentation relies heavily on the textbook, often using the same notation as presented in the book. Users of bayesm are likely to find that they need the textbook as a reference in addition to the R documentation.

R: mlogit

mlogit is a package for R which enables the estimation of the multinomial logit models with individual and/or alternative speci�c variables. The main extensions of the basic multinomial model (heteroscedastic, nested and random parameter models) are implemented. It is capable of including mixing distributions to model heterogeneity in a mixed logit model, similar to Ken Train's Matlab code.

Kenneth Train's Matlab Code

Prof. Ken Train offers his Matlab codes using both maximum likelihood estimation (MLE) and Bayesian approach for mixed logit model. He also provided the old Gauss codes using MLE for mixed logit estimation. Train's Gauss code has been modified and applied to a study about multiparty elections by Prof. Glasgow Garrett, where his code and data are provided.

Lenk's Code

Prof. Peter Lenk offers his Gauss code using Bayesian methods for discrete choice model estimation.

Biogeme

A free software package provided by Prof. Michael Bierlaire using the maximum likelihood estimation for Generalized Extreme Value (GEV) models. It can be used for Multinomial Logit models, Nested logit models and other types of GEV models.

DCM Package

DCM stands for Discrete Choice Model. It is free software package written in Ox (a substitute called OxMetrics has trial version available) which is a commercial statistics programming language. It is provided by Matias Eklof at Uppsala University and Melvyn Weeks at University of Cambridge. However, the package has not been updated since August 2005.


Software Type User Interface Design Estimation Prediction Scalability Hierarchical Cost
Excel spreadsheet graphical no yes maybe poor no $
SAS/MDC statistics package code yes yes excellent no $
Stata statistics package code no yes no no $
SPSS statistics package graphical no yes no no $
LIMDEP/NLOGIT statistics package graphical yes yes yes no $$$
Sawtooth conjoint package graphical yes yes yes excellent available $$$
R: bayesm statistics package code no yes no poor yes free
R: mlogit statistics package code no yes no poor no free
Ken Train's Code research code code no yes no yes Free - Require Matlab or Gauss
Peter Lenk's Code research code code yes free - Require GAUSS
Biogeme research program input file no yes no free
DCM package research program code no yes no free - Require Ox or OxMetrics4

External Links

Retrieved from "http://wiki.ece.cmu.edu/ddl/index.php/Software_for_discrete_choice_model_estimation"
Personal tools