William Greene
Stern School of Business
New
York
University
Discrete Choice
Modeling
Professor
Professor William. Greene
e-mail: wgreene@stern.nyu.edu
Professor’s Home Page: http://www.stern.nyu.edu/~wgreene
Abstract
This course will survey
techniques used in modeling discrete data. Discrete choice models have become
an essential tool in modeling individual behavior. The techniques are used in
all social sciences, health economics, medical research, marketing research,
transport research, and in a constellation of other disciplines. This course
will examine a large number of models and techniques used in these studies. We
will begin with a brief review of regression modeling concepts, then turn to
the fundamental building block in discrete choice modeling, the binary choice
model. Several variants and extensions will be discussed before we turn
attention to multiple equation binary choice models, ordered choice models and
models for counts. The second half of the course will be devoted to multinomial
choice models of the sort used, e.g., in modeling brand choice in marketing,
travel mode choice in transport, and a huge variety of applications in the
social and behavioral sciences.
The course will include lectures
that develop the relevant theory and extensive practical, laboratory
applications. Emphasis in the laboratory sessions will be on estimation of
discrete choice models and using them to describe behavior and to predict
discrete outcomes. Course participants will apply the techniques on their own
computers using the NLOGIT computer
program and several ‘real’ data sets that have been used in applications
already in the literature. NLOGIT is
the leading computer program for this type of estimation, so students will have
also studied the applications of the techniques using the modeling tools
familiar to researchers in the area.
Prerequisites
Prior knowledge is assumed to
include calculus at the level assumed in the first year of a Ph.D. program in
economics and a course in econometrics at the beginning Ph.D. level out of a
textbook such as Greene, W., Econometric
Analysis, 6th edition. Familiarity with NLOGIT will be helpful, but is not necessary. Another useful
reference book for the course is the primer Applied
Choice Analysis by David Hensher, John Rose and William Greene (Cambridge
University Press, 2005).
Students in this course will
obtain background in both the theory and methods of estimation for discrete
choice modeling. This course will provide a gateway to the professional
literature as well as practical application of the methods at the level of the
contemporary research in the field.
Students are assumed to have had a
previous course in Econometrics at the graduate level, using a text at the
level of Greene, W., Econometric
Analysis.
Course Outline
This is a course
in econometric analysis of discrete data. There are a huge variety of models
that are used in this context. We will focus on four which arguably comprise the
foundation for the area: the fundamental model of binary choice (and a number
of variants); models for ordered choices; the Poisson regression model for
count data; and the fundamental model for multinomial choice, the multinomial
logit model. The course will consist of discussions and laboratory sessions
which will apply the techniques to ‘live’ data sets and some time devoted to
topics, discussions and laboratory work on student projects. Discussions will
cover the topics listed below. Lab sessions will apply the techniques discussed
in the preceding sessions. Practicals will consist of directed exercises and
student assignments to be completed singly or in groups.
No specific textbook is assigned
for the course. A useful reference is
Applied Choice Analysis by Hensher, D., Rose, J. and Greene, W., (Cambridge University Press,2005).
Some of the presentation will be
based on Econometric Analysis, 6th
ed., by Greene, W. (Prentice Hall, 2003). 6 chapters are included with the
course materials: Left click to
activate. Right click to download
Chapter 9: Estimation Methods Greene-Chapter-9.pdf (Panel data methods)
Chapter 14: Estimation Methods Greene-Chapter-14.pdf (Estimation methods)
Chapter 16: Maximum Likelihood Greene-Chapter-16.pdf (Maximum likelihood estimation)
Chapter 17: Simulation Based Estimation and Inference Greene-Chapter-17.pdf (Simulation based estimation)
Chapter 23: Models for Discrete Data Greene-Chapter-23.pdf (Discrete choice models)
Chapter 25: Models for Count Data and Duration Greene-Chapter-25.pdf (Count data models)
The received literature on
discrete choice models is vast – one could easily compose a list of thousands
of articles. Your course materials include a small handful of articles, mainly
for the purpose of illustrating particular techniques or ideas. (Working papers
are attached. Most of these have been published elsewhere.) These are:
Economic
Choices, American Economic Review,
McFadden, D. (2001). McFadden’s Nobel Prize lecture.
Mixed
MNL Models for Discrete Response, McFadden, D. and Train, K., Journal of Applied Econometrics, 2000.
The
Behaviour of the Maximum Likelihood Estimator of Limited Dependent Variable
Model in the Presence of Fixed Effects, Greene, W., Econometric Journal, 2004.
Discrete
Choice Models, W. Greene, (a survey of discrete choice models) forthcoming, Palgrave Handbook of Applied Econometrics,
2009.
Functional Form and Heterogeneity
in Models for Count Data, W. Greene, (a survey of models for count data); Foundations and Trends in Econometrics,
2007..
Modeling Ordered Choices, (W.
Greene, 2008) A survey of the literature on ordered choice models. Published as
Modeling Ordered Choices, Greene, W.
and Hensher, D., Cambridge University Press, 2010.
Session
_________________________________________________________________________________________________________
Introduction: Course
description and overview
1: Methodology, software,
modeling concepts, regression basics
2: Standard models for
binary choice
3: Analysis of binary
choice. marginal effects, fit measures, prediction, hypothesis tests
4: Panel data models for
binary choice, random effects, fixed effects, Mundlak formulation, incidental
parameters problem, dynamic probit model
5: Bivariate probit models,
simultaneous equations, sample selection, multivariate probit model marginal
effects, prediction and analysis of bivariate choices
6: Panel data,
heterogeneity, simulation and latent class models,
7: Ordered choice models,
ordered outcomes, estimation and inference, generalized models, recent developments,
models for counts
8: Models for
count data
9: The multinomial logit
model, random utility models, IIA, logit modeling. fit and prediction, marginal
effects, model simulation
10: Extensions of the MNL
model, multinomial probit, heteroscedasticity, model simulation, use of the MNL
model
11: Nested logit model and
generalized nested logit models,
12: Modeling heterogeneity of preferences, simulation
based estimation, error components
13. Latent class models
14. Mixed logit and random parameter
models, simulation, generalizations of the mixed logit model
15: Repeated observations,
panel data, revealed vs. stated preference data
End:
Observations and closing remarks
Course Materials and Resources
I.
Course Timetable
II. Readings: Reading materials. These
are all PDF files
Articles about discrete choice modeling: Left click to activate. Right click to
download.
CountDataSurvey.pdf
DiscreteChoiceSurvey.pdf
OrderedChoiceSurvey.pdf
FixedEffects.pdf
McFadden-EconomicChoices.pdf
McFadden-Train.pdf
III. Assignments: Exercises
and Practicals for Discrete Choice Modeling. Left click to activate. Right click to download. Note there
are two parts to each, the .pdf file for the assignment and the .lim file that
contains the NLOGIT commands.
Assignment 1: Basic Regression (NLOGIT Commands for Assignment 1)
Assignment 2: Binary Choice Modeling (NLOGIT Commands for Assignment 2)
Assignment 3: Binary Choice Modeling with Panel Data (NLOGIT Commands for Assignment 3)
Assignment 4: Bivariate Probit and Extensions (NLOGIT Commands for Assignment 4)
Assignment 5: Heterogeneity in Binary Choice Models (NLOGIT Commands for Assignment 5)
Assignment 6: Ordered Choice Models (NLOGIT Commands for Assignment 6)
Assignment 7: Models for Count Data (NLOGIT Commands for Assignment 7)
Assignment 8: Multinomial Logit Model (NLOGIT Commands for Assignment 8)
Assignment 9: Latent Class and Random Parameters (NLOGIT Commands for Assignment 9)
Assignment 10: Revealed and Stated Preference Data (NLOGIT Commands for Assignment 10)
Assignment 11: Student Project: Model
for Moral Hazard
IV. Class Notes: These are
Powerpoint slide presentations for use during the class sessions. For
convenience, they are also stored in PDF format. Left click to activate. Right click to download.
Part 0: Introduction (PDF
format)
Part 1: Methodology (PDF
format)
Part 2: Binary Choice Estimation (PDF
format)
Part 3: Binary Choice Inference (PDF
format)
Part 4: Panel Data Models for Binary Choice (PDF
format)
Part 5: Bivariate and Multivariate Probit
(PDF
format)
Part 6: Modeling Heterogeneity (PDF
format)
Part 7: Ordered Choices (PDF
format)
Part 8: Count Data (PDF
format)
Part 9: Multinomial Logit (PDF
format)
Part 10: Multinomial Logit Extensions (PDF
format)
Part 11: Nested Logit Model (PDF
format)
Part 12: Modeling Heterogeneity in Multinomial Choice Models (PDF
format)
Part 13: Latent Class Models (PDF
format)
Part 14: Mixed Logit Models (PDF
format)
Part 15: Stated Preference Data (PDF
format)
V. Lab Notes: These are
Powerpoint slide presentations that explain using NLOGIT and how to do the
assignments with NLOGIT. Left click
to activate. Right click to download.
Lab 1: Getting Started
Lab 2: Binary Choice
Lab 3: Binary Choice with Panel Data
Lab 4: Bivariate Probit
Lab 5: Models with Heterogeneity
Lab 6: Ordered Choice Models
Lab 7: Models for Count Data
Lab 8: Multinomial Logit
Lab 9: Multinomial Probit and MNL Extensions
Lab 10: Revealed and Stated Preference
VI. Data Sets: These are
NLOGIT project (.lpj) files. Left
click to activate. Right click to download.
American Express Credit Data
Spector and Mazzeo Basic Probit Data
Shoe Brand Choices, Stated Preference, Multinomial
Choice
Multinomial Choice, Travel Mode
Dairy Farms, Panel Data
Health Care Panel Data
Labor Supply Data (Mroz Study)
Panel Data on Innovation for Probit Models
Stated Preference Data on Automobile Choices
VII. NLOGIT Software: This
section contains a brief introduction and two manuals: The short introduction
is a “getting started.” The LIMDEP manual explains the basics of using LIMDEP
and NLOGIT. (LIMDEP is embedded in NLOGIT). The NLOGIT manual contains
descriptions of how to use the special features for discrete choice modeling
with NLOGIT (as well as some additonal material on other discrete choice models
that are also contained in LIMDEP). The setup file contains an installation kit
for installing a copy of NLOGIT made specially for this course on your own
computer. You should download the setup file to your own computer and execute
it there, rather than launching it from your web browser. Left click to activate. Right click to download.
Short Introduction to NLOGIT
LIMDEP Student User’s Manual
NLOGIT Student User’s Manual
NLOGIT Software Setup for Installing NLOGIT
on Your Computer