 |
 |
 |
 |
 |
 |
|
Estimation of Discrete Choice Models using Ox
15 - 16 December 2003,
The Royal Statistical Society, 12 Errol Street, London EC1Y 8LX, U.K.
Contents
Course Description
Course Programme
Request an Enrolment Form Now
Terms and Conditions
About OxMetrics
About DCM
For many years applied workers working in the area of discrete choice have been fully versed in the trade-off between the tractable, yet restrictive, logit model, and the flexible, yet difficult to operationalise, probit model. Thus, despite multinomial probit offering the potential of improving upon the restrictions on choice behaviour imposed by the logit model - namely restrictive substitutions patterns, enforced iid over individuals faced with multiple choices over time, and a neglect of random preference heterogeneity (RPH) over a population - the logit model and other Generalised Extreme Value (GEV) variants have, in general, dominated much of the literature. Following a long lineage of earlier work (from Thurstone (1927), Hausman and Wise (1978), Manski and Lerman (1981), independent work by McFadden (1989) and Pakes and Pollard (1989), resulting in the emergence of a class of simulation-based estimators (alongside the continued growth in computer power) the well known potential of probit models has begun to offer a viable alternative to logit models. However, it might be said that the mixed logit model - in essence a weighted average of logit choice probabilities evaluated over a distribution of parameter values offers a more viable extension of vanilla logit. As with the probit model, mixed logit is capable of circumventing the fundamental limitations of vanilla logit. However, once the RPH is integrated out, a standard logit choice probability remains, with attendant tractability.
This course provides an overview of these developments in the discrete choice literature, and by introducing a new software package written in Ox, provides hands-on instruction on the issues involved in the specification, identification and estimation of this ever expanding class of models.
Objectives of The Course
The objectives of this course are to provide participants with both an introduction to the theory and application of binary and multiple response models, and an overview of recent developments in discrete choice modelling. The distinction between single and multiple index models is also made. This introduction progresses from the standard two choice logit and probit model, to models which in various ways facilitate the representation of a wide range of choice behaviour. The class of models include nested logit, mixed (random coefficient) logit, the multinomial probit model, and ordered (random coefficient) probit.
As a natural complement to the material on both the mixed logit and multinomial probit model, we also provide an introduction to the use of simulation as a tool to estimate choice probabilities. This material will include both a brief overview of the econometric theory but also a hands-on introduction to the use of simulation methods in applied work. We emphasise that although throughout the course an exposition of the underlying econometric issues will be provided, the emphasis will be upon applied issues. To this end, participants are also introduced to a new user-friendly package, DCM (Discrete Choice Models) which may be used to estimate a wide range of discrete choice models.
Course Outline - The course material will include
- An introduction to programming in Ox and PcGive, in conjunction with an overview of economic modelling using the Ox-Metrics suite of programs
- Brief historical overview of discrete choice models and associated estimators
- Extension to the vanilla logit model
- An introduction to simulation-based inference in the context of discrete choice modelling. We begin with a very simple accept-reject (AR) procedure in the guise of the well known crude frequency simulator (CFS). Smoothed versions of this algorithm along with the Geweke, Hajivassiliou, Keane (GHK) simulator, the simulator of choice for many practitioners estimating multinomial probit model, are also introduced.
- Identification of Discrete Choice Models
- Revealed versus Stated Preference Models
- Hands-on experience in the use of Ox and PcGive to estimate binary, multinomial and ordinal discrete choice models
- An introduction to DCM - a new object-oriented package for estimating a wide range of discrete choice models
Applications - Throughout the course we make use a wide range of economic applications. These will include:
- Stated Preference Models for Mobile Communications
- Discrete Choice Models of Labour Supply
- Panel Data Model of Ordered Response: The Measurement and Determinants of Institutional Change
- Discrete Choice Models of Transport Behaviour
- Ordered Response Models of Fertility Behaviour
The Principal Lecturer - The principal lecturers are:
- Dr Melvyn Weeks, Faculty of Economics, University of Cambridge
- Dr. Matias Eklof, Faculty of Economics, University of Uppsala
Who Should Attend
The course is designed to attract both academic and policy analysts working with qualitative data. Whilst the instruction makes no assumptions as to prior exposure to the econometrics literature on discrete choice, participants will require a basic familiarity with standard econometric techniques, such as multiple regression, binary choice models, and estimation methods such as maximum likelihood. Familiarity with basic programming concepts and empirical analysis will also be advantageous.
Course Texts - Throughout the course we will make use of many published articles. However, two texts which will be consulted frequently are:
- Train, K. (2003) Discrete Choice Methods with Simulation. Cambridge University Press
- Louviere, J., D. Hensher, and J. Swait, (2000). Stated Choice Methods: Analysis and Applications. Cambridge University Press
Cost - the cost of the course is
For those atteding this course only
| Non-Academic Organisations | 1st Participant | £1,000+VAT=£1,175.00 |
| 2nd+ Participants | £ 800+VAT=£ 940.00 | |
| Academic Organisations | 1st Participant | £ 700+VAT=£ 822.50 |
| 2nd+ Participants | £ 595+VAT=£ 699.13 |
For those also attending the Financial and Econometric Modelling Using PcGive and PcGets, 8 - 12 December 2003
| Non-Academic Organisations | 1st Participant | £ 800+VAT=£ 940.00 |
| 2nd+ Participants | £ 600+VAT=£ 705.00 | |
| Academic Organisations | 1st Participant | £ 600+VAT=£ 705.00 |
| 2nd+ Participants | £ 540+VAT=£ 634.50 |
The cost includes course materials, lunch, refreshments and the use of computers. The number of delegates is restricted. Please register early to guarantee your place. Further instructions will be sent with the joining instructions. If you need assistance in locating hotel accommodation in the area, request the help of our Training Department.
DAY 1
Morning
Session 1. Introduction to Ox
- Why Use Ox?
- Introduction to the OxMetrics Family
- An overview of programming in Ox including the Object Oriented approach
- Estimation and Inference Using Ox and PcGive
Session 2. Practical I
- Data input and manipulation
- A simple OLS estimator in matrix form: A first Ox program
- Input and Output: An introduction to the construction of datasets as input into estimations routines. This will include an overview of the Database and Modelbase classes in Ox.
- Calling libraries within Ox
- Estimating simple binary and multinomial choice models in Pc_Give
Afternoon
Session 3. Theory I: Simple Binary and Multinomial Choice Models
- The Random Utility Model
- Properties of Discrete Choice Models
- The Tractability of Logit
- Alternative Stochastic Specifications
- Multiple versus Single Index Models
DAY 2
Morning
Session 4. Theory Il: Extensions of the Standard Logit Model
- The Limitations of Logit
- The Curse of Dimensionality in Discrete Choice Models
- Circumventing the Curse: An Introduction to the Multinomial Pro bit and Mixed Logit Model
- Identification of Discrete Choice Models
Session 5. Practical Il:
- Stated Preference Models for Mobile Communications
- Specification and Estimation of a Conditional Logit Model
- Testing for Random Preference Heterogeneity and the Mixed Logit Model
- Specification and Estimation of a Conditional Logit Model
- A Discrete Choice Model of Labour Supply
- Specification and Estimation of a Multinomial Probit Model
- The Mixed Probit Model
Afternoon
Session 6. Practical III
- Single Versus Multiple Index Models
- Circumventing the Curse with Ordinality
- Specification and Estimation of Ordered Response Models: An Application to Finance
- An Application to Fertility with Social Interaction
- Circumventing the Curse with Ordinality
DAY 3
Morning
Session 7. Theory Ill: Panel Data Estimation of Discrete Response Models
- Unobserved Heterogeneity in Nonlinear Models
- Probit and Logit: Fixed versus Random Effects
- Panel data estimation of Mixed Logit Models
Session 8. Practical IV
- The Measurement and Determinants of Institutional Change
- A Panel Data Estimator of an Ordered Response Model
Additional Readings
- HAUSMAN, J., AND D. WISE (1978): "A Conditional Probit Model for Qualitative Choice: Discrete Decisions Recognizing Interdependence and Heterogeneous Rpeferences," Econometrica, pp. 403-429.
- MANSKI, C., AND S. LERMAN (1981): "On the Use of Simulated Frequencies to Approximate Choice Probabilities," in Structural Analysis of Discrete Data with Econometric Applications, ed. by C. Manski, and D. McFadden, pp. 305-319. MIT Press, Cambridge, Massachusetts.
- McFADDEN, D. (1989): "A Method of Simulated Moments for Estimationof Discrete Response Models Without Numerical Integration," Econometrica, 57, 995-1026.
- PAKES, A., AND D. POLLARD (1989): "Simulation and Asymptotics of Optimization Estimators," Econometrica, 54, 755-785.
- THURSTONE, L. (1927): "A Law of Comparative Judgement," Psychological Review, 34, 273-286.
Registration closes 5 calendar days prior to the start of the course.
Cancellations:
- full fee returned for cancellations made over 28 calendar days prior to start of the course
- half-fee returned for cancellations made 14 calendar days prior to he start of the course
- no fee returned for cancellations made less than 14 calendar days prior to the start of the course.
Payment of course fees required prior to the course start date
|
||