graphs/graph_enac_l.gif
English only
INTER  >  TRANSP-OR > Workshop on Discrete Choice Models
Logo TRANSP-OR
Content

Workshop on Discrete Choice Models

August 27 - 29, 2009

Ecole Polytechnique Fédérale de Lausanne - Room GC A3 30 (click here for a map)

The 2009 workshop will be organized in the same spirit as the previous ones: an informal meeting for the exchange of ideas around discrete choice models, with the objective to trigger new collaborations, or strengthen existing ones. At the end of the workshop, a list of potential collaborations will be identified, with specific objectives.

Registration

Top

Everyone interested is invited to attend. Presentations are upon invitation only. All participants, including speakers, must register with the following form.

The registration fee includes the dinner on Thursday, lunch on Friday, coffee breaks, as well as transportation on Saturday.

Click here to access to the registration form

Schedule

Top
Thursday Aug. 27, afternoon - Friday Aug. 28, morning
Presentations (see detail below)
Friday Aug 28, afternoon
Workshop meeting
Saturday August 29, morning
Hiking
Saturday August 29, 12:00
Genuine swiss raclette in a chalet up in the mountain
Thursday
13:45 14:00 Welcome
14:00 14:30 Ben-Elia
14:30 15:00 Polydoropoulo
15:00 15:30 Newman
15:30 16:00 Fosgerau
16:00 16:30 BREAK
16:30 17:00 Bierlaire
17:00 17:30 Schussler
17:30 18:00 Fetiarison
18:00 18:30 de Lapparent
Friday
09:00 09:30 Frejinger
09:30 10:00 Robin
10:00 10:30 Sivakumar
10:30 11:00 BREAK
11:00 11:30 Hurtubia
11:30 12:00 Kaplan
12:00 12:30 Chen
Lunch
14:00 16:00 Workshop meeting

Accomodation

Top

EPFL has negociated special rates with several hotels in the area.

The complete list is available here.

Ask the hotel for a free public transportation pass.

Transportation

Top

The easiest way to get to EPFL is to take the train from Geneva Airport to Renens. In Renens, take the light-rail (called M1) towards Lausanne. There is a stop at EPFL. The travel time is about 1 hour.

A map of the bus and metro network can be found here and time tables are available at the Lausanne Transport web page. Ask the hotel for a free public transportation pass.

Check the Swiss Federal Railways website.

To navigate within EPFL, use map.epfl.ch.

Consult also the page "How to get to EPFL?"

List of presentations (14)

Top

Click on the title to download the slides (if available).

  • A choice model of commuter behavior with rewards for avoiding peak-hour driving by Eran Ben-Elia
  • Experimental analysis of the implicit choice set generation using the Constrained Multinomial Logit model by Michel Bierlaire
  • A method of calculating path observation likelihood from GPS data by Jingmin Chen
  • Unknown travel times and decision under risk: modelling choices using the rank dependent expected utility theory by Matthieu de Lapparent
  • Estimation of discrete choice models: extending BIOGEME by Mamy Fetiarison
  • Choice-probability generating functions by Mogens Fosgerau
  • A dynamic discrete choice approach for modeling route choice by Emma Frejinger
  • Inferring the activities of smartphone users from context measurements using Bayesian inference and random utility models by Ricardo Hurtubia
  • Development and estimation of a semi-compensatory model with flexible error structure by Sigal Kaplan
  • Pay no attention to the alternatives behind the curtain by Jeffrey Newman
  • Modeling the Impact of Satisfaction-Happiness on Choice Behavior by Amalia Polydoropoulou
  • A new dynamic facial expression recognition framework by Thomas Robin
  • Challenges of route choice models derived from GPS observations by Nadine Schüssler
  • Modelling Urban Energy Systems and Related Travel Behaviour Research by Aruna Sivakumar

List of abstracts

Top
  1. A choice model of commuter behavior with rewards for avoiding peak-hour driving by Eran Ben-Elia (Utrecht University, Netherlands)

    'Spitsmijden' is a novel idea tested recently in The Netherlands to reward drivers for changing their commuting behavior as temporary measure for alivating congestion. Based on a 13 week RP study, different rewards were investigated and commuters' behavior was monitored using state-of-the art detection equipment. A model of departure time and mode choice was estimated using panel mixed logit. Results show that although rewards are important contributors to behavioral changes other factors such as information, previous experiences, situational and personal factors and even the weather, are significant as well. The findings provide important insights to an implementation of a reward-based strategy on a larger scale.

    Click here to download the presentation

  2. Experimental analysis of the implicit choice set generation using the Constrained Multinomial Logit model by Michel Bierlaire (TRANSP-OR, EPFL)

    Discrete choice models are defined conditional to the knowledge of the actual choice set by the analyst. The common practice for is to assume that individual-based choice sets can be deterministically generated based on the choice context and the characteristics of the decision maker. There are many situations where this assumption is not valid or not applicable, and probabilistic choice set formation procedures must be considered. The Constrained Multinomial Logit model (CMNL) has recently been proposed by Martinez et al. (2009) as a convenient way to deal with this issue, as it is also appropriate for models with a large choice set. In this paper, we analyze how the implicit choice set generation of the CMNL compares to the explicit choice set generation as described by Manski (1977). The results based on synthetic data show that the implicit choice set generation model may be a poor approximation of the explicit model. (joint work with Ricardo Hurtubia and Gunnar Flötteröd)

    Click here to download the presentation

  3. A method of calculating path observation likelihood from GPS data by Jingmin Chen (EPFL)

    A method is proposed to probabilistically map location observations to the underlying network. Instead of generating a single path as the map matching algorithms do, this method aims at calculating a likelihood for each potentially true path to have been the actual path. The result can be used in route choice modeling to avoid biases introduced by a deterministic map matching algorithm. Both spatial and temporal relationships existing in the location data trace and network are taken into account in the method. An algorithm is designed to calculate path probability, starting by defining the measurement for the topological relationship between location observation and network data. Results from the algorithm for a simulated trip are presented to demonstrate the viability of the algorithm.

    Click here to download the presentation

  4. Unknown travel times and decision under risk: modelling choices using the rank dependent expected utility theory by Matthieu de Lapparent (INRETS)

    There is a recent interest in adapting and implementing decision theories under risk (known probability distribution) or under uncertainty (unknown probability distribution) to analysis of travel choices. Amongst the wide range of existing decision theories under risk (see Cohen and Tallon (2000) for a review of additive and non additive expected utility models, and Kahneman and Tversky (1979, 1981, 1992, 2000, 2003, etc.) and Fishburn (e.g. papers from 1964 to 1994) for alternative approaches) is the rank dependent expected utility (RDEU, Quiggin (1982)) theory. Rank dependence is a realistic phenomenon. It entails the attention paid to an outcome does not depend only on how much and how likely it is but also on its relative ranking in the ordered set of outcomes. In this approach, the decision maker is characterized by two functions : a (dis)utility function on consequences measuring preferences over sure outcomes and a probability weighting function measuring the subjective weighting of occurrence probabilities of these outcomes. Descriptively a probability weighting function may be interpreted as a misapprehension about the true probability distribution. Normatively, this transformation may be seen as a conscious misappropriation of the true probability distribution. It overcomes many drawbacks of more ?simple? approaches such as the expected utility theory (and its mean-variance special case) without loosing practicability, thanks to an elegant functional representation of preferences over a set of lotteries: - it is consistent with the behavior revealed by the Allais paradox (attraction for certainty) ; - the decision maker could dislike risk (prefer to any lottery its expectation) without necessarily avoiding any increase in risk (weak and strong risk aversion are no more confused) ; - diminishing marginal utility (assuming that outcomes are gains) may coexists with risk seeking attitudes ; - decision makers with the same utility function may differ in their choices between lotteries when they have different probability weighting functions. All in all, RDEU theory is recognised for its greater capability to model heterogenous and realistic behaviours as it regards decision under risk, hence analysis of risk aversion. From an empirical perspective, the fact that RDEU theory disentangles attitude to risk from attitude to wealth by means of two distinct functions through a generic and flexible functional form makes possible to propose a series of risky choice models by building up on probabilistic assumptions that lay down standard random utility models and by investigating different mathematical formulations that define the normative behaviour of a decision maker (i.e. different specifications for the probability weighting function and different specifications for the utility function that is defined on time outcomes and other deterministic travel attributes). This is the main purpose of the presentation. The application focuses on mode choice using SP data collected in the Zürich canton. The choice sets are based on two lotteries, each with two time outcomes, that include also deterministic attributes.

    Click here to download the presentation

  5. Estimation of discrete choice models: extending BIOGEME by Mamy Fetiarison (TRANSP-OR, EPFL)

    BIOGEME is a free software package for estimating by maximum likelihood a broad range of random utility models. It can estimate particularly Multivariate Extreme Value (MEV) models including the logit model, the nested logit model, the cross-nested logit model, and the network MEV model, as well as continuous and discrete mixtures of these models. Biogeme has been designed to provide modelers with tools to investigate a wide variety of discrete choice models without worrying about the estimation algorithm itself. We present some new features and capabilities of Biogeme. To make it more flexible, we allow explicitly the user to specify the random utility model to be estimated and the associated likelihood function. With simple formulations, it will be able to handle more sophisticated models such as latent variable models, latent class models, dynamic models, etc. required by modern modeling practice.

    Click here to download the presentation

  6. Choice-probability generating functions by Mogens Fosgerau (DTU)

    This paper establishes that every random utility discrete choice model (RUM) has a representation that can be characterized by a choice-probability generating function (CPGF) with specific properties, and that every function with these specific properties is consistent with a RUM. The choice probabilities from the RUM are obtained from the gradient of the CPGF. Mixtures of RUM are characterized by logarithmic mixtures of their associated CPGF. The class of multivariate extreme value (MEV) RUM coincides with a certain set of CPGF, which is strictly smaller than the set of all CPGF. This set is closed under a number of operations that can be used to create new MEV RUM from old. The paper indicates CPGF that are new in the discrete choice literature. The notion of a CPGF leads also to a generalization of the Archimedian copula. Finally, the paper generalizes a result by Dagsvik to show that any RUM may be approximated by a cross-nested logit. (joint work with Daniel McFadden and Michel Bierlaire)

    Click here to download the presentation

  7. A dynamic discrete choice approach for modeling route choice by Emma Frejinger (Centre for Transport Studies, Stockholm)

    In this talk we present a dynamic discrete choice approach for the estimation of the parameters of a route choice model. In the dynamic modeling approach, the individual is seen as taking sequential decisions on which link to choose, and the choices are made at the nodes in the network. The obvious advantage with this approach is that the choice set at every stage is quite small and well defined, while a correlation structure is naturally imposed among different paths, even if each sequential decision follows a multinomial logit model. The utility maximising choice of path may be broken down into a sequence of link choices, where at each stage the individual considers the utility associated with downstream link choices accumulated into a value function. However, if we were to compute the value function associated with the available link choices at every stage, the complexity of the problem would be at least the same as the original path choice problem. An exact solution method to calculate the value function runs into the curse of dimensionality when solving a dynamic programming problem. The objective of this research is to test whether it is possible to generate good predictors for the value function such that the parameters of the route choice model may be estimated on link choices rather than path choices. If this turns out to be possible, then both the econometric and computational complexity of route choice modelling may be dramatically reduced (joint work with M. Fosgerau and A. Karlström).

  8. Inferring the activities of smartphone users from context measurements using Bayesian inference and random utility models by Ricardo Hurtubia (TRANSP-OR EPFL)

    Smartphones collect a wealth of information about their users. This includes GPS tracks and the MAC addresses of devices around the user, and it can go as far as taking visual and acoustic samples of the user's environment. We present a framework to identify a smartphone user's activities in a Bayesian setting. As prior information, we us a random utility model that accounts for the type of activity a user is likely to perform at any given location and time; this model was estimated for the whole population using data from the 2005 Swiss Transport Microcensus. The smartphone measurements come from a preliminary 2-month period survey, where one user carried around a phone programmed to constantly record his GPS location and other context variables, including the MAC addresses of nearby bluetooth devices. In addition to this, the user answered a daily survey, where he described and geolocated all the activities performed during this period. An analysis of the recorded data shows that the bluetooth information is useful to identify other users or devices that are frequently observed when performing specific activities. The bluetooth data is therefore used to estimate the likelihood of observing certain devices when performing certain activities. Combining the prior activity information from the random utility model with these likelihoods allows to generate improved posterior distributions of the user's activities. Due to the limited amount of available data only exemplary results are given, which, however, clearly indicate that the accuracy of the predictions can be greatly improved by using bluetooth data. (joint work with Gunnar Flötteröd and Michel Bierlaire)

    Click here to download the presentation

  9. Development and estimation of a semi-compensatory model with flexible error structure by Sigal Kaplan (Technion - Israel Institute of Technology )

    Semi-compensatory models, representing a sequence of an elimination-based choice set formation and a utility-based choice, show promise in increasing the behavioral realism of discrete choice models by integrating the choice set specification within the representation of the choice process. Alas, a disadvantage of current semi-compensatory models versus compensatory models is their behaviorally non-realistic assumption of an independent error structure. This study proposes a novel semi-compensatory model incorporating a flexible error structure. Specifically, the model represents a sequence of a conjunctive heuristic assuming correlated cut-off thresholds for choice set formation, and a utility-based choice assuming a substitution pattern with a nested structure across alternatives. Mathematically, the model jointly represents the conjunctive heuristic by a multidimensional mixed ordered-response model and the utility-based choice by alternatively (i) a nested-logit model and (ii) an error-component logit. In order to test the suggested methodology, the model was estimated for a sample of 1,893 ranked choices and respective threshold values from 631 students who participated in a web-based two-stage choice experiment of rental apartment choice. The estimated semi-compensatory models were compared to their compensatory counterparts. Results demonstrate the applicability and feasibility of representing flexible substitution patterns with the proposed semi-compensatory model, as well as promote the comprehension of the differences between semi-compensatory and compensatory models.

    Click here to download the presentation

  10. Pay no attention to the alternatives behind the curtain by Jeffrey Newman (TRANSP-OR, EPFL)

    Discrete choice modeling advances abound for scenarios where we can collect data about the behaviors of individuals, and observe their ultimate choice from among the range of possible alternatives. However, particularly in competitive private sector enterprises (e.g. airlines, hotels), it is often difficult and expensive to conduct surveys of users, particularly of the customers of one's competitors. However, there is typically a huge pool of data readily available for virtually no cost about one's own customers: what, how and when they purchase, which alternatives were made available from one's own pool of alternatives, etc. We pose the question: what can we do with this data? Traditional choice based sampling tools have problems with this scenario, as the sampling rate for the choices is typically employed as a divisor, but in this case it is, for some choices, zero. But all is not lost. We propose introducing aggregate measures of demand and market share (available from government statistics and/or competitors corporate disclosures when they are publicly traded companies) to fill in the gaps in the readily available disaggregate data. This works reasonably well when the underlying model is assumed to be MNL. Whether this strategy can still be employed when the underlying model is a more general GEV model is an open question, which we are beginning to explore.

    Click here to download the presentation

  11. Modeling the Impact of Satisfaction-Happiness on Choice Behavior by Amalia Polydoropoulou (University of the Aegean)

    Click here to download the presentation

  12. A new dynamic facial expression recognition framework by Thomas Robin (TRANSP-OR, EPFL)

    A recent interest appears in transportation for users emotion recognition. This permits to adapt car behaviors to drivers mood for safety reasons, or improve public transportation offers. Human emotions are complex and defined by several elements, such as voices intonations or facial expressions. We propose a new dynamic facial expression recognition framework based on Discrete Choice Models (DCM). The aim of the work is to model the choice of a person who is exposed to a video sequence representing a facial expression, and has to label it. The approach originality lies on the absence of ground truth and the explicit modelling of causal effects between facial features and face expression. The model is a combination of two DCMs. The first one captures the dynamic facial expression evaluation across the video frames (expression perception sub-model), and the second one concerns the frames weighting in order to determine at which moment the person decides the facial expression when looking at the video sequence (frame choice sub-model). A computer vision tool, called Active Appearance Model (AAM) is used to extract facial information in videos. Information are processed in order to obtain facial features which are injected in the expression perception sub-model according to the Facial Action Coding System (FACS). Derivatives of facial features are used in the frame choice sub-model. In addition, direct outputs of the AAM are injected in both sub-models to account for global appearances of faces (C vectors elements). The model is then estimated using videos from the Facial Expressions and Emotions Database (FEED). Expressions labels on the videos have been obtained using an internet survey available at http://transp-or2.epfl.ch/videosurvey/.

    Click here to download the presentation

  13. Challenges of route choice models derived from GPS observations by Nadine Schüssler (IVT ETH Zürich)

    With the increasing use of GPS devices in transport surveys, route choice modellers are facing new challenges in addition to old ones such as the appropriate account for route overlap. One of these challenges is the identification of the chosen route, often solved by so-called mapmatching procedures. Another one is the generation of choice sets in high-resolution networks. High-resolution networks are essential for an accurate mapmatching but substantially increase the computational costs for choice set generation procedures. The presentation will present the work based on the car trips contained in the GPS study presented in 2008 and the Swiss Navteq network, a fine-grained network covering all regions of Switzerland and containing 408636 nodes and 882120 unidirectional links. After a short presentation of the employed mapmatching procedure, different choice set generation procedures, deterministic as well as stochastic ones, will be tested to evaluate their computational performance as well as their effect on the choice set composition, the resulting route choice models and treatment of route overlap. In the choice models, the impact of different route attributes will be investigated with special attention to the formulation of the similarity factors, which account for route overlap.

    Click here to download the presentation

  14. Modelling Urban Energy Systems and Related Travel Behaviour Research by Aruna Sivakumar (Imperial College London)

    Click here to download the presentation

List of participants (18)

Top
  • Ben-Elia Eran, Utrecht University, Netherlands
  • Bierlaire Michel, TRANSP-OR, EPFL
  • Chen Jingmin, EPFL
  • de Lapparent Matthieu, INRETS
  • Denis Clément, EPFL TRANSP-OR
  • Fetiarison Mamy, TRANSP-OR, EPFL
  • Flötteröd Gunnar, TRANSP-OR, EPFL
  • Fosgerau Mogens, DTU
  • Frejinger Emma, Centre for Transport Studies, Stockholm
  • Hurtubia Ricardo, TRANSP-OR EPFL
  • Jensen Anders Fjendbo, DTU Transport
  • Kaplan Sigal, Technion - Israel Institute of Technology
  • Newman Jeffrey, TRANSP-OR, EPFL
  • Polydoropoulou Amalia, University of the Aegean
  • Ramjerdi Farideh , Institute of Transport Economics
  • Robin Thomas, TRANSP-OR, EPFL
  • Schüssler Nadine, IVT ETH Zürich
  • Sivakumar Aruna, Imperial College London