15h30 - 15h55
Capturing Endogneity of Bicycle Sharing System Infrastructure on Usage: Evidence from BIXI Montreal
With the growing installation of bicycle sharing systems across the world there is substantial interest on examining the influence of bicycle infrastructure (number of station and station capacity), land-use and built environment on bicycle usage. A majority of these studies consider bicycle infrastructure as exogenous to modeling demand. However, the decision process of bicycle infrastructure installation considers the influence of land-use and built environment. In cases where the bicycle infrastructure is closely related to the land-use and urban form it is important to recognize that developing models treating the bicycle infrastructure as exogenous to the dependent variable (bicycle demand) might lead to incorrect and biased model estimations. In this study, we address this challenge by proposing a joint decision process composed of three choice processes. The first process considers the bicycle infrastructure installation process while the second and third processes consider bicycle usage characterized as arrivals and departures respectively. In this modeling framework we allow for potential correlations across the three choice systems. We consider an ordered representation for all the variables yielding a three dimensional panel ordered formulation. In addition, in our approach, we recognize that hourly arrivals and departures at the same station are likely to be influenced by common unobserved factors. To accommodate this, we adopt a repeated observation based panel ordered logit model. At the same time, the framework developed recognizes that BSS infrastructure model is a one-time decision variable (unlike repeated observations of usage). The proposed model is estimated using data compiled from the Montreal BIXI system from April to August 2012.
15h55 - 16h20
Comparison of Route Choice Models' Predictive Performance
Route choice models are used to predict the path an individual will choose to travel on a network conditional on his origin and destination. These models are important to many transport related applications and models that are often used in practice, such as path size logit (PSL), are based on choice sets of paths. Recently, a link-based recursive logit model (RL) was proposed where route choice is modeled as a sequence of link choices without any restrictions to the network. In this talk we discuss issues related to comparing the prediction performance of PSL and RL. Moreover we present an empirical comparison using cross validation.
16h20 - 16h45
Network Capacity Control under a Non-Parametric Choice Model of Demand
One of the most powerful and simple approaches to model a customer’s choice behavior, with the aim to predict his choice decision facing different options, is non-parametric choice modeling of demand. In this approach, each arriving customer chooses from available alternatives according to an ordered preference list of products. If the customer's most preferred product is not available, he substitutes it with the next lower rank product in his ordered preference list.
In this research, we propose a new mathematical programming approach to compute optimal allocation of resources under a non-parametric choice model of demand. We develop a modified column generation algorithm to efficiently solve large scale, real world practical problems. The computational results show that the approach outperforms alternative models.
16h45 - 17h10
Decomposition Method and a Recursive Mixture Logit for Route Choice Analysis
The multinomial logit (MNL) model is in general used for analyzing route choices in real networks. Recently, Fosgerau et al. (2013) proposed the link based recursive multinomial logit (RL) which is based on the underlying assumption that any path in the network is feasible and belongs to the universal choice set. The RL model is theoretically superior to the well-known sampling approach because it can be consistently estimated and efficiently used for prediction. However, the estimation requires solving a system of linear equation per observation whch raises computational concerns. We therefore propose the decomposition approach that significantly speeds up the estimation and a mixed recursive logit model which is based on the mixed logit idea. Our numerical results are based on the Borlange network in Sweden. This network is composed of 3077 nodes and 7459 links and it is uncongested so travel times are assumed static and deterministic. The sample of real path observations consists of 1832 with 466 destinations, 1420 different origin-destination (OD) pairs and more than 37,000 link choices.