Stephane Hess

Approximation of Bayesian efficiency in experimental choice designs

This paper compares different types of simulated draws over a range of number of draws in generating Bayesian efficient designs for stated choice (SC) studies. The paper examines how closely pseudo Monte Carlo, quasi Monte Carlo and Gaussian quadrature methods are able to replicate the true levels of Bayesian efficiency for SC designs of various dimensions. The authors conclude that the predominantly employed method of using pseudo Monte Carlo draws is unlikely to result in leading to truly Bayesian efficient SC designs. The quasi Monte Carlo methods analysed here (Halton, Sobol, and Modified Latin Hypercube Sampling) all clearly outperform the pseudo Monte Carlo draws. However, the Gaussian quadrature method examined in this paper, incremental Gaussian quadrature, outperforms all, and is therefore the recommended approximation method for the calculation of Bayesian efficiency of SC designs.

Accounting for Random Taste Heterogeneity in Airport Choice Modeling

The findings from a disaggregate analysis of the choice of airport, airline, and access mode for business travelers living in the San Francisco Bay Area, California, are presented. Aside from formulation of the multidimensional choice process, the main objective is to explore random taste heterogeneity among decision makers in their evaluation of the attractiveness of the different alternatives. The results indicate that this heterogeneity is present in tastes relating to in-vehicle access time, access cost, and flight frequency. Accounting for this heterogeneity leads to gains in model fit but, more important, leads to important insights into the differences in behavior across decision makers and avoids the bias introduced into trade-offs when fixed coefficients are used in the presence of significant levels of heterogeneity. In terms of substantive results, the models also reveal a significant impact of changes in out-of-vehicle access time, indicate a preference for service on jet over turboprop flights, and show that experience plays an important role in air travel choice behavior.

Analysis of the effects of speed limit enforcement cameras: Differentiation by road type and catchment area

A detailed statistical analysis is presented of the effects of speed limit enforcement cameras on injury accident numbers. The approach used is constructed in such a way that it is possible to differentiate not only between the effects of the cameras and the effects of trend and seasonality but actually to produce estimates that are independent of any other overall time-dependent effects. Crucially, the estimates produced are also net of the effects of regression to the mean. To allow for the simultaneous treatment of the different levels of severity, weights are used that reflect the frequency of the different types of accidents. This approach is then used on a data set for all injury accidents in Cambridgeshire between 1990 and 2002, which also contains data from 49 camera sites. To quantify the range of effectiveness of the cameras, estimates of the changes in accident numbers are produced for different distances from the camera site. The analysis shows that, overall, in the immediate vicinity of the camera sites, the installation of a camera can be expected to lead to decreases in weighted injury accident numbers by an astounding 45.74%. Lower, but still significant, decreases are observed in the wider surrounding area. Finally, to provide further insight into the differences in performance on different road types, the sites are grouped together according to road category. This analysis shows that the biggest reduction in accident numbers can be obtained on roads with a higher incidence of speeding offenses.

Conditional parameter estimates from Mixed Logit models: distributional assumptions and a free software tool

A number of authors have discussed the possible advantages of conditioning parameter distributions on observed choices when working with Mixed Multinomial Logit models. However, the number of applications is still relatively small, partly due to a limited implementation in available software. To address this situation, the present paper discusses the development of a freeware software tool that allows users to compute conditional distributions independently of the software used during model estimation. Additionally, the paper looks at what impact assumptions made for the unconditional distributions have on the results obtained with conditional distributions. Here, an application using stated choice data collected in Denmark shows that while the move from unconditional to conditional distributions possibly brings results closer together, some discrepancies do remain.

Incorporating environmental attitudes in discrete choice models: An exploration of the utility of the awareness of consequences scale

Environmental economists are increasingly interested in better understanding how people cognitively organise their beliefs and attitudes towards environmental change in order to identify key motives and barriers that stimulate or prevent action. In this paper, we explore the utility of a commonly used psychometric scale, the awareness of consequences (AC) scale, in order to better understand stated choices. The main contribution of the paper is that it provides a novel approach to incorporate attitudinal information into discrete choice models for environmental valuation: firstly, environmental attitudes are incorporated using a reinterpretation of the classical AC scale recently proposed by Ryan and Spash (2012); and, secondly, attitudinal data is incorporated as latent variables under a hybrid choice modelling framework. This novel approach is applied to data from a survey conducted in the Basque Country (Spain) in 2008 aimed at valuing land-use policies in a Natura 2000 Network site. The results are relevant to policy-making because choice models that are able to accommodate underlying environmental attitudes may help in designing more effective environmental policies.

Review of Evidence for Temporal Transferability of Mode–Destination Models

One main motivation for developing travel behavior models is to use them to forecast future levels of transport demand. Given that the interest in transport planning is often in long-term forecasts, with forecast horizons of up to 30 years, it is important to consider the transferability of travel behavior models over time. The importance of model transferability has been recognized since disaggregate models were first applied in the late 1970s and early 1980s, but seems to have been largely forgotten recently, because the focus has been on the development of ever more advanced models that better explain current behavior, with a particular focus on the representation of taste heterogeneity. However, there are sufficient grounds to suspect that the model that best explains current behavior may not necessarily be the best tool for forecasting, not least because of the risk of overfitting to the base data. This paper aims to return the crucial issue of temporal transferability of travel demand models to the research agenda. First, the notion of transferability is discussed, highlighting the potential impacts of violations of the assumption of transferability, and the way transferability can be assessed is also described. Next, the most complete review of existing work investigating the temporal transferability of mode and mode-destination models to date is presented. Finally, a number of areas in which future research should be directed are identified.

Improving the quality of demand forecasts through cross nested logit: a stated choice case study of airport, airline and access mode choice

Airport choice models have been used extensively in recent years to assist the transport planning in large metropolitan areas. However, these studies have typically focussed solely on airports within a given metropolitan area, at a time when passengers are increasingly willing to travel further to access airports. This article presents the findings of a study that uses broader, regional data from the East Coast of the United States collected through a stated choice based air travel survey. The study makes use of a cross-nested logit structure that allows for the joint representation of inter-alternative correlation along the three choice dimensions of airport, airline and access mode choice. The analysis not only shows significant gains in model fit when moving to this more advanced nesting structure, but the more appropriate cross-elasticity assumptions also lead to more intuitively correct substitution patterns in forecasting examples.

Estimated Value of Savings in Travel Time in Switzerland: Analysis of Pooled Data

Reliable measures of the value of travel time savings (VTTS) are crucial inputs into policy planning in any regional or national context. However, the evidence from secondary or parallel VTTS studies often differs from that from official or national studies. This paper presents evidence from a study aimed at merging evidence from four separate studies of VTTS conducted in Switzerland. The analysis shows the potential of models estimated jointly on the basis of the combined data and produces stable results that should be more representative of those for the overall population.

EFFECTS OF SPEED LIMIT ENFORCEMENT CAMERAS ON ACCIDENT RATES

Speed limit enforcement cameras (SLECs) have been in operation in Great Britain since 1991. However, there is still considerable dispute regarding their effectiveness in reducing accident rates. The aim of this research was to analyze the effects of SLECs on accident rates in Cambridgeshire, United Kingdom, using time series data collected over an 11-year period. A time series analysis of the accident data revealed the presence of both trend and seasonality components. A method was developed to remove the influence of these two components from the data and compare mean accident levels before and after installation of the camera. The method was also constructed in such a way that it would be able to distinguish between the actual effects of the camera installation and the effects of regression to the mean. The initial investigation into the effects of SLECs showed an average decrease over sites in the monthly accident frequency by around 18%; a more detailed analysis suggested that the best approximation of the effect of the introduction of a SLEC is a decrease in injury accidents by 31.26%, thus giving clear evidence that SLECs do indeed contribute to a significant decrease in accident numbers.

Analytic approximations for computing probit choice probabilities

The multinomial probit model has long been used in transport applications as the basis for mode- and route-choice in computing network flows, and in other choice contexts when estimating preference parameters. It is well known that computation of the probit choice probabilities presents a significant computational burden, since they are based on multivariate normal integrals. Various methods exist for computing these choice probabilities, though standard Monte Carlo is most commonly used. In this article we compare two analytical approximation methods (Mendell–Elston and Solow–Joe) with three Monte Carlo approaches for computing probit choice probabilities. We systematically investigate a wide range of parameter settings and report on the accuracy and computational efficiency of each method. The findings suggest that the accuracy and efficiency of an optimally ordered Mendell–Elston analytic approximation method offers great potential for wider use.