Improving Understanding of Choice Experiments to Estimate Values of Travel Time
A Daly, RAND Europe / ITS, University of Leeds, UK
Willingness-to-pay (WTP) is a key concept in the formulation of public policy. Decision makers justify expenditure on the basis that the public would be willing to pay for infrastructure or improved services: the challenge is to quantify WTP. A typical example is to estimate the value of time (VOT) in travel, which forms the key input to for most decisions on transport policy and infrastructure.
The most common approach adopted for the estimation of population VOT (as for a number of other WTP parameters) is to conduct stated choice experiments. These experiments involve offering a sample of respondents a series of pairs of alternatives, differing in time and cost. In several studies, including some recent ones, part of the experiment has been based on an experimental design that balances the gains and losses in time and cost but does not include any other variables.
This design, though well tried and tested, has been criticised as being too simple, as not giving sufficient variety of choice and of inviting inconsistency in responses (Hess et al., 2009). Sometimes it has proved quite difficult to obtain satisfactory results from specific data sets. However, sophisticated analyses of data of this form (e.g. Fosgerau, 2006) are able to obtain quite robust estimates. It is therefore interesting to consider whether the problems that have been encountered might be due to the excessively simple form of analysis that has conventionally been used.
The conventional assumption is that all the choices are equally accurately assessed, but it appears that this assumption can cause serious problems in modelling the responses. Preliminary tests with alternative assumptions suggest that they could not only solve the difficulty of obtaining credible results but also give a much better explanation of the data. One such assumption is that of random inter-respondent heterogeneity, but other assumptions are possible.
For example, the assumption that error in the data is proportional to the cost difference between the alternatives presented leads to analyses on a basis that has been described as ?Willingness to Pay? space. This is no more complicated than the conventional analysis but, as shown by Fosgerau and Bierlaire (2008), appears to give simply better results in terms of both plausibility and fit to the data. However, a number of other assumptions are possible and it is not clear which of them would be best or what the theoretical underpinning of each of them might be.
The paper therefore addresses these issues by enumerating the possible simple assumptions that might be made, discussing what their theoretical foundations might be and assessing their success on 1-3 data sets that have been used in important VOT studies. Conclusions are then drawn for future practice. The conclusions will take account of the constraints under which practical studies have to be conducted in a commercial environment, where budget, time and even expertise are often lacking.
Fosgerau, M. (2006) Investigating the distribution of the value of travel time savings, Transportation Research Part B, 40, pp. 688-707.
Fosgerau, M. and Bierlaire, M. (2008?) Discrete choice models with multiplicative error terms, in press, Transportation Research Part B.
Hess, S., Rose, J. and Polak, J. (2009?) Non-trading, lexicographic and inconsistent behaviour in stated choice data, accepted for publication, Transportation Research Part D.
Association for European Transport