Reducing Bias in Value of Time Estimates by Joint Estimation on Multiple Datasets: a Case Study in Switzerland



Reducing Bias in Value of Time Estimates by Joint Estimation on Multiple Datasets: a Case Study in Switzerland

Authors

S Hess, Imperial College London, UK; A Erath, M Vrtic, K Axhausen, ETHZ, CH

Description

This paper discusses the estimation of state of the art value of time estimates for use in the official Swiss norm, making use of almost 60,000 observations from 5 surveys.

Abstract

The estimation of reliable measures of the valuation of travel time savings (VTTS) is one of the main topics in the area of travel behaviour research. Such estimates are a crucial input for policy decisions, for example in the context of cost benefit analysis.

Two main ways exist of improving the reliability of VTTS estimates; using better data, and using better modelling approaches. The aim of using better data can be achieved primarily by improving the design of the surveys used in the collection. However, the stability and quality of estimates can also be improved upon by making use of larger samples and by using a more cross-sectional sample of the population, i.e. making the sample population more representative of the overall population. In terms of using better modelling approaches, there are again two possible avenues for improvements, namely those of model structure and model specification. The former refers to the actual mathematical structure of the econometric model, while the latter concerns the specification of the utility function used in that model.

In this paper, we present evidence from a large-scale VTTS study conducted in Switzerland, where the aim is precisely to improve the reliability of the VTTS estimates through improving the quality of the model inputs and the model specification. The quality of the data is improved through grouping together observations from five separate SP studies conducted over recent years, leading to a very large sample size of almost 60,000 observations. This not only improves the stability of the results through estimation on a larger sample size, but the aggregation of five datasets should arguably provide a better cross-section of the population.

In terms of model specification, we make use of flexible non-linear specifications that allow for a continuous interaction between income and trip distance and the various marginal utility coefficients. This allows us to directly gauge the variations in the VTTS in relation to income and trip distance, and also proves very useful when extrapolating the results from the sample level to the population level.

On completion of the study, the results of the analysis will be fully integrated into the new official Swiss guidance for cost-benefit analysis. Initial findings from the analysis show significant differences when compared to corresponding analyses making use of a more basic utility specification. As expected, the results obtained on the overall sample differ from those obtained on the individual samples, by giving a weighted average.

Publisher

Association for European Transport