Distribution of Willingness‐to‐Pay for Speed Reduction with Non‐positive Bidders: Is Choice Modelling Consistent with Contingent Valuation?

Abstract

The paper addresses the issue of consistency between two commonly employed stated preference data—referendum contingent valuation (CV) and discrete choice modelling (CM)—with respect to estimated distributions of individual willingness‐to‐pay (WTP) for non‐market goods. The policy context is that of a local externality: effective speed reduction by means of traffic‐calming in towns crossed by fast roads. In particular, data from two independent samples of the same population are contrasted. The findings show that both methods indicate that speed reduction via traffic‐calming is valued in a polarized fashion. Results from both methods are consistent with the presence of two groups of preferences: a larger group holding positive values and a smaller one with non‐positive values. While the estimates of the relative proportions of the two groups are similar across the two data sources, once the econometric analysis of the CM responses allows for polarized preferences the estimates of the distribution of individual WTP differ substantially. The results from the choice modelling survey indicate that residents are also willing to pay for other benefits from traffic‐calming, such as noise reduction and a decreased waiting time for crossing, but preferences for these are also polarized, with WTP for aesthetic improvements being positive only for those supporting effective speed control. In comparing distributions of value estimates from CM and CV, surveys practitioners should account for the effects of taste heterogeneity over externalities and take advantage of the ability to derive individual‐specific WTP estimates from panel estimation rather than simply deriving estimates for common features of the WTP distribution.

Acknowledgements

The authors gratefully acknowledge financial support from the Engineering and Physical Science Research Council (Grant No. GR/L76792). Special thanks to Vic Adamowicz, Guy Garrod, John Loomis, Luis Rizzi, and the members of the civil engineering team supporting the study: Jessica Anderson, Michael Bell and Sergio Grosso. All remaining errors, of course, are entirely the authors’ responsibility.

Notes

1. These are also often termed ‘choice experiments’ in the environmental economics literature.

2. For example, the claim that “The standard assumptions of a multinomial choice model implies that the distribution of WTP follows the logistic, an assumption that finds little support in the CV literature” is clearly overcome by the current tendency to use‐mixed‐logit models on panel data, which allow the derivation of individual WTP estimates conditional on patterns of observed choice. Such WTP distributions are no longer constrained to be logistic at either the population level (e.g. Train, 2003) or at the sample level (e.g. Hensher and Greene, 2003; Scarpa et al., 2004; Scarpa and Thiene, 2005; Sillano and Ortúzar, 2005).

3. Another way to interpret this is that they may need compensation to accept it.

4. Indeed, this is one reason why it has long been considered good practice in CM (Louviere, 1988; Louviere et al., 2000) to include a no choice option in the set of alternatives, or why in some CE studies attribute worsening as well as improvements are considered in the experimental design (Willis et al., 2005).

5. For full details, see Scarpa et al. (2001a).

6. An algorithm was written in GAUSS to produce all the possible combinations of pair‐wise cases, and for the subsequent elimination of all cases in which the choice was dominated by one particular profile.

7. About 500 households were contacted, for a completion rate of about 83%.

8. Note that we do not use a random utility function approach, but analyse the CV responses using a variation function approach based on the expenditure function (Cameron and James, 1987). However, the scale and location parameters can be used to derive their counterparts (constant and slope) under a random utility function interpretation. However, one needs not assume an identical functional form to compare estimates of benefits from two different samples and SP methods. For an analogue example of directly modelling WTP from CM data, see Train and Weeks (2005).

9. Estimates for more than two classes and for membership probabilities conditional on socio‐economic covariates are available from the authors.

10. Note that although the unconditional estimates for spread and location parameters for the assumed normal distribution of the speed parameter imply a probability of a negative taste of nearly 51%, conditional on the patterns of observed choices the number of predicted individuals with a negative conditional WTP (as computed using equation 5) is substantially lower (only 37.77%, or 156 cases).

11. As regards the other attributes, further tests show that, in order of significance, after speed, came cost and noise.