Scale heterogeneity in the valuation of road traffic risk reductions: the case of Newfoundland's moose-vehicle collisions

ABSTRACT

The willingness to pay (WTP) for reductions in the risk of moose-vehicle collision in Newfoundland (the insular portion of Newfoundland and Labrador, Canada) is estimated using the Contingent Valuation Method. Our estimations use the information obtained from a double-bounded payment format and let us examine the existence of scale heterogeneity (dependent on response certainty levels), question effects such as anchoring and shift effects, and the effects of accounting for the former on the latter. Our main findings are that the estimated WTP tends to be lower in the models that account for scale heterogeneity but that correction makes little difference in models that fully correct for the question effects involved in the use of the double-bound payment format. Accounting for scale heterogeneity does change, however, the way in which corrections for question effects influence the size of welfare estimates. In particular, it reduces the variability of welfare measures across treatments of these question effects, yielding an estimate of WTP close to the one obtained from using the single-bounded portion of the data.

KEYWORDS:

• Moose vehicle collisions
• risk reduction
• contingent valuation
• value of a statistical life
• scale heterogeneity
• heterogeneous anchoring

Acknowledgments

We thank Henrik Andersson, Justin Quinton, and María Á. García Valiñas for their valuable contributions during the development of the survey instrument and suggestions made during preliminary data analysis.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Roberto Martínez-Espiñeira http://orcid.org/0000-0001-8022-1230

Notes

1 See, for example, Schwabe and Schuhmann (2002) and Schwabe, Schuhmann, and Tonkovich (2002) for very useful analyses of how to estimate the costs of WVCs and the value of wildlife, in their case deer in areas of the US, lost to collisions with vehicles.

2 See Section 3.2.3 in Andersson and Treich (2011) for a review.

3 Note that, in this context, the issue was insensitivity and non-near-proportionality of the respondents' WTP to the size of the risk reduction, a pervasive issue in the safety valuation literature. The quantity of the good provided is often referred to as the ‘scale’ or, less often, the ‘scope’ of the good provided. We use ‘scope’ in this paper when referring to the issue of ‘embedding’ Andersson and Svensson (2008) focused on, in order to avoid any confusion with the term that we use to refer to the scale of the WTP function.

4 Further details based on the first wave of fieldwork about the survey effort can be found in Martínez-Espineira and Lyssenko (2014).

5 We acknowledge the limited realism potentially conveyed by some of our valuation scenarios. We note, however, that our treatment of protest responses (described in detail in Martínez-Espineira and Lyssenko 2014) should have identified most, if definitely not all, of those cases in which respondents provided a negative response to the payment questions because they found the proposed risk reduction strategy implausible or unrealistic.

6 As explained in detail in Martínez-Espineira and Lyssenko (2014), in a few cases, missing responses to the follow-up question allowed to impute a doubled or halved bid value, resulting in a few cases with, for instance, the value of $500 as firstbid.

7 Further details about the distribution of response patterns by initial bid can be found in Table A1 in Appendix 1.

8 The full text of the valuation scenarios is available in Appendix 1.

9 Values of diffM and diffI equal to zero would have prevented the calculation of a logarithmic value, so we added a 0.001 to the original values of the scope variables.

10 In those private version cases in which the respondent did not provide a subjective estimate, the corresponding objective average level of baseline risk was applied instead.

11 Note that this accounts for the fact that ‘stacking’ the two DBDC responses from Versions E and F about doubled the number of observations from that version.

12 This report contains information on the data from the first wave of surveying but the differences across waves are negligible in terms of the treatment of the data.

13 However, since Halvorsen and Sælensminde (1998) did not use separate samples, the endogeneity of the open-ended question may be contaminating the comparison of the two types of question (Cameron et al. 2002).

14 Carlsson and Johansson-Stenman (2010) criticized, though, the conclusions in Haab, Ju-Chin, and Whitehead (1999) in a more recent contribution.

15 This study exploits the notion that the standard deviation of a transitory error component for each respondent is a function of the squared distance between the current bid (for each question) and average WTP, the ‘cost distance’ (Alberini, Kanninen, and Carson 1997).

16 Whether one would rather discard the information provided by the second response (at least in cases where significant question effects have been detected) is a matter of judgement. It may be still desirable to use the DBDC format, after accounting for the tradeoff between the likely downward bias generated by the question effects and the tighter confidence intervals afforded by the use of additional information from the second response (Alberini 1995).

17 After the elimination of ‘don't know’ responses.

18 We focus on point estimates throughout in order to simplify the discussion of results.

19 These tests were run on completely unweighted regressions, since a likelihood-ratio test is not as reliable when weights are used. Tests conducted forcing weights into the test resulted in different statistics but yielded similar conclusions.

20 The potential endogeneity of SUV is not considered in the paper, since we include this variable as a control rather than to make any type of causal inference about the impact of SUV on WTP. Following Angrist and Pischke (2008, 68), this type of variable can be referred to as proxy control variable, in the sense that we include it in the regression in order to serve as a measure of the observed WTP for MVC risk reductions and in order to avoid omitted variable bias. While including this variable would not generate a relevant regression coefficient, it may be an improvement over the alternative of using no control.

21 The pretest helped identify an expansion of the bid vector to better capture the distribution of WTP. Other than that, there ended up being no differences between the pretest questionnaire and the final questionnaire.

22 Because our response rate is relatively low, although in line with similar studies that used phone surveys, we are cautious about the generalizability of our results. Further, we used sampling weights to obtain results more generalizable to the provincial population.

23 As explained below, only frequent drivers were eligible for the second wave's survey instrument, so interviews with a further 39 individuals had to be terminated.

24 Those who received the private good question first were assigned a value of one for variable privatefirst.

25 With RM taking a random value from the vector [4,6,8,10,12] and MULTI taking the value of either 2, 3, or 4, as explained above.

26 This sentence in square brackets is a randomized provision constraint mechanism applied to 50% of respondents in Versions S, D, and E. This resulted in variable referendumreminder.

27 Or RM if Q12 was missing.

28 With RM taking a random value from the vector [4,6,8,10,12] and MULTI taking the value of either 2, 3, or 4, as explained above.

29 The interviewer, if asked, stressed that the level of risk would revert to Q12 (or RM is Q12 was missing) in 100,000 after discontinuing the use of the device.

30 See Footnote 29.

31 The highest three bids were only proposed in Version F, during the second wave of interviewing.

32 About half of the respondents were proposed a provincial program instead.

33 With RM taking a random value from the vector [4,6,8,10,12] and MULTI taking the value of either 2, 3, or 4, as explained above.

34 Where RI=RM $\cdot 30.$

35 About half of the respondents were proposed a program run by the Provincial government instead.

Additional information

Funding

This work was supported by the Collaborative Applied Research in Economics (CARE) initiative at the Department of Economics of Memorial University of Newfoundland under grant numbers CARE-004 and CARE-019.