Abstract
This study reports the results from a discrete choice experiment conducted in Beijing China. It aims to elicit monetary values for the value of a statistical life (VSL) and the value of a statistical illness (VSI) that can be considered for policy purposes in China, and to examine how different payment regimes influence willingness to pay (WTP) and whether WTP is age-dependent. We find that our estimates are robust between different econometric model specifications and that they are reliable when compared to previous Chinese findings. We find no evidence of any VSL–age relationship but we find that the payment scheme had an effect on the levels of the estimates of the VSL and VSI, and that taking into account the payment regimes when estimating the models improved their performance. However, levels were relatively close and not statistically significantly different for VSL which may suggest that respondents considered both schemes as similar.
Acknowledgements
We are grateful to Xuxuan Xie for his contribution on the survey and to James Hammitt, Christoph Rheinberger and Jie He for their comments. We thank all the members of the Environmental Economics and Policy Study Group (EEPS) of Peking University for data collection. Two anonymous referees provided helpful comments and suggestions. All errors are our own.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. For the full description of the models, see Bergstrom, Boyle, and Yabe (Citation2004).
2. For instance, Swallow and McGonagle (Citation2006) assumed that each respondent “interprets the use of [reallocations] as being drawn from all publicly provided goods in proportion to their existing share of the public budget” (59).
3. Here, the opportunity cost indicates which part(s) of the existing tax resources that will be reallocated, e.g. reducing funding for health, education, transport, etc., in order to increase the quantity provided of the good defined in the survey.
4. For a more detailed description see, e.g., Johansson (Citation2002).
5. Either face-to-face in individuals’ homes or a web-based survey based on stratified sampling would have been preferred modes in most contexts. However, we were concerned about the representativeness of the sample using these modes (e.g. a very low internet penetration at the time of the survey in China and also the low response rate of mail surveys). Moreover, it would have been hard to obtain authorization to conduct in home interviews based on a random or a stratified sample from the Beijing authorities, and most of the Chinese people do not like being interviewed at home by strangers due to cultural reasons and also for security reasons.
6. There is evidence that individuals have difficulties understanding small probabilities, but that visual aids can help them understand (Corso, Hammitt, and Graham Citation2001).
7. To clarify, this morbidity risk was not based on self-reported baseline levels and in the survey design Morbidity and the risk changes in ΔCold are orthogonal.
8. To construct the choice sets, the SAS program was used.
9. See Fiebig et al. (Citation2010) and Hole (Citation2008) for descriptions of these models.
10. Stata was used to run the regressions. For cases 2, 3, and 4, we used the commands clogithet (http://econpapers.repec.org/software/bocbocode/s456737.htm), mixlogit (Hole, Citation2007), and GMNL (Gu, Hole, and Knox Citation2013).
11. Latent class models have, for instance, been used in DCE to examine non-attendance to attributes (e.g. Campbell, Hensher, and Scarpa Citation2011).
12. The scoring system was created prior to the interviews being conducted. After the data had been collected, a random control was conducted by one of the involved researchers and for all checked observations no evidence was found suggesting that the scoring system did not serve its purpose.
13. These findings were robust to different specifications to how age was treated in the regressions, such as different age group definitions, continuous age variables, etc.
14. A mixed logit model based on Model Mix3 with age and income interactions did not converge, and hence we are not able to provide a mixed logit model that corresponds to Model C4.