What is a life worth? : methodological issues in estimating the value of a statistical life

Abstract: This thesis addresses methodological issues in estimating the value of a statistical life (VSL). Two main approaches have been used to estimate the VSL, the indirect and direct methods; the indirect method is based on revealed preferences, such as the wage premium demanded for a risky job. The direct method is based on surveys where respondents are asked about their willingness to pay for risk reductions.There are several methodological problems with both approaches. This thesis mainly concentrates on methodological problems with the direct approach; more specifically with the contingent valuation approach. Some problems are universal for all survey based research in economics, but there are also specific problems in using surveys to ask respondents about the willingness to pay for a small risk reduction.The first paper concerns the findings that individuals often show inadequate sensitivity to the scale of the provided good, i.e. the willingness to pay is almost the same for a risk reduction of 0.0001 as for a risk reduction of 0.00005. The results in the first paper show that there is a systematic bias with respect to scale insensitivity such that individuals with lower score on a basic cognitive ability test are more likely to not show scale sensitivity. Hence, scale bias can be considered as (at least partly) a problem based on respondents lack of understanding of the small changes in probabilities. The second paper concerns the problem of hypothetical bias, i.e. that respondents usually overstate their willingness to pay in a survey scenario. The paper proposes that more reliable VSL estimates may be calculated if only the most certain respondents are included in the final calculations. It has been shown in social psychology that certain (uncertain) respondents have a strong (weak) attitude-behavior relationship. In several recent lab and field experiments in economics; it has been shown that if the uncertain respondents that say they will buy the good are excluded: hypothetical surveys answers match individual behaviour in a “real market situation”. The results in the paper indicates that value of life estimates are approximately reduced by half if only trusting the answers from the certain respondents, and only trusting the certain respondents creates almost identical VSL estimates from two completely different surveys at around 20 million Swedish kronor (SEK).The third paper focuses on the consistent findings that the value of a statistical life is higher (lower) if the hypothetical good is described as a private (public) abstract good. We hypothesize that this is not, as previously discussed in the literature, dependent on different types of altruistic behaviour, but is dependent on pure preferences regarding the provider of the good. Results indicate that the private VSL is much higher among respondents that dislike public provision of goods in general, hence, even if the risk reductions are of equal magnitude, the goods are not considered equal dependent on the provider of the good.The fourth paper estimates the value of a statistical life based on stated willingness to pay and on stated behaviour regarding seat belt and bicycle helmet use. VSL estimates are in the range of 38 to 75 million SEK (highest for stated WTP). Further, as a validity test, it is found that individuals that state a high willingness to pay are also, as theoretically expected, more likely to wear seat belts. The fifth paper in the thesis differs from the first four, in that it rather concerns how the value of life estimates can be put to use in other contexts than only cost-benefit analysis. It estimates the relationship between changes in the unemployment rate and heart attack incidence, lethality and mortality. Results, based on regional panel data analysis in Sweden, indicate that among individuals aged 20-49 an increase in the unemployment rate increases heart attack incidence and mortality. Using value of life estimates, it is estimated that a one percentage point increase in the unemployment rate has a monetized “health cost” of 350 million SEK. The findings may be used to consider a broader societal cost of unemployment.