It is important to understand the drivers of VSL levels in order to be able to better adjust the values to reflect the relevant policy problem and geographic location. This chapter considers potential factors that could be used to adjust value of statistical life (VSL) estimates for application in specific contexts. It uses the updated meta-data to examine the relationship between income and VSL and assess its use as an adjustment factor. It also presents the results of meta-regression analyses to assess the evidence for adjusting VSL estimates based on the quality of primary valuation studies and risk type. Finally it also reports the results of several sensitivity analyses and reviews evidence from the literature regarding the impact of these and other factors on VSL estimates.
Mortality Risk Valuation in Policy Assessment
Abstract
A number of socio-demographic and study-related factors can impact VSL estimates. This chapter considers potential factors that could be used to adjust VSL estimates for application in specific contexts. Section 5.2 uses the updated meta-data to examine the relationship between income and VSL, i.e. the income elasticity of VSL, and assess its use as an adjustment factor. Section 5.3 presents the results of meta-regression analyses to assess the evidence for adjusting VSL estimates based on the quality of primary valuation studies and risk type. Section 5.3 also presents sensitivity analyses of several aspects of the data. Finally, Section 5.4 reviews evidence from the literature regarding the impact of these and other factors on VSL estimates.
As discussed in Chapter 2, willingness to pay (WTP) for mortality risk reductions is expected to increase with the income of the affected adult population. This means that VSL estimates can also be expected to increase with income, a relationship that is typically measured by what’s known as the income elasticity of VSL. This elasticity expresses the statistically derived percentage increase in VSL associated with a one percent increase in income. As noted in Chapter 2, the income elasticity of VSL is important for two reasons. First, in order to accurately assess the costs and benefits of policies that have impacts across populations (typically across countries), mortality risk values must be adjusted to reflect differences in income levels across countries. Second, for policies with long-lasting effects, it is necessary to estimate how the value of mortality risk reductions increase over time as income increases. Note that there is generally a lack of information from primary valuation studies about the mean household or individual income of the sample from which VSL estimates are derived. An alternative source of income data can be found in official statistics of gross adjusted disposable income (OECD, 2023[1]). However, these data are often not available in all years for all countries. Hence, GDP per capita or Gross National Income (GNI) per capita is often used as a proxy for income1. GDP per capita is also a commonly used measure for undertaking benefit transfer analyses across policy contexts.
5.2.1. Main estimation models
Several characteristics of the meta-data present challenges for estimating the relationship between VSL estimates and GDP per capita. For example, limited variation in income levels exists among high-income countries as compared to low-income countries (cf. Figure 5.1). At the same time, most VSL estimates are from high- or middle-income countries and may also vary for reasons other than differences in GDP. Many countries in the dataset are also characterised by no or very few combinations of VSL estimates and GDP per capita for any given year. This analysis utilises the full dataset spanning 1970 to 2023 to assess variation in income and VSL estimates over time and across countries, as both dimensions are relevant for estimating a statistical relationship between income and VSL.
Note: Points reflect study-level average VSL estimates after removing far-outs. GDP per capita and VSL are expressed in 2022 USD using the same conversion procedure as for primary VSL estimates, cf. Annex B.
Table 5.1 reports the results of five meta-regression models using the same approach as described in Section 4.4 of Chapter 4, but introducing additional explanatory regressors into the multilevel model to estimate the true population estimate of VSL as follows (see Box 4.1 in Chapter 4 for a definition of components):
Equation 5.1
,
,
,
,
.
The dependent variable in Equation 5.1 is the logarithm of VSL and the covariate is a vector of explanatory variables. All models include the logarithm of GDP per capita as an explanatory variable within this vector. Since the model is log-log, the estimated coefficients of GDP per capita in Table 5.1 can be interpreted as the income elasticity of VSL. Models 3-5 include additional controls such as types of SP and RP methods and years since data collection. Models 4 and 5 control for fixed effects at the country and year levels, respectively. The estimate of income elasticity of VSL in Model 4 is estimated using within-country variation in VSL estimates and income levels, and therefore on the basis of variation in these factors over time within each country. In Model 5, on the other hand, the estimate of the income elasticity of VSL is estimated using within-year variation across countries.
Models 1 and 2 yield similar estimates of the income elasticity of VSL of around 1.1, both of which are statistically significant. Model 3 controls for types of SP and RP methods and years since data collection and yields a statistically significant estimate of the income elasticity of VSL of approximately 1.
Model 5 yields a slightly higher, statistically significant estimate of 1.1, and Model 4 yields a higher estimate of 1.4, although this value is not statistically significant. These results suggest that the income elasticity of VSL over time may be higher than the income elasticity of VSL across countries, although this finding does not appear robust and is difficult to detect in the full dataset 1970-2023 dataset used here.
Dependent variable = log(VSL), sample of all meta-data
Note: The table displays coefficients as well as their confidence intervals (in brackets). Stars indicate statistical significance at the 10% level (*); at the 5% level (**); and at the 1% level (***). 93 negative VSL estimates are not included due to log transformation (as taking the log of negative numbers is not possible). Model 4: Elasticity estimates are identified by within-country variation of (49-18) countries, as information from 18 countries is not used. Model 5: Elasticity estimates are identified by within-year variation of (51-14) years, as information from 14 years is not used. FE stands for fixed effects. Log (GDP) is log (GDP per capita). Details regarding variable definitions are provided in Table 4.1 of Chapter 4.
5.2.2. Alternative models and robustness checks
A number of alternative estimations and robustness checks regarding the estimation of the income elasticity of VSL are included in Annex E. A summary of the most important results is provided below:
New SP and RP: Basing the estimations shown in Table 5.1 on only the new RP and SP data leads to slightly higher estimates of the income elasticity of VSL, ranging from 1.12 to 1.25, all of which are statistically significant. The exception is Model 4 (reflecting within-country variation over time), which yields an insignificant elasticity estimate of 3.5 (Table A F.1 in Annex F).
Old SP and RP: Basing the estimations shown in Table 5.1 on only the old RP and SP data yields similar estimates of the income elasticity of VSL as using the entire dataset, ranging from 0.98 to 1.14, all of which are statistically significant, except for Model 4 (reflecting within-country variation over time), which yields an insignificant elasticity estimate of 1.58 (Table A F.2 in Annex F).
Comparing SP and RP: When estimating the models using SP and RP data separately, the SP data yields a systematically higher income elasticity of VSL (1.33 to 1.45, all statistically significant) than RP (0.71 to 1.35, all statistically significant). Model 4 yields insignificant estimates of 0.16 and 3.1 for SP and RP, respectively (Table A F.3 and Table A F.4 in Annex F).
Elasticity by country and income groups: Splitting the data by the country groups defined in Section 4.5 of Chapter 4 and running one regression model for each group separately generates large variations in income elasticity estimates, although most are statistically insignificant, and the lower sample sizes in these models reduce their statistical power. For the high-income group, the estimates vary from -0.6 to 1.5, indicating a lower elasticity within this group than when considering the whole sample. Estimates of the income elasticity of VSL among low- and middle-income country groups are significant and range from 1.16 to 2.14. These results suggest that the income elasticity of VSL for lower income countries may be higher than it is for higher income countries (Table A F.5 in Annex F).
Alternative econometric specifications: Alternative model specifications, such as the inclusion of interaction effects between GDP and higher than average income (Table A F.6 in Annex F) and specifications to control for study-level effects (Table A F.7 in Annex F) were also examined, yielding relatively robust results across specifications.
From the results presented here and in Annex F, estimates of the income elasticity of VSL for the full dataset range between 0.5 to 1.5, with most estimates at or slightly greater than 1. Estimates of the income elasticity of VSL are more likely to be identified by cross-country variations (by including fixed effects as proxies for year-specific unobservable variables) and less likely to be identified by within-country variation. This is because only a subset of estimates in the meta-data can be used to estimate models reflecting within-country variation over multiple years. As a result, the estimates of income elasticity of VSL based on models reflecting variation within countries over time, such as Model 4, exploit only 40 percent of the meta-data, which come largely from a few countries in which many studies were carried out. For this reason, estimates of the income elasticity of VSL from these models should be considered less globally representative than estimates from models based on variation across countries.
Given the large variation in the estimates of income elasticity of VSL across countries as estimated in models by country and income group and the lack of sufficient variation in VSL estimates within countries over time, it is challenging to identify suitable and robust elasticity estimates that could be used for specific elasticity estimates by country group (such as those defined in Section 4.5 of Chapter 4) or that can be applied over time. This issue is discussed further with respect to other evidence from the literature in Section 5.4.2 and with respect to recommendations for the use of the income elasticity of VSL in policy assessments in Chapter 6.
5.3.1. Quality of primary valuation studies
This section presents the results of meta-regressions that assess the extent to which a number of additional factors impact VSL estimates. The results below are based on the three-level random effects meta-regression model used to estimate the relationship between income and VSL estimates (Equation 5.1) but employ a linear dependent variable VSL (rather than log) in order to retain negative VSL estimates in the analysis. Table 5.2 reports the results of six models with different sets of covariates, estimated using the pooled new SP and RP data.
As can be seen in Table 5.2, the coefficient of income (GDP per capita) is statistically significant and positive in all five models in which it is included2. Estimates from both Contingent Valuation (CV) and Consumer Markets (CM) methods yield significantly lower VSL estimates compared to Hedonic Wage (HW) methods (the base category) for all models. Estimates based on Discrete Choice Experiments(CE) methods are not statistically different from those based on HW methods. The coefficient for years since data collection is statistically significant in all four models. This indicates a trend of declining VSL estimates over time, even when controlling for GDP and valuation methods3.
Whether the study is published in a peer-reviewed journal is statistically significant and positive in Models 3-5, and whether the target population is nationwide is statistically significant and positive only in Model 6. Other statistically significant variables include the Quality Score variables 2 and 3. Quality Scores are intended to serve as a rough measure of the scientific quality of the underlying primary valuation studies (cf. Sections 4.3.3 and 4.6 of Chapter 4). The Quality Scores for SP and RP are defined as:
Where the right-hand side elements are dummy variables representing the presence or absence of various quality characteristics in SP and RP studies. , , and reflect whether the survey clearly defines the baseline risk, reports a scope test and uses visual aids to explain risk, respectively. , and indicate whether VSL estimates were estimated using an instrumental variable for fatality rate, whether the estimating regression controls for non-fatal risk of injury and whether the study relied on the Census of Fatal Occupational Injuries (CFOI) as a source for risk data. Further discussion of quality aspects is provided in Section 4.7 of Chapter 4.
The coefficients of Quality Scores taking values 2 and 3 are positive and statistically significant in Models 5 and 6. This indicates that studies with these quality characteristics tend to yield higher VSL estimates, although the reasons for this result are unclear. Whether studies target the general population (“General Population”), use random sampling strategies (”Representative”) or whether standard errors (SE) are reported (“No Reported Standard Errors”) do not have a statistically significant effect on VSL estimates across models, nor do the gender variables (“Only Male”, “Only Female”), which reflect whether the study surveyed only male or female respondents, respectively. Excluding GDP per Capita in Model 6 does not significantly change the above results, except that the variable “Nationwide” now becomes positive and statistically significant.
Dependent variable = VSL (USD million), New RP and SP data pooled
Note: This table presents regression results from Models 1-6. For each variable, the table displays its regression coefficients and corresponding confidence intervals in brackets. Stars indicate statistical significance at 10% (*); at 5% (**); at 1% (***) levels. Details regarding variable definitions are provided in Table 4.1 of Chapter 4.
5.3.2. Types of risk evaluated
Table 5.3 reports the results of six additional models estimated using the new SP and RP data that include covariates reflecting different risk types studied (Models 1-2 and 5-6), as well as information on whether the studied risks were acute (traumatic death) or chronic (a period of e.g. illness before death) (Models 3-4). Results indicate that statistically significant risk types are cancer (significant and positive across all models), military risk (significant and negative in Models 1 and 2), acute risk (significant and positive in Model 3) and mixed risk (significant and positive in Models 3 and 4). The other risk types (e.g. climate, virus, transportation-risks) are not significant in any model. The impact of traumatic/acute risk is not robust and only statistically significant at the 10% level in Model 3. GDP per capita continues to be statistically significant and positive across all models.
Dependent variable = VSL (USD million), New RP and SP dataset pooled
Note: This table presents regression results from Models 1-6. For each variable, the table displays its regression coefficients and corresponding confidence intervals in brackets. Stars indicate statistical significance at 10% (*); at 5% (**); at 1% (***) levels. Risk types covered include Cancer; Climate: extreme weather events, sea level rise, depletion of fisheries, heat stroke; Disaster: human-caused disaster (e.g. nuclear accident); Environment: environmental hazards other than climate events and natural disasters (e.g. air pollution); Health: disease other than cancer (e.g. heart failure, stroke); Mixed: a combination of any of the above risks; Natural Disaster: avalanches, earthquakes and floods; Transportation: transport-related fatalities; Virus: contagious diseases such as COVID-19, dengue fever or rabies. Details regarding variable definitions are provided in Table 4.1 of Chapter 4.
5.3.3. Sensitivity analyses
Annex F provides the results of several sensitivity analyses, including results from using all RP and SP data, using each specific dataset separately (New RP and New SP), and accounting for impacts of mortality risk types.
The results in Table A F.8 in Annex F indicate that carrying out the same regression as in Table 5.2, but using all RP and SP data (rather than new RP and SP data only), yields results that are similar to those reported in Table 5.2 in terms of statistical significance, although coefficient values differ somewhat. When SP data is assessed separately (Table A F.9 in Annex F), a number of other variables are statistically significant in addition to GDP per capita (which is significant in all models). In addition, CE methods generally yield higher VSL estimates than CV methods (statistically significant in three out of four models), as do journal publications (statistically significant in two out of four models). Random sampling, as measured by the “Representative” variable, is statistically significant and positive in one out of the two models in which it is included. The annualised risk change is statistically significant and strongly negative where it is included in Models 3 and 4. Although theory suggests that VSL estimates should be independent of this risk change (Lindhjem et al., 2011[2]), this result is typical for SP studies. Similar results are also reported in a meta-regression analysis of Canadian studies (Ginbo, Adamowicz and Lloyd-Smith, 2023[3]). The only risk type variable that is statistically significant, positive and robust across all four models is cancer risk (Models 1-2, 5-6 in Table A F.10 in Annex F). The variable for acute risk is also statistically significant and positive, but only in two out of four models in which it is included (Models 1-2 in Table A F.10 in Annex F)
A separate analysis of RP data (Table A F.11 in Annex F) indicates that GDP per capita is statistically significant and positive in all models. The only other variable that is statistically significant and robust across models (though negative) is “RP with Full Sample”. This indicates that VSL estimates are lower when estimated based on a representative sample of employed individuals compared to sample of individuals employed in specific sectors or job types. For the analysis of RP data including risk categories (Table A F.12 in Annex F), the risk types “disaster”, “health”, “environment”, “acute” and “mixed” are all significant and negative in at least two models. The reasons for these results are unclear and the results are also not robust across models.
Taken together, the meta-regression results presented above do not provide sufficiently robust results for any variables beyond income (GDP per capita) that would justify their use in adjusting base VSL estimates. One potential exception could be cancer risk. However, adjusting VSL estimates for this type of risk is difficult to operationalise based on the relatively limited information available in the meta-data (e.g. whether cancer was mentioned in the study or what type of cancer was mentioned).
This section provides an updated review of comparable meta-analyses, review papers and primary valuation studies on how factors such as age, adult vs. child mortality, pandemics and other large-scale events impact VSL estimates. The purpose of this review is to better understand the state of the current evidence on several important factors that could potentially be used to adjust base VSL estimates for application in specific policy contexts.4 This is particularly relevant considering the data limitations of the current analysis (e.g. few studies consistently report information about all of these factors) and the heterogenous nature of the meta dataset that prevents a full investigation of such factors using the meta-regressions presented in Sections 5.2 and 5.3.
It is important to note that even if certain variables have been found to have statistically significant and robust impacts on VSL estimates in one or more studies or countries, these results do not necessarily indicate that such factors should be systematically used to adjust base VSL estimates. Determining which adjustment factors to use will also depend on other concerns, such as whether information about the factors is available for the policy context or country in which VSL estimates will be applied, whether trade-offs exist between equity and ethical concerns, and whether the simplicity and transparency of the guidance is considered important. These considerations are discussed further in Chapter 6.
5.4.1. Results from other meta-analyses, reviews and primary valuation studies
Several notable quantitative reviews using meta-analysis methods have been carried out since 2012. Some studies cover individual countries, including USEPA (2016[4]) and Newbold et al. (Newbold et al., 2024[5])5 (in the United States), Ginbo et al. (2023[3]) (in Canada) and Wang et al. (Wang et al., 2024[6]) (in China). The aim of these studies is typically to derive VSL recommendations for policy evaluation within a given country, and some also investigate methodological issues (e.g. which factors help to explain variation in VSL estimates).
USEPA (2016[4]) conducts a meta-analysis study in the United States to derive updated VSL estimates for use in assessing the mortality impacts of environmental policies. The study uses various screening criteria for SP and RP studies, and different statistical (parametric and non-parametric) methods to derive mean VSL estimates. A number of procedures used were later criticised by a scientific advisory board reviewing the study (SAB/USEPA, 2017[7]). Noting this caveat, the study recommends a VSL estimate for the general adult population in the United States of USD2013 10.3 million6. USEPA (2016[4]) does not recommend any adjustments for risk characteristics or other factors (except for increasing income over time). The authors, “leave the task of estimating adjustment factors to account for the influence of risk and individual characteristics on VSL estimates, possibly through inclusion of additional control variables in the meta-regression model, for future work” (USEPA, 2016, p. 9[4]). Cropper, Joiner and Krupnick (2024[8]) review the recent HW, CM and SP literature of relevance in the United States and conclude that recent evidence is “sufficiently rich to permit a revision of EPA’s baseline estimate”.
Banzhaf (2022[9]) conducts a meta-analysis of previous meta-analysis studies from the United States to derive a VSL estimate for policy use. The baseline statistical model yields a central VSL estimate of USD2019 8 million from the included meta-analysis studies, with a 90% confidence interval of 2.4 to 14.0 million. This is somewhat lower than USEPA (2016[4]), taking into account that their estimate is in USD2013.
Ginbo et al. (2023[3]) provide an updated measure of VSL (specifically the value of reduced mortality risk, VRMR) in Canada. The analysis is based on 18 primary studies published between 1989 and 2018, yielding 158 VSL estimates. The authors select nine studies for inclusion in the meta-analysis that were considered to have samples representative of the country’s population7. The authors use a weighted least squares approach with clustered errors and panel data regression procedures. The analysis yields a mean VSL estimate of CAD2020 13 million (ca. USD2020 9.5 million), while the lower and higher values from two alternative valuation methods are about CAD 10 million and CAD 16.5 million, respectively8. This updated mean VSL estimate is 60% higher than VSL recommended by Canadian as of 2024, which according to the authors, can mainly be attributed to risk preferences that change over time and advancements in empirical methods. Further, the authors find that the levels of baseline risk and risk reduction are among the main determinants of VSL estimates. Considerable variation exists in mean VSL estimates, e.g. depending on the type of data source used. As noted in Chapter 4, Ginbo et al. (2023[3]) find that SP studies yield VSL estimates that are about two times higher than those from RP studies.
Wang et al. (2024[6]) conduct a meta-analysis of 19 primary SP surveys of the Chinese population between 1998 and 2019. For government programs in China, they recommend VSL estimates of CNY2020 5.18 million, 11.64 million and 1.68 million (equivalent to USD2020 1.24 million, 2.8 million and 0.4 million) for environmental, health and transport related mortality risks, respectively. Income, as well as the type and magnitude of risk reduction, are the most important factors influencing VSL estimates. Unlike research conducted in developed countries, they find no evidence of publication bias.
In addition to the meta-analysis studies above, several other reviews of the VSL literature are worth mentioning, notably the global systematic review by Keller et al. (2021[10]) of 120 global VSL studies (both SP and RP) published between 2009 and 2019. The authors do not use statistical meta-analysis methods to derive central tendencies and conclude that: “Estimates for VSL varied substantially by context (sector, developed/developing country, socio-economic status, etc), with the median of midpoint purchasing power parity-adjusted estimates of USD2019 5.7 million, 6.8 million, 8.7 million, and 5.3 million for health, labour market and transportation safety sectors, respectively” (Keller et al., 2021, p. 1521[10]). These estimates are roughly comparable to the raw mean and median VSL estimates at the estimate and study levels reported in Section 3.3 of Chapter 3. From these findings, the authors further conclude that the large variation observed in the reported VSL estimates depends mainly on the context rather than the method used (although this is difficult to verify without further statistical analysis, e.g. meta-regression analysis).
As noted in Chapter 1, other reviews of the VSL literature with more limited scope include Bahamonde-Birke et al. (2015[11]) focusing on road accidents, Cropper et al. (2024[8]) on US EPA-relevant studies in the United States, Kniesner and Viscusi (2019[12]) on generic VSLs, Robinson and Hammitt (2015[13]; 2016[14]) who carry out reviews with a focus on policy implications, and Ananthapavan et al. (2021[15]) who undertake a systematic review for Australia. These studies demonstrate a heterogenous and diverse literature in terms of risk causes, contexts, methods and VSL estimates. This report draws on these studies further in the following sections, especially when considering recommended VSL estimates for policy use.
5.4.2. Income
A number of studies investigate the income elasticity of VSL within and/or across countries. Masterman and Viscusi (2018[16]) estimate income elasticities of VSL of around 0.94 to 1.05 based on global SP studies. They also find evidence that elasticities may decrease with income, suggesting an elasticity of around 1 for lower-income countries and 0.55 for higher-income countries. Lindhjem et al. (2011[2]) analyse almost identical data to the new SP data used in the current analysis and find elasticities of between 0.7 and 0.9, with significantly lower elasticities for quality-screened models that use similar SP surveys.
Based on a global dataset of HW studies, Viscusi and Masterman (2017[17]) estimate an elasticity of VSL of between 0.5 and 0.7 for the United States, and just above 1 for other countries. They also find evidence that much of the observed disparity in elasticities across countries is attributable to income differences, as elasticity estimates are higher for lower-income populations.
Based on HW data, Kniesner et al. (2010[18]) also find higher income elasticity of VSL values for lower-income deciles (ranging from around 2.2 in the lowest income decile to 1.2 in the highest), except within the United States. In their discussion, Robinson, Hammitt and O’Keeffe (2019, p. 24[19]) note that while studies published before 2010 found elasticities ranging from as low as 0.1 to 2.0, more recent reviews have found elasticities closer to 1.0, and they recommend using elasticities around 0.8 for extrapolating transfers between high-income countries and between 1.0 and 1.2 for lower-income countries.
There are a number of challenges in measuring the income elasticity of VSL, especially over time, as such an analysis is ideally informed by a representative longitudinal panel of individuals whose willingness to pay for mortality risk reductions is measured over time, data that is not available. USEPA (2017[7]) discusses some of the challenges of estimating elasticities, especially over time, including the use of longitudinal vs cross-sectional data and the challenge of accurately measuring income in surveys. Another important consideration is that even if GDP per capita grows (often a proxy for income across countries), individual incomes may not change in the same way. This is the case, for example, when an increase in mean income is accompanied by an increase in income inequality.
According to Hammitt et al. (2022[20]) only two studies have estimated VSL for a population over time, and both found that VSL estimates increased more rapidly than income over time, suggesting an income elasticity of VSL larger than 1. Costa and Kahn (2004[21]) used HW data from 1940-1980 and GDP as proxy for income to estimate an implied income elasticity of 1.5 to 2.0. Hammitt et al. (2022[20]) notes that estimating the income elasticity of VSL is vulnerable to the endogeneity problem associated with an individual worker’s VSL because it reflects their choice of job risk and wage, which is determined by optimising over the set of available jobs she is qualified for.
In SP studies that were performed in China in 2005 and repeated in 2016, Hammitt et al. (2019[22]) find an implied income elasticity of 3. The elasticity is implied insofar as it assumes that the preference (utility) function is stable over time. For example, if individuals become more concerned about safety over time (unrelated to income), this may exaggerate the importance of income growth when accounting for why VSL has increased over time. Hammitt et al. (2022[20]) use HW data from Chinese Taipei from 1982-1997 to estimate income elasticities of 2 to 5, assuming that all of the observed differences in VSL estimates compared to income are attributable to income growth. Controlling for the endogeneity of the data as noted above instead yields much lower elasticity estimates of between 0.65 and 0.91.
Alberini and Ščasný (2021[23]), investigate increases in VSL and income over time by comparing two identical CV surveys carried out five years apart in the Czech Republic, estimating an income elasticity of VSL of ca. 0.77.
In summary, the cross-country results presented in Section 5.2 are similar to those found in other studies, with income elasticities of VSL ranging from 0.5 to 1.5, and the majority of estimates around 1. While there are some indications that the income elasticity of VSL may be higher for lower-income countries (and lower for higher-income countries), including in the meta-data of the present report, it is methodologically challenging to discern changes in elasticity estimates within countries over time. As is shown in the sensitivity analysis in Annex F, elasticity estimates within country groups lack robustness and depend on estimation approaches.
5.4.3. Age (of adults)
As noted in OECD (2012[24]) a significant controversy erupted in the United States over the so-called “senior discount”, i.e. the practice of downward adjustment of VSL estimates for older individuals. Because environmental policies often reduce risks more for the very young or the very old, the use of differentiated VSL estimates with respect to age arose first in this sector. The US EPA employed a lower VSL for older individuals in sensitivity analyses conducted for air pollution rules prior to 2004, including the Clear Skies Initiative, where benefits to senior citizens constituted the majority of the policy benefits (Robinson, 2007[25]). Aldy and Viscusi (2007[26]) point out that whether VSL should vary by age is not a matter of equity or political expediency, but should rather be grounded in how individuals’ WTP for risk reductions vary with age. As people age, their life expectancy shortens, but their economic resources vary (and often increase) as well, giving rise to a theoretical indeterminacy in the age-VSL relationship (see also Viscusi, 2009[95]). Discussions about this issue have arisen again recently in many countries that have considered the costs and benefits of COVID-19 policies (see below).
While there is some empirical evidence that VSL declines at older ages, this relationship is uncertain and possibly follows an inverted U-shape (Hammitt, 2007[27]; Krupnick, 2007[28]; Viscusi and Aldy, 2007[26]). More recent studies and reviews have confirmed many of the previous findings. For example Brey and Pinto-Prades (2017[29]) find that different econometric approaches to the analysis of their SP data produce similar results, namely, an inverted-U relationship between VSL and age. Utilising RP data from car purchase decisions in the United States, O’Brien (2018[30]), also find a significant inverted U-shaped relationship between age and VSL, ranging from USD2009 1.9 to 19.2 million. Aldy (2019[31]) also find an inverted U-shape with age in a US context.
Other evidence, however, does not support this relationship. Herrera-Arujo and Rochaix (2020[32]), for example, find a negative relationship between age and VSL when analysing combined data on wage-differentials and detailed data on workers’ medical histories. Mitani, Liu and Lindhjem[108] (in preparation) find only a weak U-shaped relationship in a SP survey related to pandemic risk in Japan. The authors speculate that the high accumulation of wealth among the elderly in Japan may increase WTP (and thus VSL estimates) and/or the life expectancy of adults, making it difficult to capture a (potential) negative impact of age on VSL estimates for this population.
While adjusting VSL estimates for age may be perceived to be problematic for equity reasons, the evidence on this relationship is also still not fully clear in the literature (cf. also (Kniesner and Viscusi, 2019[12])). Even if the relationship between VSL and age is characterised by an inverted U-shape in some contexts, it is difficult to determine how an age adjustment could be executed in practice. As limited age information is available in the new RP and SP data used for this analysis, this issue has not been investigated in the current meta-analysis9. OECD (2012[24]) carried out a basic analysis of the old SP meta-data and found no clear relationship between age and VSL, although for a subset of the data indications of such a relationship were found insofar as VSL estimates increased with age to about 40-50 years and then declined.
5.4.4. Adults vs. children
As noted in OECD (2012[24]), VSL estimates appear to be higher for children, with evidence from the United States and Europe indicating that such estimates can be twice as high as that of adults (OECD, 2010[33]; USEPA, 2003[34]). More generally, in cases where a policy intervention particularly affects children, either because of the nature of the policy (e.g. pesticides in school grounds) or because children are particularly vulnerable to a specific hazard (e.g. lead in drinking water), child-specific VSL estimates are likely to be useful in ensuring that resources and policy efforts are allocated efficiently. Based on these arguments OECD (2012[24]) recommended that, for policies that target reducing risks for children or that reduce these risks indirectly, the VSL estimates used for children should be 1.5 to 2.0 times higher than those used for adults.
A more recent review by Keller et al. (2021[10]) found that VSL estimates for children were up to 4.66 times higher when parents were asked to value mortality risks for their children compared to themselves. Keller et al. (2021[10]) are unable to explain the factors behind the premium. Even though some recent evidence exists for “child premiums”, a number of studies consider the evidence for such an adjustment to be too limited to draw clear conclusions. While finding evidence of premiums, reviews by Raich et al. (2018[35]) and Robinson et al. (2019[36]), also note that more research is needed to accurately estimate a potential adjustment factor for VSL estimates for children. Kniesner and Viscusi (2024[37]) review the evidence regarding adjustments for VSL estimates (including age adjustments) and conclude that there are many reasons why it may be difficult to make such adjustments. These include both potential methodological limitations in the primary valuation studies and the “implausibility” of using this reasoning in practical policy analysis, e.g. applying a percent drop in VSL at 18 (the age of legal majority in the United States) compared to age 17, in the context of the US Consumer Product Safety Commission and the US EPA.
VSL estimates for children were screened out of the meta-data used for this report (cf. Chapter 3). Because the evidence remains limited and uncertainties are likely to be even larger when considering potential adjustment factors across countries, as well as extrapolating to countries beyond those where relevant evidence exists, the approach taken in this report is to use the same base VSL estimates for all age groups.
5.4.5. Pandemics and other large-scale events
An important issue that has become increasingly relevant is mortality risk valuation in the context of COVID-19 and other extreme events that could occur in the future (e.g. pandemic, natural disaster or climate change-induced) where risks are non-marginal and disproportionately affect specific groups or areas (Hammitt, 2020[38]; Viscusi, 2020[39]; Viscusi, 2021[40]). For example, there has been an intense debate over which VSL estimate to use to evaluate the health and lockdown policies introduced in the wake of the COVID-19 pandemic (Hammitt, 2020[38]). In addition, events like the COVID-19 pandemic may influence estimates obtained from SP studies conducted during these periods (Mourato and Shreedhar, 2021[41]), or may result in persistent changes to risk preferences that should be considered when valuing mortality risks in the future (Hanaoka, Shigeoka and Watanabe, 2018[42]). When investigating the possible impact of COVID-19 on VSL estimates in the meta-regression analysis in Section 5.3, the models did not detect any such effects, neither as a cause of risk nor as differences evident in the estimates produced by studies conducted during the pandemic (cf. Table 5.3).
Regarding which VSL estimate to use to value mortality risk reductions from COVID-19 policies, the main debate has centred around how to deal with the fact that most of these risk reductions accrue in older people (Hammitt, 2020[38]; Robinson, Sullivan and Shogren, 2021[43]), as the decision of whether or not to adjust VSL estimates for older individuals is consequential for the results of any CBA of COVID-19 policies. As noted in section above, VSL estimates may follow an inverted U-shaped relationship with age (Brey and Pinto-Prades, 2017[29]). However, even if this result is empirically supported, it could be considered unethical to value the lives of older people less than those of younger people. It should also be noted that COVID-19 policies have other direct and indirect effects on different population groups and their mortality risks from other causes, e.g. due to increased social isolation and mental health problems. Generally, the same set of principles should be considered and applied when deciding how to value all mortality risk changes from COVID-19.
Viscusi (2020[39]) estimates, for example, that the global mortality cost of COVID-19 (by July 2020) was around USD 3.5 trillion, but the estimate is substantially reduced if the shorter life expectancy and the lower income of COVID-19 victims relative to the general population are taken into account. The authors acknowledge the fundamental equity concerns of such methodological choices. Robinson et al. (2021[43]) investigate the impact of using three different approaches to adjusting VSL estimates for age, namely i) an invariant population-average VSL, ii) a constant value per statistical life-year (VSLY) and iii) a VSL that follows an inverse-U pattern, peaking in middle age. The authors find that when applied to the U.S. age distribution of COVID-19 deaths, these approaches result in average VSL estimates of USD 10.63 million, 4.47 million and 8.31 million, respectively, and that the choice of which of these VSL estimates are used affect the resulting policy conclusions regarding social distancing measures.
Hammitt (2020[38]) estimates the mortality costs of COVID-19 in the United States alone at USD 1 trillion using a constant (age independent) VSL estimate of USD 10 million. The author argues that the total mortality costs of the pandemic may be too high for two main reasons, namely that i) COVID-19 impacted the elderly the most and ii) the mortality risks of COVID-19 are non-marginal, yielding a smaller rate of substitution of wealth for risk reduction. On the other hand, other researchers argue that non-marginality should yield higher mortality costs due to the willingness to accept vs. willingness to pay disparity observed when mortality risk reductions are non-marginal (Colmer, 2020[44]). Taken together, this suggests that further consideration is warranted, especially considering concerns raised in the literature regarding non-marginal risks (Adler, 2020[45]). A related issue, recently considered in more general non-market valuation of health and environmental benefits, may be that the mortality (and morbidity) costs of a pandemic of this scale will yield non-marginal impacts, requiring considerations of general equilibrium effects in the valuation of mortality risks (and other non-market impacts) (Carbone and Smith, 2008[46]; Carbone and Smith, 2013[47]). It should be noted that the concept of VSL, as presented and discussed in this report is limited to the valuation of marginal changes to mortality risk only, and does not include morbidity risk.
In addition to the issues of age and non-marginal risk changes, several studies observe the importance of other factors when considering how to value mortality risk reductions from COVID-19 policies. Examples include the health status of those affected, the size of the risk change (as noted above), the extent to which the risk is dreaded, uncertain, ambiguous and catastrophic, as well as the extent to which it is involuntarily incurred and outside of one’s control (Colmer, 2020[44]; Hammitt, 2020[38]; Robinson, Sullivan and Shogren, 2021[43]). While many of these factors could be considered likely to increase VSL estimates, Robinson et al. (2021[43]) note that their potential interactions and impact on VSL remain uncertain due to a lack of empirical evidence.
In many ways, the challenges of providing general guidance on how changes in mortality risks should be valued are the same as discussed in OECD (2012[24]) and only amplified in the face of the COVID-19 pandemic. As noted by Colmer (2020, p. 57[44]): “Almost all benefit-cost analyses apply VSL estimates to new populations. As such, it is important that VSL estimates are chosen carefully, and that the assumptions that give VSL estimates a meaningful interpretation are plausible, when applied to new contexts. These considerations are of particular importance in the context of COVID-19”.
As noted in Section 5.3.2, the results of the current meta-analysis do not indicate any effect of pandemic-related risks on VSL estimates or differences in VSL estimates for studies that were conducted during the pandemic. It is, however, possible that pandemic-specific factors could influence VSL estimates for the reasons discussed above. When studies use CE methods, for example, it is easier to investigate characteristics associated with pandemics and their trade-offs with costs. Studies using SP methods to estimate VSL during the less serious, but comparable SARS pandemic10, for example, found significantly higher VSL estimates than normal. This effect was thought to be attributed to “the high degree of concern about SARS at the time the data were collected” (Liu et al., 2005, p. 83[48]). COVID-19, and similar pandemics in the future, could be expected to generate a larger effect than SARS. Even so, given the lack of sufficient evidence from the literature, as well as other reasons (e.g. the difficulty of operationalising guidance on adjusting for pandemic- or natural disaster-related risks), the approach taken in this report is not to assume specific adjustments to base VSL estimates due to these factors.
5.4.6. Other factors
The health valuation literature is large, especially in high-income countries and a number of additional factors have been found to affect VSL estimates (cf. Chapter 2). These factors include (but are not limited to): cancer and dread of death (or risk perceptions more generally), the health status of those affected and their baseline risk (Herrera-Araujo and Rochaix, 2020[32]), the size of the risk, the timing of risk (latency), its immediate (traumatic) vs. chronic (illness) nature, individuals’ degree of control over the risk (Robinson et al., 2010[49]), and altruism and private and public risks (Andersson, Levivier and Lindberg, 2019[50]; Chanel, Luchini and Shogren, 2021[51]). OECD (2012[24]) reviews many of these factors, as do the reviews and meta-analyses referenced above. As noted above, the evidence regarding the relationship between VSL and many of these factors appear to be heterogeneous and context specific. Furthermore, even if clear evidence of robust effects for any of these factors did exist, it would nevertheless be difficult to operationalise workable and transparent guidance on how to make relevant adjustments within and across countries and populations.
Some of the factors above have been analysed at a basic level in the meta-regression analysis in Section 5.3. One of these factors is cancer, as the meta-regression models indicate a positive and statistically significant “cancer premium” for studies in which cancer is mentioned. This finding aligns with other studies, such as Olofsson et al. (2019[52]) and Viscusi, Huber and Bell (2014[53]). However, given the large range of types, degrees of seriousness and treatments for cancers, it is difficult to defend a universal “cancer premium” across all cancer types in the context of general VSL guidance intended to apply across countries and policy types. The available information about cancer types in the primary studies included in the current analysis is moreover insufficient to provide a sufficient basis for such guidance. Other risk characteristics, e.g. traumatic vs chronic death and different risk types and causes, are not significant and robust in the meta-regression results presented in Section 5.3.
In conclusion, the meta-regression analysis and review of the literature above indicate that income (GDP per capita) is the only factor whose impacts are found to be statistically significant and robust across models and heterogenous global datasets. As such, the approach taken in this report is to adjust VSL estimates on the basis of income alone.
References
[45] Adler, M. (2020), “What Should We Spend to Save Lives in a Pandemic? A Critique of the Value of Statistical Life”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3636550.
[23] Alberini, A. and M. Ščasný (2021), “On the validity of the estimates of the VSL from contingent valuation: Evidence from the Czech Republic”, Journal of Risk and Uncertainty, Vol. 62/1, pp. 55-87, https://doi.org/10.1007/S11166-021-09347-8.
[31] Aldy, J. (2019), “Birds of a feather: Estimating the value of statistical life from dual-earner families”, Journal of Risk and Uncertainty, Vol. 58/2-3, pp. 187-205, https://doi.org/10.1007/S11166-019-09303-7.
[15] Ananthapavan, J. et al. (2021), “Systematic review to update ‘value of a statistical life’ estimates for Australia”, International Journal of Environmental Research and Public Health, Vol. 18/11, https://doi.org/10.3390/IJERPH18116168.
[50] Andersson, H., E. Levivier and G. Lindberg (2019), “Private and public willingness to pay for safety: A validity test”, Accident Analysis and Prevention, Vol. 123, pp. 170-175, https://doi.org/10.1016/J.AAP.2018.11.022.
[11] Bahamonde-Birke, F., U. Kunert and H. Link (2015), “The Value of a Statistical Life in a Road Safety Context — A Review of the Current Literature”, Transport Reviews, Vol. 35/4, pp. 488-511, https://doi.org/10.1080/01441647.2015.1025454.
[9] Banzhaf, H. (2022), “The Value of Statistical Life: A Meta-Analysis of Meta-Analyses”, Journal of Benefit-Cost Analysis, Vol. 13/2, pp. 182-197, https://doi.org/10.1017/BCA.2022.9.
[29] Brey, R. and J. Pinto-Prades (2017), “Age effects in mortality risk valuation”, European Journal of Health Economics, Vol. 18/7, pp. 921-932, https://doi.org/10.1007/S10198-016-0852-8.
[47] Carbone, J. and V. Smith (2013), “Valuing nature in a general equilibrium”, Journal of Environmental Economics and Management, Vol. 66/1, pp. 72-89, https://doi.org/10.1016/J.JEEM.2012.12.007.
[46] Carbone, J. and V. Smith (2008), “Evaluating policy interventions with general equilibrium externalities”, Journal of Public Economics, Vol. 92/5-6, pp. 1254-1274, https://doi.org/10.1016/J.JPUBECO.2007.06.003.
[51] Chanel, O., S. Luchini and J. Shogren (2021), “Does charity begin at home for air pollution reductions? Unraveling intra familial altruism”, Journal of Choice Modelling, Vol. 38, https://doi.org/10.1016/J.JOCM.2021.100268.
[44] Colmer, J. (2020), “What is the meaning of (statistical) life? Benefit-cost analysis in the time of COVID-19”, Oxford Review of Economic Policy, Vol. 36, pp. S56-S63, https://doi.org/10.1093/OXREP/GRAA022.
[21] Costa, D. and M. Kahn (2004), “Changes in the value of life, 1940-1980”, Journal of Risk and Uncertainty, Vol. 29/2, pp. 159-180, https://doi.org/10.1023/B:RISK.0000038942.18349.88/METRICS.
[8] Cropper, M., E. Joiner and A. Krupnick (2024), “Revisiting EPA’s Value per Statistical Life”, Review of Environmental Economics and Policy, Vol. 18/2, pp. 193-211, https://doi.org/10.1086/730476.
[3] Ginbo, T., W. Adamowicz and P. Lloyd-Smith (2023), “Valuing Mortality Risk Reductions in Canada: An Updated Meta-Analysis and Policy Guidance”, Canadian Public Policy, Vol. 49/3, pp. 233-251, https://doi.org/10.3138/CPP.2022-052.
[38] Hammitt, J. (2020), “Valuing mortality risk in the time of COVID-19”, Journal of Risk and Uncertainty, Vol. 61/2, pp. 129-154, https://doi.org/10.1007/S11166-020-09338-1.
[27] Hammitt, J. (2007), “Valuing Changes in Mortality Risk: Lives Saved Versus Life Years Saved”, Review of Environmental Economics and Policy, Vol. 1/2, pp. 228-240, https://doi.org/10.1093/REEP/REM015.
[22] Hammitt, J. et al. (2019), “Valuing mortality risk in China: Comparing stated-preference estimates from 2005 and 2016”, Journal of Risk and Uncertainty, Vol. 58/2-3, pp. 167-186, https://doi.org/10.1007/S11166-019-09305-5.
[20] Hammitt, J., J. Liu and J. Liu (2022), “Is survival a luxury good? Income elasticity of the value per statistical life”, Journal of Risk and Uncertainty, Vol. 65/3, pp. 239-260, https://doi.org/10.1007/S11166-022-09397-6.
[42] Hanaoka, C., H. Shigeoka and Y. Watanabe (2018), “Do risk preferences change? Evidence from the Great East Japan Earthquake”, American Economic Journal: Applied Economics, Vol. 10/2, pp. 298-330, https://doi.org/10.1257/APP.20170048.
[32] Herrera-Araujo, D. and L. Rochaix (2020), “Does the Value per Statistical Life vary with age or baseline health? Evidence from a compensating wage study in France”, Journal of Environmental Economics and Management, Vol. 103, https://doi.org/10.1016/J.JEEM.2020.102338.
[10] Keller, E. et al. (2021), “How Much Is a Human Life Worth? A Systematic Review”, Value in Health, Vol. 24/10, pp. 1531-1541, https://doi.org/10.1016/J.JVAL.2021.04.003.
[37] Kniesner, T. and W. Viscusi (2024), A Tale of the Tails: The Value of a Statistical Life at the Tails of the Age Distribution, https://papers.ssrn.com/abstract=4785616 (accessed on 28 October 2024).
[12] Kniesner, T. and W. Viscusi (2019), “The Value of a Statistical Life”, Oxford Research Encyclopedia of Economics and Finance, https://doi.org/10.1093/ACREFORE/9780190625979.013.138.
[18] Kniesner, T., W. Viscusi and J. Ziliak (2010), “Policy relevant heterogeneity in the value of statistical life: New evidence from panel data quantile regressions”, Journal of Risk and Uncertainty, Vol. 40/1, pp. 15-31, https://doi.org/10.1007/S11166-009-9084-Y.
[28] Krupnick, A. (2007), “Mortality-risk Valuation and Age: Stated Preference Evidence”, Review of Environmental Economics and Policy, Vol. 1/2, pp. 261-282, https://doi.org/10.1093/REEP/REM016.
[2] Lindhjem, H. et al. (2011), “Valuing Mortality Risk Reductions from Environmental, Transport, and Health Policies: A Global Meta-Analysis of Stated Preference Studies”, Risk Analysis, Vol. 31/9, pp. 1381-1407, https://doi.org/10.1111/J.1539-6924.2011.01694.X.
[48] Liu, J. et al. (2005), “Valuation of the risk of SARS in Taiwan”, Health Economics, Vol. 14/1, pp. 83-91, https://doi.org/10.1002/HEC.911.
[16] Masterman, C. and W. Viscusi (2018), “The Income Elasticity of Global Values of a Statistical Life: Stated Preference Evidence”, Journal of Benefit-Cost Analysis, Vol. 9/3, pp. 407-434, https://doi.org/10.1017/BCA.2018.20.
[41] Mourato, S. and G. Shreedhar (2021), Conducting economic valuation surveys during extreme events, OECD.
[5] Newbold, S. et al. (2024), “A two stage-random-effects meta-analysis of value per statistical life estimates”, National Center for Environmental Economics Working Paper, US Environmental Protection Agency.
[30] O’Brien, J. (2018), “Age, autos, and the value of a statistical life”, Journal of Risk and Uncertainty, Vol. 57/1, pp. 51-79, https://doi.org/10.1007/S11166-018-9285-3.
[1] OECD (2023), Household disposable income, https://www.oecd.org/en/data/indicators/household-disposable-income.html (accessed on 29 October 2024).
[24] OECD (2012), Mortality Risk Valuation in Environment, Health and Transport Policies, OECD Publishing, Paris, https://doi.org/10.1787/9789264130807-en.
[33] OECD (2010), Valuing Environment-Related Health Impacts - with special emphasis on children (VERHI-Children).
[52] Olofsson, S. et al. (2019), “Dread and Risk Elimination Premium for the Value of a Statistical Life”, Risk Analysis, Vol. 39/11, pp. 2391-2407, https://doi.org/10.1111/RISA.13341.
[35] Raich, W. et al. (2018), Valuing Reductions in Fatal Risks to Children. Final Report by Industrial Economics to Consumer Product Safety Commission, https:// www.cpsc.gov/s3fs-public/VSL_Children_Report_FINAL_20180103.pdf. (accessed on 30 October 2024).
[25] Robinson, L. (2007), “How US Government Agencies Value Mortality Risk Reductions”, Review of Environmental Economics and Policy, Vol. 1/2, pp. 283-299, https://doi.org/10.1093/REEP/REM018.
[14] Robinson, L. and J. Hammitt (2016), “Valuing Reductions in Fatal Illness Risks: Implications of Recent Research”, Health Economics (United Kingdom), Vol. 25/8, pp. 1039-1052, https://doi.org/10.1002/HEC.3214.
[13] Robinson, L. and J. Hammitt (2015), “Research Synthesis and the Value per Statistical Life”, Risk Analysis, Vol. 35/6, pp. 1086-1100, https://doi.org/10.1111/RISA.12366.
[49] Robinson, L. et al. (2010), “Valuing the risk of death from terrorist Attacks”, Journal of Homeland Security and Emergency Management, Vol. 7/1, https://doi.org/10.2202/1547-7355.1626/MACHINEREADABLECITATION/RIS.
[19] Robinson, L., J. Hammitt and L. O’Keeffe (2019), “Valuing Mortality Risk Reductions in Global Benefit-Cost Analysis”, Journal of Benefit-Cost Analysis, Vol. 10, pp. 15-50, https://doi.org/10.1017/BCA.2018.26.
[36] Robinson, L. et al. (2019), “Valuing Children’s Fatality Risk Reductions”, Journal of Benefit-Cost Analysis, Vol. 10/2, pp. 156-177, https://doi.org/10.1017/BCA.2019.10.
[43] Robinson, L., R. Sullivan and J. Shogren (2021), “Do the Benefits of COVID-19 Policies Exceed the Costs? Exploring Uncertainties in the Age–VSL Relationship”, Risk Analysis, Vol. 41/5, pp. 761-770, https://doi.org/10.1111/RISA.13561.
[7] SAB/USEPA (2017), Review of EPA’s Proposed Methodology for Updating Mortality Risk Valuation Estimates for Policy Analysis.
[4] USEPA (2016), Valuing mortality risk reductions for policy: a meta-analytic approach. White Paper, https://www.epa.gov/system/files/documents/2025-04/vsl-white-paper_final_020516-1.pdf.
[34] USEPA (2003), “Children’s Health Valuation Handbook”, http://yosemite1.epa.gov/ee/epa/eed.nsf/3cdbd09d7c867d9785256c9200548b12/6ed3736d44c87a4a85256dc1004da4ac/$FILE/handbook1030.pdf (accessed on 25 October 2024).
[40] Viscusi, W. (2021), “Economic lessons for COVID-19 pandemic policies”, Southern Economic Journal, Vol. 87/4, pp. 1064-1089, https://doi.org/10.1002/SOEJ.12492.
[39] Viscusi, W. (2020), “Pricing the global health risks of the COVID-19 pandemic”, Journal of Risk and Uncertainty, Vol. 61/2, pp. 101-128, https://doi.org/10.1007/S11166-020-09337-2.
[26] Viscusi, W. and J. Aldy (2007), “Labor market estimates of the senior discount for the value of statistical life”, Journal of Environmental Economics and Management, Vol. 53/3, pp. 377-392, https://doi.org/10.1016/J.JEEM.2006.10.005.
[53] Viscusi, W., J. Huber and J. Bell (2014), “Assessing whether there is a cancer premium for the value of a statistical life”, Health Economics (United Kingdom), Vol. 23/4, pp. 384-396, https://doi.org/10.1002/HEC.2919.
[17] Viscusi, W. and C. Masterman (2017), “Income Elasticities and Global Values of a Statistical Life”, Journal of Benefit-Cost Analysis, Vol. 8/2, pp. 226-250, https://doi.org/10.1017/BCA.2017.12.
[6] Wang, Y. et al. (2024), “Valuing mortality risk reductions in a fast-developing society: A meta-analysis of stated preference studies in China from 1998 to 2019”, Social Science & Medicine, Vol. 363, p. 117471, https://doi.org/10.1016/J.SOCSCIMED.2024.117471.
[55] World Bank (2025), GDP per capita, PPP (constant 2021 international $), https://data.worldbank.org/indicator/NY.GDP.PCAP.PP.KD (accessed on 27 March 2025).
[54] World Bank (2025), GNI per capita, PPP (constant 2021 international $), https://data.worldbank.org/indicator/NY.GNP.PCAP.PP.KD (accessed on 27 March 2025).
← 1. The difference between these two measures is marginal in practice (World Bank, 2025[54]; World Bank, 2025[55]).
← 2. Note that this coefficient cannot be interpreted as elasticity because the income and VSL values are not log transformed.
← 3. The covariate “Years since data collection” takes the value zero at 2022 and increases sequentially to reflect data collected in prior years.
← 4. Certain factors, such as age and income, would be good candidates for their own systematic reviews, which is outside of the scope of this review.
← 5. Newbold et al. (2024[5]) focuses on methodological issues to a greater extent, investigating a two-stage random effects meta-analysis model and alternative specifications, and does not provide VSL estimates per se. This analysis is referred to in Section 4.4 of Chapter 4 regarding meta-analysis approaches.
← 6. This estimate is not directly comparable with the base value estimated for the United States in Section 4.5 of Chapter 4. One reason is the use of income adjustment of VSL estimates before deriving central tendency. Another is that the body of studies is much smaller and older.
← 7. It is unclear whether other screening criteria (e.g. for outliers) are applied to studies for inclusion in the analysis.
← 8. One reason for these relatively high values is likely to be that the authors adjust VSL estimates to 2020 proportionally with growth in real income (i.e. assuming an income elasticity of VSL of 1).
← 9. While mean age of the main survey sample of some studies is reported in some cases, such information is typically not reported for subsamples used for sensitivity analyses from which VSL estimates are also derived.
← 10. An outbreak of the severe acute respiratory syndrome (SARS) viral respiratory disease occurred between 2002 and 2004.