The value of a statistical life in Sweden: A review of the empirical literature

Health Policy 108 (2012) 302–310

Contents lists available at SciVerse ScienceDirect

Health Policy journal homepage: www.elsevier.com/locate/healthpol

The value of a statistical life in Sweden: A review of the empirical literature Lars Hultkrantz a , Mikael Svensson a,b,∗ a b

Department of Economics and Statistics, Örebro University, Sweden Department of Economics and Statistics, Karlstad University, Sweden

a r t i c l e

i n f o

Article history: Received 21 March 2012 Received in revised form 18 September 2012 Accepted 20 September 2012 JEL classification: D61 I18 H51 Keywords: Willingness to pay Value of a statistical life Stated preferences Revealed preferences Economic evaluation

a b s t r a c t Recent focus on cost-benefit/socio-economic assessment of government “life-saving” programmes within public health, pharmaceutics, transport, and civil contingencies has spurred a wave of empirical research on the value of a statistical life (VSL) in Sweden. This paper provides an overview of the received evidence from a range of studies in one country and over a relatively short time period. A literature search was conducted in Econlit, Pubmed, Google Scholar and in bibliographies of published papers. Twelve studies on VSL with a total of 48 VSL estimates, published with data from Sweden from 1996 onwards, were identified. Among all estimates VSL varies from 9 to 1121 million SEK (D 0.9–121 million). Based on a set of additional quality inclusion criteria, as used also in a recent global review of VSL studies, the sample is restricted to 9 studies with a total of 29 VSL estimates with VSL varying from 9 to 98 million SEK (D 0.9–10.6 million). The raw mean among these estimates is 34.6 million SEK (D 3.7 million) and the median is 23 million SEK (D 2.5 million). Currently, official authorities in Sweden recommend a VSL of 22 million Swedish kronor (D 2.4 million). We also point out important concerns regarding validity of these estimates: primarily the problem that VSL is significantly related to the size of the mortality risk reduction showing significant scale insensitivity, in contrast to theoretical assumptions but in line with previous empirical findings. © 2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Several Swedish government authorities use costbenefit analysis (CBA) to make trade-offs between public expenditure and effects of public programmes on mortality risks. Since the early 1980s, the Swedish Transport Administration has used explicit economic values for prevention of fatalities and injuries from road and rail accidents, and for mortality and morbidity effects of air pollution in the planning of transport infrastructure investments and traffic policies (such as speed limits). Likewise, but more on a

∗ Corresponding author at: Karlstad University, 651 88 Karlstad, Sweden. Tel.: +46 54 700 25 08. E-mail address: [email protected] (M. Svensson). 0168-8510/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthpol.2012.09.007

case-by-case basis, the Swedish Civil Contingencies Agency and the Environmental Protection Agency conduct costbenefit analyses of policies and investments that involve health and health risk effects. In a CBA all the relevant benefits and costs of a proposed policy are expressed in monetary terms (discounted to present value if necessary). If the monetized benefits are larger than the monetized costs, the policy is said to increase societal economic efficiency/welfare. In CBA the economic parameters of policies and programmes affecting health and mortality risks are intended to reflect tax payers’ willingness to pay for public health-enhancing efforts. The key parameter is the value of a statistical life (VSL), i.e. the value of preventing a fatality. In CBA within the transport sector and in accident prevention within others sectors, other health related values, such as values

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

of preventing different kinds of injuries and health benefits from reduction of air pollution, are derived as proportions of the VSL. In Sweden, VSL has also been used in the decision making process for subsidization of medical technologies; for example, one (out of a few) suggested threshold values of the maximum accepted cost per a quality-adjusted life year (QALY) gained is based on the VSL [1].1 The increased use of VSL in public sector planning and policy-making in Sweden has spurred a wave of empirical research in recent years. As a result, there is a unique cluster of empirical VSL estimates from different studies, to some extent based on different methods, in one country within a time span of 14–15 years. The purpose of this study is to provide an overview of the main results of these studies. It is hoped that this may synthesize future research on methodological improvements in the field, and provide insight into policy making. Using studies from all over the world, a recent meta-analysis of VSL estimates commissioned by the OECD was based on a set of studies carried out between 1985 and 2009 (including two Swedish studies) [2,3]. In contrast, our study focuses on studies conducted in one country and over a shorter time period. Since public preferences are contextdependent and change over time, we argue that we to a larger degree avoid the possible confounding problems connected to multi-country comparisons and the use of data from a longer time period. The rest of this paper is structured in the following way. In Section 2 we introduce the concept of the value of a statistical life and the main approaches to empirically estimate VSL. Section 3 contains an overview of studies estimating VSL in Sweden from the year 1995 onwards. A discussion of the empirical estimates and policy-relevant weaknesses with these estimates concludes the paper in Section 4. 2. The value of a statistical life 2.1. The marginal willingness to pay for a mortality risk reduction Up until the 1960s the standard approach to estimate the value of prevented fatalities was the human capital approach that “values life” by future production potential, calculated by summing all future income streams of the “saved individual” to a present value (using an appropriate discount rate). This does not take into account lost leisure and other non-market activities, risk aversion and preferences for life and health per se. Hence, a relevant economic approach using the willingness to pay (WTP) approach was developed initially by Dreze [4] and further elaborated by e.g. Schelling [5] and Jones-Lee [6]. In the WTP approach VSL is a measure of the population mean marginal rate of substitution between mortality risk and wealth. This can be illustrated using an atemporal model within an expected utility framework. It is assumed that individuals face and maximize an expected

1 Where VSL was used to derive the willingness to pay for a QALY by dividing VSL with the (discounted) quality-adjusted expected lifeexpectancy.

303

utility function in a certain period, as seen in Eq. (1), where w, p, and us (w), s ∈ {a, d}, denote wealth, baseline probability of death, and the state-dependent utilities, respectively, with subscripts a and d denoting being alive or dead [7]. EU(w, p) = pud (w) + (1 − p)ua (w)

(1)

Totally differentiating Eq. (1) with expected utility held constant gives us the marginal WTP for a mortality riskreduction, i.e. the value of a statistical life (VSL). EU(w, p) = pud (w) + (1 − p)ua (w) VSL = =



dw   dp EU constant

ua (w) − ud (w) pud (w) + (1 − p)ua (w)

(2)

In Eq. (2) prime denotes first derivative. Under the reasonable assumptions (which are standard in the literature) that ua (w) > ud (w), ua (w) > ud (w) ≥ 0 and us (w) ≤ 0 for s ∈ {a, d}, it can be shown that VSL is positive and increasing in wealth (w) and baseline mortality risk (p) [8–10]. VSL increases with wealth since a higher wealth is associated with a higher utility if alive and lower expected utility decrease for a given level of resources used to reduce mortality risks. VSL also increases with baseline risk since with a higher baseline mortality risk it is more likely that resources used to reduce mortality risks will “come from” the state of death (low cost).2 In a health policy context it is also of relevance how VSL varies with age and health status. However, there are no clear results from the literature on these relationships. It has been shown that the effect of health status on VSL is ambiguous since there are effects affecting both the numerator and denominator in Eq. (2) [10,12]. Regarding age and VSL it has been shown theoretically that VSL may show different patterns with respect to age using different (but all plausible) assumptions. Many authors argue that most empirical results support the view that VSL declines with age or has an “inverted U-shape” with regards to age, i.e. increases up to ages around 40–45 after which it decreases [13–15]. 2.2. Empirical approaches: how to empirically estimate VSL VSL can be estimated with the WTP approach using either experimental or observational data, i.e. either from decisions made by subjects in contingent valuation (CV)/choice experiments (CE) involving trade-offs between money and health risk, or from observation of behaviour in real life choice settings. While randomized experimental data is often preferred to observational data due to (usually) clearer identification, this is not necessarily so in the context of this literature since it is difficult to conduct experiments involving a real money trade-off with risk (although there are a few such studies [16]). Hence, subjects participating in health risk CV and CE studies are usually asked to make hypothetical choices. Hypothetical choice experiments are used in the stated preference (SP)

2

The so-called “dead-anyway effect” [11].

304

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

approach, while observational data is the basis for the socalled revealed preference (RP) approach. In SP-applications it is not feasible to estimate the marginal WTP (Eq. (2)), instead respondents are asked for their WTP for a small finite mortality risk reduction (p) and VSL is then calculated as VSL = WTP/p. In order to give a more specific example of how the SP-approach works we exemplify with a recent Swedish study that conducted a survey using a random sample of respondents in a Swedish municipality. Preferences were elicited for a local public road-safety programme that would reduce traffic mortality risks by 50%, from 6 to 3 fatalities per year in a population of approx. 80,000 implying that p = 3/80,000 [17]. The respondents voted “yes” or “no” to have this hypothetical programme implemented for a given cost/price. Randomly assigning different prices for the good to different respondents makes it possible to estimate the WTP for the road safety programme and subsequently the VSL. The SP method can be tailor-made to any given typical risk scenario and is therefore very flexible, but has several limitations. A major concern regards how WTP varies with the size of the risk reduction (often referred to as scope, or here scale (in)sensitivity/bias). Theoretically WTP should increase close to proportionally (“near-proportionality”) with the magnitude of the risk reduction, but several studies have shown it to lack significant sensitivity to scale [18].3 This is problematic since it implies that VSL will be relatively higher in SP studies that use relatively smaller risk reduction (and the difference may be quite large). Whether this is mostly due to poor survey designs, cognitive constraints among respondents to comprehend small probability changes [20], incomplete and not well-behaved preferences [21] or to “rational inattention” [42] is still being debated and we will return to this issue later on in the paper. Another major issue with SP studies is hypothetical bias, which refers to that individuals tend to overestimate their WTP in surveys when no real transactions are made, maybe by as much as a factor of two to three [22,23]. However, within the general value eliciting literature several approaches have shown tentatively positive results in significantly reducing this bias. These approaches include ex-ante elaboration of the survey design, for instance by budget-constraint reminders, communication of the bias problem [24], and/or use of pivotal/referencing design that exploits a real choice situation for the SP survey [25], ex-post trimming of data, e.g., by elimination of outliers, protest bids, and/or responses from responses for which respondents report less than full confidence [26], statistical calibration (using data from economic experiments with real payments), and estimation on combined SP and RP data.

3 Two different type of tests has been suggested to test for the “nearproportionality” in empirical applications when estimating non-marginal WTP: a “weak” and a “strong” scale sensitivity test [19]. The “weak” test refers to that WTP should be positively related to the size of the risk reduction and the “strong” test refers to that WTP should be “near-proportional” to the size of the risk reduction.

In contrast to the SP approach, RP approaches use actual behaviour among individuals to estimate their implicit WTP for mortality risks. The most common RP source is labour market data. The idea is that a more dangerous job (higher morbidity and mortality risks) all else equal has to offer a higher wage, and by controlling for other wagedetermining factors VSL can be calculated by estimating the average wage premium offered for a riskier job [27]. Another RP method is the “consumer behaviour” approach [28], of which one example is a study that investigates the price premium consumers pay for safer cars [29]. Even though RP approaches have the advantage of being based on actual behaviour there is a range of confounding issues related to data that is non-experimental. For example, in labour market studies estimating the marginal WTP for mortality risk reductions may be inappropriate for a policy-estimate of VSL if individuals with risker jobs have significantly different (and unobserved) mortality risk/wealth preferences compared to the average individual. In consumer behaviour studies we need to implicitly assume that there are not systematic misperceptions with regards to different mortality risks associated with different consumer choices; which may be a rather strong assumption. 2.3. Use of VSL for cost-benefit analysis in Sweden In Sweden, the National Road Administration started using CBA for evaluation of national road investment objects in the 1960s. Following a parliament decision in 1978 that the national traffic policy should be guided by economic principles, routines were developed for economic evaluation of road and rail infrastructure projects. The first official recommendations on economic parameters for these assessments were published in 1995, and have subsequently been revised in 1999, 2002, 2008, and 2012 [30]. As regards VSL, the 1995 version based recommendation on results from an unpublished SP study [31]. Although a few more studies were available in 1999 and 2002, the traffic authorities choose not to change (except for inflation adjustment) because of concerns about the validity of these. However, in 2008 the authorities felt enough evidence had accumulated for making a confident revision. The new recommended value remains after the latest revision in 2012 and is at 22.3 million SEK (D 2.4 million). Road safety has for long been a major concern of the Swedish traffic policy. In 1997 the national parliament adopted the so called Vision Zero, aiming for zero fatalities and serious injuries in traffic. Initially this was by some commentators seen as a break with the CBA approach to transport planning, but this criticism faded out as it was made clear that Vision Zero was a target for the very long term. Also, it turned out that many of the Vision Zero suggested actions were supported in the economic evaluations. In fact, a recent appraisal of Swedish road-safety research find that a range of major road-safety programmes pursued under the Vision Zero banner are associated with high benefit-cost ratios, also when account is taken of the considerable research cost that was laid down before the measures were implemented [32].

15 Percent 10 5

Economic tools for making priorities have more recently also come into regular use by Swedish authorities in the health sector. From 2002 The Dental and Pharmaceutical Benefits Agency determines which prescription drugs that should be subsidized in the national health system, partly based on cost-effectiveness/utility grounds. The principal indicator is the cost per gained quality-adjusted life-year (QALY). There is not a definite cut-off value for the maximum cost per QALY accepted, but a range of values that are “inspired” by the VSL used by the traffic authorities [1].

305

20

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

3.1. Main results In September 2011 (with an updated search in June 2012) we performed relevant searches in the databases Econlit, Pubmed, Google Scholar, and the bibliographies in published papers on the value of a statistical life in Sweden identified in the first round of searches. We only include papers from 1995 and onwards, since as stated in the introduction, we wanted to analyze papers within a reasonable similar context (same country and relatively reasonable time period). However, this also includes almost all of WTP-based VSL studies elicited with a Swedish population sample and, to the best of our knowledge, all that have been published in academic journals. Keywords in initial searches included value of (statistical) life, value of prevented fatality (mortality) risk reduction, willingness to pay, stated preferences, contingent valuation, and choice experiment, in combination with Sweden/Swedish. Abstracts of identified papers were screened in the first phase to select papers for full reading in a second phase. We also searched all the bibliographies in papers that were read in full in order to identify additional papers not found in the database searches (we did not find any additional papers based on the bibliographies). Table 1 summarizes all papers eliciting VSL by using a WTP approach that were published using Swedish data 1995–2011 as we identified in the searches. It shows whether a private or public good is used to elicit VSL, risk context/domain, the risk reduction for which WTP is estimated, and VSL in million SEK and Euro. All values are converted to the same price level (2010).4 Some of the studies report one or several estimates for the full sample. In some cases the differences is clear in the Table (e.g. a private good versus a public good was valued in the study), other differences not included in the study regards that the authors of a study often use several statistical approaches (e.g. parametric and non-parametric approaches) to estimate VSL. We comment on this in the text when we see it as important differences. In total there are 48 estimates from 12 different studies shown in Table 1 and as can be seen most studies use a SP method. The lowest value of VSL is 9 million SEK (D 0.9 million) and the highest value is 1121 million SEK (D 121 million). The raw mean and median estimate of VSL is 105.6 million SEK (D 11.4 million) and 40.9 million SEK (D 4.4

4

Converted from SEK to Euros using exchange rate: D 1 = SEK9.25.

0

3. Results and analysis 0

50 100 VSL Estimates in million Swedish Kronor

150

Fig. 1. Histogram of VSL estimates. Note: Few estimates over 150 million SEK (see Table 1) excluded for sake of clarity.

million), respectively. Fig. 1 also shows the distribution of VSL estimates as reported in Table 1 (although excluding the few estimates above 150 million SEK in order for clarity in the graph). It shows, as outlined in the raw median above, that most studies estimate VSL to be at least lower than 45–50 million SEK and there is a relatively long right tail (even though we do not show all of it in Fig. 1). The oldest papers in this review as listed in Table 1 are two studies by Johannesson et al. [33,34], with different co-authors, with one estimate based on a road-safety context and the other based on a health-care context. The VSL estimates from one of these studies are shown in an interval (36–70 and 20–58 million SEK) since the specific point estimates were not reported in the paper. Like the first of these studies, the majority of subsequent works, as can be seen in Table 1, have estimated VSL in a road safety context. These also include Persson et al. [35] who report three different estimates, from 15 to 35 million SEK, which differ with respect to the size of the risk reduction used in the survey (three different magnitudes). The highest estimate (35 million SEK) is based on the smallest risk reduction. Another road safety valuation study is Hultkrantz et al. [36] who differ with respect to the risk reduction provided by either a private or a public good; the privately provided risk reduction is valued significantly higher. Andersson [10] provide several different estimates, which differ depending on parametric or non-parametric assumptions, and with different distributional assumptions. Andersson and Lindberg [37] also estimate VSL for publicly and privately provided goods and report separate VSL estimates for these different elicitation contexts. Svensson [38] provides two estimates that differ with respect to the sample used (based on two different data collections). Krüger and Svensson [17] also estimate VSL in a road safety context, with two estimates that differ depending on whether or not the realoption value is included. There are also two studies using RP methods in a road safety context. Andersson [39] examines the price premium paid for safer cars and arrives at a VSL estimate of 14 million SEK (D 1.5 million). Svensson estimates VSL in a road safety context dependent on consumer behaviour of seat belt and bicycle helmet use [40].

306

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

Table 1 Published WTP studies empirically estimating VSL using Swedish data since year 1995. Author

Year

Method

Type of good

Risk domain

risk

VSL million SEK

VSL million D

Johannesson et al. [33] Johannesson et al. [33] Johannesson et al. [34] Johannesson et al. [34] Persson et al. [35] Persson et al. [35] Persson et al. [35] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Carlsson et al. [42] Andersson [39] Andersson [39] Andersson [39] Andersson [39] Hultkrantz et al. [36] Hultkrantz et al. [36] Andersson [10] Andersson [10] Andersson [10] Andersson [10] Andersson [10] Andersson [10] Andersson [10] Andersson [10] Svensson [40] Svensson [40] Svensson [40] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Krüger and Svensson [17] Krüger and Svensson [17] Andersson and Lindberg [37] Andersson and Lindberg [37] Andersson and Lindberg [37] Andersson and Lindberg [37] Carlsson et al. [41] Carlsson et al. [41] Carlsson et al. [41]

1996 1996 1997 1997 2001 2001 2001 2004 2004 2004 2004 2004 2004 2004 2004 2005 2005 2005 2005 2006 2006 2007 2007 2007 2007 2007 2007 2007 2007 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2010 2010 2010

SP SP SP SP SP SP SP SP SP SP SP SP SP SP SP RP RP RP RP SP SP SP SP SP SP SP SP SP SP RP RP SP SP SP SP SP SP SP SP SP SP SP SP SP SP SP SP

Private Public Private Private Private Private Private Private Private Private Private Private Private Private Private Private Private Private Private Private Public Private Private Private Private Private Private Private Private Private Private Private Public Public Public Public Public Public Public Public Private Private Public Public Unclear Unclear Unclear

Road Road Health care Health care Road Road Road Air travel Air travel Air travel Air travel Taxi travel Taxi travel Taxi travel Taxi travel Road Road Road Road Road Road Road Road Road Road Road Road Road Road Seat belt Bic. helmet Road Road Road Road Road Road Road Road Road Road Road Road Road Road Fire Drowning

8.11/100,000 8.11/100,000 2/10,000 2/10,000 1.8/100,000 2.4/100,000 5/100,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 0.5/1,000,000 – – – – 4.1/100,000 4.1/100,000 1/100,000 1/100,000 1/100,000 1/100,000 4/100,000 4/100,000 4/100,000 4/100,000 11.6/100,000 1.3/100,000 3/100,000 3.6/100,000 3.6/100,000 3.6/100,000 4.1/100,000 4.1/100,000 4.1/100,000 3.6/100,000 3.6/100,000 2.5/100,000 2.5/100,000 2.5/100,000 2.5/100,000 3.3–10/100,000 3.3–10/100,000 3.3–10/100,000

36–70 20–58 37 52 35 28 15 473 1121 381 892 385 163 151 128 14 14 10 9 58 22 153 138 128 73 46 45 41 30 21 37 82 31 21 23 53 42 21 23 43 66 98 22 56 20 13 13

3.8–7.6 2.2–6.3 4.0 5.7 3.8 3.0 1.6 51.1 121.0 412 964 41.6 17.6 16.3 13.8 1.6 1.5 1.0 0.9 6.2 2.4 16.5 14.9 13.8 7.9 5.0 4.9 4.4 3.3 2.3 4.0 8.9 3.4 2.3 2.4 5.8 4.6 2.3 2.5 4.6 7.1 10.6 2.4 6.1 2.2 1.4 1.4

Note: All values are in 2010 price levels.

Regarding the studies with a non-road safety context this includes Johannesson et al. [34], which uses a nonspecific health-care context. Further, Carlsson et al. [41] report three estimates, which differ with respect to the risk context, the highest being for road safety, and fire and drowning safety not significantly different from each other. Finally, Carlsson et al. [42] estimate VSL in both road (taxi) and air travel contexts and find, in comparison with the other studies, a very high estimate for air travel safety. A significant difference with this study is also the very small mortality risk reductions used when eliciting WTP. The first study used here was published 20 years after the seminal work in the UK by Jones-Lee [6]. From a methodological point of view, all included studies have therefore been able to draw on a considerable previous value-elicitation literature. All studies were partly

motivated by an interest in methodological and other general issues left open by previous research and many have found their way into leading academic journals within the field. However, it is fair to say that they all have struggled with the methodological challenges and limitations that characterize elicitation of VSL; for SP studies above the already mentioned problems with scale and hypothetical bias and also growing public resistance to responding to surveys with low response rates. All SP studies acknowledge the problems in understanding small risk probabilities, and use different ways to help respondents with this, for instance diagrams and community analogies. However, those studies that performed tests for scale bias were not able to reject the presence of such a bias; except for the study by Andersson [39]. Hypothetical bias was mitigated by ex-ante budget reminders and ex-post trimming

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

307

Table 2 Published estimates of VSL: Only including studies meeting our specified additional inclusion criteria. Author

Year

Method

Type of good

Risk domain

risk

VSL million SEK

VSL million D

Johannesson et al. [33] Johannesson et al. [33] Johannesson et al. [34] Johannesson et al. [34] Andersson [39] Andersson [39] Andersson [39] Andersson [39] Hultkrantz et al. [36] Hultkrantz et al. [36] Svensson [40] Svensson [40] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Svensson [38] Krüger and Svensson [17] Krüger and Svensson [17] Andersson and Lindberg [37] Andersson and Lindberg [37] Andersson and Lindberg [37] Andersson and Lindberg [37] Carlsson et al. [41] Carlsson et al. [41] Carlsson et al. [41]

1996 1996 1997 1997 2005 2005 2005 2005 2006 2006 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2010 2010 2010

SP SP SP SP RP RP RP RP SP SP RP RP SP SP SP SP SP SP SP SP SP SP SP SP SP SP SP

Private Public Private Private Private Private Private Private Private Public Private Private Public Public Public Public Public Public Public Public Private Private Public Public Unclear Unclear Unclear

Road Road Health care Health care Road Road Road Road Road Road Seat belt Bic. helmet Road Road Road Road Road Road Road Road Road Road Road Road Road Fire Drowning

8.11/100,000 8.11/100,000 2/10,000 2/10,000 – – – – 4.1/100,000 4.1/100,000 11.6/100,000 1.3/100,000 3.6/100,000 3.6/100,000 3.6/100,000 4.1/100,000 4.1/100,000 4.1/100,000 3.6/100,000 3.6/100,000 2.5/100,000 2.5/100,000 2.5/100,000 2.5/100,000 3.3–10/100,000 3.3–10/100,000 3.3–10/100,000

36–70 20–58 37 52 14 14 10 9 58 22 21 37 31 21 23 53 42 21 23 43 66 98 22 56 20 13 13

3.8–7.6 2.2–6.3 4.0 5.7 1.6 1.5 1.0 0.9 6.2 2.4 2.3 4.0 3.4 2.3 2.4 5.8 4.6 2.3 2.5 4.6 7.1 10.6 2.4 6.1 2.2 1.4 1.4

of outliers [36], while one study [38] estimates VSL only on responses from subjects stating full confidence in their own answers to the willingness-to-pay question. 3.2. Results using more restrictive inclusion criteria In a second step we partly follow the procedure in a recent global meta-analysis of VSL estimates to show VSL estimates meeting a certain number of additional inclusion criteria [2] argued to be in line with some relevant quality characteristics. These were (1) sample size of at least 200, (2) explicit risk reduction, (3) and representative population. We also add a fourth inclusion criteria, namely that (for SP studies) elicitation was conducted based on a discrete choice question (yes/no) rather than an open-ended VSL in Euro 2002−prices (PPP adjusted) Portugal Spain Japan Greece Poland Czech republic Belgium France Denmark Italy Austria Ireland Australia Germany Finland Canada New Zealand Netherlands Sweden Switzerland Great Britain Norway United States

55812 150253 181920 206087 221392 316334 462717 589177 679737 791748 899014 1170695 1193378 1266000 1273372 1408630 1704131 1741000 1954000 2010000 2107000

4. Concluding discussion

2707000 3189000

0

1,000,000

2,000,000

question (“how much are you maximum willing to pay?”) since this is in line with general recommendations for conducting CV studies (e.g. regarding incentive compatibility) [43,44]. Based on these inclusion criteria, we are left with the following VSL estimates as shown in Table 2. Based on the set of papers in Table 2, VSL is now in the range between 9 and 98 million SEK (D 0.9–10.6 million) with a raw mean of 34.6 million SEK (D 3.7 million) and a raw median at 23 million SEK (D 2.5 million). Hence, with these additional inclusion criteria the range of estimates becomes significantly smaller. This is primarily due to the fact that the study of VSL in an air and taxi safety context is not included; which is a study based on a much smaller mortality risk reduction compared to the other studies and also based on an open-ended WTP question. Another work (with high VSL estimates) not included using the additional inclusion criteria is Andersson [10] due to the fact that it uses open-ended WTP questions.

3,000,000

Fig. 2. Official monetary valuation of a road accident fatality in selected countries. Euro in 2002 prices.

This paper reviewed published estimates of VSL in Sweden between 1995 and onwards. 12 independent studies containing 48 VSL estimates were identified in this review. Among these 48 estimates, VSL varied between 9 million SEK (D 0.9 million) and 1121 million SEK (D 121 million) with a raw mean of 105.56 million SEK and a median of 40.89 million SEK (D 11.4 and D 4.4 million). In a second step we added additional inclusion criteria in the same line as a recent international meta-analysis together with a criteria that WTP/VSL (for SP studies) should have been elicited using a dichotomous choice question (or

308

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

referendum format). This restricted the number of studies to 9 and the number of VSL estimates to 29 (Table 2). Based on these estimates VSL varied between 9 million SEK to 98 million SEK (D 0.9–10.6 million) with a raw mean at 34.6 million SEK and a median at 23 million SEK (D 3.7 million and D 2.5 million). This is not too far away from currently recommended VSL estimates for CBA in the Swedish public sector, which is at 22.3 million SEK (D 2.4 million) [45]. This correspondence is not so surprising since the recommended VSL for public policy is anchored on some of the more recent studies included in this review. Even though VSL estimates vary substantially in this review, especially based on all estimates as listed in Table 1, the range is not as large as is found in global reviews of VSL, especially if we look at the studies with the additional inclusion criteria (Table 2). For example, in a recent study of global VSL, estimates vary from 4450 USD to 197,000,000 USD [2]. Thus, a large part of the global variation is probably due to differences of countries, time periods and survey methodology that may be difficult to properly control for in a meta-analysis. The variation of VSL estimates may also be interesting to examine in light of the large spread of the official monetary valuation of preventing road fatalities as shown in Fig. 2. As can be seen, the Swedish official value is at the upper end for European countries, but considerably lower than the US value. The OECD recently made a recommendation for VSL in the range of 1.8–5.4 million USD [3], with a preferred point estimate at 3.6 million USD, all in 2005 prices. Comparable values5 of our estimated mean and median VSL for Sweden are 4.1 and 2.7 million dollar, respectively; i.e. both within the recommended range and centred on the point estimate. Regarding the variation found in this study, the highest estimate is based on air safety, which may be caused by individual preferences differing significantly due to the air travel context. It has been shown that preferences may differ according to e.g. seriousness of risk, dread, the respondents’ own exposure to the risk, voluntariness of the risk and the controllability of the risk [46,47]. With air travels it may be argued that dread (high) and controllability (low) may influence the VSL estimates to become significantly higher. Further reasons for the variation in VSL, and the very high VSL estimate in the air-safety study, may be due to different mortality risk reductions used (among the SP studies). It has been shown that VSL studies, especially when mortality risk reductions are very small, tend to suffer from scale insensitivity, i.e. WTP does not increase proportionally or “near-proportionally” with the risk reduction [18]. Considering the relatively low number of studies and observations in our review, we have refrained from conducting formal regression meta-analyses of the data, but simple correlations and regressions still indicate that there is a significant relationship between the mortality risk

5 The values were deflated from 2010 to 2005 with Swedish CPI and converted from SEK to USD using exchange rate USD1 = SEK7.81 (July 1, 2005).

reduction and VSL in our sample. This is hardly surprising since proportionality is generally rejected in the included studies in the primary data. A significant relationship between the mortality risk reduction and VSL was also more formally established in the global meta-analysis of VSL estimates (including two of the studies surveyed in this paper) [2] as well as in a number of other previous empirical studies [18]. The problem with scale insensitivity is of course also potentially important for the VSL estimates included in Table 2 in general, i.e. the (in international comparisons) relatively small range of VSL estimates may be due to the fact that most studies have used reasonably similar mortality risk reductions, and results could have been quite different with a larger spread in the used mortality risk reductions. Although, it should be mentioned that recently there have been some more optimistic results showing the theoretically expected scale sensitivity in estimating preference for live-saving policies based on relative person trade-offs (rather than the standard approach as included the papers in this review) [48].6 If scale bias is primarily explained by cognitive constraints to comprehend small changes in probabilities person trade-off questions may be easier for respondents to comprehend and result in more “wellbehaved” preferences.7 Another consistent finding in the empirical literature is that when formulated as privately provided risk reductions, VSL estimates tend to be higher [2,49]. Yet again we tested for this informally in our data, and we have the same indications among the included studies here. Also, some of the included studies specifically tested for this effect using split-sample designs, and confirmed the finding that VSL is higher when elicited using a private good context [36,37]. From a theoretical point of view it would be more plausible for values to be higher when using a public good context (when the risk reduction does not only benefit the respondent, but also all other individuals). One reason for this finding may be that public safety investments especially in a road-safety context may be accompanied by negative consequences (e.g. road bumps). If respondents “anchor” their responses on these types of public safety investments, this could lead to a downward adjustment in their preferences [50]. Interestingly, the meta-analysis commissioned by the OECD finds that the statistical significance of the variable representing a public good (as opposed to a private good) context vanishes when control is made of whether the study frame was road safety [2,3]. Still, in Swedish studies the effect has been shown to be quite large also in studies using a road safety context [36,37,50].

6 In this approach respondents are asked to choose between different policies with the same cost but with different “benefits” (e.g. the number of lives saved may or may not be equal, the risk context may differ, the age-profiles of those saved may differ, etc.). 7 Recent economic research based on insights from experiments in psychophysics and neurobiology suggests that this and other choice “anomalies” may be understood as “rational inattention”. The capacity limitation of the nervous system requires choice to be based on a compressed representation of the real choice set [42], which seems to blur the distinction between different magnitudes of very small risk probabilities.

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

Apart from the methodological aspects and lessons from this review, there are also policy-relevant implications that may be drawn. For example, the range of VSL estimates surveyed in this study is also interesting for comparison with implicit values of life from life-saving policies in Sweden. There is (naturally) no all-encompassing data on cost per life saved for Swedish life-intervention policies. However, given the range of VSL reported above there is still abundant evidence indicating inefficient use of money for life-saving interventions. It has been shown that policies implemented in Sweden can have a cost per life saved ranging from 50,000 SEK (or lower) up to several billion SEK per life saved [51,52]. A recent case in the high end is cleaning up of arsenic-contaminated sites in Sweden [52], which was estimated to cost up to 9 billion SEK per life saved. Examples such as the latter cannot find economic support from any estimate as included in this review. The review in this study can also be informative for economic evaluations of medical technologies, usually conducted with cost-effectiveness/utility analysis with the relevant outcome being the cost per incremental quality adjusted life year (QALY) gained. In a recent paper it was shown, using more plausible assumptions than in earlier work, how VSL in a quite straightforward way can be converted into a WTP for a QALY [53]. There is an increasing discussion regarding the relevant WTP per QALY (or “threshold” or “lambda” value) in many countries, such as the UK, Australia, Sweden, and Norway, since it is quite often the case that reimbursement or subsidization decisions for new medical technologies are conducted subsequentially: “one technology at a time”. Using the main approach in the recent work cited above using the median VSL estimate from Table 2 in this study of 23 million SEK (D 2.4 million) gives a WTP per QALY at 1.2 million SEK (D 130,000) using a discount rate of 3%. This is clearly higher than most informal “thresholds” regarding the value of a QALY in Sweden [1,54]. The threshold value used in the UK by the National Institute for Health and Clinical Excellence (NICE) for recommendations on methods to be used by the national health services is often referred to also in other countries. This value (GBP 20,000–30,000) is less than a third of the WTP per QALY that is inferred by our study [55]. Finally, there are some areas that still seem to be very “under-researched” in the VSL literature. There is still scarce knowledge on how society values prevented fatalities in different risk contexts. Previous research has suggested that factors such as voluntariness and controllability of risks may affect VSL [46], even though one of the studies surveyed in this paper specifically tested for this without finding relevant correlations. Other recent work has also found that the “irresponsible” behaviour significantly affects the valuation, e.g. fatality risks that is due to a higher degree of “irresponsible behaviour” gives lower VSL estimates [48]. In the UK the National Institute for Health and Clinical Excellence (NICE) argues that the degree of blame in different risk contexts should not play a role in the decisions of allocating scarce health care resources [48,56]. There have been no such formal policy statement in the Swedish health/risk policy context, and as can be seen in this review there is further only scarce evidence regarding

309

if the population has any preferences over this issue (and if so, how large should such differences be). This becomes an important issue as the use of costbenefit analysis is increasingly being spread to more sectors so VSL estimates are used in contexts other than road safety. Further, in other applications/context the age characteristics of patients/victims may be distinctly different from those of a typical road safety measure that can be expected to save lives of all ages. For instance, lowering of exposure in urban environments to particle matter pollutants seems to reduce premature mortality of the elderly from cardiovascular diseases and lung cancer [57]. Such measures therefore save a limited number of remaining years of life of each potential victim, while an avoided traffic-accident fatality on average saves 35–40 years of remaining life time. To account for such differences, multiplying VSL with the proportion of live years saved is sometimes preferred to using the road-traffic average number. Some support for these practices has recently been provided in a study where it is shown that saved lives of younger individuals are consistently given higher values than older lives saved. For example, saving one 30-year-old driver is on average equivalent to saving 2.5 drivers aged 70 [47]; still, it is unclear how the age-dimension should be dealt with in cost-benefit analysis of life-saving policies.

Conflict of interest No conflicts of interest are present.

Acknowledgements Financial support from the Swedish Civil Contingencies Agency is gratefully acknowledged. We wish to thank two anonymous reviewers for helpful comments and suggestions.

References [1] Persson U, Hjelmgren J. Hälso- och sjukvården behöver kunskap om hur befolkningen värderar hälsan. Läkartidningen 2003;100(43):3436–7. [2] Lindhjelm H, Navrud S, Braathen NA, Biausque V. Valuing mortality risk reductions from environmental, transport, and health policies: a global meta-analysis of stated preference studies. Risk Analysis 2011;31(9):1381–407. [3] Biausque V, Braathen NA, Lindhjelm H, Navrud S. Mortality risk valuation in environment, health and transport policies. Paris: OECD; 2012. [4] Dreze JH. L’utilité sociale d’une vie humaine. Revue Francaise de Recherche Opérationnelle 1962;22:93–118. [5] Schelling TC. The life you save may be your own. In: Chase SB, editor. Problems in public expenditure analysis. Washington, DC: The Brookings Institution; 1968. p. 127–62. [6] Jones-Lee M. The value of life: an economic analysis. Chicago: University of Chicago Press; 1976. [7] Rosen S. The value of changes in life expectancy. of Risk and Uncertainty 1988;1(3):285–304, Journal http://dx.doi.org/10.1007/BF00056139. [8] Jones-Lee MW. The value of changes in the probability of death or injury. Journal of Political Economy 1974;82(4):835–49. [9] Weinstein MC, Shepard DS, Pliskin JS. The economic value of changing mortality probabilities: a decision-theoretic approach. Quarterly Journal of Economics 1980;94(2):373–96, http://dx.doi.org/10.2307/1884546.

310

L. Hultkrantz, M. Svensson / Health Policy 108 (2012) 302–310

[10] Andersson H. Willingness to pay for road safety and estimates of the risk of death: Evidence from a Swedish contingent valuation study. Accident Analysis and Prevention 2007;39:853–65. [11] Pratt JW, Zeckhauser RJ. Willingness to pay and the distribution of risk and wealth. Journal of Political Economy 1996;104(4):747–63. [12] Hammitt JK. QALYs versus WTP. Risk Analysis 2002;22(5):985–1001. [13] Evans MF, Smith VK. Do we really understand the age-VSL relationship? Resource and Energy Economics 2006;28:242–61, http://dx.doi.org/10.1016/j.reseneeco.2006.02.004. [14] Aldy JE, Viscusi K. Age differences in the value of a statistical life: revealed preference evidence. Review of Environmental Economics and Policy 2007;1(2):241–60. [15] Krupnick A, Alberini A, Cropper M, Simon N, O’Brien B, Goeree R, Heintzelman M. Age, health and the willingness to pay for mortality risk reductions: a contingent valuation survey of Ontario residents. Journal of Risk and Uncertainty 2002;24(2):161–86, http://dx.doi.org/10.1023/A:1014020027011. [16] Shogren J, Shin S, Hayes DJ, Kliebenstein JB. Resolving differences in willingness to pay and willingness to accept. The American Economics Review 1994;84:255–70. [17] Krüger N, Svensson M. The impact of real options on willingness to pay for mortality risk reductions. Journal of Health Economics 2009;28:563–9. [18] Hammitt JK, Graham JD. Willingness to pay for health protection: inadequate sensitivity to probability? Journal of Risk and Uncertainty 1999;18(1):33–62, http://dx.doi.org/10.1023/A:1007760327375. [19] Corso PS, Hammitt JK, Graham JD. Valuing mortality-risk reduction: using visual aids to improve the validity of contingent valuation. Journal of Risk and Uncertainty 2001;23(2):165–84, http://dx.doi.org/10.1023/A:1011184119153. [20] Andersson H, Svensson M. Cognitive ability and scale bias in the contingent valuation method. Environmental and Resource Economics 2008;39:481–95. [21] Fischhoff B, Frederick S. Scope (in)sensitivity in elicited valuations. Risk, Decision, and Policy 1998;3:109–23. [22] List JA, Gallet CA. What experimental protocol influence disparities between actual and hypothetical stated values? Environmental and Resource Economics 2001;20(3):241–54, http://dx.doi.org/10.1023/A:1012791822804. [23] Harrison GW, Rutstrom EE. Experimental evidence on the existence of hypothetical bias in value elicitation methods. In: Plott C, Smith V, editors. Handbook in experimental economics results. New York: Elsevier Science; 2008. p. 752–67. [24] Cummings RG, Taylor LO. Unbiased value estimates for environmental goods: a cheap talk design for the contingent valuation method. American Economic Review 1999;89(3):649–65. [25] Hensher DA. Hypothetical bias, choice experiments and willingness to pay. Transportation Research Part B 2010;44(6):735–52. [26] Hedemark Lundhede T, Olsen SB, Jacobsen JB, Thorsen BJ. Handling respondent uncertainty in choice experiments: evaluating recoding approaches against explicit modelling of uncertainty. Journal of Choice Modelling 2009;2(2):118–47. [27] Viscusi WK, Aldy JE. The value of a statistical life: a critical review of market estimates throughout the world. Journal of Risk and Uncertainty 2003;27(1):5–76, http://dx.doi.org/10.1023/A:1025598106257. [28] Blomquist GC. Value of life saving: implications from consumption activity. Journal of Political Economy 1979;87:540–58. [29] Atkinson SE, Halvorsen R. The valuation of risks to life: evidence from the market for automobiles. Review of Economics and Statistics 1990;72(1):133–6. [30] ASEK. ASEK (Arbetsgruppen samhällsekonomiska kalkyler). Borlänge: The Swedish Transport Administration; 2012. [31] Persson U, Cedervall M. The value of risk reductions: results of a Swedish sample survey. In: IHE Working Paper 1991, vol. 6. 1991. [32] Elvik R, Kolbenstvedt M, Elvebakk B, Hervik A, Braein L. Costs and benefits to Sweden of Swedish road safety research. Accident Analysis and Prevention 2009;41:387–92. [33] Johannesson M, Johansson P-O, O’Conor RM. The value of private safety versus the value of public safety. Journal of Risk and Uncertainty 1996;13(3):263–75. [34] Johannesson M, Johansson P-O, Löfgren K-G. On the value of changes in life expectancy: blips versus parametric

changes. Journal of Risk and Uncertainty 1997;15(3):221–39, http://dx.doi.org/10.1023/A:1007705309543. [35] Persson U, Norinder A, Hjalte K, Gralén K. The value of a statistical life in transport: findings from a new contingent valuation study in Sweden. Journal of Risk and Uncertainty 2001;23(2):121–34, http://dx.doi.org/10.1023/A:1011180018244. [36] Hultkrantz L, Lindberg G, Andersson C. The value of improved road safety. Journal of Risk and Uncertainty 2006;32(2):151–70, http://dx.doi.org/10.1007/s11166-006-8291-z. [37] Andersson H, Lindberg G. Benevolence and the value of road safety. Accident Analysis and Prevention 2009;41:286–93. [38] Svensson M. The value of a statistical life in Sweden estimates from two studies using the certainty approach calibration. Accident Analysis and Prevention 2009;41:430–7. [39] Andersson H. The value of safety as revealed in the Swedish car market: an application of the hedonic pricing approach. Journal of Risk and Uncertainty 2005;30(3):211–39, http://dx.doi.org/10.1007/s11166-005-1154-1. [40] Svensson M. Precautionary behavior and willingness to pay for a mortality risk reduction: searching for the expected relationship. Journal of Risk and Uncertainty 2009;39:65–85. [41] Carlsson F, Daruvala D, Jaldell H. Value of statistical life and cause of accident: a choice experiment. Risk Analysis 2010;30(6): 975–86. F, Johansson-Stenman O, Martinsson P. [42] Carlsson Is transport safety more valuable in the air. Jourof Risk and Uncertainty 2004;28(2):147–63, nal http://dx.doi.org/10.1023/B:RISK.0000016141.88127.7c. [43] Carson RT, Flores NE, Meade NF. Contingent valuation: controversies and evidence. Environmental and Resource Economics 2001;19(2):173–210, http://dx.doi.org/10.1023/A:1011128332243. [44] NOAA. Report of the NOAA panel on contingent valuation 58. Federal Register 46; 1993. [45] SIKA. Samhällsekonomiska principer och kalkylvärden för transportsektorn: ASEK 4. SIKA PM 2008:3; 2008. [46] Cropper M, Subramanian U. Public choices between lifesaving programs: the tradeoff between qualitative factors and lives saved. Journal of Risk and Uncertainty 2000;21:117–49. [47] Johansson-Stenman O, Martinsson P. Are some lives more valuable? An ethical preferences approach. Journal of Health Economics 2008;27(3):739–52. [48] Covey J, Robinson A, Jones-Lee M, Loomes G. Responsibility, scale and the valuation of rail safety. Journal of Risk and Uncertainty 2010;40:85–108. [49] de Blaeij A, Florax RJGM, Rietvald P, Verhoef E. The value of statistical life in road safety: a meta-analysis. Analysis and Prevention 2003;35(6):973–86, Accident http://dx.doi.org/10.1016/S0001-4575(02)00105-7. [50] Svensson M, Vredin Johansson M. Willingness to pay for private and public road safety in stated preference studies: why the difference? Accident Analysis and Prevention 2010;42:1205–12. [51] Sjöberg L, Ogander T. Att rädda liv Kostnader och effekter (Ds 1994:14). Finansdepartementet ESO; 1994. [52] NIER. Does remediation save lives? On the cost of cleaning up arseniccontaminated sites in Sweden. The National Institute of Economic Research, Working Paper No. 108. Stockholm; 2009. [53] Mason H, Jones-Lee M, Donaldsson C. Modelling the monetary value of a QALY: a new approach based on UK data. Health Economics 2009;18(8):933–50. [54] Socialstyrelsen Nationella riktlinjer för prostatacancersjukvård – Medicinskt och hälsoekonomiskt faktadokument. Stockholm: Socialstyrelsen; 2007. [55] NICE. Guide to the methods of technology appraisal. Available at: http://www.nice.org.uk/page.aspx?o=201974; 2004 [August 2010]. [56] NICE. Social value judgments: principles for the development of NICE guidance, second ed. Available at: http://www. nice.org.uk/media/C18/30/SVJ2PUBLICATION2008.pdf; 2008. [57] Nafstad P, Håheim LL, Wislöff T, Gram F, Oftedal B, Holme I, Hjermann I, Leren P. Urban air pollution and mortality in a cohort of Norwegian men. Environmental Health Perspectives 2004;112(5): 610–5.