Cancer values of prevented fatalities (VPFs), one size does not fit all: The benefits of contaminated site cleanups in Italy

Abstract

The authors conducted a survey based on conjoint choice experiments in Milan, Italy, about mortality risk reductions delivered by hypothetical private behaviors and public programs, and used it to estimate the value of a prevented fatality (VPF) when the cause of death is cancer. Their estimate of the VPF is €4.2 million. The VPF is about €1 million larger when the risk reduction is delivered by a public program, but further analyses reveal that it is so only when the respondent believes that public programs are effective at reducing this particular type of mortality risk. This estimate of the VPF is higher than generic European Union–wide figures recommended by the European Commission Directorate-General for Environment (DG Environment) for environmental policy analyses, and is comparable to other VPFs that are appropriate for Italy, hazardous waste regulations, and enforcement-based cleanup programs. The authors use their VPF to compute the benefits of addressing leaking landfills, illegal disposal of hazardous wastes, and poor hazardous waste management practices in the provinces of Naples and Caserta in southern Italy. The authors also examine the importance of the discount rates, since the mortality benefits of remediation begin in 20 yr and are assumed to continue over 30 yr.

Implications:

Cost-benefit analysis that includes monetized nonmarket goods and services such as adverse health and premature mortality effects has become a standard tool in project appraisal and policy decision-making. The authors' estimates of the mortality benefits of cleaning up hazardous waste sites in a region of southern Italy are much larger than previous calculations that relied on the guidelines by DG Environment, raising concerns that using one VPF for all member countries of the European Union, and/or VPF figures estimated in a private risk reduction setting, might considerably underestimate the benefits of cleanup policies.

Introduction and Motivation

Despite considerable progress in remediation efforts over the last decade, recent estimates suggest that there are roughly 250,000 sites in the European Union (EU) member countries where pollutants have severely contaminated soil and groundwater, and cleanup is needed (European Environment Information and Observation Network, 2007). This estimate accounts for some 8% of the total 3 million sites where potentially polluting activities have taken place, and is expected to grow as investigation activities, monitoring, and data collection become more accurate and more evenly implemented throughout Europe. As of 2006, only about one-third of these 250,000 sites had actually been cleaned up (EIONET, 2007).

Remediation is undertaken to prevent or stop exposure to contaminants that may cause adverse health effects to people. Pollutants often found at contaminated sites include, among others, oil and heavy metals, polycyclic aromatic hydrocarbons (PAHs), aromatic hydrocarbons from the decomposition of petroleum products (BTEX), phenols, and chlorinated hydrocarbons (CHCs), all of which can cause cancer or other serious long-term illnesses.

Many EU countries support the “polluter pays” principle established by the EU's Common Strategy for Soil Protection (COM(2006)231 final) and the proposal for a Soil Framework Directive (COM(2006)232 final), and require the party who has caused the contaminated site to pay for cleanup (European Environment Agency, 2000; Van-Camp et al., 2004). In practice, implementing this principle has proven difficult and onerous.

When the parties that are responsible for the contaminated sites and should pay for cleanup cannot be identified, or are insolvent and cannot pay for cleanup, governments have had to finance cleanups using public monies from waste taxes, loan systems, or voluntary agreements with industry. When government-supervised and -funded remediation reduces the risk of dying from cancer attributable to contaminated site exposures, the monetized benefits of such a program are equal to the number of lives saved (premature deaths avoided) by the program, multiplied by a summary measure of the value that people place on mortality risk reductions. This summary measure is termed the value of a prevented fatality (VPF), also known as the value of a statistical life (see Viscusi, 1993).

Estimating the value of a prevented fatality for the purpose of computing the benefits of contaminated site remediation programs, however, is no easy task. To begin with, mortality risks are not traded in regular markets, and so the VPF must be estimated using nonmarket valuation methods. Second, many observers have argued that since contaminated site exposures cause cancer and other long-term illnesses, it is inappropriate to use estimates of the VPFs derived in the context of workplace or road traffic accidents. (Workplace accidents are the source of VPF estimates used in U.S. environmental policy analyses. In the United Kingdom, much research has been done in the area of transportation accidents. The European Commission Directorate-General for Environment [DG Environment] [2001] adapted the latter to the environmental policy context.) Cancer and long-term, chronic illnesses are highly dreaded, cause pain and suffering prior to death, and are experienced years after exposure, which may imply different VPFs than for accidental deaths.

Third, contaminated site exposures are likely involuntary, and earlier literature shows that people place a higher value on reducing death risks that they consider involuntary (e.g., Revesz, 1998; Cookson, 2000). Finally, to avoid double counting, recent stated-preference studies elicited the willingness to pay to reduce private mortality risks (see Krupnick, 2007), but with contaminated site policies the risk reductions are delivered by public programs, and estimates of the VPF based on private risk reductions may be too conservative.

In late November and December 2008, we conducted a stated-preference survey in Milan, Italy, where we asked respondents to answer a series of conjoint choice experiment questions. We wished to contrast cancer with mortality risk reductions due to noncancer chronic illness, and with accidental death risks. Since cancer risks are often latent, we devised a study design that allowed us to disentangle cancer effects from latency effects. Finally, we wished to get respondents to trade off private risk reductions with public-program risk reductions within the same survey, and we wanted to examine the opposite ends of the age spectrum—children and the elderly.

Accordingly, the alternatives in the conjoint choice experiments were risk reduction alternatives defined by five attributes: (i) the size of the risk reduction; (ii) the cause of death (cancer, respiratory illnesses, or road traffic accidents); (iii) the number of years that must elapse before the risk reduction begins; (iv) whether the risk reduction would be delivered by a public program or by a private action (e.g., medical test or product); and (v) cost to the respondent.

Survey participants were selected to be representative of the population of Milan residents aged 20–60 and with at least one child aged 17 or younger. Half of the respondents were to answer the choice questions assuming that the risk reduction applied to himself or herself, and the other half as if the risk reduction applied to one of his or her children. This study design allows us to estimate the cancer VPF, check whether the VPF is different for children and adults, and establish by how much public-program VPFs differ from VPFs in a private risk reduction context.

Using the responses to the choice questions, we estimate the cancer VPF to be €4.2 million (2008 euro). The VPF does not differ across children and adults, and is significantly larger in a public-program context, but only if respondents believe that the public program is potentially effective in reducing risks.

We use our estimates of the VPF to compute the benefits of eliminating mortality health risks attributable to uncontrolled hazardous wastes in the provinces of Naples and Caserta in southern Italy. Guerriero and Cairns (2009) estimate that 803 lives are lost every year as a result of improper landfilling of municipal solid waste and illegal disposal of hazardous waste in those areas. To calculate the mortality benefits of policies that address the uncontrolled disposal of wastes, they use an estimate of the VPF, combined with assumptions about latency, the horizon over which the risk reductions delivered by the policy would take place, and the discount rate.

Guerriero and Cairns (2009) write that “the WTP approach has not been used to estimate the VPF in Italy, nor in the context of waste exposure,” and so they use the VPF suggested by the European Commission DG Environment for cost-benefit analysis purposes—both the “generic” VPF as well as the one specific for cancer deaths. Contrary to their claims, a number of original studies are available for Italy that estimate the VPF or related metrics using the total willingness to pay (WTP) approach, including Alberini et al. (2007b), which is specific to the hazardous wastes and contaminated sites context, and Alberini and Chiabai (2007a), where attention is restricted to cardiovascular disease, which has been linked with heavy metal exposures. These studies were based on broad national samples in Italy, used current stated-preference methods, and focus on causes of death that match more closely those truly associated with hazardous waste site exposure.

We recalculate the benefits of addressing improper landfilling and uncontrolled hazardous waste disposal in the provinces of Naples and Caserta using our VPF figure plus figures from these other original studies in Italy, and show that the earlier Guerriero and Cairns analysis vastly underestimated the mortality benefits of remediation. The context- and cancer-specific VPF figures for Italy are at least twice as large as the figures recommended by DG Environment, and are the reason why we obtain much higher mortality benefits. This questions the use of one-for-all European Union–wide VPF estimates.

The approach implemented in this paper could be applied at a variety of locales, as well as to regulations that prospectively regulate hazardous waste management and handling. Since it requires estimates of the mortality risks associated with hazardous waste exposures, and it focuses exclusively on mortality risks, it is different from that suggested in Kiel and Zabel (2001) or Greenstone and Gallagher (2008), which use changes in property values associated with discovery of contamination and cleanup to estimate the benefits of Superfund and related cleanup programs. Gayer et al. (2000, 2002) and Davis (2004)infer the value of a statistical case of cancer from the depreciation and appreciation in property values observed as information is released about risks at specific locations. Because of its focus on the beneficiaries of the risk reductions, rather than on the agencies making decisions, our approach is also different from the analyses of agency cleanup decisions in Viscusi and Hamilton (1999) or Gupta et al. (1996)

Attention in this paper is restricted to enforcement-based cleanup programs, but we wish to emphasize that implementing such policies (as well as prospective regulation of hazardous waste disposal and treatment) has potentially important implications for real estate transactions (e.g., Segerson, 1994, 1997; Howland, 2002) and for the redevelopment of potentially contaminated properties (Adams et al., 2001). The remainder of this paper is organized in follows. The second section presents the relevant concepts and metrics, discusses reasons for the existence of a “cancer premium,” and reviews the relevant literature. The third section describes our study design. The fourth section presents the model. The fifth section presents data and results. The sixth section presents benefits calculations, and the seventh section concludes.

Background and Previous Literature

Institutional context

In Italy, the first piece of legislation addressing hazardous waste sites—the Waste Act—was passed in 1997 (Gazzetta Ufficiale, 1997). The statute spells out regulations for handling hazardous wastes so that they do not end up posing risks to human health. Ministry Decree No. 4/1/1999 addresses existing contaminated sites, requiring cleanup if the concentrations of certain pollutants exceed the maximum contaminants limits set by the law for soil and water (regardless of actual exposure). A recent law (Legislative Decree 152/2006) requires that risk assessments be conducted at sites where pollutants exceed the maximum concentration limit, and that remedial plans be based on such risk assessments. Remediation is recommended when excess lifetime cancer risk exceeds 10:5 (Gazzetta Ufficiale, 2006).

Only contaminated sites on the National Priorities List (NPL) qualify for cleanup paid for by the national government; for instance, remediation is paid for with public monies if the responsible parties cannot be tracked down or no longer exist. At this time, cleanup costs for the first 40 sites enlisted in the Italian NPL are expected to be about €3 billion and the costs for the remaining 14 is several billion euros (Bottarelli, 2008), but the available public funding is only €541 million. This means that to prioritize cleanup effort, it is useful and important to estimate the benefits of cleanup. When attention is restricted to mortality outcomes, the benefits are equal to the monetized value of the mortality risk reductions delivered by cleanup over the appropriate population. The benefits start with a delay relative to cleanup, due to the latent effects of most pollutants at contaminated sites, and can continue for decades. We discuss what this monetized value is, and how it is typically estimated, in the next sections.

What is the VPF?

The value of a prevented fatality (VPF; also known as the value of a statistical life or VSL) is the marginal value of a reduction in the risk of dying, and is therefore defined as the rate at which people are prepared to trade off income for a risk reduction, holding utility constant, as shown in eq 1:

(1)

where R is the risk of dying. The VPF can be equivalently described as the total willingness to pay (WTP) by a group of N people experiencing a uniform reduction of 1/N in their risk of dying. To illustrate, consider a group of 10,000 individuals, and assume that each of them is willing to pay €30 to reduce his, or her, own risk of dying by 1 in 10,000. The VPF implied by this WTP is €30/0.0001, or €300,000.

As shown in this example, in practice an approximation to the VPF is often computed by first estimating the WTP for a specified risk reduction, ΔR, and then by dividing WTP by ΔR. The mortality benefits are then computed as VPF × L, where L is the expected number of lives saved by the policy.

There is a reasonable degree of consensus in academic and policy circles that, when conducting cost-benefit analyses of proposed environmental or safety regulations, the benefits of reducing mortality risks should be valued using the VPF (Office of Management and Budget, 2003; U.S. Environmental Protection Agency [EPA], 2000, 2010a). However, the terms VPF and VSL sometimes lead to confusion and alarm among policy-makers and the general public. For this reason, the EPA recently proposed to use WTP estimates for a risk reduction that are not summarized in terms of a statistical death avoided (EPA, 2010b; Science Advisory Board, 2011). The proposed alternative measure is the “value of a mortality risk,” which measures WTP in terms of a unit risk reduction (e.g., 1 in a million) that is relevant to the policy being evaluated.

Methods for estimating the VPF

People do not directly trade mortality risks in marketplaces, and so it is not possible to infer the value that people place on improved safety from prices and quantities. To estimate the VPF, it is necessary to use nonmarket methods. One approach is to observe the compensation required by workers for them to accept riskier jobs (Aldy and Viscusi, 2007; Viscusi, 1993; Viscusi and Aldy, 2003). Despite econometric difficulties and recent evidence questioning the interpretation of the results from compensating wage studies (Black and Kniesner, 2003; Hintermann et al., 2010), the VPF figures currently used by the EPA in its environmental policy analyses reflect primarily this approach (EPA, 2010b).

Alternatively, it is possible to infer the VPF by observing the expenditures incurred by people to reduce their risks of dying in an accident (e.g., Jenkins et al., 2001), the prices of vehicles with additional safety features (Andersson, 2005), the time spent engaging in protection (Blomquist, 1979), and other consumer behaviors (reviewed in Blomquist, 2004). Stated-preference studies, such as contingent valuation or conjoint choice experiments, have proven useful in the absence of sufficient variation in observed risks, or when observed risks, prices, behaviors, and populations do not match those addressed by the policy of interest. They rely on individuals saying what they would do under hypothetical circumstances, rather than observing actual behaviors in marketplaces (Alriksson and Öberg, 2009).

In contingent valuation surveys, respondents are asked to report information about their WTP for a hypothetical risk reduction that is specified to them in the course of a survey. Dekker et al. (2010) review studies that have elicited the VPF using contingent valuation, and present a meta-analysis of such VPF estimates. Among other things, they find that studies that posited larger risk reductions elicited higher VPF figures, a result that is consistent with economic theory, and that studies that deployed scenarios with public programs tended to elicit lower VPF estimates.

In conjoint choice experiments, respondents are shown K (K ≥ 2) alternative variants of a hypothetical good or policy described by a set of m attributes, and are asked to choose the one they most preferred (Bateman et al., 2002; Hanley et al., 2001). The alternatives differ from one another in the levels taken by two or more of the m attributes. Price (or cost to the respondent) is usually one of the attributes, which allows the analyst to estimate the value people ascribe to the good or the monetized benefits of the policy.

The choice responses are assumed to be driven by an underlying random utility model. Alberini et al. (2007b) review basic econometric models used with conjoint choice experiments, and Swait (2007) presents more elaborate models that allow for preference heterogeneity (e.g., mixed logit and latent class models). Tsuge et al. (2005), Alberini et al. (2007a), and Alberini and Ščasný (2010, in press) are recent examples of studies that have estimated the VPF from conjoint choice experiments.

Cause and context

Many observers question whether the VPF in an accidental death context should be applied when the cause of death is cancer or another illness with different characteristics, since the attributes of the risk itself may matter. This concern is reported in government sources as well (e.g., Office of Management and Budget, 2003, p. 29). Cancer is usually delayed with respect to environmental exposures, associated with suffering and pain, and highly dreaded (see Starr, 1969; Fischhoff et al., 1978; Slovic, 1987; and Chilton et al., 2006), which is often taken to imply that the VPF should be greater when the cause of death is cancer. Surveys where subjects were asked to allocate funds or otherwise state their preferences for public programs that save lives have often found that respondents give higher priority to programs where the cause of death is cancer (e.g., Jones-Lee et al., 1985; McDaniels et al., 1992; and Savage, 1993).

Earlier research has focused on the importance of voluntariness, finding that the more a risk is perceived as voluntary, the lower the willingness to pay to reduce it (Cookson, 2000; Revesz, 1998). This consideration suggests that, all else the same, people might be willing to pay more for a reduction in the risk of dying from cancer due to air pollution or contaminated sites (which are largely involuntary risks) than from cancer caused by, for example, smoking. The literature also suggests that the higher one's personal exposure to the risk, the greater the willingness to pay to reduce it (Chilton et al., 2002, 2006; Savage, 1993). For some individuals, such as smokers, the offsetting effects of voluntariness and higher risks may, however, coexist (see, e.g., Dickie and Gerking, 2010).

In policy practice, the ExternE project series considered the use of a cancer premium for fatal outcomes due to heavy metals and radionuclides; the 2005 update of the methodology suggested a 50% premium for fatal cancer (European Commission, 2010). A similar cancer premium was adopted by (DG Environment, 2001). By contrast, the EPA applies a single VPF in its policy analyses, regardless of the cause of death (see EPA, 2011); however, different U.S. agencies apply different VPF values that might reflect the effect of different contexts (see, e.g., Robinson and Hammit, 2011).

Despite these arguments and debate in policy and academic circles, there is surprisingly little empirical evidence that people are prepared to pay more to reduce the risk of dying from cancer than that from other causes. Magat et al. (1996) find that the median survey participant was indifferent between reducing the risk of terminal lymph cancer and reducing the risk of automobile death, implying that the VPF for the former is the same as that for the latter. Hammitt and Liu (2004) elicit WTP for reductions in the risks of acute and latent cancer and noncancer illnesses affecting the lung or the liver. WTP to reduce cancer risks is about 40% larger than WTP to reduce a risk of a similar chronic, degenerative disease (with a VPF of around $2.88 million [2008 US dollars] for acute lung cancer, or of $2.05 million for acute lung noncancer), but the coefficient on the cancer dummy was significant only at the 10% level.

Tsuge et al. (2005) conduct conjoint choice experiments where they vary the cause of death, the size of the risk reduction, and the degree of latency, and conclude that it is unnecessary to adjust the VSL according to the differences in the type of risk, if the VSL is calculated by using an “adequate approach.” Hammitt and Haninger (2010) find no discernible difference between the WTP for cancer and noncancer diseases, and that for reducing the risk of dying in a motor-vehicle crash.

If the evidence about the cancer VPF is ambiguous, even less is known about the VPF for other causes of death that have been linked with environmental exposures, such as cardiovascular illness. We are only aware of few studies in this area—for instance Alberini and Chiabai (2007a, 2007b) and Tsuge et al. (2005).

People may be willing to pay more for a public-program risk reduction due to their altruistic considerations (see, e.g., Viscusi et al., 1988; Zhang et al., 2006; or Andersson and Lindberg, 2009. However, Jones-Lee (1991) shows that individual WTP for reducing the risks of others should be included among the benefits in cost-benefit analyses if and only if altruism is exclusively safety-focused, that is, the individual cares about the individual health of others but is completely indifferent to other determinants of the well-being of others. Unfortunately, in applied work it is very difficult to observe the nature of each respondent's altruism (paternalistic vs. pure), and efforts to tell respondents what to assume about other people's payments have proven awkward and confusing (Johannesson and Meltzer, 1998). This has prompted many researchers to turn to valuing private risk reductions to avoid double counting (e.g., Dickie and Gerking, 1996; Krupnick et al., 2002; see Krupnick, 2007).

Earlier estimates of the VPF for Italy

In this section, we review recent studies that we deem suited for valuing the mortality benefits of contaminated site remediation programs in Italy. We identified studies that met at least two of the following four criteria: They (i) estimate the VPF directly from the potential beneficiaries of mortality risk reductions using surveys; (ii) were conducted in Italy; (iii) present scenarios that entail hazardous waste sites; or (iv) value reductions in the risk of dying from causes that have been linked with hazardous waste exposures (e.g., cancer, chronic illnesses, cardiovascular diseases).

Alberini et al. (2006) conduct a contingent valuation survey in several Italian cities that elicits the WTP for a reduction in the risk of dying of either 1 or 5 in 1000 over 10 yr. The risk reduction covers any (nontrauma) cause of death, and is delivered by an unspecified “product” and an abstract scenario (see Alberini et al., 2004; Krupnick et al., 2002). The VPF is €1.163 million (2008 euro) or €2.577 million (2008 euro), depending on whether median or mean WTP is used.

Alberini and Chiabai (2007a, 2007b) survey individuals in five cities in Italy (Venice, Milan, Genoa, Rome, and Bari) in May 2004 using a survey instrument that draws from Krupnick et al. (2002) and Alberini et al. (2006), but focuses on the risk of dying for cardiovascular causes and uses a greater range of risk reductions (up to 12 in 1000 over 10 yr). Independent samples of respondents consider either a hypothetical preventive medical intervention (or diagnostic test) or a completely abstract risk reduction. For a risk reduction of 1 in 10,000 a year—which is close to the annual mortality risks attributable to uncontrolled wastes or other environmental exposures (see later section)—the VPF for cardiovascular disease for persons aged 30–49 is €2.486 million (if median WTP is used) or €5.299 million (if mean WTP is used) (all figures in 2008 euro). For comparison, for persons aged 60–69 the respective VPF is €1.264 million or €2.696 million, and if they already have a cardiovascular condition, the VPF is €1.770 million or €3.774 million (2008 euro). Alberini and Chiabai (2007b) further ask people to report information about their WTP now for a future risk reduction, and estimate the discount rate implicit in people's responses, which is 0.3–1.7%, depending on whether different WTP responses within the same individuals are allowed to be correlated or modeled as statistically independent.

Alberini et al. (2007a) conduct conjoint choice experiments about enforcement-based cleanup programs for contaminated sites. The alternatives in the conjoint choice experiments are stylized public programs that would address uncontrolled hazardous waste sites (including poorly managed landfills, industrial plants, etc.) and are described by five attributes. These are (i) the annual risk reduction, expressed as the number of lives saved in a million; (ii) the size of the population that would benefit from this risk reduction (0.5, 1, and 2 million); (iii) the latency period until the risk reductions begin (2 or 10 yr); (iv) the years over which the risk reductions would be experienced (T = 20, 30, 40, and 45 yr); and (v) the cost of the policy to the respondent's household, which would be incurred immediately and paid one time only. The causes of death whose risk would be reduced are cancers and other serious long-term illnesses associated with exposures that originate from contaminated sites. The study, however, does not distinguish between them.

The survey respondents were residents of the cities of Venice, Milan, Naples, and Bari, all of which have contaminated sites on the Italian National Priorities List. As such, the respondents may have been beneficiaries of the risk reductions proposed in the survey—together with other people living in those areas.

Alberini et al. (2007a) estimate the VPF to be €5.976 million (standard error around the VPF €0.826 million; 2008 euro) for an immediate risk reduction. No differences in VPF were detected across the cities where the study was conducted. Since the discount rate implicit in the respondents' choices is estimated to be 7.41%, it follows that the VPF would be only €2.849 million if the latency period is 10 yr, €1.350 million if the latency period is 20 yr, and €0.647 million if the latency period is 30 yr (all figures in 2008 euro).

Tonin et al. (2011) estimate the value of a statistical case of cancer (VSCC), namely the willingness to pay for a marginal change in the risk of developing cancer (which may or may not be fatal). They deploy conjoint choice experiments, and the sample of respondents is selected among the residents living within specified distances of a major Superfund site in Italy, the Marghera chemical complex, which is on the mainland side of Venice. The VSCC is €2.696 million (standard error €0.283 million; all 2008 euro), is highest among those respondents who live closest to the contaminated sites, and increases with income.

Study Design

Our conjoint choice experiments

In late November to mid-December 2008, we conducted a survey of residents of Milan, Italy (see Alberini and Ščasný, 2010, in press) and asked them to engage in several conjoint choice tasks. Half of the respondents were to assume that the alternatives in these choice tasks would apply to them, and the other half that these choice tasks would apply to one of their children (selected at random from the respondent's children). Assignment to one or the other variant of the questionnaire was random.

The attributes we used to describe the hypothetical alternatives were (i) the size of the mortality risk reduction; (ii) the cause of death to which the risk reduction applies (cancer, respiratory illnesses, road traffic accidents); (iii) whether the risk reduction would be delivered by a public program or would be privately undertaken (in which case it would apply exclusively to the respondent or his/her child); (iv) latency (0, 2, 5, and 10 yr); and (v) a one-time cost to the respondent, to be incurred now. Attributes and attribute levels are summarized in Table 1.

Table 1. Summary of attributes and attribute levels in the conjoint choice experiments

Attribute	Number of Levels	Levels
		Cancer
Context (cause of death)	3	Road traffic accidents
		Respiratory illnesses
		Private good (no other beneficiaries);
Private good or public program	2	Nationwide public program (other beneficiaries)
Latency	4	0, 2, 5, 10 yr
Size of the risk reduction	4	2, 3, 5, 7 in 10,000 over 5 yr
Cost to the respondent (one-time)	4	200, 500, 1000, 2000 euros

Respondents viewed, and answered choice questions about, a total of five pairs of alternatives; for each pair, about half of the subjects were asked to indicate which they preferred—alternative A or B (a forced choice question). This question was followed by another choice task where the respondent was asked to pick the most preferred among alternative A, alternative B, and the status quo (which entails no risk reduction, and no cost to the respondent). The other half of the respondents were asked directly to choose among alternative A, alternative B, and the status quo. Assignment to the former (TFORMAT = 1) or latter group (TFORMAT = 2) is random.

To ease the respondents' task, in the first two pairs of alternatives, alternatives A and B focused on the same mortality risk (e.g., cancer). Moreover, the latency period (namely, the number of years that must elapse before the risk reduction is experienced) was the same across alternatives A and B within the same pair (but varied across pairs and respondents).

The questionnaire was self-administered by the respondents using the computer, and resulted in a total of 1906 completed questionnaires. In what follows, attention is restricted to the subsample of n = 647 respondents who valued cancer risk reductions in the first two screens of the conjoint choice portion of the interview. We choose to restrict attention to the first two questions to eliminate the possibility that subsequent answers were conditioned by the order in which the cause of death appeared in the questionnaire and fatigue effects, if any. Of these, n = 638 respondents provided valid (i.e., nonmissing) answers to the conjoint choice questions, for a total of N = 1931 usable responses.

Selection of attribute and attribute levels

Public versus private risk reductions (item ii above) were presented to the respondents with a reminder that the former imply that there are other beneficiaries of the risk reduction beyond the respondent, whereas the respondent is the sole beneficiary of the risk reduction when the action is private. We described the public programs as being “nationwide.”

Our interest in the public versus private nature of risk reduction is driven by two reasons. First, earlier stated-preference research has asked people to value private risk reduction in hopes of avoiding double counting (Johannesson and Meltzer, 1998), but environmental and other safety regulations are part of public programs, and one wonders how widely the estimates of the VPF based on private risk reductions differ from those where the scenario centers around a public program. The former are generally regarded as conservative estimates of the VSL, but we are aware of at least one study (Johannesson et al., 1996) that actually found them to be larger than the VSL estimate for a comparable risk reduction in a public-program context. Second, private behaviors versus public programs provide us with an opportunity to study if controllability of risks—which depends on how effective people judge the risk-reducing measures—influences the WTP to reduce them.

We note here that when stated-preference methods are used to elicit the WTP for risk reductions, differences in the WTP for a public-program and a private risk reduction may, all else the same, be due to strategic behaviors. Conjoint choice experiments—the valuation method we use in this paper—are not incentive compatible. However, our experience with conjoint choice experiments is that people are so busy trading off the alternatives that they do not think about strategic behaviors. Moreover, a public program is more realistic for the specific context of this paper.

Latency is expressed as the number of years that elapse before the risk reduction begins. To avoid confounding between the cause of death and the latency aspect, we used latency levels of 0 (= immediate risk reduction), 2, 5, and 10 yr, and we varied this attribute independently of the context of death and the other attributes.

We used four possible levels for the risk reduction, namely 2, 3, 5, and 7 in 10,000 over 5 yr. Finally, each alternative risk reduction plan had a price tag. This cost would be incurred by the respondent's household immediately and would be paid this one time only. We used four possible cost amounts ranging from €200 to €2000 (see Table 1). Under alternate assumptions about the discount rates, these cost amounts correspond to VSL of a few hundred thousand to several million euros. Each respondent was randomly assigned to a set of five pairs of risk reduction profiles. There were a total of 32 possible sets, which were constructed by (i) creating the full factorial design; (ii) forming all possible pairs of alternatives, but discarding those that contained identical or dominated alternatives; and (iii) selecting five pairs at random without replacement from the universe in ii. Procedure iii was repeated 32 times (see Alberini and Ščasný [in press] for more details about the design).

In this paper, however, we use the responses to only the first two pairs and limit the sample to those respondents for whom such pairs were cancer risks (about one-third of the total number of subjects). In practice, depending on the random assignment of the respondent to TFORMAT = 1 or TFORMAT = 2, the sample used in this paper is based on two or four choice responses per person (see Table 2). The purpose of this split-sample treatment was to check whether the forced choice exercise influences the estimates of the coefficients, possibly because of violations of the IIA assumption implicit in the conditional logit model. In practice, we found that it did not, and for this reason in this paper we pool the responses from both versions of the questionnaire.

Table 2. Summary of pairs and choice responses used in this paper

		Choice Set
		Treatment TFORMAT = 1	Treatment TFORMAT = 2
Pair 1 (alternatives A and B)	First choice task	A vs. B	—
	Second choice task	A vs. B vs. status quo	A vs. B vs. status quo
Pair 2 (alternatives C and D)	First choice task	C vs. D	—
	Second choice task	C vs. D vs. status quo	C vs. D vs. status quo
Total number of choice responses per person		4	2

Structure of the questionnaire

Respondents self-administered the questionnaire using the computer. The computer “interview” starts with the respondent entering his or her gender, age, and the name, age, and gender of each of his or her children. The computer program then selects at random one of the respondent's children aged 17 and younger for future questions.

Section A asks questions about the health status of the selected child, and section B about the respondent's own health. Section C of the questionnaire elicits extensive information about lifestyle, environment, genetic predisposition to cancer, and familiarity with it. The purpose of this section is to understand the salience of certain risks to the respondent and to get a sense of the exposure to certain risk factors.

Section D contains a probability tutorial. We start with a simple and intuitive presentation based on tossing a coin or casting a die, but point out that the notion of chance also applies in other familiar situations (e.g., the weather forecast and the chance of rain). This is followed by a simple quiz to make sure that people have grasped the basics of probability. We then move on to the notion of mortality risks. We use two visual representations of risk: (i) a grid with 10,000 squares, which we use when attention is restricted to a reference group or population; and (ii) bar charts, which we use when we want to show how risks vary across age groups (and hence change as a person ages).

In section E of the questionnaire, we inform the respondent that it is possible to reduce one's own risk of dying in many ways. Using respiratory illnesses, cancer, and road traffic accidents as examples, we explain that risk reductions may result from individual actions (e.g., getting a flu shot, purchasing a car with safety equipment) and government programs (e.g., an air pollution control program). We also emphasize that some actions are specific for men (e.g., prostate cancer tests), some are specific for women (e.g., Pap smears), and others apply only to children (child seats in cars).

Section F contains exercises to make respondent focus on the magnitude of the risks. In section G of the questionnaire, we zero in on three causes of death, namely cancer, respiratory illnesses, and road traffic accidents. We describe them briefly and then ask for the respondents' subjective assessments of the comparative magnitude of these risks for people their age. In section H, the respondents express their opinions on the effectiveness of private actions and public programs in reducing the risk of dying for each of the three causes of death studied in this project, and also assess the futurity versus immediateness of risks.

Section I is dedicated to the conjoint choice questions. In the first screen of this section, we summarize the five attributes of the alternatives being compared, remind respondents of their own baseline risks, and point out that the choice experiments will contain relatively small risk reductions.

After the conjoint choice questions, we ask debriefing questions and explore reasons for the observed choices. The final section of the questionnaire elicits information about the respondent's socioeconomic circumstances.

Survey administration

Our questionnaire is self-administered by the respondent using the computer. Respondents took the survey in two facilities in Milan, Italy, and were paid €10 for their participation in the survey. The final survey was preceded by two pilot studies, which were conducted in Milan in June (n = 200) and September (n = 100) 2008, respectively. Both pilots used the same sampling frame as the final survey (see Data and Results).

We chose to restrict attention to Milan, rather than to a nationally representative sample, for two reasons. First, limiting the survey to a single city was significantly less expensive and allowed us to increase the sample size, albeit at the expense of national representativeness. Second, there are many hazardous waste sites (usually shuttered or active industrial plants) in Milan, and the residents are well informed about bad air pollution episodes, which are generally covered by the local news, and about the health effects of air pollution and other types of environmental exposures. Data from the Italian Health Ministry and the Italian Statistics Agency show that cancer mortality is comparatively high in the Milan area.

The Model

To model the responses to the first two choice questions, we rely on a random utility framework, which posits that the respondent's indirect utility is:

(2)

where DR is the discounted risk reduction (see below), α is the marginal utility of a unit of risk reduction, β is the marginal utility of income, (y − C) is residual income, since C is the cost of alternative j, and subscripts i and j denote the individual and the alternative, respectively.

Assuming constant exponential discounting, the discounted risk reduction is defined as:

(3)

where ΔR is the risk reduction, L is the number of years that elapse before the risk reduction begins, and δ is the discount rate.

On appending an error term, which captures aspects of the indirect utility that are known to the respondent but not the analyst, we obtain the random utility model:

(4)

We assume that the respondent will choose the alternative that yields the highest indirect utility. If we further posit that the error terms in eq 4 are independent and identically distributed (i.i.d.) and follow a standard type I extreme value distribution, the probability that the respondent chooses alternative k is:

(5)

Omitting the subscript i to avoid notational clutter, the expression in eq 5 is the contribution to the likelihood of a conditional logit model (see Train, 2003). K is the number of alternatives that the respondent is to examine in each choice task, and is equal to 2 when the respondent must indicate the most preferred alternative between program A and program B, and to 3 when he or she is to select the most preferred option among A, B, and the status quo (no risk reduction, no payments). (We remind the reader that all respondents receive choice questions with K = 3. Only about half of the respondents receive choice questions with K = 2, followed by a choice question with the same two program plus the status quo, where K = 3.)

Once maximum likelihood estimates of parameters α, β, and δ are obtained, we can use them to compute the VPF, which is . In other words, the VPF is the marginal utility of a unit of mortality risk reduction, converted into euros through division by the marginal utility of income.

Since by design we vary the risk reduction and the latency periods across and within respondents, it is possible to estimate the discount rate directly from the survey responses. If the discount rate is found to be equal to zero, indirect utility in eq 2 are simplified to:

(6)

2–6 assume that the VPF is constant for all individuals in the sample, but the model is easily amended by entering interactions between certain attributes, and between the risk reduction and individual characteristics of the respondent, so that we can check whether certain attributes and respondent characteristics affect the VPF. We also allow for random coefficients, examine whether it is appropriate to enter both (the discounted flow of) risk reductions and cost linearly in the utility function, and check whether a status quo–specific intercept should be entered in the model.

Data and Results

The data

Our target population was Milan residents aged 20–60 who had at least one child of aged 17 or younger. We asked our survey firm to make sure that the sample be evenly divided among three age groups, namely persons aged 20–34, 35–44, and 45–60. The sample was to have an even number of mothers and fathers, and to mirror the city's population in terms of education and income. For example, in Milan, roughly 23% of the residents aged 20–60 have a college degree, and mean (after tax) household income is about €30,000 a year. We also specified that no more than 20% of the mothers in the sample should be homemakers. As shown in Table 3, the sample used in this paper is consistent with the sampling plan.

Table 3. Descriptive statistics of the sample

Variable	Mean	SD	Min	Max
Female (dummy)	0.492	0.500	0	1
Age (yr)	39.691	10.288	20	59
Age2034 (20–34 age group) (dummy)	0.355	0.479	0	1
Age3544 (35–44 age group) (dummy)	0.306	0.461	0	1
Age4560 (45–60 age group) (dummy)	0.338	0.474	0	1
Matura (high school diploma) (dummy)	0.370	0.483	0	1
Somecollege (some years of college) (dummy)	0.060	0.237	0	1
College (college degree or better) (dummy)	0.251	0.434	0	1
Highincome (after-tax household income >€30,000/yr)	0.415	0.493	0	1
Income (after-tax household income, euro/yr)	30,619	11,398	5000	80,000
Homemaker (dummy)	0.060	0.238	0	1

By design, the attributes were balanced across alternatives. About 43% of the responses were in favor of alternative A, 38% were in favor of alternative B, and 19% preferred the status quo. This breakdown is similar across all latency levels. To illustrate, 18% of the respondents selected the status quo when the latency level were 0, 2, and 5 yr, and 20% selected the status quo when the latency period was 10 yr. This suggests that (1) short latency periods for cancer were not judged unrealistic by our respondents (had this been the case, pairs with shorter latencies result in more frequent status quo choices than pairs with longer latencies); and (2) people do not discount future risk reduction heavily (else, they would have chosen the status quo more frequently for longer latency periods).

Estimation results

In preliminary runs based on cancer-only program pairs, we found that limiting the sample to the early conjoint choice responses (those for the first two pairs of programs) resulted in estimates of the VPF that were slightly smaller (€0.736 million less) than those based on all cancer choice tasks. This effect is not statistically significant. In the remainder of this paper, we report results based on a sample that uses only the responses to the first two pairs of risk reduction profiles to reduce fatigue or conditioning on earlier tasks. (We note that in the literature, empirical evidence of learning and fatigue is study specific (Holmes and Boyle, 2005). In analyses that exploited the full sample in the present research project, we found that the VPF was remarkably stable across choice tasks (Alberini and Ščasný, 2010). Brouwer et al. (2010) report similar results, and Boeri (2011) finds that learning and fatigue effects are limited to relatively small shares of the respondents, with over three-quarters of them displaying stable preferences.

Estimation results based on the random utility model of eqs 2–3 and the responses for the first two pairs of alternatives are presented in Table 4. The key coefficients have the expected sign and are statistically significant. They imply a VPF of over €4 million. It is interesting that the discount rate is small and statistically insignificant when we use all responses, and becomes even smaller when attention is restricted to those respondents who valued own risk reductions. A similar result is observed when the sample includes only those respondents who were examining risk reduction for one of their children (not reported).

Table 4. Estimation results, RUM of eqs 2–3, choice responses from the first two pairs of alternative, cancer risks only

	All (Adult and Child Risk Reductions)		Adult Risk Reductions
Coefficient Label (Variable)	Coefficient	t Statistics	Coefficeint	t Statistics
α (DR, discounted flow of risk reduction)	0.2298***	11.291	0.2011***	7.238
β (cost)	−0.0005***	−9.96	−0.0005***	−6.915
δ (discount rate)	0.0201	1.531	−0.0063	−0.352
N	1931		959
Log L	−1705.69		−881.327
Value of a prevented fatality (VPF) (million euros)	4.594		4.179
Standard error around VPF (million euros)	0.418		0.566
Note: ***P-value < 0.01.

This suggests that it is acceptable to impose the restriction that δ = 0, and to reestimate the model after the RUM is simplified to eq 6. Results from these runs are displayed in Table 5.

Table 5. Conditional logit and mixed logit models of the responses to the conjoint choice question

Variables	Variable definition	A	B	C	D	E
Reduction	reduction in mortality risk	0.2116 *** (0.016)	0.2047 *** (0.022)	0.1989 *** (0.016)	0.1325 *** (0.034)	0.1395 *** (0.037)
Rpublic	risk reduction × public initiative			0.0476 *** (0.013)	−0.1524 *** (0.044)	−0.2238 *** (0.060)
Rgenetic	risk reduction × predisposition cancer				0.0269 (0.030)	0.0300 (0.0343)
Rfamily	risk reduction × a family member has had cancer				0.0167 (0.025)	0.0152 (0.0282)
Rchild	risk reduction × reduction is for the respondent's child				0.0025 (0.022)	0.0039 (0.0248)
Rsalience	risk reduction × respondent has cancer				0.0908 * (0.049)	0.0989 * (0.063)
Rfriends	risk reduction × someone close to you has cancer				0.0269 (0.023)	0.0365 (0.0263)
Rfamilies	risk reduction × someone in most families gets cancer				0.0008 (0.024)	−0.0025 (0.0267)
Rsmoker	risk reduction × smoker				−0.0113 (0.028)	−0.0042 (0.0306)
Rsmokacanc	risk reduction × smoking causes cancer				0.0459 (0.029)	0.0553 * (0.0317)
Rpubleff	risk reduction × public initiative × program effective				0.0556 *** (0.012)	0.0815 *** (0.0178)
Cost	One-time cost to household	−0.0005 *** (0.000)	−0.0005 *** (0.000)	−0.0005 *** (0.000)	−0.0005 *** (0.000)	−0.0006 *** (0.0001)
Std Dev Rpubleff	Standard deviation of the coefficient on rpubff					0.0716 *** (0.0237)
Std Dev Rsmoker	Standard deviation of the coefficient on rsmoker					−0.0019
						(1.7242)
Notes: ***P < 0.01; **P < 0.05; *P < 0.1; standard errors in parentheses.

The results generally indicate that the responses to the conjoint choice questions are internally valid. The coefficient on the risk reduction is always positive and significant, and that on the cost of the program negative and significant, as expected. The estimates in column A, which are based on the entire sample, imply that the cancer VPF is €4.164 million, with a standard error of €0.276 million (2008 euro). This figure is about twice as large as large as the central estimate for cancer recommended by the European Commission and updated to 2007 euro by Guerriero and Cairns (2009). Even if we reduced it by about 30% to account for the fact that our sample is relatively young (it is limited to persons aged 60 or less) when compared with the likely beneficiaries of environmental policies (Krupnick, 2007), the resulting VPF would be higher than that the figure used in Guerriero and Cairns (2009).

Since the risk reductions in this sample would be incurred by both children or adults, in column B we present estimation results when the sample is restricted to those who valued own (adult) risk reductions only. The VPF is virtually the same (€4.252 million, standard error of €0.420 million).

If the model is reestimated after including an alternative-specific intercept for the status quo, the key results are qualitatively similar, but (1) the VPF is smaller (€2.693 million); and (2) the alternative-specific intercept is negative and statistically significant, suggesting that perhaps some form of “action bias” might be at play.

In the models discussed until now, the risk reduction (or, in Table 4, the discounted flow of risk reductions) and cost were entered linearly in the model. To check if these assumptions are acceptable, we estimated models with (i) cost, cost square, and cost cube; (ii) risk reduction size dummies in lieu of a continuous risk reduction variable; and (iii) log risk reduction in lieu of risk reduction. All other terms were kept in linear form.

In specification i, the coefficients for the higher-order cost terms were statistically insignificant. Specifications ii and iii suggest that the VPF is higher at smaller risk reduction. For example, based on iii, we estimate the VPF to be €6.159 million for a 2 in 10,000 risk reduction, €4.106 million for a 3 in 10,000 risk reduction, €2.463 million for a 5 in 10,000 risk reduction, and €1.760 million for a 7 in 10,000 risk reduction. At the average risk reduction, however, the VPF is similar to the VPF from the linear model.

In column C of Table 5 we enter an interaction between risk reduction and a “public program” dummy to see if that changes the VPF, whereas column D enters interactions between risk reduction and individual characteristics, risk perceptions, and attitudes. We point out that all covariates in our models are interacted with the risk reduction. In other words, as economic theory would suggest, we posit that our respondent choose between risk reduction packages, that is, risk reductions that come with a variety of attributes. Results for an alternate model with a “main effect” for a public versus private program (i.e., one where individuals obtain utility even without any reduction in health risk) are not reported here and are available from the authors upon request. The econometric model is a conditional logit in columns A–D, and a mixed logit in column E where we allow for selected coefficients to be random variables in an effort to capture unobserved heterogeneity in tastes among our respondents. Using a normal mixing distribution has negligible effect when distributions are tight. For good measure, we also attempted other distributions that impose bounds, such as a uniform, and obtained similar results.

As shown in column C, people value cancer risk reductions more when they are delivered by a public program. The cancer VPF is about €0.950 million higher when the risk reduction comes from a public program. In columns D and E, however, we show that people value public risk reductions more only when they believe that public programs are “effective” at reducing cancer (where by “effective” we mean at least 4 on a scale from 1 to 5, where 1 = not effective at all and 5 = very effective). The magnitude of the negative coefficient on the public program dummy implies that it takes an effectiveness rating score of 2.74 (about the sample average) or higher for respondents to be willing to pay more for public risk-reducing programs.

Turning to individual circumstances that might affect the perceived risk of developing cancer, columns D and E show that people are prepared to pay the same amounts of money to reduce risks in their children as for themselves. We had expected thinking that cancer runs in the family, and having a family member (parent, grandparent, and sibling) who has or has had cancer to be positively associated with the VPF. This expectation is borne out in the data, but the coefficients, although positive, are not significant at the conventional levels.

We measure salience as (i) whether the respondent has or has had cancer, and/or (ii) has been hospitalized or taken to the emergency room for it. An interaction between risk reduction and the salience dummy suggest that persons who have first-hand experience with cancer hold a much higher VPF than the others (by almost €2 million), an effect that is significant at the 10% level. Another possible measure of experience and familiarity with cancer is whether a spouse or a close friend has had cancer, and this is likewise associated with a higher VPF (about €0.400 million more) but this effect is insignificant at the conventional levels. Smokers also are not statistically different from nonsmokers, but it is interesting that the coefficient on the interaction between risk reduction and being a smoker is negative (in other words, smokers appear to be more tolerant of cancer risks).

We also wished to check if agreement with the statement that “there will be a case of cancer in almost all families” (i.e., that cancer is very widespread) influences the VPF, but we find no evidence of such an association. Agreement with the statement that “Smoking is one of the major causes of cancer” tends to be associated with a higher VPF (about €0.800 million), although the effect is not significant at the conventional levels. Responses, and so WTP, may be affected by the respondent's altruistic consideration, risk perceptions, and risk aversion, but we are unable to control for these factors.

In addition to checking for heterogeneity that is systematically related to attributes of the alternatives or respondent characteristics, we also allowed for unobserved heterogeneity by fitting mixed logit. We experimented with letting different coefficients be random variables, but we found little evidence of random parameters. In the end, we settled for treating the coefficient on [risk reduction × effectiveness rating of public programs] and that on [risk reduction × being a smoker]. We selected the former because of the possibility that our other interaction terms do not fully capture the taste heterogeneity for public programs, and the latter because we expect smokers' perceived risk of dying from cancer to be high, which should increase their willingness to pay to reduce them. This expectation is based on earlier research that has shown that smokers (1) are aware of the cancer risks created by smoking; (2) in fact overestimate their risk of dying from lung cancer (Viscusi, 1990); (3) have a greater WTP to reduce their cancer risks (Cameron et al., 2009); and, based on compensating wage studies, (4) have a higher VSL than nonsmokers throughout their lives (Viscusi and Hersch, 2008). We further conjecture that some smokers may consider the proposed risk reductions (or risk reduction associated with public programs) as substitutes for their own effort. We posited that these coefficients follow independent normal distributions. The results in column E of Table 5 show that there is indeed heterogeneity across respondents in the marginal utilities of these interactions, since the estimated standard deviations of these marginal utilities are significant. All other coefficients, however, are treated as fixed and their estimates are similar to their counterparts in column D.

Benefits in the Naples and Caserta Provinces

Guerriero and Cairns (2009) estimate that a total of 848 lives are lost every year in the provinces of Naples and Caserta in southern Italy because of exposure to uncontrolled hazardous wastes (e.g., poorly constructed and managed landfills, illegal disposal of hazardous wastes, etc.). Out of these, 403 are cancer deaths.

To calculate the (avoided) mortality benefits of policies that address the uncontrolled disposal of wastes, Guerriero and Cairns use an estimate of the VPF, combined with assumptions about latency, the horizon over which the risk reductions delivered by the policy would take place, and the discount rate. Guerriero and Cairns use the VPF suggested by the European Commission DG Environment for cost-benefit analysis purposes—both the “generic” VPF as well as the one specific for cancer deaths.

Table 6 reports their estimates of the benefits that would be incurred if these excess risks were eliminated (through cleanup and better waste disposal practices in the future). Their VPF figures are taken straight out of DG Environment (2001) and simply updated to 2007 euro. We further update them to 2008 euro. For any fatal illness, they use a central VPF of €1.427 million, and low and high values of €0.968 million and €3.777 million, respectively (2008 euro). For fatal cancer, they apply a 50% premium, which results in VPF figures equal to €2.141 million (central estimate), €1.448 million (low), and €5.568 million (high), respectively (2008 euro). They further assume a latency period of 20 yr, that the risk reductions would occur for 30 yr thereafter, and that the discount rate is 0.04, the official discount rate used by the European Commission in its policy analyses.

Table 6. Mortality benefits of eliminating exposures to uncontrolled wastes in the provinces of Naples and Caserta

	Lives Lost Attributed to Waste Per Year	Benefits (Billion 2008 Euros) Central Estimate of the VSL (Lower Bound, Upper Bond)
All causes mortality	848	9.58
		(6.42, 25.49)
All fatal cancers	403	6.83
		(4.59, 17.33)
All cause mortality adjusted for cancer premium	848	11.83
		(7.95, 30.99)
Notes: Source: Guerriero and Cairns, 2009; European Environment Agency, 2000. 2007 euro (2008 euro).

We begin our recalculation of the mortality benefits of cleanup and proper waste management in the provinces of Naples and Caserta by selecting the appropriate Italy-based and context-appropriate VPF figures, which we display in table 7. Specifically, we select the VPF for a 30–39-year old from the Alberini-Chiabai studies (Alberini and Chiabai, 2007a, 2007b) because this would seem to be the age group that would be most likely affected if the (physical) risk reduction benefits begin in 20 yr and continue for 30 yr thereafter. Since this VPF is for cardiovascular illness, we will use it only in the calculations that do not distinguish for cancer deaths. We also select the Alberini and Ščasný (in press) estimates, which cover all fatal illnesses associated with contaminated sites, including cancer. Our final VPF selection, which is specific for cancer deaths, is the one from the 2008 Milan survey described above in this paper.

Table 7. Italy- and cancer/waste-specific VPF figures for computing the mortality benefits of cleanup and improved waste management

Reference	Context and Cause of Death	VPF (Million 2008 Euros)	S.E. Around VPF (Million Euros)	95% Lower Bound	95% Upper Bound	Year of the Surey	Estimated Discount Rate
Alberini and Chiabai (2007a, 2007b)	Cardiovascular illness, 30–49 yr old now	2.486	n.a.	n.a.	n.a.	2004	0.003
Alberini et al. (2007b)	All fatal illnesses associated with contaminated site exposures	5.976	0.826	4.358	7.594	2005	0.074
This paper	VPF for cancer, no specific context	4.164	0.276	3.623	4.704	2008	0.000

Table 8 reports the mortality benefits of cleanup and improved waste management based on the Italy- and waste/cancer-specific values listed in Table 5. The benefits are based on the formula:

Table 8. Mortality benefits of cleanup and waste management; central value in billion 2008 euros

Latency = 20 yr Duration = 30 yr	Guerriero and Cairns (2009)	Alberini and Chiabai (2007a, 2007b)		Alberini et al. (2007)		This paper
	δ = 0.04	δ = 0.04	δ = 0.003	δ = 0.04	δ = 0.074	δ = 0.04	δ = 0
All causes mortality (N = 848)	9.58	16.55	56.95	39.78	13.85	n/a	n/a
All fatal cancers (N = 403)	6.83	—	—	—	—	13.17	50.34
All cause mortality adjusted for cancer premium	11.83	—	—	39.78	13.85	—	—
Note: Since calculating the value of eq 7 from the Alberini et al. (2007) study requires the marginal utilities of risk reduction and income, and the discount rate estimated from that study, and these estimates are correlated, we used the original maximum likelihood estimation routine, and a simulation procedure based on 20,000 replications to compute the standard errors around (as shown in eq 7).

(7)

where N is the number of deaths avoided, L is latency (here set to 20 yr), T is the duration of the risk reduction in years (here, T = 30 yr). If δ = 0, the benefits are simplified to eq 8:

(8)

To avoid clutter, we only report central values in Table 8. We use both the discount rate used by the European Commission (4%) as well as the respondents' implicit discount rates as estimated in the three studies listed in Table 7.

The results of this exercise show clearly that when Italy- and waste- or cancer-specific VPF figures are used, the benefits are generally larger than those computed by Guerriero and Cairns (2009), because the VPFs we use are all greater than the €1.427 million (all fatal illnesses) or €2.141 million (cancer) (2008 euro) recommended by DG Environment (2001) and adopted by those authors. This highlights the importance of using estimates of the VPF that match the area and the context closely.

The only case in Table 8 where the benefits are close to the Guerriero and Cairns (2009) figures is when we use the Alberini et al. (2007a) study, and we use the discount rate exhibited by respondents in that study, which is about 7%. Indeed, the 95% confidence interval around this estimate of the benefits overlaps with the low-to-high range of benefits in and Guerriero and Cairns (2009).

Conclusions

We have conducted conjoint choice experiments in Italy about cancer mortality risk reductions delivered by private behaviors or public programs, and used the responses to estimate the cancer VPF. We have found that our respondents (residents of Milan, Italy, aged 20–60 and parents of at least one child aged 17 or younger) held the same cancer VPF (about €4 million) for themselves and for their children. The VPF was almost €1 million larger when the risk reduction was delivered by a public program, but further analyses revealed that people were prepared to pay more for risk reductions in public-program settings only when they believed that government programs were effective at reducing this particular type of risk.

The results of this survey, and VPF values from earlier literature that is specific to Italy and/or to the hazardous waste context, point to higher VPFs than the ones suggested for fatal illnesses and fatal cancers by DG Environment (2001). Our estimate is, however, very in line with a range on VPF estimates found for health context in two more recent meta-analyses. The first of them (Dekker et al., 2010), based on a review of overall 98 VPF estimates, reports its range between €0.12 million and €4.53 million for air pollution context, and between €0.49 million and €8.40 million for a general context (2008 euro). The second one (Braathen et al., 2008) reviewed 900 VPF estimates and based on that report the mean VPF value for health context about €3.66 million (2008 euro).

When we use Italy- and context-specific VPFs to recalculate the mortality benefits of cleanup and better waste management in the provinces of Naples and Caserta (previously quantified by Guerriero and Cairns [2009]), we obtain much higher benefits figures.

We recognize that this exercise does not change the conclusions in Guerriero and Cairns (2009) that the mortality benefits in that area greatly exceed the costs of remediation. However, it is important to realize that the VPF figures recommended by DG Environment are usually 50% or less than the estimates of the VPF for chronic conditions (e.g., cardiovascular illnesses), cancer, and for all (chronic) causes of death associated with exposures to hazardous wastes estimated in this paper (or in previous survey-based studies in Italy). At other locales in Italy with different remediation costs, using the locale- and context-appropriate VPF figures might entirely change the outcome of the cost-benefit analysis.

Since the benefits of remediation of contaminated sites begin in the future, even if cleanup is done now, and continue over a long time horizon, the mortality benefits depend crucially on the choice of the discount rate. The European Commission uses a discount rate of 4%. The previous studies we reviewed in this paper and the new survey we use to obtain a cancer VPF all were able to infer the beneficiaries' own discount rate by observing the tradeoffs between immediate payments and future risk reductions. These respondent-based estimates of the discount rate range from 0% to 7.41%, and have a potentially important effect on the estimates of the mortality benefits of remediation.

The cancer VPF estimates of this paper could also be used to estimate the monetized benefits of regulations that impose higher emission standards on hazardous or solid waste incinerators (Zambon et al., 2007), benefits of various Pay-As-You-Throw policies, which reduce the amount of land filled and/or incinerated waste (Šauer et al., 2008) or attempt to reduce the risk of industrial accidents where carcinogens are released into the environment (Pesatori et al., 2009).

Acknowledgments

This research was conducted as part of the VERHI–Children Research Program, funded by the European Commission, SSPE-CT-2005-6529, http://www.oecd.org/site/0,3407,en_21571361_ 36146795_1_1_1_1_1,00.html. The views and conclusions expressed in this article are solely those of the authors and do not necessarily reflect those of their affiliated universities and organizations.