NCEE Working Paper
What's in a Name? A Systematic
Search for Alternatives to "VSL"
Chris Dockins, Kelly B. Maguire, Steve
Newbold, Nathalie B. Simon, Alan Krupnick,
and Laura O. Taylor
Working Paper 18-01
April, 2018
U.S. Environmental Protection Agency |k|f*CC iff
National Center for Environmental Economics uw
https://www.epa.aov/environmental-economics e n v' r o n m e n ta^econ o m i cs
-------
What's in a Name? A Systematic Search for Alternatives to "VSL"
Chris Dockins, Kelly B. Maguire, Steve Newbold, Nathalie B. Simon,
Alan Krupnick, and Laura 0. Taylor
Abstract: Benefit-cost analyses of environmental, health, and safety regulations often rely
on an estimate of the "value of statistical life," or VSL, to calculate the aggregate benefits of
human mortality risk reductions in monetary terms. The VSL represents the marginal rate
of substitution between mortality risk and money, and while well-understood by
economists, to many non-economists, decision-makers, media professionals, and others,
the term resembles obfuscated jargon bordering on the immoral. This paper describes a
series of seven focus groups in which we applied a systematic approach for identifying and
testing alternatives to the VSL terminology. Our objective was to identify a term that better
communicates the VSL concept. Specifically, a list of 17 alternatives to the VSL term was
developed and tested in focus groups that culminated in a formal ranking exercise. Using a
round-robin tournament approach to analyze the data, and our qualitative judgments, we
identify "value of reduced mortality risk" as the best alternative to replace the VSL.
JEL Classification: Q51, J17
Keywords: value of statistical life, mortality risk valuation, terminology, focus group
1
-------
What's in a Name? A Systematic Search for Alternatives to "VSL"
Chris Dockins,* Kelly B. Maguire,* Steve Newbold,* Nathalie B. Simon,1
Alan Krupnick,** and Laura 0. Taylor*"
I. Introduction
Benefit-cost analyses of environmental, health, and safety regulations often rely on an
estimate of the "value of statistical life," or VSL, to calculate the aggregate benefits of human
mortality risk reductions in monetary terms. Based on individuals' willingness-to-pay for
small risk reductions, the VSL represents the marginal rate of substitution between mortality
risk and money. This is well-understood by economists who are familiar with these concepts,
but to many non-economists, decision-makers, media professionals, and others, the term
resembles obfuscated jargon bordering on the immoral (Cameron 2010).
It is not hard to guess the source of confusion. In regulatory contexts, where the VSL is most
commonly applied, individual reductions in future mortality risk associated with a policy—
which typically are relatively small—are often aggregated over the affected population and
reported in terms of "lives saved" per year. The mortality risk valuation estimates used to
monetize the risk changes for a benefits analysis, derived from published stated preference
and hedonic wage studies of willingness to pay for small changes in individual risk, are in
turn aggregated to match the "lives saved" calculated in the risk assessment. In so doing, the
notion of reporting incremental risk reductions affected through a given policy as well as
their monetization is obscured. What is left to the untrained eye is the appearance of placing
a value on human life using the "value of statistical life," at times leading to incredulity and
consternation. Cameron (2010) discusses confusions that often surround the VSL
terminology in more detail. These misunderstandings can hamper the ability to effectively
communicate the impact of government policies and regulations.
The VSL terminology was first introduced fifty years ago as a marriage between interest in
life-saving policy measures and the individual risks on which they are based. Schelling
(1968) is credited with highlighting the fact that a policy aimed at reducing mortality may in
fact "save" two lives in a city of 500,000, for example, but still only reduce individual risks by
1 Corresponding author, National Center for Environmental Economics, US EPA, Mail Code 1809T, 1200
Pennsylvania Avenue, NW, Washington, DC 20460. Phone: 202-566-2347. Email: simon,nathalie@epa,gov.
* US Environmental Protection Agency.
" Resources for the Future
"*North Carolina State University.
2
-------
0.0002. Economists were not "valuing life" per se. Rather, they were estimating the number
of "lives saved" (i.e., the difference between the number of expected deaths in a particular
year with versus without the policy) by aggregating individual risks, and they were valuing
the policy by aggregating the associated willingness-to-pay for the small reductions in risks
spread over a large population. Hence, the VSL terminology was born. Banzhaf (2014)
provides a detailed exposition of the history and use of the term.
The use of the word "statistical" in the VSL terminology was intended to distinguish it from
the notion of a "value of life." However, as benefit-cost analysis became more commonplace
in regulatory decision-making, an increase in confusion over the meaning of the term and
how it is applied ensued, as discussed at length by Cameron (2010). Two incidents in
particular illustrate how mischaracterization in the media contributed to widespread
misunderstanding. In 2003, environmental organizations criticized the U.S. Environmental
Protection Agency (EPA) for an adjustment made to the VSL in an alternative analysis
intended to account for remaining life expectancy (NRDC 2003). The adjustment, dubbed
the "senior death discount" by the press, was covered by a number of media outlets including
National Public Radio,2 the New York Times3, the Wall Street Journal4 and the Washington
Post.5 Ultimately, public protests mounted by members of the American Association of
Retired Persons (AARP) prompted congressional action in the form of an amendment to the
2004 Appropriations Bill, prohibiting the EPA from making such adjustments. Congressional
debate regarding the amendment did little to clarify what the EPA was monetizing with the
VSL: "You may or may not agree with putting dollar values on a life, but that's what the
agency does" (Senator Waxman, Congressional Record).
Similarly, an Associated Press (AP) story in 2008 entitled "An American Life Worth Less
Today," reported on the U.S. Environmental Protection Agency's (EPA) use of a lower VSL
estimate than it had used in the past, drawn from more contemporary, published meta-
analyses of the literature. While the article provided a brief description of the risk-income
tradeoffs underlying the VSL, and used the VSL term a few times, most of the language
confounded valuing risks with valuing life; using statements such as "agencies put a value on
human life" and "the value of life fell." The article was carried by hundreds, if not thousands,
of media outlets at the time with the change even satirized in a comedy news segment by
Stephen Colbert.6 The impact of the AP article was widespread and long-lasting, spurring
numerous public comments, additional confusion about "valuing life," and further threat of
congressional action (Cameron 2010). While there are legitimate criticisms of these EPA
analytic decisions, the fault was not with the act of monetizing small risk reductions and
misunderstanding seemed to crowd out more reasoned discussion of the economics.
2 Joseph Shapiro, "EPA Criticized for Plan to Reduce Value of Seniors Lives," National Public Radio's Morning
Edition, May 5,2003. Available at: https://www.npr.org/templates/story/story.php?storyld=1252109
3 Katharine Q. Seelye and John Tierney. "E.P.A. Drops Age-Based Cost Studies," New York Times, May 8, 2003.
4 John J. Fialka. "Chief U.S. Regulator Attempts to Find Value of Human Life," Wall Street Journal, May 30, 2003.
5 Cindy Skrzycki, "Under Fire, EPA Drops the 'Senior Death Discount,"' Washington Post, May 13, 2003.
6 The Colbert Report, "The Word - Priceless," July 14, 2008. Available at: http://www.cc.com/video-
clips/e8zxmm/the-colbert-report-the-word—priceless.
3
-------
It was in the wake of the AP article that other terms began to surface in economic analyses
produced in the economics literature and by government agencies alike. Researchers
introduced terms ranging from "willingness to swap" and "value of a micromort" (Cameron
2010); "value of a risk reduction" (Scotton and Taylor 2011; Hensher etal. 2011) and "value
of prevented fatality" (used in European Union economic analyses) to describe the monetary
value associated with risk-money trade-offs either in an attempt to avoid the mine-field
associated with VSL or as a matter of policy. Even within U.S. federal agencies there is
variation in terminology. For example, some analyses by the U.S. EPA refer to the "value of
mortality risk" rather than VSL (US EPA, 2016) - an alternative term offered to its Science
Advisory Board as one facet of proposed changes to its mortality risk valuation guidance but
never formally adopted (USEPA 2010b). While part of the motivation behind the use of these
alternatives may be to employ more intuitive and less provocative terminology than VSL, to
the best of our knowledge these alternative terms are based on researcher or analyst
intuition about what "sounds good." There is no information to indicate a more robust or
scientific process was used to derive any of these alternative terms and likewise no
indication that any of them would perform better than VSL. Indeed, EPA's SAB cautioned
against adopting new terminology in the absence of further testing "to explore the language
that best communicates this concept to the public" (USEPA 2011).
In this paper we apply a systematic approach for identifying and testing alternatives to the
VSL terminology. The main objective is to identify a term that may better communicate the
concept of a marginal rate of substitution between mortality risk and money to a broad group
of people, including members of the general public. Using results from a series of focus
groups and consultations with stakeholders, we develop and test a list of viable options. This
paper describes the process we follow and the replacement terms that emerge from our
results. Section II describes the methodology used to develop and examine alternative
replacement terms including a summary of the focus groups and ranking exercises; and
Section III describes the analysis of the ranked data, followed by a discussion of our
evaluation of alternative terms in Section IV. Section V concludes and adds a discussion of
the limitations of our approach, and possible additional steps that could be taken to further
vet any new terminology before it is broadly applied.
II. Methods
We employed a suite of methods to identify an alternative term or terms to the VSL starting
with focus group results and finishing with a ranking exercise of options that were identified
by focus group participants and experts. The focus groups were generally structured so as to
start with a general discussion, and then proceeded with an increasingly more focused
discussion as participants provided input on new terms and reactions to existing ones. At
the conclusion of the seven focus groups, we conducted a ranking exercise to identify
preferred terms using quasi-quantitative tools.
Focus groups are generally useful for providing a broad exploration of topics for which little
empirical or quantitative information exists. Within the economics literature, for example,
the use of focus groups and cognitive interviews typically precedes stated preference survey
design. In this study, budget and other constraints precluded the implementation of a full
4
-------
survey of a random sample of individuals. However, the use and qualitative assessment of
the results provide relevant information for identifying alternatives to the VSL terminology.
Our research design proceeded as follows. First, we conducted three focus groups with
members of the general public to systematically identify alternative terms. We then
conducted four focus groups with a convenience sample of non-economists employed at the
U.S. EPA to rank the alternative terms that were identified in the focus groups with the
general public. We develop our recommendations from an assessment of the preference
rankings. Each step is described in more detail below.
A. Public focus groups (Focus Groups 1-3)
We conducted a series of three focus groups in February 2017 in the greater Washington, DC
metropolitan area with members of the general public to develop and examine alternatives
to the VSL terminology.7 Each group consisted of 9 to 10 participants, aged 21 years of age
or over. Participants were recruited to provide an even split across genders, with broad
racial representation. While participants in the first focus group were limited to individuals
with a bachelor's degree or higher, this minimum education criterion was relaxed for the
second and third focus groups. Table 1 provides a summary of the demographic
characteristics of our public focus group participants compared to the U.S. population. As
indicated, focus group participants tended to be older, more racially diverse, and with higher
incomes compared to the U.S. population. All focus groups were moderated by members of
the research team who have professional experience in moderating group discussions.
Table 1: Demographic Characteristics, Focus Groups 1-3
Mean
U.S. population8
Age
58
37.9 (median; entire
population in 2016")
Gender
54% female
50.8% female
Race
46% non-White
13.1% non-White
Education
89% some college education
60.1% some college
education
Income
$97,3849
$55,322 (median")
N10
26
7 Approval for the conduct of the public focus groups, required under the Paperwork Reduction Act, was
provided by OMB on January 24,2017. OMB control number 2090-0028; expires 9/30/2018.
8 U.S. Census Bureau, Quick Facts for 2017 unless otherwise noted.
9 Average income is based on the mid-point of range categories.
10 There were 29 participants across Focus Groups 1-3. However, we lack demographic information for 3
participants.
5
-------
Focus Groups 1 and 2
Following introductions, each focus group included a warm-up activity to familiarize
participants with the subject matter. Participants in Focus Groups 1 and 2 were asked to read
a modified version of a Wall Street Journal article titled "Rail Safety and the Value of Life,"
which discussed tradeoffs between money and risk for a rail transportation project and
specifically used the term "value of statistical life" (see Appendix A).11 The purpose of this
exercise was to learn how non-economist readers interpreted the term when encountered
in a non-technical publication. We intentionally selected an article that discussed fatal
transport accidents rather than environmental risks so that the discussion could focus on
wealth-risk tradeoffs without complications from co-morbidities or environmental policy.12
We then led the participants in a discussion about the article in general, what they thought
about making trade-offs between money and risk, and how they interpreted the VSL term.
Participants had questions about the article and wanted more information about the
probability of dying from proposed rail transportation projects. Specifically, they posed
questions about the frequency of rail use, costs associated with project implementation, and
prioritization of various alternatives. When asked about the terminology, some participants
accepted that money-risk trade-offs are a necessary ingredient (e.g., "...in order to do an
analysis of anything you need to attach a value to it"). Others thought that all of the programs
should be pursued, rejecting the idea that any risks could be tolerated ("...they should just
allocate the money....and start repairing things all over the country that need to be done").
Notably, several participants equated the estimate in the article to the value of a specific life
("...they were equating one individual life to be worth $9.1 million").
We then turned to the portion of the article focusing on the VSL terminology. Most
participants indicated a general familiarity with the basic concept, with some relating it to
actuary tables that are used in setting insurance rates or used in wrongful death suits. There
was some discussion about whether the value would vary if people are willingly taking on a
risk (e.g., parachuting out of an airplane) versus taking a risk for which they have less control
(e.g., using a train to commute). Through this discussion most understood that the term, as
it was used in the Wall Street Journal article, reflected an average value across many
individuals and not the value for a specific individual, although some thought the concept
could be intentionally misleading or confusing, "...the core issue is value of life. But that's a
very emotional thing to say. So they call it...value of statistical life because it (is) more of an
abstract concept." The group wanted to know more about how the value was calculated. One
participant pointed out that the article indicates that the value represents how much people
value lowering the risk of death. "So, it's not about life. It's about lowering the risk of
death....and what we are willing to spend for not dying." Even so, some found the term
confusing. One participant stated, "I thought it was a number that was... out of the air." And,
11 In particular, the original article was reduced from about 2300 words to just over 500, and the discussion
was focused on the definition and use of VSL.
12 In this regard we are consistent with the basis of virtually all guidance on valuing mortality risk for policy
analysis, which are generally based on hedonic wage studies of workplace accidents or stated preference
studies where risk reductions are often associated with transportation risks (e.g., car accidents) or are framed
in more generic terms divorced of any specific context (Robinson, 2007).
6
-------
"[T]here also seems to be a kind of twisted assumption that by spending money I can save
lives..." Also, "[I]fs a little confusing."
Participants in Focus Group 1 expressed an interest in knowing more about the VSL
calculation, so in Focus Group 2 we introduced a short summary (see Appendix B). We
specifically chose an example that addressed safety, but again was unrelated to
environmental exposures—provision of lifeguards at public beaches. While it helped
participants understand the concept of VSL, the example itself became an unintended focus
of the discussion, with the conversation turning to recreation choices and how to avoid
drowning.
In Focus Groups 1 and 2, we asked participants to provide terms they thought might better
describe the VSL concept.13 Terms included:
• Accident reduction cost
• Cost of risk reduction
• Cost risk analysis
• Health risk valuation
• Investment risk analysis marginal utility (of something)
• Marginal safety risk
• Mortality risk reduction benefit
• Preventative cost expenditure
• Price of mortality risk reduction
• Risk reduction cost
• Risk reduction value
• Risk reduction value benefit (or analysis)
• Safety risk variable/valuation
• Safety/Health benefit cost
• Value of fatality prevention
• Value of mortality risk reduction
• Value of risk reduction14
• Willingness to pay for a service at a level of safety
• Willingness to pay for services
• Willingness to spend money to reduce the risk of dying
13 Throughout the focus groups we remained agnostic about the "units" associated with candidate terms. That
is, we did not constrain the choices to ones that would only be associated with a "life" or conversely, units that
would only be associated with actual risk reductions (such as, say, a l-in-1,000,000 reduction in the risk of
dying or micro-mort]. One feature of the VSL that may reinforce the impression that it represents the value of
"whole lives" is its implied measurement units, which are dollars per mortality per year. Alternative terms that
are agnostic about measurement units would remove this erroneous connotation and the measurement units
could then be selected flexibly on a case by case basis.
14 This term is similar to that suggested by the Environmental Economic Advisory Committee of EPA's Science
Advisory Board (SAB-EEAC). The SAB-EEAC 2017 report includes: "The SAB finds that a term such as "Value of
Risk Reduction for Mortality" (VRRM) may be a better term than "Value of Statistical Life" (VSL) for
communication with non-economists." (US EPA 2017).
7
-------
Participants also identified specific words they preferred be part of the term for the risk-
money tradeoff, including "risk," "reduction," and "cost." Participants specifically wanted to
omit the words "value" and "life" because "those are very emotional discussion points."
Perhaps for similar reasons, the term "death" did not feature prominently in their discussion
of preferred terms. But, they also recognized that alternatives might "lead people not to
realize what they are talking about." Other words participants thought should be avoided
included "marginal" and "price."
In Focus Group 2 after participants suggested terms, they voted on their preferred term and
identified "safety risk valuation" as the group favorite. Overall, participants in this group had
more difficultly with the concept and suggesting alternative terms than participants of Focus
Group 1, but ultimately were able to offer input on words they found more or less descriptive.
Focus Group 3
In Focus Group 3, we made a small modification to the Wall Street Journal article and
substituted the VSL term in half the articles provided to participants with an alternative—
"mortality risk reduction benefit," or MRRB—a term suggested in the previous two focus
group discussions that seemed to satisfy the preferences we heard in both groups. It included
the words "risk" and "reduction" to indicate direction, "mortality" to elicit less of an
emotional reaction, and "benefit" as this seemed to resonate among the participants as a
better substitute than words such as "value" and "price." Some Focus Group 3 participants
who received the version with MRRB seemed to understand what was being communicated,
"...they're trying to figure out what the risk of death and what the benefits are from that." On
the other hand, some participants found it to be too technical. "...[F]or the average public I
don't like reading those kinds of words" and "...sounds like you're in economics class..."
Others had negative reactions to the concept. "To me it sounded a little bit cold when talking
about human lives." For those who had the VSL version of the article the reactions were
similar. "I don't like the word statistical."
We then presented each participant with a random selection of five alternative terms from a
list of terms identified in the previous two focus groups. The choices varied across the
handouts to obtain information on a wide variety of options. For example:
For this purpose, the federal government has adopted a
measurement known as the " "
—roughly speaking, the amount of money Americans are
willing to spend for reductions in the risks of death.
marginal value of a risk reduction
benefit of mortality risk reduction
risk reduction value
value of mortality risk reduction
risk reduction unit benefit
8
-------
We then asked each participant to rank their top three terms based on which they thought
most clearly and accurately conveyed the concept.
Across the focus group participants there was wide variation in the first-choice terms. Recall,
each participant had a different set of terms, so variation is expected. The top ranked terms
along with some of the reasoning in parentheses are as follows:
• willingness to pay to reduce the risk of death ("...simple words... and straight to the
point...")
• value of prevented fatality ("...simple, clear, done..." "...value is more positive...")
• value of mortality risk reduction ("...I preferred ...a technical term versus an
emotional term... [not] anything related to fatality, life...")
• benefit of mortality risk reduction ("...straightforward...non-emotional...")
• benefit of reducing risk ("...plain, simple...")
• marginal value of risk reduction
• risk reduction value
• population willingness to pay to reduce death risks.
During a break, we tallied the information across the participants and presented the group
with their aggregated rankings across the group. We then guided the group through an
iterative process of removing the term that had received the least number of votes and then
re-ranking the remaining terms as the list was narrowed. As the discussion progressed,
participants seemed to reach agreement that the term should include something about the
nature of the risk, such as "mortality" or "fatality," and not just the word "risk" generically.
In addition, they felt an action word should be included, such as "prevented" or "reduced."
After some discussion, the participants came to understand that policy decisions do not
"prevent [specific] fatalities" but rather reduce their risk. They closed the discussion by
focusing on the monetary component. Should "value," "willingness to pay," "benefit," or some
other word or phrase be used to represent this aspect of the concept? Ultimately, the group
agreed (by majority consent if not consensus) that "value of reducing fatality risk" was the
most descriptive and comprehensible term.
Conclusions from Focus Groups 1-3
We drew three main conclusions from the results of the general public focus groups. First,
the words "value," "statistical," and "life" taken together are not preferred by most
respondents; they do not accurately describe the concept being conveyed to the general
public. The words "value" and "life" provoke an unintended emotional response and suggest
one is valuing a specific, individual life. Second, the term ultimately chosen should be concise,
but clear and relatively non-technical. Finally, the term should include words that describe
the monetary component, convey the fatal nature of the risk, and indicate the risk is being
reduced not prevented.15
15 Technically, what matters is that the term indicates a change in risk which is, in principle, positive or negative.
Our primary focus is on risk-reducing policies, but there can be risk increases associated with policy actions as
well. We return to this point in the concluding discussion.
9
-------
B. Systematic development of alternative terms
Informed by the focus groups with the general public, we revisited and revised the list of
candidate replacement terms, requiring that each term include words to describe three
important constituent or "building block" concepts:
• Monetization;
• Endpoint being valued (i.e., mortality risks); and
• Small change in risk associated with the endpoint, not the elimination of risk.
We began by developing lists of terms that fit into each category and then asked experts
within EPA to provide (1) feedback on the terms in each category (e.g., were any missing or
unacceptable), (2) a ranking of the terms developed from the building blocks, and (3) any
additional terms that should be considered.
Table 2 provides the final list of monetization building block words we considered. We
assessed each of the monetization words on three factors: (1) whether it was already used
in the context of benefit cost analyses so as to be potentially confusing; (2) whether it was
stated in plain language (i.e., less jargon-filled); and (3) whether it had been identified in
Focus Groups 1-3 as viable. Check marks in Table 2 indicate the term "passes" the criteria
(i.e., does not have a rival use in BCA, represents plain language, and/or accepted by focus
group participants). To narrow the list, we started by eliminating those with rival uses in
economic analyses. For example, "investment" typically conveys use of money in exchange
Table 2. Monetization building blocks
No rival use
Plain
Accepted by
Previous
Term
in BCA
language
Focus Groups
(1-3)
Value
V
V
V
Benefit
V
V
V
Willingness to pay
V
V
Willingness to spend
V
V
Willingness l<> swap
~
Willingness l<> accept
~
Cost
~
~
Expenditure
~
~
Investment
~
~
Price
~
~
Implicit price
Shadow price
10
-------
for some type of monetary profit. And, "expenditure" usually refers to payment for some
type of good or service. We consider these to be uses of the term that could complicate their
use in describing trade-offs between money and risk. Of the remaining terms, we only
retained those that were in "plain language" or had been identified in the previous three
focus groups as acceptable. Based on these assessments, we determined that "value,"
"benefit," "willingness to pay," and "willingness to spend" were the most suitable candidate
phrases for the monetization component of the candidate replacement terms.
Table 3 shows the endpoint building block words we considered. We assessed each option
on (1) whether it reflects the endpoint of interest (i.e., death and its synonyms/antonyms);
(2) whether it captured the statistical component; and (3) whether it was in plain language.
We only retained those options that satisfied all three factors. In short, we found that
"probability," "chance," and "risk" best reflect the statistical component of the term. "Death,"
"dying," "mortality," "fatality," and "survival" seem to best reflect the endpoint. We also found
in many cases that the ordering of the statistical component and the endpoint mattered, and
favored the statistical component coming before the endpoint for its improved fluidity (i.e.,
probability of death as opposed to death probability). This was a subjective judgement based
on our own assessment of what was likely to be considered a more common ordering of
terms. We explore this latter point more fully below as we discuss the next set of focus
groups.
Informed by the focus group discussions, we decided that including the direction of the
change in the terminology enhances clarity of the term and corresponds more directly to
the nature of the changes that are typically associated with most regulations. The
fundamental concept the term should capture is the value associated with a change in risk
or survival probability, so it seems reasonable to include this notion in our terminology.
That is, "reduce," "reduction," or "decrease" match the fact that governmental policies are
generally designed to reduce risks or improve survival probability. While we recognize that
there may be instances in which governmental action increases some risks, such as in the
case of offsetting risks, or in deregulatory settings, we nevertheless contend that including
a directional component in a new term is important to improving the clarity of the term for
most scenarios. In addition, we assessed whether the "change" term was applicable to
regulatory scenarios, that is, whether it could be paired in a meaningful way with the
endpoint. We eliminated several words that we determined would be misleading,
including "prevent," "prevention," "save," and "averting," in that they seemed to convey a
total elimination of the risk rather than an incremental reduction. Table 4 shows the
"change" terms we considered.
11
-------
Table 3. Endpoint Building Blocks
Term
Endpoint of
Interest
Statistical
Component
Plain Language
Probability of death
V
V
V
Chance of death
V
V
V
Risk of death
V
V
V
Probability of dying
V
V
V
Chance of dying
V
V
V
Risk of dying
V
V
V
Probability of mortality
V
V
V
Chance of mortality
V
V
V
Risk of mortality
V
V
V
Probability of fatality
V
V
V
Chance of fatality
V
V
V
Risk of fatality
V
V
V
Probability of survival
V
V
V
Chance of survival
V
V
V
Risk of survival
y
V
Death probability
¦/
¦/
Death chance
-------
Human life
V
Mort
V
Services
Accident
Safety
Life
V
Table 4. Change building blocks
Conveys
Term
Direction
Applicability
Reduce
V
V
Reduction
V
V
Decrease
V
V
Improvement
V
V*
Changing
¦/
Change
Prevent
Prevention
Save
Averted
Motes: * Only if paired with "survival"
Finally, we considered several other terms that identify the magnitude of the change. We
regarded these as optional, but they were included in some of the candidate terms. The
magnitude words we retained for consideration are:
• Micro
• Milli
• Marginal
• Small
• Incremental
• Standardized unit
Using the building blocks in Tables 2-4, we constructed a set of candidate replacement terms
by assembling various combinations of "monetization," "endpoint," and "change"
components. In some cases, a "magnitude" component was also included. The list of
candidate terms in Table 5 does not exhaust all possible combinations because these are far
too numerous to be manageable in a focus group setting. In selecting these 17 terms from
the larger set of possibilities, we tried to achieve a reasonably balanced representation of
most or all components listed in Tables 2-4 above, while omitting those we thought would
have a high likelihood of being dominated by other terms on the list.
13
-------
Table 5. Candidate terms (and acronyms) as an alternative to VSL
Term
Acronym
Value of Mortality Risk Reduction
VMRR
Value of Reduced Mortality Risk
VRMR
Value of Reduced Fatality/Fatal Risk
VRFR
Value of Decreased Mortality Risk
VDMR
Value of Mortality Risk Change
VMRC
Value of Micro Risk Reduction (for mortality)
VM RRmortality
Value of a Standardized Risk Unit (for mortality)
V S RU mortality
Value of Reduced Risk (of mortality)
VRRmortality
Value of Reduced Chance of Death/Dying
VRCD
Value of Improved Probability of Survival
VIPS
Value of Improved Chance of Survival
vies
Willingness to Pay for Decreased Mortality Risk
WTPdmr
Willingness to Pay for Reduced Mortality Risk
WTPrmr
Marginal Benefit of Mortality Risk Reduction
MBMRR
Marginal Value of Mortality Risk Reduction
MVMRR
Benefit of Micro Risk Reduction
BMRR
Willingness to Spend for Mortality Risk Reductions
WTSmrr
As noted above, the lists of building block terms and our revised list of candidate terms were
shared with eight senior staff economists representing various offices across the EPA for
input prior to the EPA focus groups. Specifically, these experts were asked to review each
list, identify building blocks and candidate terms they felt should be added or removed, and
to provide justification for their suggestions. At the conclusion of their reviews, these experts
suggested no new building block words, but did identify two additional terms for
consideration in the focus groups: "willingness to pay for mortality risk reduction" and "unit
benefit of mortality risk reduction." Input from these same experts led us to remove four
terms from further consideration: "value of reduced fatality/fatal risk," "value of reduced
chance of death/dying," "value of improved probability of survival," and "willingness to
spend for mortality risk reduction." The final list emerging from these consultations
contained 16 terms, including the value of statistical life.
We subsequently learned that contributors to the Guidelines for Benefit-Cost Analysis
project, organized by the Harvard School of Public Health and funded by the Bill and Melinda
Gates Foundation, were considering using the term "value of a standardized mortality unit"
14
-------
in lieu of "value of a statistical life." We chose to add this term to our list for further testing,
bringing our list of candidates back up to 17 terms.16
C. EPA employee focus groups (Focus Groups 4- 7)
To evaluate the 17 terms, we organized a second round of focus groups. All participants in
these focus groups were EPA employees, but none were professional economists or analysts
who regularly work on economic analyses. The only other requirement we imposed was that
participants should hold a Bachelor's degree. We recruited EPA staff members by (1) request
to other economists within the Agency to suggest staff who fit the inclusion criteria (i.e., were
not economists and did not work on benefit-cost or regulatory impact analyses); and (2)
through personal contacts of the research team. We recognize that this convenience sample
is not representative of the general population and a few participants did acknowledge
concerns unique to EPA (e.g., how terms related to other regulatory language or reporting
conventions)., Still, we believe that the preferences and opinions of this sample provide
useful input for our deliberations regarding the 17 candidate terms.
Focus Groups 4 through 7 were held at EPA Headquarters in Washington, DC in August and
September 2017. Focus Groups 4, 5, and 6 included ten participants each, and Focus Group
7 included six participants. Participants were recruited using personal contacts across the
agency. Specifically, we asked contacts to provide suggested names of colleagues or staff
who met the following criteria: non-economists who did not conduct or use benefit-cost
analysis in their daily work and held a Bachelor's degree or higher. We did not collect any
additional demographic information on participants in oder to maintain as much privacy and
anonymity as possible since it was conceivable we would encounter these participants in
other work-related settings.
The format of this series of focus groups was different than the first three focus groups with
members of the general public. We condensed the presentation and background section to
focus on the critical elements of the exercise in part to reduce the length of the focus group
to 60 minutes. Each focus group began with a 10-minute presentation and discussion of
mortality risk reductions and trade-offs between money and risk. Following the introductory
material explaining the concept of reducing risks and communicating valuation information,
each participant was provided a set of 17 index cards and was asked to rank the cards
according to his or her judgments about how accurately and clearly the terms conveyed the
target concept. The specific details of the ranking exercise varied slightly across the four
focus groups.
Focus Groups 4-6
In Focus Group 4, we asked participants to indicate their rankings by placing the 17 cards on
the table in front of them in order vertically from most preferred to least preferred, with ties
represented by placing the cards in the same row. We then discussed how and why each
participant ranked the outcomes as they did. The "value of statistical life" was included in
16 More information about the Harvard School of Public Health Benefit Cost Analysis Guidelines project can be
found at https://sites.sph.harvard.edu/bcaguidelines/.
15
-------
the list of terms, but we did not reveal to participants that this was the term currently in use.
Instead, we included it without comment amongst the terms they were asked to rank. Next,
we provided participants with a green strip of paper and asked them to place it among their
ranked cards such that they judged the terms above the strip to be "acceptable" and those
below the strip "unacceptable." All but one of the participants placed the terms that began
with "willingness to pay" or "value of reduced mortality..." at or near the top of their ranked
list. Terms with "marginal" or "standard unit" and "value of statistical life" most frequently
appeared below the participants' green lines (i.e., were identified as unacceptable). One
participant who preferred the "marginal value" terms also included "value of reduced
mortality risk" near the top of their list of preferences. Participants rejected the terms with
statistical jargon because they felt these terms were more difficult to understand. Most
participants ranked VSL at or near the bottom of their list.
Focus Groups 5 and 6 proceeded similarly except that VSL was initially excluded from the
list of options. After participants ranked the terms and placed the green line to demarcate
acceptable versus unacceptable terms, we gave them a card with "value of statistical life" and
revealed that this was the currently used term. We then asked them to place "value of
statistical life" where they thought it ranked among the other terms. In these groups,
participants consistently ranked "value of improved chance of survival" or a version of "value
of reduced mortality risk" near the top. Participants preferred these terms because they
were less "sciencey" or "jargony" and simpler. VSL consistently ranked at or near the bottom.
One participant said the VSL seemed "hypothetical." Participants described the terms with
"unit," "micro," and "standard" as "too awkward" and "cold." One participant noted that they
preferred the word "death" to "mortality." Another said they did not like the word "survival"
because it sounds like "we are on a sinking ship."
Focus Group 7
After reviewing the results from Focus Groups 5 and 6, we became concerned that the terms
presented to participants were imbalanced in that only one contained "survival" while most
of the others contained "mortality" or something similar. Those who preferred "survival" had
only one term to choose from while those who preferred "mortality" were able to split their
vote among many similar alternatives. Therefore, we developed a modified set of candidate
terms for participants to rank in Focus Group 7. This modified list was intended to provide a
more balanced set of terms, including alternatives to "mortality" (e.g., death, fatality) and
more options containing "survival."
The set of terms used in Focus Group 7 were:
• Value of decreased mortality risk
• Value of decreased fatality risk
• Value of decreased risk of death
• Value of increased survival probability
• Value of mortality risk reduction
• Value of fatality risk reduction
16
-------
• Value of reduced risk of death
• Value of survival probability improvement
• Value of reduced risk of mortality
• Value of reduced risk of fatality
• Value of reduced chance of death
• Value of improved chance of survival
Following the same procedure as Focus Groups 4-6, participants provided their initial
rankings, positioned the line demarcating acceptable and unacceptable terms, and then
placed the current VSL term among them. We then asked the participants to consider the
remainder of the terms that were used in Focus groups 4-6. Participants considered each
and included those terms they found acceptable among their ranked terms.
Considering their full rankings, participants in Focus Group 7 appeared to prefer terms that
provided a more positive connotation, like "survival," and preferred "mortality" over
"fatality" and "death." One participant thought the terms "mortality" and "fatality" were
aimed toward someone with a higher education and the terms with "marginal" and
"standard" sounded too much "like math." Another participant noted that they liked the
terms that include "mortality" because they would be consistent with other documents, like
a risk assessment, that would be included with the information accompanying a given
regulation aimed at reducing risks. Several participants indicated that they preferred terms
with simpler acronyms, recognizing that such an acronym would often be used rather than
the full term.
III. Analysis of focus group preference rankings
Focus Groups 4-7 produced 36 complete rankings of a common set of 17 candidate
replacement terms. For purposes of analyzing the data, we recorded each focus group
participants' preference rankings (without individual attribution) using the following
protocol: Terms were scored from 1 to n above the green acceptability line and from -1 to -
m below the line.
We analyzed these scores using a round-robin tournament approach, which involves head-
to-head comparisons of all pairwise combinations of the 36 terms using the scores for all
participants to infer majority-vote winners (see Appendix C for details). The round-robin
tournament was used to identify dominant (sets of) terms, which are those terms that defeat
all other terms not in the dominant set in a head-to-head vote. If only a single term comprises
the dominant set, then no voting cycles are possible. If more than one term comprises the
dominant set, then a voting cycle is possible and so a single candidate that is "preferred by
the group" cannot be definitively identified from among the terms in the dominant set.
The results of the candidate term preference ranking analysis are shown in Tables 6 through
8. Table 6 includes three sets of results for the 17 candidate terms that were ranked by all
17
-------
Table 6. Focus Groups 4-7 ranking aggregation results, using original set of 17
terms.
w/
FG7
w/o
FG7
w/o
VICS
1
Benefit of micro risk reduction
0
0
0
2
Marginal benefit of mortality risk reduction
0
0
0
3
Marginal value of mortality risk reduction
0
0
0
4
Unit benefit of mortality risk reduction
0
0
0
5
Value of a standard mortality unit
0
0
0
6
Value of a standardized risk unit ffor mortality")
0
0
0
7
Value of a statistical life
0
0
0
8
Value of decreased mortality risk
0.018
0.03
0.276
9
Value of improved chance of survival
0.962
0.825
—
10
Value of micro risk reduction ffor mortality")
0
0
0
11
Value of mortality risk change
0
0
0
12
Value of mortality risk reduction
0.008
0.019
0.017
13
Value of reduced mortality risk
0.066
0.246
0.814
14
Value of reduced risk (of mortality")
0.011
0.019
0.122
15
Willingness to pay for decreased mortality risk
0.001
0.001
0.015
16
Willingness to pay for mortality risk reduction
0
0
0
17
Willingness to pay for reduced mortality risk
0.001
0.001
0.015
Notes: Bold cells indicate members of the dominant set. Italics indicate the second
most favored term. Values indicate the frequency that each term is a member of the
dominant set based on re-sampling individual rankings with replacement. The first
column of results based on Focus Groups 4-7. The second column excludes Focus
Group 7. The third column excludes "value of improved chance of survival."
participants of the EPA focus groups. The terms that are members of the dominant set are
indicated by a bold font number in the corresponding row.
To examine the robustness of our results, we used a bootstrap approach (Efron 1979) to
estimate the frequency that each candidate term would appear in the dominant set in
repeated studies with the same sample size of focus group participants. The numbers in each
row indicate the frequency that each candidate term was a member of the dominant set
among 10,000 bootstrap samples of focus group participant rankings drawn with
replacement from the original dataset. For example, "value of improved chance of survival"
was in the dominant set in 96.2 percent of the bootstrap samples. The first column of results
is based on Focus Groups 4-7. The second column excludes Focus Group 7, because this
group followed a slightly different protocol than the other three EPA focus groups, as
explained above. The third column includes data from Focus Groups 4-7 but excludes "value
of improved chance of survival (VICS)," to examine how the other terms ranked when the
dominant term was removed. In the first and second cases, "value of reduced mortality risk"
(VRMR) was the second most frequent term in the dominant set. When VICS was excluded
from consideration, VRMR was the most frequent term in the dominant set.
Table 7 complements these results, showing the frequency distribution of the size of the
dominant set of terms among the bootstrap samples. In a large majority of the 10,000
18
-------
bootstrap samples, 96.5 percent, the dominant set included only one candidate term. In 1.7
percent of the bootstrap samples the dominant set included two candidate terms. In 91
percent of these cases (153 out of 168 bootstrap samples), the members of the dominant set
were "value of improved chance of survival" and "value of reduced mortality risk." In 95
percent of all bootstrap samples where the dominant set included more than one candidate
term (314 out of 332 bootstrap samples), "value of improved chance of survival" was a
member of the dominant set.
Table 8 includes results for the second set of 12 candidate replacement terms that were
ranked only by participants of Focus Group 7. "Value of improved chance of survival" is the
sole member of the dominant set in this case as well. However, in this case the estimated
frequencies of dominant set membership are larger and more evenly distributed among the
remaining 11 terms. The second highest frequency of membership in the dominant set (in
italics) is "value of reduced risk of death," but, as noted above, the second-best performer in
Focus Groups 4-7 is "value of reduced mortality risk."
Table 7. Frequency distribution of the size of (number of candidate terms in)
the dominant set among the bootstrap samples.
Dominant
set size
Frequency
1
0.967
2
0.017
3
0.007
4
0.003
5
0.005
6
0.001
Table 8. Focus Group 7 ranking aggregation results, using new set of 12 terms.
FG7
1
Value of decreased fatality risk
0.254
2
Value of decreased mortality risk
0.256
3
Value of decreased risk of death
0.279
4
Value of fatality risk reduction
0.223
5
Value of improved chance of survival
0.822
6
Value of increased survival probability
0.363
7
Value of mortality risk reduction
0.249
8
Value of reduced chance of death
0.279
9
Value of reduced risk of death
0.375
10
Value of reduced risk of fatality
0.256
11
Value of reduced risk of mortality
0.230
12
Value of survival probability improvement
0.193
19
-------
IV. Synthesis and Discussion
"Value of improved chance of survival" (VICS) was the only dominant term from the
systematic rankings made by participants in Focus Groups 4-7, suggesting that it may most
effectively convey the risk-income tradeoff concept to non-economists within the EPA.
However, no participants in Focus Groups 1-3, which included members of the general
public, suggested this or any similar terms framed as improvements in survival probability,
which suggests it may be less well-suited for communicating across a broader spectrum of
people.
In addition, the term "improved chance of survival" is often used as a term of art in other
contexts, which may lead to unintended confusion. For example, many scholarly articles in
health-related fields refer to "improved chance of survival" as an outcome metric when
comparing treatments for life-threatening conditions, most often related to overcoming
disease or injury or improving access to health care. Any search for the exact term will show
hundreds of articles per year in the medical literature covering topics from survival of
severely ill infants (Aramburo, et al. 2017) to CPR use (Fukuda, et al 2016) to battlefield
resuscitation (Davis, et al 2017). What much of this literature has in common is a focus on
treatments to increase survival probability for those already suffering substantial mortality
risk or severe health detriments. The VSL - and any preferred replacement term - is
generally applied for marginal risk reductions from a relatively small baseline.
Another, less common, source of potential confusion is the use of the term "improved chance
of survival" in legal parlance related to the "loss of a chance" (or "lost chance") doctrine as
applied in medical malpractice. The Oxford Handbook of US Health Law, for example,
explains this loss of a chance doctrine using the terminology "...the loss of an improved
chance of survival..." (Furrow 2016, p 428). In light of these considerations, we have some
concern that using "value of improved chance of survival" in place of "value of statistical life"
could merely trade one set of confusions with another, through unintended associations with
distantly related medical or legal concepts. Adding these additional considerations to the
qualitative findings from Focus Groups 1-3, it is not clear to us that "value of improved
chance of survival" should be preferred to other terms.
If we consider alternatives after removing VICS from the list of candidate terms, "value of
reduced mortality risk" (VRMR) is the dominant term from the systematic rankings.17 VRMR
is similar to terms that were more frequently preferred in Focus Groups 1-3 (e.g., "value of
mortality risk reduction," "value of reducing fatality risk"), suggesting that it could prove
effective in communications aimed at a broad spectrum of the general public, including those
without an economics background and individuals not employed by the EPA. In addition, this
term does not appear to be used as a term of art in other domains or have a specific legal
definition that might lead to unintended confusions. Another potential advantage of this
term is its similarity to alternative terms that have already been used in the economics
17We aggregated across like terms in Table 6 in different combinations and found similar results. VICS remained
the dominant term in all cases when included, regardless of how other terms were combined. "Value of
decreased mortality risk" became the dominant term when terms with "value" or "WTP" and "mortality," "risk,"
and "reduce/reduction/reduced" were combined. Results are available from the authors upon request.
20
-------
literature as a substitute for the "value of statistical life" (e.g., "value of risk reduction,"
Scotton and Taylor, 2011).
V. Conclusions
Thomas Schelling could not have anticipated how often the public and the press
misunderstand and misconstrue the term he coined 50 years ago—The Value of Statistical
Life. Indeed, the focus groups we describe in this paper provide further evidence that this
term has to be replaced with something more readily understood by a wider audience.
In this paper, we described evidence we gathered through the use of focus groups and a
formal ranking exercise regarding the understanding and acceptance of many alternative
terms to communicate the concept of trading off money and small reductions in mortality
risks. Many individuals in our sample accepted the term "value," but balked at "statistical"
and "life." Overall, participants expressed a preference for replacing "statistical" with "risk"
or "chance and "life" with "mortality" or "survival." Many also preferred to see directionality
in the risk, as in "reducing risk of mortality" or "increasing the chance of survival"—an idea
previously endorsed by the EPA's Science Advisory Board (US EPA 2011).
Thus, our top contenders were "value of improved chance of survival" and the "value of
reduced mortality risk." Although the former lead in our formal ranking exercise, we chose
the latter term on the basis of our qualitative synthesis of three public focus groups and
because the phrase "improved chance of survival" is used by professionals in other settings
to refer to distinct concepts. Accordingly, we conclude that the "value of reduced mortality
risk" is the most effective and readily understood alternative to the "value of statistical life"
of the many terms examined in our focus groups.
Of course, our conclusions should be qualified by the limitations of our analysis. We had a
small number of subjects - some picked randomly from the DC area and others selected from
a convenience sample of non-economists working at the EPA; neither should necessarily be
considered representative of the broader public across the US. Although we used reactions
of subjects to sample texts embedding the VSL term and concept, we have not formally
surveyed a representative sample and swapped the VSL term with our candidate
replacement terms (or other terms) to gauge understanding and acceptance. It would also
be informative to elicit responses from journalists who cover related topics or government
communications professionals who interact directly and regularly with the media about how
they view alternative terms. If they can understand and accept a replacement term, that
would go a long way to gaining the same response by the public. Ultimately, for the
betterment of economic literacy, it will be academic and government professionals who need
to replace the VSL term in their publications and public speaking with a generally accepted
alternative.
21
-------
Appendix A
Focus Groups 1-3 Handout
This is an edited version of an original article
that appeared in June 18, 2013, on page A1 in
the US, edition of The Wall Street Journal,
Rail Safety and the Value of a Life
by Ted Mann
Next month, a major bridge over the Schuylkill
River just outside Philadelphia will be declared too
unsafe for trains to use. Its wood ties are rotten and
officials fear the rails, expanding in the summer
sun, will pull the trestle apart.
A handful of serious rail crashes—including a 2008
head-on collision that killed 25 people—prompted
Congress to require upgraded rail anti-crash
technology, called Positive Train Control. PTC is
designed to automatically stop a train even before
it runs a red signal or gets into other dangerous
situations.
The Federal Railroad Administration says the
upgrades could prevent 52 accidents a year,
ranging from nonfatal rail-yard mishaps to deadly
train crashes. The FRA puts the cost of upgrades at
up to $13 billion for passenger and freight
railroads.
On Wednesday the Senate Commerce Committee
will hold a hearing on railroad safety, including the
progress on installing anti-crash gear.
Central to the debate is the delicate matter of
putting a dollar value on saving a life. It is an age-
old regulatory predicament—namely, whether or
not improvements in public safety are worth their
costs or whether they steer money away from
addressing more serious threats elsewhere.
Executive orders signed by Presidents since
require federal agencies to perform cost-benefit
analyses when imposing new rules and mandates.
For regulations designed to prevent fatalities, that
means calculating the economic benefit of
preserving a life.
For this purpose, the federal government has
adopted a measurement known as the "value of
statistical life," or VSL—roughly speaking, the
amount of money Americans are willing to spend
for reductions in the risks of death. To estimate the
VSL, economists observe the prices consumers pay
for safety features such as air bags, and the
differences in workers' wages between more and
less risky jobs, and deduce from these data how
much people value lowering their risks of death.
From there, economists extrapolate the VSL, the
economic value of saving a single life. Back in 2009,
the Departmentof Transportation putthat number
at $6 million; today it is calculated at $9.1 million.
The benefit-cost analyses mandated by all
presidents since Ronald Regan are intended to
inform difficult choices related to public safety,
such as whether government regulators should
require railroads to upgrade their anti-crash
technologies, by comparing the costs of proposed
safety improvements to their benefits. These
benefits, in turn, are represented in dollar terms by
multiplying the VSL by the number of lives that
would be saved each year with the new safety
measures in place.
According to a National Transportation Safety
Board official, "There's always arguments about,
'The technology is not there,' or, 'The money's not
there. But at the end of the day, we have to make a
choice between the cost of the upgrades and the
safety improvements they offer. We see this over
and over again in all modes of transportation."
22
-------
Appendix B
Focus Group 2 Handout
Accidental drowning is a risk faced by everyone who visits beaches. There are several ways
to reduce the risk of drowning, such as teaching swimming skills, wearing a life preserver,
and swimming with supervision, in particular when a life guard is present. Life guards can
reduce the risk of drowning by providing fast response when someone is at risk of drowning.
We would like to reduce the risk of drowning in a popular beach community. One way to
reduce the risk is to hire additional life guards to patrol the beach. Hiring, training, and
paying wages to additional life guards for the beach costs money, including installing
additional life guard stands, and providing the life guards with equipment, like flotation
devices.
But, we also know that having more life guards reduces the risk of dying from drowning and
therefore has some benefit. The life guards cannot guarantee they will completely eliminate
drownings. Accidents may still occur. For example, an expert swimmer could swallow water
and get a cramp and drown before a life guard will reach them. But, having additional life
guards will reduce the risk of drownings by providing more eyes on the water to respond
quickly in case of danger.
We want to know how much people value the additional lifeguards so that we can compare
it to the cost of adding guards to the beach. We do a study and determine that on average
people are willing to pay $10 as a beach access fee for an additional life guard at their beach.
Suppose we estimate that the additional lifeguards will reduce the risk of drowning each
year from 5 in 100,000 to 4 in 100,000 at this beach, a 1 in 100,000 reduction in the risk of
death from drowning.
If 100,000 people visit the beach each year, then on average there will be 1 fewer drowning
among the population. We don't know who will be affected, but we know the beach will be
safer with the presence of additional lifeguards.
We want to know how much people value the risk of death by drowning by 1 in 100,000 -
the result of the program to hire more lifeguards. Say we do a study and determine that on
average people are willing to pay a beach access fee of $8 if they are told there would be
additional lifeguards hired with that money, and that it would reduce the risk of drowning
death by 1 in 100,000.
If each of the 100,000 visitors is willing to pay $8 towards additional lifeguards, then this
group of 100,000 people is willing to pay in total $800,000 per year [because 100,000 people
x $8 per person per year = $800,000] to reduce the expected number of deaths due to
23
-------
drowning among them by 1 per year. This is sometimes referred to as "saving one statistical
life" per year. So in this case the "value of statistical life" would be $800,000.
How much one person is willing
to pay, on average, to reduce his
or her chance of death by 1 in
100,000
(A]
Number of people to
consider who are at risk
of dying from drowning
(B)
Value of statistical life
(VSL)
= fA)XfB)
$8
100,000
$800,000
This is referred to the "value of a statistical life" (VSL), even though what people are valuing
is a small change in the chance of dying, or the risk of dying. The VSL is calculated by
summing these small values over a large number of people who share the same risks.
We are now able to compare the benefit of hiring additional life guards to the costs of hiring
and training them.
24
-------
Appendix C:
Summary of Ranked Data Analysis
Our objective in analyzing the focus group participants' ranking data was to identify a single
candidate term (if one exists) that is unambiguously preferred by the group in the sense that
it could defeat all alternatives in a head-to-head vote. This is the so-called "Condorcet
winner" and is always the same regardless of the voting system used. That is, the Condorcet
winner, if it exists, is robust to any intentional or un-intentional vagaries of the order in
which a series of majority-winner votes among the candidates might be arranged. If a
Condorcet winner does not exist, then we would like to identify the set of two or more
candidate terms that cannot be defeated by all other candidates. (For more details on the
logic of the approach, see Stahl and Johnson 2017 Section 3.6.)
To analyze the ranked data, we first combined the rankings of all candidate terms by all 36
participants of Focus Groups 4-7 into a single dataset. Next, we conducted a round-robin
tournament by simulating a series of head-to-head votes among all (172-17)/2 =136 possible
combinations of pairs of terms. Term i defeats term j if i appears strictly higher than j in
more of the focus group participants' rankings than does j appear strictly higher than i;
participants who were indifferent between i and j were ignored in the head-to-head vote.
The results of this step were summarized in a 17 x 17 matrix R in which element Rtj = 1 and
element Rj t = 0 if candidate i defeats j and element Rtj = Rj t = 0 in the cases of ties. Figure
1 provides an example to illustrate how R would be constructed from a set of voter rankings
of all terms.18
Next we used the matrix R to identify the dominant set of terms. Each of the terms in the
dominant set defeat each of the terms not in the dominant set in a head-to-head vote, while
the dominant set itself may contain a voting cycle (Stahl and Johnson 2017). The dominant
set of terms in the example of Figure 1 is u, x, and y. The reader can confirm this result by
inspection of R: u, x, and y each defeat v, w, and z, but u does not defeat x ory, x does not
defeat u ory, andy does not defeat u or x.
We identified the dominant set of candidate VSL replacement terms based on the rankings
elicited from the focus group participants using an iterative procedure that checked each
possible dominant set in ascending order of sizes, from k = 1 to K, where ^=16. For each
possible size of the dominant set we constructed all K-choose-k trial sets. In the example in
Figure 1 there are 5 possible trial sets: e.g., u compared to v, w, x, y, and z is one trial set; u
compared to v,w, then v,x, then v,y, etc. is a second trial set, For each of these trial sets, we
used R to determine if each member of the trial set would defeat each member of the
complement set in a head-to-head vote by checking whether the corresponding element of
R is equal to 1. Because the dominant set is unique, the algorithm can stop as soon as this
iterative procedure ascending from k = 1 to K identifies a dominant set. The code used to
analyze the ranking data is included in Appendix D.
18 In this illustrative example we ignore the indifference line (i.e., none of the rankings are less than 0).
25
-------
To examine the robustness of our results, we used a bootstrap approach (Efron 1979) to
estimate the frequency that each candidate term would appear in the dominant set in
repeated studies with the same sample size of focus group participants. The basic intuition
behind the bootstrap approach is that it uses the empirical distribution of the sample data as
a surrogate for the true population distribution of interest (Efron 1999). That is, if our
sample of focus group participants is representative of the larger population of potential
focus group participants, then the empirical distribution of rankings in our sample can be
used as a surrogate for the distribution of rankings in the larger population. Following the
bootstrap logic, we calculated the dominant set frequency estimates by finding the dominant
set of candidate terms for 10,000 bootstrap datasets using the algorithm described above.
Each bootstrap dataset was constructed by taking N = 36 random draws of focus group
participant rankings with replacement from the original dataset. The number of times each
term appeared in the dominant set among the bootstrap datasets, divided by the number of
bootstrap datasets, formed our estimate of the sampling distribution of dominant set
membership for each candidate term.
Voter rankings
1 2 3 4 5
Number of votes
for i over j
u v w x y z
R
u v w x y z
u
4
4
5
5
5
u
0
5
5
2
2
5
u
0
1
1
0
0
1
V
3
1
2
1
3
V
0
0
2
0
0
3
V
0
0
0
0
0
1
w
2
2
3
2
2
-» w
0
3
0
0
0
3
-» w
0
1
0
0
0
1
X
5
6
4
4
5
X
2
5
5
0
1
5
w
0
1
1
0
0
1
y
6
5
4
4
5
y
2
5
5
1
0
5
y
0
1
1
0
0
1
z
1
3
1
3
1
z
0
2
2
0
0
0
z
0
0
0
0
0
0
Figure 1. Illustrative example of round-robin tournament based on the
rankings by five voters (1 through 5) of six terms [u through z).
26
-------
Appendix D. R code to identify dominant set of candidates from a set of voter
rankings of all candidates.
#==============================================================================
# RANKING.R
#==============================================================================
#
# PRELIMINARIES:
#
{
# install.packages('readxl')
library(readxl)
setwd("~/WorkActive/VRR focus groups")
# Create output file:
date.time <- gsub(" Sys.time())
date.time <- gsubdate.time)
date.time <- gsubdate.time)
filename <- paste("RANKING_output_",date.time,".out",sep="")
outfile <- file.create(filename)
set.seed(12345) # Set random number generator seed
}
#
# SUB-FUNCTIONS:
#
robin.f <- function(data){ # Round robin tournament
# INPUTS:
#
# OUTPUTS:
#
K <- length(data[,1]) # Number of rows in data matrix (candidates)
N <- length(data[1,]) # Number of columns in data matrix (voters)
robin <- matrix(0,K,K) # Initialize K x K matrix to hold results of all pair-
# wise votes
for (k in 1:K){ # Rows of robin matrix
for (q in 1:K){ # Columns of robin matrix
k.votes <- 0 # number of votes for candidate k over candidate q
q.votes <- 0 # number of votes for candidate q over candidate k
for (n in 1:N){
if (data[k,n]>data[q,n]){
k.votes <- k.votes + 1
}
if (data[q,n]>data[k,n]){
q.votes <- q.votes + 1
}
}
if (k.votes > q.votes){robin[k,q] <- 1}
if (q.votes > k.votes){robin[q,k] <- 1}
}
}
return(robin)
dominant.f <- function(robin){
# https://en.wikipedia.org/wiki/Smith_set
# Each member of the dominant set (aka Smith set or top cycle) defeats
# every other candidate outside the set in a pairwise election,
for (k in 1:K){ # For each possible size of the dominant set...
27
-------
trial.sets <- combn(K,k) # construct all possible trial dominant sets
M <- length(trial.sets[1,]) # M is the number of trial sets of size k
for(m in 1:M){ # for each trial dominant set of size k
# Compare each candidate in trial set with each candidate out of
# trial set
in.set <- trial.sets[,m] # candidates in trial dominant set
out.set <- setdiff(1:K,in.set) # candidates out of trial dominant set
dominant <- 1 # initialize dominant to true
for(candidatel in in.set){ # for each candidate in trial dominant set
for(candidate2 in out.set){ # for each candidate out of trial dominant s
# if candidate in trial dominant set fails to defeat candidate not in
# trial dominant set, set dominant to false
if (robin[candidatel,candidate2]==0){
dominant <- 0
}
# if even one candidate in trial dominant set fails to defeat
# even one candidate not in trial dominant set, then trial
# dominant set cannot be the true dominant set, so exit both inner
# loops
if(dominant==0){break}
}
if(dominant==0){break}
}
# if all candidates in trial dominant set defeat all candidates not in
# trial dominant set, then exit m and k loops (since true dominant set
# will be unique)
if(dominant==l){break}
}
# define dominant.set equal to trial dominant set on exit of m loop
dominant.set <- in.set
if(dominant==l){break}
}
return(dominant.set)
#
# MAIN PROGRAM:
#
# SIMULATE DATA (FOR TESTING):
{
N <- 5
K <- 7
data <- matrix(0,K,N)
for (n in 1:N){
for (k in 1:K){
data[k,n] <- ceiling(runif(1)*K)
}
}
data[which(data==0)] <- K+l
-------
# IMPORT DATA FROM EXCEL FILE:
{
# Initial 17 terms FG4-7:
data <- as.matrix(read_excel('FG_ranking_data.xlsxOriginal 17 termsCI:AL18'))
# Initial 17 terms FG4-7 excluding 'Value of improved chance of survival'
# data <- data[c(1:8,10:17),]
# New 12 terms FG7:
# data <- as.matrix(read_excel('FG_ranking_data.xlsx','New 12 terms','CI:H13'))
# Test terms 1 (based on Stahl and Johnson 2007 Fig 3.3):
# data <- as.matrix(read_excel('FG_ranking_data.xlsx','Test data l','C2:G8'))
# Test terms 2 (randomly generated):
# data <- as.matrix(read_excel('FG_ranking_data.xlsx','Test data 2','C2:G8'))
N <- length(data[1,])
K <- length(data[,1])
}
# ROUND ROBIN TOURNAMENT:
robin <- robin.f(data)
# FIND DOMINANT SET:
dominant.set <- dominant.f(robin)
# BOOTSTRAP SAMPLING DISTRIBUTION OF DOMINANT SET:
{
Z <- 10000
dominant.setBS <- matrix(0,Z,K)
cat('Bootstrap reps:\n',sep='',file=filename,append=T)
start.time <- proc.time()
for (z in 1:Z){
dataz <- data[,sample(1:N,N,replace=T)]
robinz <- robin.f(dataz)
dominant.setz <- dominant.f(robinz)
dominant.setBS[z,dominant.setz] <- 1
# Print boostrap result to output file:
cat (sprintf('%6.Of',z),sep=' ',file=filename,append=T)
for(k in 1:K){
cat(sprintf('%3.Of',dominant,setBS[z,k]),sep='',file=filename,append=T)
}
cat ( '\n',sep=' ',file=filename,append=T)
if(floor(z/100)==z/100){
now.time <- proc.time()
cat('\rCompleted rep ',
sprintf('%-.Of',z),
' of ' ,
sprintf('%-.Of ',Z),
'[ Time remaining = ',
sprintf('%5.If', (now.time[3]-start.time[3])/z*(Z-z)/60) ,
' minutes. ]',sep='')
}
}
freg <- colMeans(dominant.setBS)
dominant.set.size <- rowSums(dominant.setBS)
# PRINT SUMMARY RESULTS TO SCREEN:
29
-------
print(dominant.set)
print(freq)
# PRINT SUMMARY RESULTS TO OUTPUT FILE:
{
cat('\n\nDominant set:\n',sep='',file=filename,append=T)
for(i in 1:length(dominant.set)){
cat(sprintf('%3.Of',dominant.set[i]),sep='',file=filename,append=T)
}
cat('\n\n',sep='',file=filename,append=T)
cat('Frequency that each term appears in dominant
set:\n',sep='',file=filename,append=T)
for(k in 1:K){
cat(sprintf('%2.Of ',k),sep='',file=filename,append=T)
cat(sprintf('%7.4f\n',freq[k]),sep='',file=filename,append=T)
}
cat('\n\n',sep='',file=filename,append=T)
set.sizeBS <- rowSums(dominant.setBS)
cat('Frequency distribution of size of dominant set:\n',sep=' ' , file=filename,append=T)
for(i in 1:max(set.sizeBS)){
cat(sprintf('%3.Of ',i),sep='',file=filename,append=T)
cat(sprintf('%7.4f\n',sum(1*(set.sizeBS==i))/Z),sep='',file=filename,append=T)
}
cat('\n\n',sep='',file=filename,append=T)
30
-------
References:
Aramburo, A., J Todd, EC George, S Kiguli, P Olupot-Olupot, RO Opoka, C Engoru, SO Akech,
R Nyeko, G Mtove, DM Gibb, AG Babiker, and K Maitland. 2018. "Lactate Clearance as
a Prognostic Marker of Mortality in Severly 111 Febrile Children in East Africa," BMC
Medicine 16:37.
Banzhaf, Spencer. 2014. Retrospectives: The Cold-War Origins of the Value of Statistical Life.
Journal of Economic Perspectives. 28(4): 213-226.
Cameron, Trudy. 2010. Euthanizing the Value of a Statistical Life. Review of Environmental
Economics and Policy. 4(2): 161-178.
Davis, BL, MJ Martin, M Schrieber. 2017. "Military Resuscitation: Lessons from Recent
Battlefield Experience." Current Trauma Reports. 3(2):156-163.
Efron, B. 1979. Bootstrap Methods: Another Look at the Jacknife. Annals of Statistics. 7(1):
1-26.
Efron. B. 1999. The Bootstrap and Modern Statistics. Technical Report #206. Stanford
University, Division of Biostatistics: Stanford, CA. August.
Fukuda T, N Ohashi-Fukuda, H Kobayashi, M Gunshin, T Sera, Y Kondo, and N Yahagi. 2016.
"Conventional versus Compression-only versus No Bystander Cardiopulmonary
Resuscitation for Pediatric Out-of-hospital cardiac arrest." Circulation. 134(25):2060-
2070.
Furrow. B. 2016. Medical Malpractice Liability. In Oxford Handbook of U.S. Health Law, edited
by I Glenn Cohen, Allison Hoffman, and William M. Sage, 421-443. Oxford University
Press.
NRDC 2003. "Cheapening the Value Of Life: The Bush Administration's Death Discount April
1, 2003. Available at: https://www.nrdc.org/media/20'
Robinson, Lisa A. 2007. "How US Government Agencies Value Mortality Risk Reductions."
Review of Environmental Economics and Policy 1(2): 283-299.
Schelling, Thomas C. 1968. The Life You Save May Be Your Own. In Problems in Public
Expenditure Analysis, edited by Samuel B. Chase, Jr., 127-62. Washington, DC:
Brookings Institution.
Scotton, Elizabeth, and Laura Taylor. 2011. Valuing risk reductions: Incorporating risk
heterogeneity into a revealed preference framework. Resource and Energy Economics.
3(2): 381-397.
31
-------
USEPA. 2010a. Guidelines for Preparing Economic Analyses. Office of the Administrator,
Office of Policy. EPA 240-R-10-001. December. R-10-001. December. Available at:
http://yosemite.epa.gov/ee/epa/eed.nsf/webpages/ Guidelines.html.
USEPA. 2010b. Valuing Mortality Risk Reductions for Environmental Policy: A White Paper.
SAB Review Draft. Office of Policy, National Center for Environmental
Economics. December 10. Available at:
http://yosemite.epa.gov/sab/sabproduct.nsf/c91996cd39a82f6485257424
00690127/34D7008FAD7FA8AD8525750400712AEB/$File/White+Paper+(Dec.+2
010).pdf.
USEPA. 2011. Review of Valuing Mortality Risk Reductions for Environmental Policy: A
White Paper (December 10, 2010). Office of the Administrator, Science Advisory
Board. EPA-SAB-11-011. July 29. Available at:
http://yosemite.epa.gov/sab/sabproduct.nsf/298ElF50F844BC23852578DC0059A
616/$File/EPA-SAB-ll-011-unsigned.pdf
USEPA. 2016. Economic Analysis of the Formaldehyde Standards for the Composite Wood
Products Act Final Rule. July 2016. Available at:
https://www.regulations.gov/document?D=EPA-HQ-OPPT-2016-0461-0037
USEPA. 2017. SAB Review of EPA's Proposed Methodology for Updating Mortality Risk
Valuation Estimates for Policy Analysis. January 2017. Available at: https://yosemite.epa.gov/
sab/sabproduct.nsf/436C15984A487ElF852580D0004429CF/$File/EPA-SAB+2017-005
+Unsigned.pdf
32
------- |