Professional Documents
Culture Documents
U3 - 1993 (Viscusi) The Value of Risks To Life and Health
U3 - 1993 (Viscusi) The Value of Risks To Life and Health
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Economic Literature.
http://www.jstor.org
By W. KiP Viscusi
Duke University
1912
TABLE 1
off involving mortality risks and wages.
These labor market studies have ad-
Annual Fatality dressed the implicit values of life of
Source of Risk Risk workers in many countries, including the
United States, the United Kingdom,
Cigarette smoking
(per smoker) 1/150
Australia, Canada, and Japan. Straight-
Cancer 1/300 forward extensions of these models have
Motor Vehicle Accident 1/5,000 included a measure of nonfatal risks faced
Asteroid (doomsday rock) 1/6,000 by the worker, enabling analysts to im-
Work Accident (per worker) 1/10,000 pute an implicit value per statistical in-
Home Accident 1/11,000
Poisoning 1/37,000
jury in the workplace. Economists have
Fire 1/50,000 also analyzed the price-risk tradeoff for
Aviation Accident a variety of consumer products. In situa-
(passenger deaths/ tions in which no market data are avail-
total population) 1/250,000 able, such as some environmental risks,
Source: National Safety Council (1990); and further cal-
one can use surveys to derive a market
culations by Viscusi (1992a, 1992b): the smoking risk value if a market for the good existed.
estimates are averaged over the entire smoking popula- This paper explores these different ap-
tion. The average smoker consumes 1.5 packs per day. proaches to establishing appropriate eco-
nomic values for risks to life and health.
tions save lives at $89.3 million each; En- 2. Estimating the Value of Life Using
vironmental Protection Agency (EPA) as- Labor Market Data
bestos regulations save lives at $104.2
million each; and a proposed OSHA The dominant approach to obtaining
formaldehyde standard cost $72 billion estimates of the risk-dollar tradeoff uses
per life saved (John Morrall 1986).3 labor market data on worker wages for
Which of these different policies should risky jobs to infer attitudes toward risk.
be pursued, and which provide benefits The theory underlying this analysis ex-
that are not commensurate with their tends back to Adam Smith (1776), who
costs? observed that risky or otherwise unpleas-
In a democratic society, the appropri- ant jobs will command a compensating
ate starting point for analyzing these wage differential.
tradeoffs is the value individuals bearing Basic Elements of the Hedonic Wage
the risk place on the improved safety.4 Methodology
Over the past two decades, there has
developed a substantial literature on the The main empirical approach to assess-
value of these risk-money tradeoffs. The ing risk tradeoffs in the labor market has
greatest emphasis has been on the trade- utilized a methodology known as "he-
donic" (i.e., quality-adjusted) wage
equation.5 Controlling for other aspects
3These estimates reported by Morrall(1986) are
for new government regulations. For example, the
1986 OSHA asbestos standardthat cost $89.3 million 5A forerunnerof this line of work is the research
per life was more stringent than the 1972 OSHA on hedonic price indexes by Zvi Griliches (1971).
standard,which cost $7.4 million per life. Sherwin Rosen (1986) provides a survey of this ap-
4This principle is the same as in other benefit proach focusing on nonpecuniaryjob attributes in
valuation contexts. The primary matter of interest general, where health and safety risks represent a
is society's willingness to pay for the benefits gener- special case. See also Richard Thaler and Rosen
ated by the policy. ThomasSchelling (1968)first pre- (1976), Robert Smith (1979), and Viscusi (1979).
sented the willingness-to-payapproach in the life- Schelling (1968)first outlined the proper use of value
saving context. of life estimates.
entire market. In effect, the hedonic local tradeoff rate would characterize
wage studies fit a curve XX through every worker's wage-risk tradeoff at that
points such as these and estimate the particular risk level. Similarly, consider
market locus of wage-risk tradeoffs. the case of homogeneous firms, where
The observed (pi,wi) reflect the influ- all firms have offer curves FF. If there
ence of both supply and demand on the are heterogeneous workers, the market
market equilibrium for the entire set of tradeoffcurve XX would approximate the
workers.8 The estimated rate of tradeoff firm's offer curve, and its slope would
aw/ap equals the slope of constant ex- approximate the marginal cost of altering
pected utility loci that are tangent to XX, job risks at that risk level.
thus providing a local measure of the With heterogeneous workers and het-
wage-risk tradeoff for marginal changes erogeneous firms, as in Figure 1, XX
in risk. For any given worker located does not provide estimates of either the
along XX, the estimated slope simulta- offer curves or constant expected utility
neously reflects the marginal willingness loci. Rather, XX reflects only a set of tan-
to accept risk and the marginal willing- gencies between different firms' offer
ness to pay for greater safety. The points curves and different workers' constant ex-
on XX also represent the points of tan- pected utility loci. The value of awilapi
gency of firms' offer curves with workers' at any given point (pi,wi) is the local
constant expected utility loci. The slope tradeoff that is pertinent to the particular
for the firm reflects both the marginal worker and firm located at that risk level.
cost of greater safety and the marginal The estimated tradeoff rate at different
cost reductions from an incremental in- levels of risk reflects other job-worker
crease in risk. The slope at any point matches.
awilapi consequently represents the mar- The estimated local tradeoffs may be
ginal supply price as well as the marginal a misleading index of the wage differen-
demand price of risk for both the worker tials required to maintain a worker's con-
and firm located at that point. Economet- stant expected utility in the presence of
ric models that estimate a linear XX as- a major change in risk because workers'
sume that the observed tradeoff rates are risk preferences may differ. Worker 2 is
the same at all levels of risk. willing to accept risk P2 for W2(p2). How-
The shape of the estimated locus of ever, worker 1 will require a higher
tradeoffsdepends on the mix of firms and amount of wage compensation WI(p2)
workers. The situation illustrated in Fig- along EU1 to face the risk level P2 than
ure 1 consists of heterogeneous workers worker 2 requires on EU2. If the esti-
and firms. If all workers were homoge- mated wage-risk tradeoff curve XX for
neous and, for example, had a constant the market were linear, then the esti-
expected utility locus EU1, then the ob- mated rate of tradeoffwould be the same
served market combination (pi,wi) would for all workers whose indifference curves
consist of a series of points along EU1 are tangent to XX. However, even for a
that were tangent to different firms' offer linear locus of tangencies XX, for changes
curves. The resulting estimates of XX of more than a marginal amount from the
would then approximate EU1, and the current risk level, the worker's wage-risk
tradeoffwill not be the same because the
8Inframarginalworkers earn an economic rent. pertinent tradeoff value must be mea-
The wage-risktradeoffof the marginalworker is in- sured along a constant expected utility
strumental in establishing the wage rate the firm
must pay and consequently the value of the risk re- locus, not the estimated market tradeoff
duction. curve.
reflect subjective assessment of the fatal- different measures differs, as will be ex-
ity risk of the job by both the worker plored below.
and the firm. In practice, researchers A fundamental issue is how systematic
have a less perfect measure. Most studies biases in individual risk perception affect
have used information from available na- the market processes that give rise to
tional data sets that typically provide in- the compensating wage differential esti-
formation on several thousand workers mates.'6 A sizable literature in psychol-
and their jobs. These data sets include ogy and economics has documented bi-
detailed information pertaining to the ases in individual assessments of risk.
worker's demographic characteristics Individuals tend to overestimate low
(age, race, sex, years of schooling, health probability events, such as the chance
status, marital status, etc.), nature of em- of being struck by lightning, and to un-
ployment (wage rate, hours worked, in- derestimate risks of high probability
dustry, occupation, union status, years events, such as the chance of dying from
of experience, etc.), and place of resi- heart disease (see Baruch Fischhoff et al.
dence. The University of Michigan Sur- 1981).
vey of Working Conditions and Quality Because these biases are systematic,
of Employment Survey also included in- we know a great deal about their conse-
formation on the worker's job attributes, quences. In particular, the relationship
as perceived by the worker. Among these between perceived risks and actual risks
variables is whether the worker believes is similar to that displayed in Figure 2.
he faces hazards on his particular job. The perceived probability line CD lies
Usually, researchers have matched ob- above the actual probability level for
jective measures of the industry or occu-
pation risk to the worker based on his 16 Labor market estimates focus on the wage that
ings) that takes into account the probabil- choice situation should prefer the less
ity that the worker will actually collect precisely understood risk because they
the benefits. If the worker faces a job can quit if they acquire sufficiently unfa-
that poses zero risk, then workers' com- vorable information about the risk, and
pensation benefits offer no expected wel- they can remain on the job if they acquire
fare benefit, and consequently there will favorable information. Employers may
be no compensating differential. also respond to this quitting by raising
A related issue is the role of worker worker wages to retain experienced
uncertainty. Apart from the fact that we workers who are aware of the risk. Em-
do not know exactly what workers' risk pirically, the net effect is that more expe-
perceptions are, there is the additional rienced workers on hazardous jobs re-
problem that these perceptions may not ceive higher compensating differentials
be precise. Thus, workers have a subjec- (see Viscusi and Moore 1991).
tive risk perception pertaining to the haz-
Recognition of the Duration of Life
ards posed by a job, but these percep-
tions may not be as tight as an objective The standard hedonic wage equation
probability. includes the probability of death, but the
The main consequence of this uncer- amount and quality of life at risk differs.22
tainty for workers is to increase the likeli- For the typical healthy worker, the major
hood of a worker quitting once he learns difference across individuals will be in
about the adverse properties of a job and terms of the quantity of life at risk. A
revises his prior risk beliefs. This quit 20-year-old worker faces a more substan-
effect can be viewed as a generalization tial loss from a given fatality risk than a
of the theory of compensating differen- 60-year-old worker. An offsetting influ-
tials to a situation of worker uncertainty ence that should also be taken into ac-
and adaptive behavior (Viscusi 1979; Vis- count is that there may be age-related
cusi and O'Connor 1984). One measure differences in the proclivity toward risk
of the magnitude of this effect is that if taking, some of which may be attribut-
all industries eliminated their job risks, able to differences in family structure.
holding constant other aspects of the job Age clearly is a factor that may potentially
including wages, it would reduce the affect where along the market equilib-
manufacturing industry quit rates by rium curve XX a worker is located. If
one-third.21 On a theoretical basis, the XX is nonlinear, then age may also affect
opportunity for learning and adaptive be- the slope. Worker age may also influence
havior should lead workers to demand the offer curves workers face as well.23
less compensation per unit risk than they The simplest approach to addressing
would if they were fully informed about the life duration issue is to include a fatal-
the probabilities even if this information ity risk variable interacted with worker
did not alter the assessed risk level (see age, i.e., p x worker age. This approach
Viscusi 1979). The reasoning behind this is used in Thaler and Rosen (1976) and
result is that workers in a sequential job
22 RichardZeckhauserand Donald Shepard(1976)
develop a quality-adjustedvalue-of-life concept to
21Wages, of course, would also change in a com- recognize quantity and quality differences. Econo-
petitive market.This estimate is based on the impli- mists have had more success at estimating quantity
cations of quit rate regressions using data from the differencesthan quality differences.
University of Michigan Panel of Income Dynamics, 3 Althoughthe earliermodels with age interaction
as reported in Viscusi (1979). The one-third figure terms did not attempt to sort out both sets of influ-
(moreprecisely, 35%is calculatedusing these results ence, the structuralmodels discussed below attempt
in Viscusi 1983, pp. 67, 182). to do this.
risk tradeoffs, the presence of hetero- to zero and assume that the worker faces a time-
invariantsequence of lotteries on life and death. To
geneity of worker preferences will make recognize the dependence of the job risk data on
this estimate a nonlinear weighted worker i's reported industry j, pi will be used to
average of the individual workers' pref- denote the pertinent fatalityrisk level. Workeri se-
lects the optimal death risk pij from the available
erences.24 opportunitieslocus w(pi;) to maximizediscounted ex-
pected lifetime utility
24 Some of the more importantsources of hetero- 00
utility function (i.e., a constant relative term, and the cxkare coefficients on the
risk aversion utility function, a + bwc), taste-shifter variables to be estimated.
Viscusi and Moore (1989) estimate a two- The value of a Inwilapijis computedfrom
equation structural system based on the the first stage market wage equation.
local tradeoffimplied by the optimization This general estimation approach follows
problem in which the worker selects the the procedure advocated by Kahn and
optimal fatality risk, and a wage equation Lang (1988) and Biddle and Zarkin (1988)
characterizing the market opportunities to estimate structural hedonic systems.
locus.28 For worker i in industry j with In all such models, the worker selects
market opportunities affected by vari- the optimal job risk pij from among a
ables xim, the market implicit price equa- schedule of wage-risk combinations in
tion takes the form the workplace. From this optimization
4 problem is derived an explicit functional
In wi = > (@kRikPij + BkRIkPi1) form that relates the worker's rate of
k=l tradeoffawilapij to variousaspectsof the
M
job choice problem, including the job
+ > +11xi, + {i (2) risk pij, the discount rate, and in models
ni1=1
based on a finite time horizon, the work-
where the Rik are regional dummy vari- er's remaining life. Some models also in-
ables, the first summation is over the four clude a probability of death from causes
geographical regions, El is a random error other than the job to reflect the fatality
term, and 4)k Bk, *Pm are coefficients to risks that a worker faces throughout his
be estimated.29 The worker decision life. This nonlinear equation (3) is cou-
equation generated by this particular pled with a second market wage equation
model is (2) to complete the structural equation
In w system. The estimated discount rates
In wi= (1 1 Pij) range from 1 to 14 percent, which are
broadly consistent with financial rates of
N
return that one might use as a reference
+ > OtkXik+ E2, (3) point in assessing the rationality of inter-
n=l
temporal choices.
where v is the discount factor 1/(1 + r)
to be estimated, E2 is a random error Estimation of Utility Functions
' Identificationremainsan issue. Regionaldummy Knowledge of the shape of worker util-
variablesare used to identify the marketwage equa- ity functions rather than a local tradeoff
tion based on the assumption that these variables rate along a constant expected utility lo-
indicate geographicallydistinct labor markets, but cus would provide the basis for more de-
do not affectworkerpreferences.These regionalvari-
ables include interactionswith the linear and quad- tailed judgments. For example, it makes
ratic job risk variables. This identification issue is possible analysis of variations in the value
present in hedonic price models as well. See Dennis of life with respect to income levels and
Epele (1987).
2 This equation differs from equation (1) in that assessments of valuations of nonmarginal
the death riskand (deathrisk)2variablesare included changes in risk. The utility function mod-
by region. The nonpecuniaryrisk variable q and its els are based on two different state-de-
interactionwith workers'compensationdoes not ap-
pearbecause the NIOSH riskdatapertainto fatalities pendent utility functions. In the good
only. More generally, the hedonic wage equation health state 1, the utility function is U(w),
focuses on both supply and demand factors,whereas and in the ill health state, the utility func-
structuralmodels attempt to distinguish factors re-
flecting tastes and opportunitiesusing separateequa- tion is V(y), where y is the benefit paid
tions for each. upon death. One can make y a function
has no beneficiaries. -
B
Wia-- - - - - - -
Estimating utility functions involves a
quite different estimation procedure than EU I I
wa I I
the hedonic wage equation approach,
and it utilizes a different type of data as l I
well. The concern is no longer with trac- AII
ing out the locus of tangencies involving
a firm's offer curve and an individual qaJobRiskqb
worker's constant expected utility locus. Figure 3. The Market Offer Curve and the
Worker'sExpected Utility Locus
Rather, the focus is on information pro-
vided by two or more points along a par-
ticular worker's constant expected utility and Yb, can be computed using Wa, Wb,
locus. Because natural market experi- and workers' compensation benefit for-
ments do not provide such information, mulas for the worker's state of residence.
researchers have used survey evidence The survey addresses the components of
regarding the stated compensating differ- the following equality:
entials that the worker would require if
(1 - qa) U(Wa) + qaV(Ya)
faced with a change in job risk. This pro-
cedure leads to estimates of the state- = (1 - qb)U(wb) + qbV(yb). (4)
dependent, von Neumann-Morgenstern All the elements of equation (4) are ob-
utility functions U(w) and V(y) up to a servable except for U and V. One must
positive linear transformation. impose some structure on the utility
Viscusi and Evans' (1990) procedure functions to make estimations of them
uses worker survey data from four chemi- feasible. If we assume a specific func-
cal firms that provides informationon two tional form for the utility functions (e.g.,
equivalent jobs a and b along a constant logarithmic) or use a Taylor's series ap-
expected utility locus, such as EU in Fig- proximation to the general utility func-
ure 3. This curve is tangent to the market tion, then we can solve equation (4) for
offer curve ABC. The worker reports his the wage increase required by the worker
current wage rate Wa and his assessed to face the new risk, yielding an equation
job risk qa using a linear scale comparable that can be estimated with nonlinear re-
to the BLS injury risk metric. The worker gression methods.3' If we observe more
is then given a hazard warning for a than two points on a constant expected
chemical and told that this chemical
would replace the chemicals with which 3' Let Wb = (1 + 8)wa, where 8 is the percentage
he now works.30 The worker then as- wage premium for the higher risk on job b. The
sesses the risk qb associated with the dependent variable in the model is 8. For the first-
transformed job and the wage rate Wb order Taylor'sseries expansionvariantof the model,
the parameters to be estimated are 1 = U(wa) -
he would require to remain on EU. The V(Wa), 2 = U'(Wa), andP3 = V'(Wa). Thus,the esti-
income replacement after an injury, ya mationfocuses on three parametersthat characterize
the nature of worker preferences. With no loss of
generalityone can set I2 = 1 because von Neumann-
'The chemicals used were TNT, asbestos, chlo- Morgenstern utility functions are not altered by a
roacetophenone,and sodium bicarbonate.The warn- positive linear transformation,leaving two parame-
ings conformedwith current industrypractice given ters to be estimated. One can also make I, and 3
the propertiesof these chemicals.The originalsurvey functions of personal characteristics,such as educa-
results appearin Viscusi and O'Connor(1984). tion. See Viscusiand Evans (1990)for furtherdetails.
ings yielded by the structural estimation for whether the worker's job poses any
models. health and safety risks, can be interacted
Although many of the studies listed in with some objective measure of the risk
Table 2 consist primarily of replications po to create a potentially more refined
and consistency checks on earlier results, estimate of the risk variable (p8 x po),
there have been a number of important as in Viscusi (1978a, 1979) and Moore
innovations in the literature. The col- and Viscusi (1988b).
umns in Table 2 summarize several sa- Gegax, Gerking, and Schulze (1991)
lient dimensions of these studies. All ex- use a continuous measure of the worker's
cept four of the studies rely upon large subjective risk of fatality derived from
U.S. data sets, and all but two of these an interview approach analogous to the
studies use national surveys of worker Viscusi and O'Connor (1984) study of
behavior. nonfatal job injuries. However, the risk
scale they presented to workers ranged
Choice of the Job Risk Variable from 1/4,000 to 10/4,000.37 Because the
average U.S. fatality risk of 1/10,000 lies
Because all of these surveys include outside the range, the risk interval used
most of the demographic and job charac- was likely to generate overestimates of
teristic information needed to estimate the job risk. Not surprisingly, their re-
the wage equation, the main distinguish- spondents' assessed job risks were very
ing feature is the manner in which the high; their white-collar subsample as-
risk variable is created. None of the sets sessed the annual job fatality risk as
of survey data listed includes any objec- 1/2,000.38
tive measure of the risks posed by the The dominant approach followed in the
worker's particular job. Nor do the risk literature is to rely upon some published
data distinguish which injuries are attri- measure of the risk level by occupation
butable to the job environment and or industry, and then to match this risk
which arise solely from worker actions.35 variable to the worker in the sample us-
The University of Michigan's Survey of ing information provided by the respon-
Working Conditions and Quality Em- dent. The study by Thaler and Rosen
ployment Survey each include a subjec- (1976) used Society of Actuaries data per-
tive risk variable, which ascertained taining to the risk associated with differ-
whether the worker views the job as pos- ent occupations, as did Charles Brown
ing dangerous or unhealthy conditions.36 (1980) and Arnould and Nichols (1983).
The subjective risk perception variable These data pertain primarily to high risk
PS, which is a 0-1 dummy variable occupations with average annual risks of
death on the order of 1/,000-roughly
`5Making such distinctions is difficult because ten times the average for the U.S. work-
workplace technologies and worker safety precau-
tions jointly determine the risk. If all accidentswere
due to worker carelessness and would have arisen
in all other contexts in which the worker was em-
ployed, then no wage premiums for risk would be 3 Respondents could select one of ten integer re-
observed. sponses rangingfrom 1/4,000 to 10/4,000. The aver-
' These data sets have been analyzed by Viscusi age U.S. job fatalityrisk is not in this range.
(1978a, 1979), Dillingham (1985), Leigh (1987), and 3 The upwardbias in risk assessments in turn will
Moore and Viscusi (1988b), but not all of these re- tend to produce low estimates of the value of life,
searchershave used the subjectiveriskvariable.Also as appears to be the case. The higher value-of-life
see the nonfatal risk study of Biddle and Zarkin estimates for changes in risk generated by the same
(1988). None of the data sets include measures of survey instrument, as reported in Gerking, deHaan,
the employer's perceptions, which are relevant to and Schulze (1988), are also consistent with overesti-
the shape of the offer curve. mation of the base job risk level.
Dt
;r;EIb
H00 s e N (M
? Q; s z o csCg ? G ? z z x ~~~~~~~~~~~~~~~z
z t-Z O
3 z3z z z z z z z z z z o
z m ~~~~~~~~~~~~z
m m z z zm m m
D~~~~~~~~C E CD CD
igSBaS C)
2 E > U g E s-s c s _
<CD 8D - 4
CD C cD CD E
cll,
C05 C05 cl C;
cq cq
-4 z z te z z
cli
0 0 0 0 0 0 0 0
z z z z z z z
0 0 0 0 0 0 0 0
z z z z z z z z
jf cll,
c)
O
C) z O CD
z C)
O C)
C;
> 7a
$.. >1
0
0 C) 0
o - 0
C)
0 0 0 0 0 0 (1)
C) C) 0 ; " (1)'Z 7
0 z
z 0 C>z -) (11
(1) o
-a -5 >, C) bJD
-1
o
0
C) -00
4-i ; 0
1-4 E 4
-
0
z z
(M
C) C) 0 0 0
cq cq cq cq
oo 0
o 0 0 0 0
bJD to 4
(M AS - --4
u
0
0 lyC) 2 (M
01 ol P-4 4
0 bJD
0 0 0 bJD
bb >
0 0
bJD 0 0
0--
0 -
0--
(1)
u
0
; -
>
C,,
-
0 0 C)
b-C
C) ;2 z
place. To the extent that workers who over time.40 After the advent of the
select themselves into high risk jobs have OSHA, the reporting system for all inju-
a lower risk-dollar tradeoff than workers ries changed so that BLS risk data begin-
in higher risk jobs, one would expect to ning in 1972 are not comparable with the
obtain lower value-of-life estimates in earlier data. Studies such as Robert S.
studies that use these data, as Thaler and Smith (1974, 1976) and Viscusi (1978a,
Rosen (1976) recognize.39 In addition, 1979) use the pre-OSHA industrial fatal-
the Society of Actuaries data pertain to ity data, whereas more recent studies us-
all incremental mortality risks associated ing BLS data have relied upon the post-
with people in 37 different occupations, OSHA data.
not simply the job-specific risk. Thus, The main deficiency of industry-based
this variable also reflects risks other than data is that they pertain to industry-wide
those on the job, which would not be averages and do not distinguish among
compensated through the wage mecha- the different jobs within that industry;
nism. Actors, for example, have a very perceptional differences in risk are also
high mortality rate. not recognized. To promote greater per-
The lower value-of-life estimates ob- tinence of the risk measures to the jobs
tained using the Society of Actuaries data in the survey, some researchers have re-
rather than data for workers in less risky stricted the sample composition by, for
jobs is consistent with the self-selection example, limiting it to males or blue-col-
of individuals with low risk-dollar lar workers, for whom the risk data are
tradeoffs into the most hazardous pur- more relevant."1Additional limitations of
suits. The substantial variation in the the BLS data are that the reporting may
value-of-life estimates in Table 2 with the not be complete, and occupational dis-
risk level, which is a consequence of the eases are underrepresented.
joint influence of worker and firm hetero- The National Institute of Occupational
geneity, suggests that one should exer- Safety and Health sought to reduce the
cise substantial caution in extrapolating measurement error associated with the
estimates across risk ranges. industry level fatality data through its Na-
Twelve of the studies listed in Table tional Traumatic Occupational Fatality
2 use the BLS risk data based on the Survey. This survey yielded new data on
risks associated with different industries. industrial fatality rates that have been
The BLS risk measure is positively cor-
related with workers' subjective risk as-
sessments in the Survey of Working Con- 40An interesting and so far unexplored issue is
whether the change in the BLS reporting system
ditions. Moreover, the BLS objective altered the risk data in a manner that affectedwork-
industry risk measure and the worker- ers' risk beliefs. The most that is availableis a com-
specific subjective risk variable yield esti- parison of the wage premiums generated by the dif-
ferent risk measures.
mates of the annual risk premium that 41 Manystudies have obtained significantestimates
are not significantly different from one of wage premiums without such restrictions. How-
another (see Viscusi 1979a; Viscusi and ever, there remainsan importantneed both for better
risk measures as well as more detailed assessments
O'Connor 1984; and Gerking, deHaan, of the value of other nonpecuniary aspects of the
and Schulze 1988). job so that the estimates will represent premiums
The BLS risk measure has changed for risk ratherthatjob attributescorrelatedwith risk.
The Quality of Employment Survey and Survey of
WorkingConditionsdo provide detailed nonpecuni-
39 Some initial efforts to address the role of these ary attributedata, but the risk variableis categorical
worker differences included the interaction of the (does the worker'sjob expose him to dangerous or
risk variable with demographiccharacteristicvaria- unhealthy working conditions?)rather than continu:
bles. ous.
TABLE 3
is consistent with the effect of greater
FATALITY RATE DATA BY INDUSTRY
random measurement error with the
BLS measure offsetting the influence of
BLS Fatality NIOSH Fatality the underreporting of injuries with the
Rate per Rate per BLS data.42
100,000 100,000
Industry Employees Employees Nonfatal Risks and
Other Job Attributes
Mining 35.4 30.1
Construction 23.9 23.1 An important dimension on which the
Agriculture, Forestry, studies in Table 2 differ is the set of other
and Fisheries 17.5 20.3 job characteristic variables included in
Transportation,
Communication, 16.8 19.5
the equation. Omission of nonpecuniary
Electricity attributes of the job may bias the esti-
Manufacturing 4.2 4.2 mated risk coefficient. Most studies at-
Retail Trade 2.9 1.8 tempt to control for these influences us-
Wholesale Trade 2.9 1.1 ing sets of occupational or industry
Services 2.4 2.9
Finance, Insurance,
dummy variables. In addition, several of
Real Estate 1.9 1.3 these studies include a measure of the
Public Sector NA NA nonfatal risk associated with the job.
The studies by Viscusi (1978a, 1979)
Source: Bureau of Labor Statistics (1985), using data were the first to obtain an estimate of a
for 1982-1983 combined; and data from the National
Institute for Occupational Safety and Health (1987).
statistically significant value of compen-
NA = Not Available sation for injuries as well as fatalities.
These studies also included a compre-
hensive set of nonpecuniary job charac-
used in four recent studies by Moore and teristics, including whether the worker
myself, which are reported at the end was a supervisor, the speed of work,
of Table 2, and by Kniesner and Leeth whether the worker made decisions on
(1991). the job, whether the job requires the
Table 3 summarizes the BLS and worker not make mistakes, job security,
NIOSH fatality rates per 100,000 work- overtime work, worker training, and a
ers for majorindustry groups. The overall dummy variable for the worker's occupa-
size of the risk is on the order of 1/10,000, tion.
with agriculture, mining, construction, The chief recent addition to the wage
and transportation representing above equation has been the inclusion of a
average risk industries. The direction of workers' compensation variable, begin-
the risk differences between the BLS and ning with the studies by Butler (1983)
NIOSH data varies by industry. Differ- and by Arnould and Nichols (1983). In
ences in sampling procedure may ac- practice, inclusion of this variable has
count for the discrepancy. The BLS data raised the estimated wage-risk tradeoff.
are based on a sample of industry reports Although most studies in the literature
to the agency, whereas the NIOSH data have used state average benefit mea-
are based on a comprehensive census of
death certificates to identify job-related 42
See Moore and Viscusi (1988a). Suppose that
fatalities. In practice, the main difference we observe q * instead of measuring the actual risk
is that estimates of the wage-risk tradeoff of qi, where q * = qi + vi. The usualrandommeasure-
ment errormodel assumptionis that vi has zero mean
based on NIOSH data are roughly twice and constant variance, in which case the coefficient
as large as those of the BLS data, which on q * will be biased downwardfrom its true value.
sures, Viscusi and Moore (1987) and their spect to earnings, which was approxi-
subsequent work included an individual- mately 1.0. The calculations below as-
specific workers' compensation variable sume that a unitary income elasticity also
calculated based on the state benefit for- pertains to aw/lp.
mulas in conjunction with the worker's To see how one might apply the value-
demographic characteristics. This vari- of-life estimates to a different group, sup-
able was interacted with the risk on the pose that we are valuing the benefits
worker's job so that it was the expected from improved aviation safety. The aver-
workers' compensation benefit (or more age passenger on a U. S. airline has a me-
specifically, the expected rate of replace- dian income level of $32,840, which is
ment of lost earnings), which is a more considerably higher than the income lev-
pertinent measure than the overall state els listed in Table 2 (see The Gallup Or-
benefit average. ganization 1989). Extrapolating Thaler
and Rosen's (1976) values to this income
Survey Differences
group would yield a value per life of $1.0
in Worker Earnings Levels
million.43 Those in Viscusi (1978a, 1979)
As the average earnings level data in would rise to $5.7 million. The results
Table 2 indicate, the sample composition from the U. K. by Marin and Psacha-
has varied considerably. This distinction ropoulos (1982) would be even high-
is important because what these studies er-$8. 1 million-even though their es-
yield is an estimate of the implicit wage- timates for workers in the U.K. are lower
risk tradeoffthat is pertinent to a particu- than in most studies of U. S. workers. The
lar segment of the population and cannot final column in Table 2 summarizes the
necessarily be generalized to the popula- implied value-of-life estimates for the
tion at large. The standard ex post mea- typical airline passenger.
sure of economic damages for acci-
The Value-of-Life Range
dents-the present value of earnings-
varies proportionally with income. One As the implicit value-of-life estimates
would also expect some earnings varia- in Table 2 indicate, the estimated wage-
tion in the wage-risk tradeoffs, which are risk tradeoff varies considerably across
related to the compensation required for data sets and methodologies. Some het-
injury prevention. In particular, if w, p, erogeneity is expected. The value of life
and q are defined in terms of annual earn- is not a universal constant, but reflects
ings and annual risk, then the tradeoff the wage-risk tradeoff pertinent to the
awlap is the implicit value per statistical preferences of the workers in a particular
life and awlaq is the implicit value per sample. The mix of workers in these sam-
statistical injury. Viewed somewhat dif- ples is quite different. The majority of
ferently, the statistical value of life is the the estimates in Table 2 are in the $3
total amount of compensation n workers million-$7 million range.
would require to face one expected death The results that I place the greatest
from their group, where n is a large num- reliance on for the typical worker are
ber. The implicit values of life evaluated those in Viscusi (1978a, 1979), which in-
at the sample mean risk levels appear clude the most comprehensive set of non-
in the second-to-last column of Table 2.
Using survey data that provided informa- 43These calculationsused the income levels and
tion on the two points along EU indicated value of life estimates reported in Table 2 and scaled
up the estimates proportionallywith the income of
in Figure 3, Viscusi and Evans (1990) cal- airline passengers relative to the average sample
culated the elasticity of awlaq with re- member's income.
0; Y; N S K S s E 0n x cg o
mc c m m C
E to m;_E
8 ce
cr E r=g%
4-1
01 U
0 1-4 0
-q r 06
c, G -0-4 Z
1-4 -9
(:5
1-4
z
0 10
C) O cd
00
0
-:z cq :z Cd
Z,
0 ;,
--q
r, :Z,
Cd C) '34 lo 1=4 -cd
(D 00
lo a) 0 00
--4
0 0 C)
(m C) 0 jM4 t- bk 00
Cd 4-1
1"4 06 OC)
lo Id cd
134 0
(1)
$..
cd c)
Z, 0 6 r" 06 k C) Ss 'd
4-1 06
--q m
00
0) 00 01 00
cd
o > (m
(M OC)
OC) t-
O OC)
O
4 Ss cl CO CO
foD- foD- foD- foD- foD- 6&
CL.
0
0
0 0 0 0 0
z z z z z
CL.
&Z -cd
Cd
0
0 0 0 0 0
z z z z z
06
O
O
m m -9
z
0 -I 4-1 Cd 0
Z c) O
cd Cd bt 7:3 Cd
$.. cd
-4
(D C) v,
4-1 -4 -4
$-. r,
4-1 4-1 Co 4-1
C) (1) Cd
cd >, -i-,
CZ 0 (U -0
.a C." 4-, 4-1
0 Cd Cd
C-4 Cd 0 (1) 0
> 0 -
C) .. r, rj, r.
4
o
4 =
Id -4--, (51, "O (1)
0 .>,co 0
0
4-1
0 -4
0 0
--4
0 0 4-1 4-1
-4
4-1
1=4
C..
cd cd
cd o > co
cd
>
7cd 00 0
Cd 0
4-1
>1 r. t- 4-1
OC)
co 0 OC) = 0
=4-1 = (M
t-
0 OC)
(U 00
0
0 4 >
OC) >
00
0
x -4 OC) eq
00 OC)
u u
vere injuries may be included. There are donic wage equation approach are those
other differences as well, as some sets that have attempted to explore more fully
of injury data pertain to average reported the character of individuals' utility func-
industry risk levels, whereas others are tions. Biddle and Zarkin (1988) at-
subjectively assessed injury risks. tempted to impose greater structure on
The first of the injury variables used the estimation process by taking into ac-
was the BLS injury rate data gathered count the constraints imposed by the tan-
before the advent of the OSHA and the gency of individual utility functions with
institution of the new reporting system the market offer curve. They jointly esti-
(see Viscusi 1978a, 1978b, 1979). The mate a two-equation structural system
second injury variable used is the total similar in spirit to that described above
BLS reported accident rate. Studies us- for Viscusi and Moore (1989). The first
ing these data capture all job injuries, equation is the hedonic income locus-
including those that are not severe the envelope of the firms' isoprofit curves
enough to lead to the loss of a day of for the annual income offers Y for jobs
work (see Viscusi 1981; Olson 1981; of different risk. The second equation is
V. K. Smith 1983; Leigh and Folsom the first-order condition that equates dYl
1984; Viscusi and Moore 1987; and Garen ap for the hedonic income locus and the
1988). To capture injuries of greater se- worker's utility function, which they as-
verity, some studies have used only the sume to be a translog utility function.
lost workday injury component of the re- The other nonfatal risk study that does
ported BLS nonfatal accident statistics not consider a standard wage equation
(see Viscusi and Moore 1987; and Knies- is Viscusi and Evans (1990). They explic-
ner and Leeth 1991). itly estimate individual utility functions
Two studies have used subjective risk in good and ill health following Equation
perception variables based on workers' (4) using survey data in which responses
assessed risk, where the risk scale pre- to a hazard warning and baseline job in-
sented to the workers was patterned after formation make it possible to observe two
the BLS objective risk measure de- points along a constant expected utility
scribed above. Viscusi and O'Connor's locus.
(1984) reference scale was based on the Several additional insights are pro-
overall reported BLS injury rate, and vided by knowledge of the individual
Joni Hersch and Viscusi's (1990) scale was utility functions. First, job accidents
based on the BLS lost workday accident lower the marginal utility of income.48
rate. These two studies provide the val- Job injuries consequently alter the struc-
ues of workers' subjective risk percep- ture of preferences and cannot be treated
tions that are the counterparts of the two as tantamount to a monetary loss. Their
currently maintained BLS injury rate estimates imply that less than full insur-
series.47 ance of income loss (i.e., 85 percent re-
The two exceptions to the standard he- placement rate) is optimal. Second, dif-
ferences between willingness-to-accept
47 Twootherstudieshaveusedotherriskdatathat values for risk increases and willingness-
are more specific in nature. Butler (1983) analyzed to-pay amounts for risk reductions of a
employment data for South Carolinaworkers using magnitude of.01 are very minor-under
workers'compensationdata for injuries that are se-
vere enough to be filed in the workers'compensation
system in South Carolina.French and Kendall(1992) 4 For the logarithmicutility functioncases Viscusi
relied upon Federal RailroadAdministrationinjury and Evans (1990) found that U(w) = 1.077 log w
rate data to derive estimates of the implicit value of and V(y) = log y, where the coefficient for log y
job injuries. was constrained to be unity (no loss of generality).
TABLE 5
SUMMARY OF VALUE OF LIFE STUDIES BASED ON TRADEOFFS OUTSIDE THE LABOR MARKET
component of the tradeoff. These studies of seven different studies in the litera-
consequently provide a less direct and ture. The tradeoffs involve the choice of
probably less reliable measure than labor highway speed, installation of smoke de-
market estimates. tectors, cigarette smoking, property val-
Nevertheless, even if the labor market ues, and automobile safety. Many of
estimates are more accurate reflections these choices involve discrete safety de-
of the market tradeoff, the evidence from cisions. Will the consumer purchase a
product markets is valuable as well. Ob- smoke detector? Such studies provide a
taining estimates of the value of life and lower bound on the value of life, but will
health in a variety of risk contexts should not provide information about the con-
enhance our confidence in the existence sumer's total willingness to pay for safety,
of such tradeoffs. Moreover, because because with such discrete decisions con-
these different risk contexts often involve sumers are not pushed to the point where
individuals with different preferences the marginal cost of greater safety equals
facing different magnitudes of risk than its marginal valuation.
those posed by jobs, this evidence is of A major difference among the studies
independent interest. is the observability of the monetary com-
Table 5 summarizes the components ponent of the tradeoff. An example of a
study that closely parallels the labor mar- rate, he estimates the annual disutility
ket analysis in terms of having reliable cost to be $45.S2 Blomquist's value-of-life
information on the monetary component estimate is lower than most labor market
of the tradeoffis that of Scott E. Atkinson estimates, perhaps in part because of the
and Robert Halvorsen (1990). The de- presence of other nonpecuniary costs
pendent variable in their hedonic price (e. g., discomfort) and possible driver un-
model is the car's purchase price, which derestimation of the risk reduction of seat
is the analog of the wage variable. The belt use. For example, Arnould and
explanatory variables consist of product Henry G. Grabowski (1981) find that
market counterparts of the job character- these precautions are suboptimal, given
istics (e.g., Consumer Reports ratings of the benefits and costs of seat belt use.53
comfort, EPA fuel efficiency ratings) and One explanation for possibly suboptimal
individual characteristics (e.g., age and precautionary behavior is that the per-
gender of drivers). The risk variable is ceived risk function is flatter than the
the occupant fatality rate for the automo- actual risk perception function in Figure
bile model, and the coefficient of this 2. Risk reductions from Bo to Ao are
variable defines the price-risk tradeoff, viewed as being a smaller amount-from
which they estimate to be $4.0 million B1 to A1. Such misperceptions will lead
per life. Their study provides the most individuals to take a suboptimal amount
comprehensive analysis of risk-dollar of precautions. Market estimates will un-
tradeoffs outside the labor market. derstate the implicit value of life that
Many of the other studies use imputed would prevail if individuals were fully ra-
values of the monetary component of the tional or informed.
risk tradeoff, potentially introducing an- Capital market contexts also may pro-
other source of error. Consider, for ex- vide evidence on the value of life. Ivy
ample, Glenn Blomquist's (1979) imagi- Broder (1990) found that industrial fatali-
native analysis of the decision to wear a ties such as airplane crashes and hotel
seat belt. The risk involved pertains to fires were valued by stockholders at $50
the reduced risk of fatality associated million per death. This high estimate re-
with the wearing of seat belts. Before flects private valuations of risk by con-
the advent of state mandatory seat belt sumers of the firm's products, the total
laws, only 17.2 percent of the population cost of tort awards, and possibly a low-
always used seat belts, 9.7 used the belt ered assessment of the overall quality of
most of the time, 26 percent used the the firm's operations as well.
belts sometimes, and 46.6 percent never
used seat belts.5" The major issue in this 6. Surveys and Contingent Valuations
analysis is the value attached to the time
Market-based evidence on risk
and inconvenience costs of wearing a seat
tradeoffs offers the considerable strength
belt, which are not directly observable.
that it is based on the actual risk-taking
The monetary component of the risk
decisions individuals make. Revealed
tradeoff analyzed by Blomquist (1979) is
preferences toward risk are a potentially
the value of time required to buckle a
seat belt, which he estimates at eight sec-
onds per use. Valued at the driver's wage 52
This approachis similarin spirit to the approach
used in highway speed-travel time tradeoffanalysis
of Ghosh, Lees, and Seal (1975).
51
These 1983 data are for persons five years old 5 Their analysisdoes not addressthe nonpecuniary
and over. See the U.S. Department of Commerce, costs of seat belts, however, and consequently is not
(1986), p. 604. conclusive.
useful basis for inferring the price that crease in the risk level may avoid these
individuals attach to improved safety. problems. Figure 3 illustrates a market
The above review of these estimates wage opportunities frontier ABC and the
identified a series of potential shortcom- worker's highest constant expected util-
ings. Chief among these are the follow- ity locus EU, which is tangent to the mar-
ing. First, the tradeoff values are perti- ket opportunities frontier. The most that
nent only in a local range.54 Analysis of can be achieved with a well designed la-
nonincremental risk changes or other bor market study is an evaluation of the
fundamental policy questions, such as local tradeoff at a point such as B. In
how the local tradeoff rate will change contrast, a survey that asks an individual
if the individual's base level of risk is al- what wage increase is needed to bear an
tered, cannot be addressed. Second, increase in risk level from qa to qb will
there remain substantial estimation is- provide information on two points B and
sues regarding the identification and D on a constant expected utility locus.
meaning of the risk premium estimates. The wage increment in this context will
The usual studies of a single risk-wage truly be a compensating differential that
tradeoff, for example, ignore the substan- maintains the individual's utility at a con-
tial heterogeneity across individuals in stant level. Moreover, the risk incre-
their attitudes toward risk. Third, there ments between qa and qb can be de-
is an important class of econometric prob- signed to analyze risk changes of any
lems pertaining to whether the re- magnitude.
searcher has in fact isolated the risk- Perhaps most important, information
money tradeoff. In the case of labor mar- pertaining to two or more points along
ket studies, there are other nonpecuniary a constant expected utility locus permits
aspects of the job correlated with riski- the estimation of the utility functions
ness that one must also take into account governing behavior. With knowledge of
to isolate the risk tradeoff. Many of the these utility functions, all pertinent ques-
product market studies encounter similar tions regarding risk valuation may be ad-
difficulties, with the added complication dressed, thereby greatly extending the
that it is often necessary to impute the range of issues that can be explored.
monetary component of the tradeoff. The character of the influence of health
Furthermore, all of these results are impacts on the utility function is an im-
premised on an assumption of individual portant matter of concern. If, for exam-
rationality. If individuals do not fully un- ple, adverse health effects lower the mar-
derstand the risk and respond to risks ginal utility of income in the ill health
in a rational manner, then the risk state, then less than full income replace-
tradeoff that people are actually making ment following these losses is optimal.
may not be those that researchers believe The entire structure of the optimal social
they are making based on objective mea- insurance efforts consequently hinges on
sures of the risk. Finally, market studies the character of utility functions.
of risk are limited to a narrow range of Viscusi and Evans' (1990) results men-
health outcomes. tioned earlier found that the typical job
Survey methods that elicit individual injury lowered both the absolute level
willingness to pay for greater safety or of utility and the marginal utility of in-
compensation required to bear an in- come. Moreover, the character of this
effect differed from what would have oc-
54It should be noted, however, that use of the curred if the job injury was tantamount
two-stage structuralhedonic approachwith market
data can address nonmarginalrisk changes as well. to a monetary loss equivalent (i.e., V(w)
TABLE 6
SUMMARY OF VALUE OF LIFE ESTIMATES BASED ON SURVEY EVIDENCE
Implicit
Average Value
Income of Life
Author (Year) Nature of Risk Survey Methodology Level ($ millions)
Consider two different valuation stud- with a similar class of injuries but ad-
ies of the same health outcome using sur- dressed a much more comprehensible
veys involving different risk levels. The risk level-on the order of 15/10,000 an-
first of the studies reported in Table 7, nually. The value of the morbidity effects
by Viscusi, Magat, and Huber (1987), fo- such as skin poisonings and chloramine
cuses on individuals' valuations of the gassing is in a more reasonable range,
risks from bleach and drain opener, chlo- as the health effects assessed in this study
ramine gassings, child poisonings, and range in value from $700 to $3,500.
hand burns, among others. These mor- The discrepancy in the studies can be
bidity effects are by no means cata- traced to the difficulties posed by the
strophic, but the estimated values the small risks in the first study. Individuals
respondents attach to them are in excess who are willing to pay one dollar to
of $1 million in three cases. reduce the risk of bleach gassing by
Viscusi, Magat, and Huber (1987) dealt 1/1,000,000, will exhibit an implicit value
TABLE 7
SUMMARY OF VALUATIONS OF NONFATAL HEALTH RISKS
Average Value of
Income Health
Author (Year) Survey Methodology Level Nature of Risk Outcome
TABLE 7 (Continued)
Average Value of
Income Health
Author (Year) Survey Methodology Level Nature of Risk Outcome
Magat, Viscusi, & Risk-risk computer $35,700 Environmental risk of $1.6 million (nerve dis-
Huber (1991) survey at mall, nonfatal nerve dis- ease), $2.6 million
1990 ease, fatal lym- (nonfatal lymphoma),
phoma, nonfatal $4.1 million (fatal
lymphoma lymphoma)
Viscusi, Magat, & Risk-risk and risk- $41,000 Environmental risk of .32 fatality risk or
Huber (1991) dollar computer severe chronic $904,000 risk-risk;
survey at mall, bronchitis morbid- $516,000 risk-dollar
1988 ity risk
Alan Krupnick & Viscusi-Magat-Hu- $39,744 Environmental risk of $496,800-$691,200
Maureen Crop- ber (1991) survey severe chronic (median)
per (1992) for sample with bronchitis morbid-
chronic lung dis- ity risk
ease, 1989
for the injury of $1 million, but this re- can also be addressed by using a survey
sponse may not reflect the underlying mechanism that is designed to elicit a
risk-dollartradeoff so much as it does the truthful expression of preferences, such
inability of individuals to deal with ex- as hypothetical voting on a political refer-
tremely low probability events. As the endum.
risk levels in Table 1 indicate, the usual
7. Policy Implications
range of experience with risk is with haz-
ards of much greater frequency. The evi- Although the value-of-life literature is
dence in the psychology and economics now roughly two decades old, the essen-
literature sketched in Figure 2 indicates tial approach became well established in
there is a tendency to overestimate the the 1970s. The appropriate measure of
magnitude of very low probability the value of life from the standpoint of
events, particularly those called to one's government policy is society's willing-
attention. Survey methodologies may ness to pay for the risk reduction, which
elicit individual valuations as perceived is the same benefit formulation in all pol-
by the respondent, but one must ensure icy evaluation contexts.
that what is being perceived is accurate. Economists have had the greatest suc-
Errors in risk perceptions may be a par- cess in assessing the risk-money tradeoff
ticularly salient difficulty. using labor market data. Although the
A final concern with respect to survey tradeoff estimates vary considerably de-
valuations is whether the respondents pending on the population exposed to the
are giving honest and thoughtful answers risk, the nature of the risk, individuals'
to the survey questions. In practice, income level, and similar factors, most
truthful revelation of preferences has of the reasonable estimates of the value
proven to be less of a problem than has of life are clustered in the $3 million-$7
the elicitation of meaningful responses million range. Moreover, these estimates
because of a failure to understand the are for the population of exposed work-
survey task. Strategic misrepresentation ers, who generally have lower incomes
than the individuals being protected by tions has been in excess of $100 million
broadly based risk regulations. Recogni- (see Viscusi 1992a). It is in addressing
tion of the positive elasticity of the value the most extreme policy errors that the
of life with respect to worker earnings estimates will be most useful, as opposed
will lead to the use of different values to pinpointing the value of life that
of life depending on the population being should guide policy decisions.
protected. Taste differences may also en- A needed major change is to establish
ter, as smokers and workers in very haz- an appropriate schedule of values of life
ardous jobs, for example, place lower that is pertinent for the differing popula-
values on health risks. tions at risk. The quantity of life at risk
The 1980s marked the first decade in often varies quite widely, as do individual
which use of estimates of the value of attitudes toward these risks. Policies that
life based on risk tradeoffs became wide- protect groups who incur risks volun-
spread throughout the Federal govern- tarily should be treated quite differently
ment. Previously, agencies assessed only from policies that protect populations
the lost present value of the earnings of who bear risks involuntarily or who have
the deceased, leading to dramatic under- a very high aversion to incurring health
estimation of the benefit value. In large risks. Differences also arise on a temporal
part through the efforts of the U. S. Office dimension. Valuation of health risks to
of Management and Budget, agencies future generations is assuming greater
such as OSHA and EPA began incorpo- policy importance, but these values will
rating value-of-life estimates in their ben- not be the same as for those currently
efit evaluations.55 Policy makers' recogni- alive.
tion of the nonpecuniary aspects of life Broad gaps in our knowledge remain,
is an important advance. particularly with regard to risks other
Given the range of uncertainty of the than accidental fatalities. How, for exam-
value-of-life estimates, perhaps the most ple, should we value genetic risks and
reasonable use of these values in policy increased life extension for AIDS vic-
contexts is to provide a broad index of tims, as compared with other health out-
the overall desirability of a policy. In comes? The class of health outcomes of
practice, value-of-life debates seldom fo- policy interest is much broader than
cus on whether the appropriate value of acute fatal injuries and lost workday acci-
life should be $3 million or $4 million- dents, which have been the main targets
narrow differences that cannot be distin- of analysis. Survey evidence on attitudes
guished based on the accuracy of current toward risk can potentially expand the
estimates and the potential limitations of range of health outcomes that we can
individual behavior underlying these es- value, but there is a continuing need to
timates. However, the estimates do pro- assess the validity of these benefit mea-
vide guidance as to whether risk reduc- sures.
tion efforts that cost $50,000 per life
REFERENCES
saved or $50 million per life saved are
warranted. The threshold for the Office ACTON, JAN P. Evaluating public programs to save
of Management and Budget to be suc- lives: The case of heart attacks, R-950-RC. Santa
Monica: The Rand Corporation, 1973.
cessful in rejecting proposed risk regula- ARNOULD, RICHARD J. AND GRABOWSKI, HENRY G.
"Auto Safety Regulation: An Analysis of Market
55Indeed, the U.S. Office of Management and Failure," Bell J. Econ., Spring 1981, 12(1), pp.
Budget (1990, pp. 661-63) has developed explicit 27-48.
ARNOULD, RICHARD J. AND NICHOLS, LEN M. "Wage-
guidelines on the use of implicit values of life and
health. Risk Premiums and Workers' Compensation: A Re-
finement of Estimates of CompensatingWage Dif- and the Endogeneity of Job Riskiness,"Rev. Econ.
ferential,"J. Polit. Econ., Apr. 1983, 91(2), pp. Statist., Feb. 1988, 70(1), pp. 9-16.
332-40. GEGAX, DOUGLAS; GERKING, SHELBY AND SCHULZE,
ATKINSON, ScoTTE. AND HALVORSEN, ROBERT. "The WILLIAM. "PerceivedRiskand the MarginalValue
Valuationof Risksto Life: Evidence from the Mar- of Safety," Rev. Econ. Statist., Nov. 1991, 73(4),
ket for Automobiles," Rev. Econ. Statist., Feb. pp. 589-96.
1990, 72(1), pp. 133-36. GERKING, SHELBY; DEHAAN, MENNO H. AND
BERGER, MARK C. ET AL. "ValuingChangesin Health SCHULZE, WILLIAM. "The MarginalValue of Job
Risks: A Comparison of Alternative Measures," Safety: A Contingent Valuation Study," J. Risk Un-
Southern Econ. J., Apr. 1987, 53(4), pp. 967-84. certainty, June 1988, 1(2), pp. 185-99.
BIDDLE, JEFF E. AND ZARKIN, GARY. "WorkerPrefer- GHOSH, DEBAPRIYA; LEES, DENNIS AND SEAL, WIL-
ences and Market Compensation for Job Risk," LIAM. "OptimalMotorwaySpeed and Some Valua-
Rev. Econ. Statist., Nov. 1988, 70(4), pp. 660- tions of Time and Life," Manchester Sch. Econ.
67. Soc. Stud., June 1975, 43(2), pp. 134-43.
BLOMQUIST, GLENN. "Valueof Life Saving:Implica- GRILICHES, ZVI, ed. Priceindexesand qualitychange.
tions of Consumption Activity," J. Polit. Econ., Cambridge: Harvard U. Press, 1971.
June 1979, 87(3), pp. 540-58. HAMERMESH, DANIEL S. AND WOLFE, JOHN R. "Com-
BRODER, IvY E. "The Cost of Accidental Death: A pensating Wage Differentials and the Duration of
Capital Market Approach,"J. Risk Uncertainty, Wage Loss," J. Labor Econ., Jan. 1990, 8(1), Part
Mar. 1990, 3(1), pp. 51-63. 2, pp. S175-97.
BROWN, CHARLES. "Equalizing Differencesin the La- Smok-
HERSCH, JONI AND VISCUSI, W. KIP. "Cigarette
bor Market,"Quart. J. Econ., Feb. 1980, 94(1), ing, Seatbelt Use, and Differences in Wage-Risk
pp. 113-34. Tradeoffs,"J. Human Res., Spring1990, 25(2), pp.
BROWN, JAMES N. "StructuralEstimationin Implicit 202-27.
Markets," in The measurement of labor cost. Ed.: HERZOG, HENRY W., JR. AND SCHLOTrMANN, ALAN
JACK E. TRIPLETT. Chicago: U. of Chicago Press, M. "ValuingRiskin the Workplace:MarketPrice,
1983, pp. 123-51. Willingness to Pay, and the Optimal Provision of
BUTLER, RICHARDJ. "Wage and Injury Rate Re- Safety,"Rev. Econ. Statist., Aug. 1990, 72(3), pp.
sponses to Shifting Levels of Workers' Compensa- 463-70.
tion," in Safety and the work force. Ed.: JOHN IPPOLITO, PAULINE M. AND IPPOLITO, RICHARD A.
D. WORRALL. Ithaca: ILR Press, 1983, pp. 61-86. "Measuring the Value of Life Saving from Con-
COUSINEAU, JEAN-MICHEL; LACROIX, ROBERT AND sumer Reactions to New Information," J. Public
GIRARD, ANNE-MARIE. 'Occupational Hazard and Econ., Nov. 1984, 25(1/2), pp. 53-81.
Wage Compensating Differentials." U. of Montreal JONEs-LEE, MICHAEL W. The value of life: An eco-
Working Paper, 1988. nomic analysis. Chicago: U. of Chicago Press,
DARDIS, RACHEL. "The Value of a Life: New Evi- 1976.
dence from the Marketplace," Amer. Econ. Rev., . The economics of safety and physical risk.
Dec. 1980, 70(5), pp. 1077-82. Oxford:Basil Blackwell, 1989.
DILLINGHAM, ALAN. "The Influence of Risk Variable KAHN, SHULAMIT AND LANG, KEVIN. "EfficientEsti-
Definition on Value-of-Life Estimates," Econ. In- mation of Structural Hedonic Systems," Int. Econ.
quiry, Apr. 1985, 23(2), pp. 277-94. Rev., Feb. 1988, 29(1), pp. 157-66.
DORSEY, STUART AND WALZER, NORMAN. "Workers KNIESNER, THOMAS J. AND LEETH, JOHN D. "Com-
Compensation, Job Hazards, and Wages," Ind. pensating Wage Differentials for Fatal Injury Risk
Lab. Relat. Rev., July 1983, 36(4), pp. 642-54. in Australia, Japan, and the United States," J. Risk
EPPLE, DENNIS. "Hedonic Prices and Implicit Mar- Uncertainty,Jan. 1991, 4(1), pp. 75-90.
kets: Estimating Demand and Supply Functions KRUPNICK, ALAN J. AND CROPPER, MAUREEN L. "The
for DifferentiatedProducts,"J. Polit. Econ., Feb. Effect of Information on Health Risk Valuations,"
1987, 95(1), pp. 59-80. J. Risk Uncertainty, Feb. 1992, 5(1), pp. 29-
EVANS, WILLIAM N. AND VISCUSI, W. KIP. "Estima- 48.
tion of State-Dependent Utility Functions Using LEIGH, J. PAUL. "Gender, Firm Size, Industry, and
Survey Data," Rev. Econ. Statist., Feb. 1991, Estimates of the Value-of-Life," J. Health Econ.,
73(1), pp. 94-104. Sept. 1987, 6(3), pp. 255-73.
FISCHHOFF, BARUCH ET AL. Acceptable risk. Cam- LEIGH, J. PAUL AND FOLSOM, ROGER N. "Estimates
bridge: CambridgeU. Press, 1981. of the Value of Accident Avoidanceat the Job De-
FRENCH, MICHAEL T. AND KENDALL, DAVID L. "The pend on the Concavity of the Equalizing Differ-
Value of Job Safety for RailroadWorkers,"J. Risk ences Curve," Quart. Rev. Econ. Bus., Spring
Uncertainty, May 1992, 5(2), pp. 175-85. 1984, 24(1), pp. 56-66.
GALLUP ORGANIZATION. Air travel survey, 1989, pro- MAGAT, WESLEY A.; VISCUSI, W. KIP AND HUBER,
duced for Air TransportAssociation of America. JOEL. "The Death Risk Lottery Metric for Health
Princeton:Gallup Organization,1989. Risks:Cancerand Nerve Disease." Duke U. Work-
GARBACZ, CHRISTOPHER. "SmokeDetector Effective- ing Paper, 1991.
ness and the Valueof Savinga Life,"Econ. Letters, MARIN, ALAN AND PSACHAROPOULOS,GEORGE. "The
Dec. 1989, 31(3), pp. 281-86. Reward for Risk in the Labor Market: Evidence
GAREN, JOHN E. "CompensatingWage Differentials from the United Kingdom and a Reconciliation
with Other Studies,"J. Polit. Econ., Aug. 1982, lic Policy: A Review," Ind. Lab. Relat. Rev., Apr.
90(4), pp. 827-53. 1979, 32(3), pp. 339-52.
MILLER, TED. "The Plausible Range for the Value SMITH, V. KERRY, "The Role of Site and Job Charac-
of Life-Red Herrings Among the Mackerel,"J. teristics in Hedonic Wage Models," J. Urban
Forensic Econ., Fall 1990, 3(3), pp. 17-39. Econ., May 1983, 13(3), pp. 296-321.
MILLER, TED AND GURIA, JAGADISH. "The Value of SMITH, V. KERRY AND GILBERT, CAROL. "The Implicit
Statistical Life in New Zealand." Report to the Risksto Life: A ComparativeAnalysis,"Econ. Let-
Ministry of Transport, Land Transport Division, ters, 1984, 16, pp. 393-99.
1991. THALER, RICHARD AND ROSEN, SHERWIN. "The Value
MISHAN, EzRA J. "Evaluationof Life and Limb: A of Saving a Life: Evidence from the Market,"in
TheoreticalApproach,"J. Polit. Econ., July/Aug. Household production and consumption. Ed.:
1971, 79(4), pp. 687-705. NESTOR E. TERLECKYJ. Cambridge:NBER, 1976,
MOORE, MICHAEL J. AND VISCUSI, W. KIP. "Doubling pp. 265-98.
the Estimated Value of Life: Results Using New TORRANCE, GEORGE W. "Measurement of Health
OccupationalFatality Data,"J. Policy Anal. Man- State Utilities for Economic Analysis:A Review,"
age., Spring 1988a, 7(3), pp. 476-90. J. Health Econ. Mar. 1986, 5(1) pp. 1-30.
. "The Quantity-Adjusted Value of Life," U.S. BUREAU OF LABOR STATISTICS. Occupational in-
Econ. Inquiry, July 1988b, 26(3), pp. 369-88. juries and illness in the United States by industry,
. "PromotingSafety through Workers'Com- 1983. Washington, DC: U.S. GPO, 1985.
pensation," Rand J. Econ., Winter 1989, 20(4), U. S. DEPARTMENT OF COMMERCE. Statisticalabstract
pp. 499-515. of the United States. Washington,DC: U.S. GPO,
. Compensation mechanisms for job risks. 1986.
Princeton:Princeton U. Press, 1990a. U. S. OFFICE OF MANAGEMENT AND BUDGET. Regula-
. "Discounting Environmental Health Risks: tory program of the United States Government,
New Evidence and Policy Implications,"J. Envi- April 1, 1988-March 31, 1989. Washington:U. S.
ron. Econ. Manage., Mar. 1990b, 18(2) Part 2, GPO, 1988.
pp. S51-62. . Regulatory program of the United States
. "Models for Estimating Discount Rates for Government, April 1, 1990-March 31, 1991.
Long-Term Health Risks Using Labor Market Washington, DC: U.S. GPO, 1990.
Data,"J. Risk Uncertainty, Dec. 1990c, 3(4), pp. VISCuSI, W. KIP. "Labor Market Valuations of Life
381-401. and Limb:EmpiricalEstimatesand Policy Implica-
MORRALL, JOHN F. III. "A Review of the Record," tions," Public Policy, Summer 1978a, 26(3), pp.
Regulation,, Nov./Dec. 1986, 10(2), pp. 25-34. 359-86.
NATIONALINSTITUTE FOR OCCUPATIONALSAFETYAND . "Wealth Effects and Earnings Premiums for
HEALTH. "National Traumatic Occupational Fatali- Job Hazards," Rev. Econ. Statist., Aug. 1978b,
ties, 1980-1984." Apr. 27, 1987. 60(3), pp. 408-16.
NATIONAL SAFETY COUNCIL. Accidentfacts. Chicago: . Employment hazards: An investigation of
National Safety Council, 1990. market performance. Cambridge: Harvard U.
OLSON, CRAIG A. "An Analysis of Wage Differentials Press, 1979.
Received by Workerson Dangerous Jobs,"J. Hu- . "OccupationalSafetyand Health Regulation:
man Res., Spring 1981, 16(2), pp. 167-85. Its Impact and Policy Alternatives,"in Research
PORTNEY, PAUL R. "Housing Prices, Health Effects, in public policy analysis and management, Vol.
and ValuingReductionsin Riskof Death,"J. Envi- 2. Ed.: J. CRECINE. Greenwich, CT: JAI Press,
ron. Econ. Manage., Mar. 1981, 8(1), pp. 72-8. 1981, pp. 281-99.
ROSEN, SHERWIN. "TheTheory of Equalizing Differ- ._ Risk by choice: regulating health and safety
ences," in Handbook of labor economics. Eds.: in the workplace. Cambridge:HarvardU. Press,
ORLEY ASHENFELTER AND RICHARD LAYARD. Am- 1983.
sterdam: North-Holland, 1986, pp. 641-92. . "The Valuation of Risks to Life and Health:
. "The Value of Changes in Life Expectancy," Guidelines for Policy Analysis,"in Benefits assess-
J. Risk Uncertainty,Sept. 1988, 1(3), pp. 285-304. ment:the state of art. Eds.: JUDITH D. BENTKOVER,
SCHELLING, THOMAS C. "The Life You Save May Be VINCENT COVELLO, AND JERYL MUMPOWER. Dor-
Your Own," in Problems in public expenditure and drecht, Holland:Reidel Publishers, 1986, pp. 193-
analysis. Ed.: SAMUEL B. CHASE, JR. Washington, 210.
DC: BrookingsInstitute, 1968, pp. 127-62. "ProspectiveReference Theory: Towardan
SMITH, ADAM. The wealth of nations. New York: Explanationof the Paradoxes,"J.Risk Uncertainty,
Modern Library, [1776] 1937. Sept. 1989, 2(3), pp. 235-63.
SMITH, ROBERT S., "TheFeasibility of an 'InjuryTax' . "Sources of Inconsistency in Societal Re-
Approach to Occupational Safety," Law Contemp. sponses to Health Risks,"Amer. Econ. Rev., May
Probl., Summer-Autumn 1974, 38(4), pp. 730- 1990, 80(2), pp. 257-61.
44. . Fatal tradeoffs: public and private responsi-
. The occupational safety and health act. bilitiesfor risk. New York:OxfordU. Press, 1992a.
Washington, DC: American Enterprise Institute, . Smoking:making the risky decision. New
1976. York:OxfordU. Press, 1992b.
"CompensatingWage Differentialsand Pub- VISCuSI, W. KIP AND EVANS, WILLIAM N. "Utility
Functions That Depend on Health Status: Esti- viron. Econ. Manage., July 1991, 21(1), pp. 32-51.
mates and Economic Implications,"Amer. Econ. Viscusi, W. KiPAND MOORE, MICHAEL J. "Workers'
Rev., June 1990, 80(3), pp. 353-74. Compensation: Wage Effects, Benefit Inadequa-
Viscusi, W. KiP AND MAGAT, WESLEY A. Learning cies, and the Value of Health Losses," Rev. Econ.
about risk:Consumerand workerresponsesto haz- Statist., May 1987, 69(2), pp. 249-61.
ard information. Cambridge: Harvard U. Press, . "Rates of Time Preference and Valuations
1987. of the Duration of Life," J. Public Econ., Apr.
Viscusi, W. Kip;MAGAT, WESLEY A. AND FORREST, 1989, 38(3), pp. 297-317.
ANNE. "Altruisticand Private Valuationsof Risk . "Worker Learning and Compensating Differ-
Reduction,"'J.PolicyAnal. Manage., Winter 1988, entials," Ind. Lab. Relat. Rev., Oct. 1991, 45(1),
7(2), pp. 227-45. pp. 80-96.
Viscusi, W. Kip; MAGAT, WESLEY A. AND HUBER, Viscusi, W. KIP AND O'CONNOR, CHARLES. "Adaptive
JOEL. "An Investigationof the Rationalityof Con- Responses to Chemical Labeling: Are Workers
sumer Valuationsof Multiple Health Risks,"Rand Bayesian Decision Makers?"Amer. Econ. Rev.,
J. Econ., Winter 1987, 18(4), pp. 465-79. Dec. 1984, 74(5), pp. 942-56.
. "Pricing Environmental Health Risks: ZECKHAUSER, RICHARD AND SHEPARD, DONALD.
Survey Assessments of Risk-Risk and Risk- "Where Now for Saving Lives?" Law Contemp.
Dollar Trade-Offsfor Chronic Bronchitis,"J. En- Probl., Autumn 1976, 40, pp. 5-45.