MORRIS,
USING
HEALTHLIFE
ECONOMICS
HAMMITT
EXPECTANCY TO COMMUNICATE BENEFITS
Using Life Expectancy to
Communicate Benefits of Health Care
Programs in Contingent Valuation Studies
Jill Morris, PhD, James K. Hammitt, PhD
Background. There is growing interest in the use of contin-
gent valuation (CV) to estimate the monetary value of health
program benefits. Ideally, CV could be used to value a specific
shift in survival curve. However, a shift in survival curve may
prove too complex for widespread use in CV instruments. To
facilitate the use of CV in valuing longevity benefits, research-
ers need alternative summary measures that describe the lon-
gevity benefit in a single number that is more readily commu-
nicated in a CV context. Methods. The authors compare 2
methods for communicating longevity benefits in a CV sur-
vey. Random subsamples of respondents valued a longevity
benefit expressed either as a continuing reduction in annual
mortality risk or as a gain in life expectancy. To compare the
validity of the alternative descriptions, the authors evaluate
1. Introduction
The rapid growth of medical technologies offers an
expanding array of interventions that extend human
lives. An increasing number of preventive programs,
such as environmental regulations to reduce air pollu-
tion, are also aimed at prolonging lives. Resource con-
straints preclude implementation of all possible life-
extending opportunities and necessitate comparison of
the anticipated costs and benefits of alternative inter-
ventions. Benefit-cost analysis requires credible values
of longevity benefits if it is to contribute in a meaning-
ful manner to the resource allocation debate.
There is growing interest in the use of contingent
valuation (CV) to estimate the monetary value of health
program benefits. CV uses surveys to elicit risk-dollar
trade-offs in hypothetical decisions. A recent review
suggests that the use of CV for evaluation of health care
programs is in a developmental stage that would bene-
fit from research to enhance the validity of the meth-
ods.1 We offer this study as a contribution to the devel-
468 · MEDICAL DECISION MAKING/NOV–DEC 2001
HEALTH ECONOMICS
willingness to pay (WTP) estimates for consistency with theo-
retical predictions. Results. It is found that WTP for a longev-
ity benefit is sensitive to the framing of the benefit, with re-
spondents expressing higher WTP for the benefit expressed
as a life expectancy gain. The life expectancy format per-
forms better than the risk reduction format in one important
regard—sensitivity to scope of the benefit—and no worse
than the risk reduction format in other regards. Conclusion.
Expressing longevity benefits in terms of life expectancy ap-
pears to hold promise as a method for enhancing the validity
of economic evaluation of health care programs. Key words:
risk communication; contingent valuation; life expectancy;
willingness to pay; pneumonia vaccine. (Med Decis Making
2001;21:468–478)
opment of CV methods for estimating monetary values
of longevity benefits.
An intervention that extends life alters the benefi-
ciary’s survival curve. Ideally, we could use CV to elicit
a monetary value for the specific shift in the survival
curve. However, a survival curve is a complex con-
struct, and describing benefits as a shift in the survival
curve might prove too cognitively demanding for use in
CV applications. We evaluate alternative summary
Received 17 August 2000 from the Center for Risk Analysis, Harvard
School of Public Health, Boston, Massachusetts (JM, JKH); and the
Centers for Disease Control and Prevention, Atlanta, Georgia (JM). Re-
vision accepted for publication 25 June 2001. This work was sup-
ported by the U.S. Environmental Protection Agency (R 825312-01-0).
The research presented herein does not represent the opinions of the
Centers for Disease Control and Prevention or the U.S. Department of
Health and Human Services.
Address correspondence and reprint requests to Dr. Morris: Centers
for Disease Control and Prevention, 4770 Buford Highway NE, MS K-
28, Atlanta, GA 30341-3724; telephone: (770) 488-3262; e-mail:
jmorris1@cdc.gov.
 USING LIFE EXPECTANCY TO COMMUNICATE BENEFITS
measures that collapse the information into a single
number that can be more easily communicated.
A variety of summary measures may be used in lieu
of presenting alternative survival curves to respon-
dents. Two common measures are expressing the bene-
fit as a reduction in mortality risk or as a gain in life ex-
pectancy. With the risk reduction method, the benefit is
usually expressed as an absolute reduction in probabil-
ity of death (or the complementary increase in survival
probability) over a specified time period. However, any
reduction in mortality risk implies a gain in life expec-
tancy, and it is relatively simple to convert a reduction
in mortality risk to the equivalent life expectancy gain.
Both methods are potentially useful summaries be-
cause each collapses a shift in the survival curve into a
single number. However, both methods have limita-
tions—expressing risk reduction as a probability may
be cognitively difficult, and expressing risk reduction
as a change in life expectancy does not describe a
unique change in the survival curve.
Over the past 20 years, economists have gained con-
siderable experience with the risk reduction method by
asking survey respondents to value small changes in
their risk of death in a specified time period (often 1
year). A significant body of research suggests that peo-
ple have difficulty thinking about and distinguishing
among small quantitative probabilities,2-6 and this diffi-
culty is evident in CV surveys of mortality risk reduc-
tions. Although most CV studies do not directly test re-
spondents’ comprehension of basic probability
concepts, 2 studies that did found that only about 50%
to 60% of respondents could correctly identify the
larger of 2 risk reductions expressed as probabilities.7,8
The cognitive difficulties associated with small
probabilities have led to recent attempts to use life ex-
pectancy gain as an alternative for eliciting the value of
longevity benefits. Johannesson and Johansson9,10 ex-
amined willingness to pay now for an additional year
of life expectancy conditional on reaching age 75, and
Johnson and others11 estimated willingness to pay for
an “added year of life” with certainty. To our knowl-
edge, these studies are the first to ask respondents to
value benefits described in terms of life extension, with
the Johannesson and Johansson studies being the only
ones to explicitly use a life expectancy framing. Al-
though these studies are promising in that respondents
appear willing to answer questions about the value of
longevity benefits, they did not examine whether val-
ues were sensitive to the choice of framing or whether a
life expectancy framing is a valid alternative to tradi-
tional risk reduction descriptions.
This article addresses the question of how to best to
communicate a life extension benefit in the context of
HEALTH ECONOMICS
CV. We compare 2 methods of expressing a life exten-
sion benefit by asking half of a random sample to value
a life extension benefit expressed as a continuing re-
duction in annual mortality risk and the other half of
the sample to value the equivalent benefit expressed as
a gain in life expectancy. There are 2 issues on which
the study sheds light. First, we examine whether will-
ingness to pay (WTP) for a longevity gain is sensitive to
the framing of the benefit. That is, we ask whether dif-
ferent methods of expressing an equivalent benefit
yield different estimates of WTP. Second, we compare
the validity of WTP estimates generated by alternative
methods by using consistency of the estimates with
theoretical predictions as a (construct) validity check.
For example, one would expect that WTP for longevity
benefits would be larger for people with higher in-
comes and for benefits of greater magnitude (or scope).
One might also expect WTP for longevity benefits to in-
crease with the quality of life expected during the
added longevity. In this study, we use sensitivity of
WTP to scope as a primary test of validity. Sensitivity of
WTP to income, predicted quality of life, and other
sociodemographic variables are used as secondary
tests of validity.
The remainder of the article is structured as follows.
In section 2, we present background information on the
use of sensitivity to scope as a tool for evaluating the va-
lidity of CV instruments. Section 3 describes survey de-
sign and methods for using survey data to test the re-
search hypotheses. In section 4, we present survey
results and an evaluation of the communication meth-
ods based on sensitivity to scope and the secondary va-
lidity criteria. We conclude with a discussion of the re-
sults and proposals for future research in section 5.
2. Background
A life expectancy framing may seem preferable to a
risk reduction framing because it avoids difficulties
with explaining small probabilities, yet it is not obvi-
ous that a life expectancy framing is superior for com-
municating longevity benefits. There is no empirical
evidence comparing the methods in this context and
very little evidence about the public’s understanding of
the life expectancy concept in general. Given that we
face a choice of methods for communicating a life ex-
tension benefit, how should we choose between them?
One way to gain insight into this question is to elicit
values for the same good using both methods. Similar
results would obviate the need for a choice and provide
evidence of convergent validity of the valuation meth-
ods. If the methods yield dissimilar results, criteria are
required for selecting the more appropriate method.
469
 MORRIS, HAMMITT
One criterion is to compare CV estimates with theo-
retical predictions about how the estimates should
vary according to factors logically related to WTP. One
theoretical prediction is that WTP should be sensitive
to scope; that is, it should depend on the size of the ben-
efit. Many CV studies have been criticized for insensi-
tivity to scope, meaning that respondents’ WTP for
mortality risk reductions did not increase with the size
of the risk reduction. In most cases where WTP did in-
crease with the size of the risk reduction, it did not in-
crease as much as economic theory predicts it should.†
Under reasonable assumptions,‡ economic theory
predicts that WTP for a larger quantity of a good ex-
ceeds WTP for a smaller quantity. For the special case
in which the good in question is a short-term reduction
in the probability of death, WTP should be nearly pro-
portional to the change in probability. That is, if a re-
duction in annual mortality risk from 3 in 100,000 to 2
in 100,000 is valued at $50, then a larger reduction from
3 in 100,000 to 1 in 100,000 should be valued at about
$100. WTP may deviate from proportionality if the pay-
ment constitutes a sizable portion of the respondent’s
income or yields a large change in mortality risk,12-14
but these effects are negligible in the context of WTP for
small reductions in mortality risk.15,16
In the manner just described, economic theory pro-
vides a criterion for defining adequate sensitivity of
WTP to scope for short-term mortality risk reductions.
However, the good that we are interested in valuing—a
continuing risk reduction—is more complex. The in-
tervention provides a series of annual risk reductions.
WTP for a series of risk reductions can be evaluated us-
ing a simple economic model, as described in Appen-
dix A.
3. Methods
3.1. SURVEY DESIGN
We employed a professional survey firm to adminis-
ter a 2-part CV survey from December 1998 to March
1999. The 1st component of the survey was a national
random-digit-dial phone interview used to recruit sub-
jects and gather socioeconomic data. The 2nd compo-
nent of the survey consisted of a follow-up phone inter-
view administered after respondents received a mailed
packet of material, although none of the mailed mate-
†
The issue of inadequate sensitivity to scope is reviewed in detail
in Hammitt and Graham.8
‡
These assumptions include nonsatiation and that more of a good
is preferred to less.
470 · MEDICAL DECISION MAKING/NOV–DEC 2001
rial was used for this study. Multiple follow-up at-
tempts were made to reach each respondent.
The CV question was based on WTP for a hypotheti-
cal pneumonia vaccine. Half of the survey respondents
received a version in which the vaccine benefit was ex-
pressed as a life expectancy gain (hereafter referred to
as the life expectancy subsample) and half received a
version in which the equivalent benefit was expressed
as a reduction in average annual chance of death (here-
after referred to as the risk reduction subsample). Exact
wording of the CV question for each subsample is given
in Appendix B.
To test for sensitivity to scope, each subsample was
further split and asked to value goods of different mag-
nitude, resulting in a total of 4 subsamples. Two of the 4
subsamples received a version in which they were
asked WTP now for a vaccine given at age 60, and the
other 2 subsamples received a version in which they
were asked WTP now for a vaccine given at age 70. Be-
cause receiving the vaccine at age 60 includes the bene-
fit of receiving the vaccine at age 70 as well as addi-
tional benefit between ages 60 and 70, it is
unambiguously preferred, regardless of time prefer-
ence. Respondents whose age exceeded the age of vac-
cination in the survey version to which they were ran-
domized were not asked a WTP question.
The magnitude of the benefit depends on whether
the respondent received the version with vaccination
at age 60 or with vaccination at age 70, and the framing
of the benefit depends on whether the respondent re-
ceived the version with benefit expressed as life expec-
tancy change or as a reduction in the annual probability
of dying. Based on pretesting of the survey instrument,
we elected to use a hypothetical pneumonia vaccine
with a benefit about twice as large as the actual benefit
estimated from known pneumonia mortality rates and
vaccine effectiveness.
For respondents randomized to the risk reduction
subsample, baseline risk was expressed as an average
annual chance of dying and the vaccine benefit was ex-
pressed as an absolute reduction of 0.2% in the annual
average chance of dying. Respondents valuing the vac-
cine at age 60 were told it would reduce their average
annual mortality risk from 4.8% to 4.6% for the rest of
their life; those valuing the vaccine at age 70 were told
it would reduce their average annual mortality risk
from 7.0% to 6.8%. In both cases, the complementary
annual survival probabilities were also stated. Respon-
dents randomized to the life expectancy subsample
were told that the benefit of receiving the vaccine at age
60 (70) was an increase in life expectancy from 21 years
to 21 years, 11 months (from 14 years to 14 years, 5
months).
 USING LIFE EXPECTANCY TO COMMUNICATE BENEFITS
The survey question consisted of 2 parts. After pre-
senting information about baseline risk, the respondent
was asked whether he or she would consider getting
the vaccine (either at age 60 or at age 70, depending on
the subsample). If the respondent answered “yes,” he
or she was presented with a WTP question. If “no,” he
or she was presented with a follow-up question with re-
gard to the reason for not considering the vaccine.
Respondents who indicated that they would con-
sider getting the vaccine were presented with a double-
bounded dichotomous-choice WTP question. With
this format, each respondent is randomly assigned an
initial bid. If the respondent answers “no” to the initial
bid, a follow-up question with a predetermined lower
bid is presented. If the respondent answers “yes” to the
initial bid, a follow-up question with a predetermined
higher bid is presented. Three bid vectors of (initial/
higher/lower) 220/$700/$40, 400/$1500/$80, and 750/
$3500/$130 were selected to span the range of WTP es-
timates observed in pretesting. The payment vehicle
was specified as an out-of-pocket expense to be paid
now for the pneumonia vaccine to be received either at
age 60 or at age 70. Respondents were reminded to con-
sider their current budget and expenses.
The survey also included questions in which the re-
spondents were asked to rate on a scale from 1 to 10
their perceived quality of life at their current age and
their predicted quality of life at the age of 65.
3.2. RESEARCH HYPOTHESES
We examine 3 primary hypotheses. First, we test for
sensitivity of WTP to the framing of the vaccine benefit.
That is, we ask whether estimated WTP depends on
whether the benefit was described as an increase in life
expectancy or a reduction in annual probability of dy-
ing. Our null hypothesis is that WTP is insensitive to
the framing of the benefit.
Second, we test for sensitivity of WTP estimates to
scope within the life expectancy and risk reduction
subsamples. That is, we ask whether the ratio of WTP
for the vaccine at age 60 to WTP for the vaccine at age 70
depends on the framing of the benefit. For each
subsample, the null hypothesis is that WTP estimates
are not sensitive to scope, that is, that there exists no
difference in WTP for vaccine at age 60 versus vaccine
at age 70.
Third, we test for associations between WTP and
factors that are logically related to WTP for longevity
benefits, such as income and predicted quality of life.
We predict that, controlling for other socio-
demographic characteristics, WTP should increase
with these factors.
HEALTH ECONOMICS
3.3. STATISTICAL METHODS
We estimated parametric regression models using
WTP as the dependent variable. Because we did not
elicit point estimates of WTP, each respondent’s WTP is
unobserved but bounded by the bids presented in the
double-bounded dichotomous-choice question. This
format results in either interval-censored data, in
which we know both the lower and upper bounds of
the dependent variable, or right-censored data, in
which we know only the lower bound. We assume all
WTP values are greater than or equal to zero. Using SAS
PROC LIFEREG,17 we fit a variety of models using the
method of maximum likelihood. In preliminary model
fitting, we examined both log normal and Weibull dis-
tributions. Model results were similar for both distribu-
tions, with median WTP estimates slightly lower using
the log normal distribution. A Weibull distribution was
assumed in all subsequent analyses.
4. Results
4.1. SAMPLE CHARACTERISTICS
Of 1456 participants who were recruited, 1104 par-
ticipated in the 2nd telephone interview (response rate
of 75%). Of these, 134 respondents were excluded be-
cause their age exceeded the age of vaccination in the
survey version to which they were randomized. Thus, a
total of 970 respondents completed the pneumonia
vaccine questionnaire.
Of the 970 respondents who received the pneumo-
nia vaccine question, 681 (70%) said they would con-
sider getting the vaccine. The remaining 289 indicated
that they would not consider getting the vaccine or did
not know whether they would consider it. Only re-
spondents who indicated that they would consider the
vaccine were presented with the WTP question. Re-
spondents who indicated that they would not consider
getting the vaccine were similar to those who indicated
they would, except they were younger (mean age 39 vs.
41 years), more likely to have children younger than 18
(50% vs. 42%), and more likely to be white (83% vs.
79%). There were no significant differences in income,
years of education, percentage employed, percentage
male, percentage married, or predicted quality of life at
age 65 (using a 2-sample t test with P = 0.05). Respon-
dents who received the life expectancy version of the
survey were more likely to indicate that they would not
consider getting the vaccine (33% vs. 26%, P = 0.01).
About 70% of respondents who would not consider the
vaccine reported that vaccine benefits are “too small”
or “uncertain.” Another 10% cited safety concerns, 3%
471
 MORRIS, HAMMITT

Table 1 Sample Characteristics Table 2 Crude Willingness to Pay Responses

for Pneumonia Vaccine: Joint Frequency
Life Risk
Expectancy Reduction
Frequency of Response:
Variable Subsample Subsample
Number (Proportion)
n 332 349
Life Risk
Percentage receiving version with
Bids 1st (2nd) Expectancy Reduction
vaccine at age 60 49 48 Responses 1st/2nd n Subsample Subsample
Percentage receiving version with
vaccine at age 70 51 52 $220($700/$40) 231a
Agea 40.8 40.4 No/no 22 11 (0.10) 11 (0.09)
Percentage male 45 42 No/yes 82 34 (0.30) 48 (0.40)
Incomeb $50,600 $45,500 Yes/no 87 41 (0.37) 46 (0.39)
Educationc 14.5 14.3 Yes/yes 40 26 (0.23) 14 (0.12)
Percentage employed full-time 63 60 $400($1500/$80) 225
Percentage married 58 52 No/no 46 20 (0.18) 26 (0.23)
Percentage with children < 18 43 41 No/yes 95 55 (0.49) 40 (0.36)
Percentage white 81 75 Yes/no 60 27 (0.24) 33 (0.29)
Mean predicted quality of life at Yes/yes 24 11 (0.10) 13 (0.12)
age 65 (1 to 10 scale) 7.4 7.4 $750($3500/$130) 225
a. Measures respondent’s median age within the following categories: (18- No/no 55 24 (0.22) 31 (0.26)
24), (25-29), (30-34), (35-39), (40-44), (45-49), (50-54), (55-59), (60-64), (65- No/yes 99 43 (0.40) 56 (0.47)
69), and (70+). Age 70 was used for the upper bound.
b. Measures respondent’s median household income within the following Yes/no 54 27 (0.25) 27 (0.23)
categories: (<$15K), ($15 to $25K), ($25K to $40K), ($40K to $50K), ($50K to Yes/yes 17 13 (0.12) 4 (0.03)
$75K), ($75K to $100K), and (>$100K). $15K and $100K were used for the
lower and upper bounds, respectively. a. For example, 231 respondents received an initial bid of $220, of which
c. Measures respondent’s years of education, where less than high school = 104 indicated that they would not be willing to pay $220 for the vaccine and
10; graduated high school = 12; some college = 14; graduated college = 16; 127 indicated that they would be willing to pay $220 for the vaccine. The
and graduate or professional school = 18. 104 respondents who declined were subsequently offered a lower bid of
$40. Of these, 22 indicated that they would not be willing to pay $40
whereas 82 respondents indicated that they would be willing to pay $40 for
the vaccine. The 127 respondents who initially expressed willingness to
pay $220 were then presented with a larger bid of $700. Of these, 87 indi-
voiced a general dislike for vaccines, and 17% pro- cated that they would not be willing to pay $700 and 40 respondents indi-
vided no reason. cated that they would be willing to pay $700 for the vaccine.

Table 1 presents sample characteristics for the life

expectancy and risk reduction subsamples. The full
sample had a mean age of about 41 and was predomi-
nantly white (78%), female (56%), married (55%), and
employed full-time (62%). The mean annual house-
hold income ($48,000) and years of education (14.4)
were above the national averages of $44,000 and 12.7
years, respectively.18 The subsamples were similar,
with no significant differences between sample means
except for percentage white, which was greater in the
life expectancy subsample (P = 0.05).
4.2. WTP FOR PNEUMONIA VACCINE
Table 2 summarizes the crude responses for the life
expectancy and risk reduction subsamples. Pooled
crude responses are presented graphically in Figure 1.
The step functions shown in Figure 1 are
nonparametric Kaplan-Meier estimators that represent
the probability of saying “yes” to a WTP bid as a func-
tion of bid amount (shown on the horizontal axis). The
upper (lower) curve represents the probability of ac-
cepting the bid based on the upper bound (lower
bound) of each respondent’s WTP interval. As ex-
pected, the fraction accepting is inversely related to the
bid amount. The smooth curve between the step func-
tions represents the Weibull model fit to the unob-
served WTP estimates. The Weibull model lies almost
entirely between the upper and lower bounds, suggest-
ing an adequate fit to the data.
Table 3 presents estimated median WTP for the
pneumonia vaccine for each subsample defined by the
framing and magnitude of the benefit. Medians were
estimated independently for each subsample using a
regression model with no covariates and assuming a
Weibull distribution. Across the 4 subsamples, median
WTP for a pneumonia vaccine ranged from $280 to
$451.

472 · MEDICAL DECISION MAKING/NOV–DEC 2001

 USING LIFE EXPECTANCY TO COMMUNICATE BENEFITS
1
0.9
0.8
0.7
P(ACCEPT BID)
0.6
0.5
0.4
0.3
0.2
0.1
0
0 500 1000 1500 2000 2500
WTP BID
Figure 1. Kaplan-Meier (nonparametric) estimates of willingness to
pay (WTP) compared with fitted Weibull model.
4.3. SENSITIVITY OF WTP TO BENEFIT FRAMING
The last row of Table 3 presents the ratio of median
WTP in the life expectancy subsample to median WTP
in the risk reduction subsample (denoted WTPLE/
WTPRR) for vaccination either at age 60 or at age 70. If
WTP estimates were insensitive to the framing of the
benefit, the value of these ratios would be 1. For vacci-
nation at age 60, WTPLE/WTPRR = 1.52, indicating that
those who received the benefit expressed as a change in
life expectancy had a median WTP 52% greater than
those who received the same benefit expressed as an
absolute risk reduction (P < 0.01). For vaccine at age 70,
WTPLE/WTPRR = 1.04, indicating that those who re-
ceived the benefit expressed as a change in life expec-
tancy had a median WTP 4% greater than those who re-
ceived the equivalent benefit stated in terms of absolute
risk reduction, although this difference was not statisti-
cally significant. Thus, the hypothesis that WTP is in-
sensitive to the framing of the benefit can be rejected for
the subsample with vaccination at age 60 (P < 0.01) but
not for the subsample with vaccination at age 70 (P =
0.88).
4.4. SENSITIVITY OF WTP TO SCOPE
The last column of Table 3 presents the ratio of me-
dian WTP for vaccination at age 60 to median WTP for
vaccination at age 70 (denoted WTP60/WTP70) for the
life expectancy and risk reduction subsamples. If WTP
estimates were insensitive to the scope of the good, the
value of these ratios would be 1. That is, WTP for the
good of greater magnitude, vaccination at age 60,
HEALTH ECONOMICS
Table 3 Median Willingness to Pay (WTP) for
Pneumonia Vaccine by Subsample
WTP60 WTP70 WTP60/WTP70
Life expectancy (LE) $451 $291 1.55*
UPPER-BOUND WTP Risk reduction (RR) $296 $280 1.06
LOWER-BOUND WTP
Ratio WTPLE/WTPRR 1.52** 1.04
Weibull
*P < 0.05. **P < 0.01.
would be no greater than WTP for the good of lesser
3000 3500
magnitude, vaccination at age 70.
For the life expectancy subsample, WTP60/WTP70 =
1.55, indicating that median WTP for vaccination at age
60 was 55% greater than median WTP for vaccination
at age 70 (P < 0.05) among those who received the bene-
fit expressed as a change in life expectancy. For the risk
reduction subsample, WTP60/WTP70 = 1.06, indicating
that median WTP for vaccination age 60 is 6% greater
than median WTP for vaccination at age 70 among
those who received benefit expressed as a reduction in
probability of dying, although this difference was not
statistically significant. Thus, the hypothesis that WTP
is insensitive to scope can be rejected for the life expec-
tancy subsample (P < 0.05) but not for the risk reduc-
tion subsample (P = 0.72).
For the life expectancy subsample, the WTP esti-
mates are consistent with the prediction that WTP for a
larger good should exceed WTP for a smaller good. As
noted earlier, economic theory also provides insight
into how much larger WTP for the larger good is ex-
pected to be. Using a simple model in which WTP is
proportional to discounted quality-adjusted life expec-
tancy (Appendix A), we derive the predicted ratio of
WTP for the vaccine at age 60 to WTP for the vaccine at
age 70. The ratio is independent of the respondent’s
age.
For simplicity, we weight all years equally, so WTP
is proportional to the gain in discounted life expec-
tancy. As shown in Figure 2, the predicted WTP ratio is
a function of the discount rate, r.§ For a discount rate of
0, WTP60/WTP70 is simply the ratio of life expectancy
gains calculated from age 60 (or before), about 2.6.¶ Be-
cause the vaccine at age 60 provides greater near-term
§
We assume the utility discount rate and consumption discount
rate are equal, as is conventional in cost-effectiveness analysis. If the
consumption discount rate exceeds the utility discount rate, the cal-
culated ratio will be larger.
¶
The value can be calculated as the ratio of the life expectancy
gain from age 60 to the life expectancy gain from age 70, equal to 11
months divided by (5 months × 0.85), where 0.85 is the probability of
surviving from age 60 to age 70.
473
 MORRIS, HAMMITT

4 Table 4 Regression of Willingness to Pay

(WTP) for Pneumonia Vaccine on
Magnitude of Benefit, Framing
of Benefit, and Selected Sample Covariates
3
Life Risk
Full Expectancy Reduction
Variable Sample Subsample Subsample

2 n 618 301 317

Intercept 5.67 5.86 5.36
(0.550) (0.779) (0.725)
VACC60 0.914* 1.34* 1.09
1 (0.450) (0.633) (0.584)
LIFE_EXP 0.039a N/A N/A
–0.1 –0.05 0 0.05 0.1
(0.121)
Discount rate LIFE_EXP*VACC60 0.447a N/A N/A
(0.236)
Figure 2. Predicted ratio of willingness to pay for vaccine at age 60 SRQoL (1-10) 0.077* 0.0481 0.091
to willingness to pay for vaccine at age 70. (0.034) (0.0479) (0.048)
Age 0.00853 0.00256 0.018
(0.0074) (0.0110) (0.010)
benefits, the ratio increases with the discount rate. For AGE*VACC60 –0.0245* –0.0239 –0.029*
a 3% rate, the ratio is about 2.9. For discount rates less (0.011) (0.0154) (0.014)
than 0, the ratio falls to a minimum of about 2.1. Includ- Income ($1000) 0.00677** 0.00836* 0.00552
ing health-related quality-of-life weights that decline (0.0025) (0.00359) (0.00349)
with age has an effect similar to increasing the discount Education (years) –0.0360 –0.032 –0.0295
rate, and yields a larger ratio. (0.0290) (0.0414) (0.0391)
As shown in Table 3, we estimated WTP60/WTP70 ra- WHITE –0.406** –0.474* –0.436*
tios of 1.55 and 1.04 for the life expectancy and risk re- (0.149) (0.230) (0.193)
duction subsamples, respectively. These values are Gender (1 = male) 0.101 0.527** –0.321
smaller than the minimum value implied by our sim- (0.125) (0.176) (0.169)
ple model, which suggests that the WTP estimates are Log likelihood –781.86 –383.07 –391.18
inadequately sensitive to scope. Note: SRQoL = predicted quality of life at age 65 on a scale from 1 ( worst
quality imaginable) to 10 (best quality imaginable). Standard error is in
Table 4 presents statistical results for the estimated parentheses.
regression equations for the full sample (i.e., pooling a. LIFE_EXP and LIFE_EXP*VACC60 joint P < 0.05.
responses across vaccination ages) and for the life ex- *P ≤ 0.05. **P ≤ 0.01.

pectancy and risk reduction subsamples. The full-

sample model includes an indicator variable
(LIFE_EXP) for framing of the longevity benefit, which
is equal to 1 for the respondents to whom the benefit
was described as a life expectancy gain and 0 for the re-
spondents to whom the benefit was described as a re-
duction in annual mortality risk.
The full-sample and subsample models include an
indicator variable for magnitude of benefit. VACC60 is
an indicator variable equal to 1 for the subsample pre-
sented with vaccination at age 60 (the larger good) and
equal to 0 for the other subsample. AGE*VACC60 rep-
resents the interaction between age and magnitude of
benefit.
The coefficient on VACC60 is positive in both
subsamples but significant only in the life expectancy
subsample, indicating that only the life expectancy
subsample shows significant sensitivity to scope. This
result is consistent with Table 3. Note, however, the
negative interaction between age and VACC60 for each
subsample (P < 0.05 for the full sample and the risk re-
duction subsample). This interaction implies declining
sensitivity to scope with increasing age.†† The effect is
illustrated in Table 5, which shows WTP (as predicted
by the regression models in Table 4) for a 55-year-old
white man and a 25-year-old white man (holding other
covariates constant at the sample mean). Sensitivity to
scope can be assessed by examining the ratio WTP60/
††
This pattern was also evident when we examined median WTP
in subsamples younger and older than 50 years.

474 · MEDICAL DECISION MAKING/NOV–DEC 2001

 USING LIFE EXPECTANCY TO COMMUNICATE BENEFITS
Table 5 Predicted Median Willingness to Pay
(WTP) for Selected Ages
Ratio WTP60/
Subsample Age WTP60 WTP70
Life expectancy 55 $361 $352
Life expectancy 25 $685 $326
Risk reduction 55 $164 $278
Risk reduction 25 $234 $163
Note: For white male with other covariates equal to full sample mean.
WTP70 in the last column. If WTP were proportional to
discounted life expectancy, the value of this ratio
would be approximately 2.9 (assuming a 3% discount
rate). For both ages, sensitivity to scope is greater for the
life expectancy subsample than for the risk reduction
subsample, but in each subsample the ratio WTP60/
WTP70 for the 25-year-old is closer to the predicted op-
timal value than the corresponding ratio for the 55-
year-old. This finding runs counter to our a priori ex-
pectation that, all else equal, older respondents would
be likely to exhibit greater sensitivity to scope because
the benefits are closer in time and likely to be more sa-
lient to them.
4.5. VARIATION IN WTP WITH
SOCIODEMOGRAPHIC CHARACTERISTICS
Consistent with the hypothesis that a longevity ben-
efit is a normal good, income is positively associated
with WTP in all subsamples, although it is statistically
significant (P < 0.05) in only the full sample and the life
expectancy subsample (Table 4). The estimated income
coefficient (0.0068 in the full sample) implies that for
an additional $1000 of household income, WTP for a
pneumonia vaccine is $2.30 greater on average (at the
mean sample income). Although this effect appears
small, the implied income elasticity of WTP is about
0.30 (at the sample mean income), a value consistent
with reported empirical income elasticities for WTP for
mortality risk reduction.
Predicted quality of life at age 65 (SRQoL), as mea-
sured on a scale of 1 (worst quality imaginable) to 10
(best quality imaginable), is positively associated with
WTP for pneumonia vaccine for all subsamples, al-
though the effect is statistically significant only for the
full sample. The estimated full-sample regression coef-
ficient implies that WTP for the pneumonia vaccine is
about $23 greater if predicted quality of life is 1 unit
larger (at the sample mean SRQoL of 7.4). A related
question about the respondent’s perceived chance of
HEALTH ECONOMICS
surviving long enough to benefit from the vaccine had
no significant predictive effect (results not shown).‡‡
Years of education is negatively but not significantly
associated with WTP in each subsample. We found no
WTP70 interaction between education and magnitude of the
benefit, suggesting that greater educational attainment
1.03
is not associated with improved sensitivity to scope. To
2.10
the extent that sensitivity to scope is a proxy for com-
0.59
prehension of the magnitude of the goods being valued,
1.44
this result suggests that greater educational attainment
is not related to better comprehension of either a life ex-
pectancy or a risk reduction framing.
5. Discussion
Obtaining credible estimates of the value of longev-
ity benefits is important if we wish to efficiently allo-
cate resources among competing programs that extend
human lives. By comparing 2 major approaches to ex-
pressing longevity benefits, we offer the present study
as a contribution to the ongoing search for methods that
enhance the validity of CV in the context of valuing
longevity benefits.
Results from this study suggest that WTP for a lon-
gevity benefit is sensitive to the framing of the benefit.
Respondents valued benefits expressed as a life expec-
tancy gain more highly than identical benefits ex-
pressed as a reduction in annual mortality risk. That al-
ternative methods for expressing an identical benefit
yield dissimilar WTP estimates leads logically to the
question of which method yields the “better”
estimates.
To compare the validity of the alternative methods,
we evaluated the respective WTP estimates for consis-
tency with theoretical predictions. We invoke sensitiv-
ity of WTP to scope as the primary criterion for assess-
ing the construct validity of the alternative methods.
We observed increasing WTP with increasing magni-
tude of longevity benefit in the life expectancy
subsample but not in the risk reduction subsample.
Thus, our primary test of validity suggests that the life
expectancy method demonstrates greater validity than
the risk reduction method.
We examined the relationship between WTP and
other factors such as income and predicted quality of
life as secondary tests of validity. The methods per-
formed comparably with regard to these factors. Both
methods yield the expected positive relationship be-
‡‡
The question asked: “How would you rate your chance of living
to age 60 (70) compared with others of your age and gender? Is your
chance much higher, a little higher, about the same, a little lower, or
much lower?”
475
 MORRIS, HAMMITT
tween income and WTP, with implied income elastici-
ties within the range observed in the literature. Both
methods exhibit a positive association of similar mag-
nitude between predicted quality of life at age 65 and
WTP, although the effect is statistically significant only
for the full sample. Likewise, the methods produce
similar relationships between WTP and age and be-
tween WTP and education. Thus, the secondary tests of
validity suggest that the methods are of comparable
validity.
We conclude that the life expectancy method ap-
pears to perform better than the risk reduction method
in at least one important regard—sensitivity scope—
and no worse than the risk reduction method in other
regards. To the extent that enhanced sensitivity to
scope reflects better comprehension of the valuation
task, we have partial evidence that a life expectancy
framing is more readily understandable. If, indeed, the
life expectancy framing is more comprehensible, then
the monetary value of longevity benefits is more likely
to reflect true preferences if elicited using a life expec-
tancy framing.
This promising finding is tempered by several limi-
tations. First, a life expectancy framing appears to work
less well for older respondents in that scope sensitivity
exhibited by the life expectancy subsample is stronger
for younger respondents than for older respondents.
Although the life expectancy format appears to work
better than the risk reduction format in all cases, its su-
periority is less marked among older respondents.
Thus, our a priori expectation—that greater sensitivity
to scope would be seen among older respondents be-
cause they may find the good more salient and thus
consider the valuation question more carefully—is not
borne out by the empirical results.
Second, the sensitivity to scope exhibited by the life
expectancy subsample falls short of the degree of sensi-
tivity suggested by a simple economic model. The
model suggests that the ratio of WTP for the vaccine at
age 60 to WTP for the vaccine at age 70 should be larger
than 2, but our empirical estimate is only 1.55.
Third, respondents who received the life expec-
tancy version of the sample were more likely to indi-
476 · MEDICAL DECISION MAKING/NOV–DEC 2001
cate that they would not consider getting the pneumo-
nia vaccine. To the extent that this negative response
reflects a general rejection of the CV questionnaire
rather than a genuine response to the question, it can be
argued that greater scenario rejection occurred using
the life expectancy framing.
Despite these limitations, we believe that expressing
longevity benefits in terms of life expectancy holds
promise for enhancing the validity of health care CV
applications. Future research should focus on improv-
ing our understanding of how the public comprehends
the life expectancy concept, which should enable us to
improve risk communication tools by identifying and
“designing out” common misperceptions, as well as of-
fering insight into methods for obtaining appropriate
scope sensitivity. In addition, it would be helpful to
compare the sensitivity of WTP to differences in lon-
gevity benefit described as a gain in life expectancy or
as a reduction in annual mortality risk holding the time
period over which the risk is reduced constant and
varying the magnitude of the reduction in probability
of death. Future research should also compare the life
expectancy method with other risk communication
methods not examined here, such as using relative ver-
sus absolute risk reduction,19 using cumulative rather
than annual mortality risk reduction,20 or using an al-
ternative metric such as “number needed to treat” (the
reciprocal of absolute risk reduction).21,22
The growing popularity of health care CV studies
leads to the broader consideration of the need for crite-
ria by which to judge discrepant WTP estimates. We
suggest using sensitivity to scope as the primary crite-
rion and sensitivity to factors such as income and qual-
ity of life as secondary criteria for comparing the valid-
ity of alternative CV methods. As the number of health
care CV applications grows and the methods evolve, it
will be important to examine the merits of various
types of CV validity tests. Advances in both validity
testing and risk communication should be integrated
into the conceptual framework for health care CV
studies.23
The authors thank Phaedra Corso for assistance with survey de-
sign and administration and Alan Krupnick for helpful comments.
 USING LIFE EXPECTANCY TO COMMUNICATE BENEFITS

APPENDIX A
Simple Economic Model of Willingness to Pay
for Continuing Reductions in Mortality Risk

To examine the theoretically predicted sensitivity of will- Using equation (2), the ratio of WTP at age a for the vaccine
ingness to pay (WTP) to the magnitude of the benefit from the taken at age 60 to WTP at age a for the vaccine taken at age 70
pneumonia vaccine, consider a simple model in which WTP is
for an increase in longevity is proportional to the gain in dis-
counted quality-adjusted life expectancy: ∞ q 60 − p (3)
a
1+ d
10 ∑t = 0 (t1 + r )tt ut
= 
v −a ∞ WTP60 m60
pv  qtv / pv − pt / pv  
WTPva = 
1  .
 mv ∑ ut , (1) a  1+ r  q 70 − p
pa  t =v
WTP70 m70 ∞
1+ d (1 + r )t −v  ∑t = 0 (t1 + r )tt ut
where WTPva is the individual’s WTP at age a for the vaccine
taken at age v. The probability of surviving from birth to age t Note that the WTP ratio is independent of the respondent’s
is pt in the absence of a vaccine andqtv if the vaccine is taken at age. For simplicity, we assume that the health-related quality
age v. Note that qtv = pt for all t < v and that the probability of of life is 1 for all ages (equivalently, that WTP is proportional
surviving from age a to age v equals pv /pa. to discounted life expectancy). Using values of ut that decline
The bracketed term in equation (1) is the individual’s WTP with age would yield a larger WTP ratio. We also assume the
at age v. It is the discounted quality-adjusted life expectancy marginal utility of income is equal at ages 60 and 70, so m60/
(ut is the health-related quality-of-life weight for age t and r is m70 = 1. Because mt is the inverse of the marginal utility of in-
the utility discount rate) multiplied by the inverse of the mar- come, diminishing marginal utility of income would imply
ginal utility of income at age v, mv . We allow for the possibil- that m60/m70 > 1 if income declines from age 60 to 70 and m60/
ity that the consumption discount rate d differs from the util- m70 < 1 if income increases. Consistent with standard practice
ity discount rate r. To calculate WTP at age a, we discount for in cost-effectiveness analysis, we assume the individual dis-
the time until the benefit is received (v – a) using the con- counts income and utility at the same rate, so that d = r. The
sumption discount rate d (because the benefit at age v is val- plausible alternative assumption is that d > r, which yields a
ued in monetary terms) and multiply by the probability of larger WTP ratio.
surviving from age a to age v. We set the baseline survival probabilities pt using the 1995
24
Making use of the fact that qtv = pt for t < v and canceling U.S. life table. The survival probabilities with the vaccine
the pv terms yields are modeled under the assumption that the vaccine reduces
the age-specific mortality rate by a constant factor, which is
v −a ∞
1+ r  qtv − pt
WTPva = 
mv chosen to yield the specified life-expectancy gain. The vac-
1+ d

pa
∑ (1 + r )t
ut . (2)
cine taken at age 60 reduces annual mortality risk by about
t =o
10%, and the vaccine taken at age 70 reduces annual mortal-
ity risk by about 5%. The resulting WTP ratio is plotted as a
function of the discount rate in Figure 2.

APPENDIX B
Contingent Valuation Questions

LIFE EXPECTANCY METHOD
(WITH BID VECTOR 1: $220/$700/$40)
Part 1
On average, a person aged 60 (70) has a life expectancy of
21 (14) years. That is, the average 60-year-old (70-year-old)
will live to age 81 (84). Suppose that a pneumonia vaccine
will be available to you when you reach age 60 (70). The vac-
cine is perfectly safe and if you get vaccinated when you are
60 (70), your life expectancy will increase from 21 years to 21
years and 11 months (from 14 years to 14 years and 5 months).
Would you consider getting the vaccine at age 60 (70)? [If re-
spondent answers “yes,” continue with part 2.]
Part 2
Now assume that you have to pay some money this year to
have the vaccine available to you when you reach age 60 (70).
Assume that you would need to pay this cost out of your own
pocket; it would not be covered by insurance. Also assume
that no better vaccine would become available before you
reach age 60 (70). Considering your current income and ex-
penses, would you pay $220 this year to have the vaccine
available for you when you reach age 60 (70)? [If respondent

HEALTH ECONOMICS 477

 MORRIS, HAMMITT
answers “yes,” present higher bid of $700; if respondent an-
swers “no,” present lower bid of $40.]
RISK REDUCTION METHOD
(WITH BID VECTOR 1: $220/$700/$40)
Part 1
On average, a person aged 60 (70) has a 4.8% (7%) proba-
bility of dying each year from all causes. That is, the average
chance of living at least 1 more year is 95.2% (93%). Suppose
REFERENCES
1. Diener A, O’Brien B, Gafni A. Health care contingent valuation
studies: a review and classification of the literature. Health Econ.
1998;7:313–26.
2. Gomez UAL. Communicating very low probability events. Unpub-
lished dissertation; Carnegie Mellon University; 1990.
3. Fisher A, Chestnut LG, Violette DM. The value of reducing risks of
death: a note on new evidence. J Policy Analysis Manage. 1989;8:88–
100.
4. Baron J. Biases in quantitative measurement of values for public
decisions. Psychol Bull. 1997;122:72–88.
5. Baron J. Confusion of relative and absolute risk in valuation. J Risk
Uncertainty. 1997;14:301–9.
6. Tversky A, Kahneman D. Rational choice and the framing of deci-
sions. J Business. 1986;59(suppl):S251–S278.
7. Jones-Lee MW, Hammerton M, Phillips PR. The value of safety: re-
sults of a national survey. Econ J. 1985;95:49–72.
8. Hammitt JK, Graham JD. Willingness to pay for health protection:
inadequate sensitivity to probability? J Risk Uncertainty. 1999;18:33–
62.
9. Johannesson M, Johansson PO. To be or not to be, that is the ques-
tion: an empirical study of the willingness to pay for an increased life
expectancy at an advanced age. J Risk Uncertainty. 1996;13:163–74.
10. Johannesson M, Johansson PO. Quality of life and the WTP for an
increased life expectancy at an advanced age. J Public Econ.
1997;65:219–28.
11. Johnson FR, Desvousges WH, Ruby MC, Stieb D, De Civita P.
Eliciting health state preferences: an application of willingness to
pay for longevity. Med Decis Making. 1998;18(suppl):S57–S67.
12. Jones-Lee MW. The value of changes in the probability of death or
injury. J Polit Econ. 1974;82:835–49.
478 · MEDICAL DECISION MAKING/NOV–DEC 2001
that a pneumonia vaccine will be available to you when you
reach age 60 (70). The vaccine is perfectly safe, and if you get
vaccinated when you are 60 (70), your annual probability of
dying each year would decrease from 4.8% to 4.6% (from 7%
to 6.8%). That is, your annual probability of surviving each
year will increase from 95.2% to 95.4% (from 93% to 93.2%).
Would you consider getting the vaccine at age 60 (70)? [If re-
spondent answers “yes,” continue with part 2.]
Part 2
Same as above.
13. Weinstein MC, Shepard DS, Pliskin JS. The economic value of
changing mortality probabilities: a decision-theoretic approach. Q J
Econ. 1980;94:373–96.
14. Pratt JW, Zeckhauser RJ. Willingness to pay and the distribution
of risk and wealth. J Polit Econ. 1996;104:747–63.
15. Hammitt JK. Valuing mortality risk: theory and practice. Environ
Sci Tech. 2000;34:1396–1400.
16. Hammitt JK. Evaluating contingent valuation of environmental
health risks: the proportionality test. Association of Environmental
and Resource Economists Newsletter. 2000;20(1):14–9.
17. SAS Institute. SAS procedures guide, version 6. 3rd ed. Cary, NC:
SAS Institute Inc; 1990.
18. U.S. Bureau of the Census. Statistical abstract of the United
States: 1999. 119th ed. Washington, DC: U.S. Bureau of the Census;
1999.
19. Stone ER, Yates JF, Parker AM. Risk communication: absolute ver-
sus relative expressions of low-probability risks. Organ Behav Hu-
man Decis Processes. 1994;60:387–408.
20. Krupnick A, Alberini A, Cropper M, et al. Age, health, and will-
ingness to pay for mortality risk reductions: a contingent valuation
survey of Ontario residents. Discussion paper no. 00-37; Resources
for the Future; 2000.
21. Walter SD. Choice of effect measure for epidemiological data. J
Clin Epidemiol. 2000;53:931–9.
22. McQuay HJ, Moore RA. Using numerical results from systematic
reviews in clinical practice. Ann Intern Med. 1997;126:712–20.
23. O’Brien B, Gafni A. When do the “dollars” make sense? Toward a
conceptual framework for contingent valuation studies in health
care. Med Decis Making. 1996;16:288–99.
24. National Center for Health Statistics. Vital statistics of the United
States, 1995. Hyattsville, MD: National Center for Health Statistics;
1998. (Preprint of vol 2, Mortality, pt A, sec 6, Life Tables).