Professional Documents
Culture Documents
Wiktor Adamowicz (corresponding author) is at Department of Resource Economics and Environmental Sociology, University of Alberta, T6G 2H1, Edmonton, Alberta, Canada (vic
.adamowicz@ualberta.ca). Mark Dickie is at Department of Economics, University of Central
Florida, P.O. Box 1400, Orlando, FL 32816-1400, United States. Shelby Gerking is at Department of Economics, University of Central Florida, P.O. Box 1400, Orlando, FL 328161400, United States, and Department of Economics and Tilburg Sustainability Center, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, the Netherlands. Marcella Veronesi is at
Department of Economics, University of Verona, Vicolo Campoore 2, 37129, Verona, Italy,
and Institute for Environmental Decisions, Professorship of Environmental Policy and Economics, ETH Zurich, Universittstrasse 22, 8092 Zurich, Switzerland. David Zinner is at Department of Resource Economics and Environmental Sociology, University of Alberta, T6G 2H1,
Received November 29, 2013; Accepted September 18, 2014; Published online November 6, 2014.
JAERE, volume 1, number 4. 2014 by The Association of Environmental and Resource Economists.
All rights reserved. 2333-5955/2014/0104-0002$10.00
http://dx.doi.org/10.1086/679255
481
482
December 2014
Edmonton, Alberta, Canada. The US Environmental Protection Agency (US EPA) partially
funded this research under RD-3159201-0 and RD-83326301-0. The research has not been
subjected to US EPA review and may or may not reect the views of the agency; no ofcial
endorsement should be inferred. We thank seminar participants at the Tinbergen Institute, US
EPA National Center for Environmental Economics, ETH Zurich, and the Norwegian University of Life Sciences, as well as Glenn Blomquist, Alan Krupnick, Martina Menon, Eleonora
Matteazzi, Federico Perali, the editor, and two anonymous reviewers for helpful comments on
previous drafts.
1. Parents willingness to pay for reduced risks to their childrens health can be included in
monetary benets to determine the socially optimal level of risk reduction if parents are
safety-focused altruists ( Jones-Lee 1991) but not if their altruism takes the form of benevolence (Bergstrom 2006).
Adamowicz et al.
483
for virtually all existing estimates of parents willingness to pay to reduce risks to childrens health (Gerking and Dickie 2013).
The collective model (Blundell, Chiappori, and Meghir 2005) looks within the
household to examine the interaction of two utility-maximizing parents with different
preferences as they allocate resources between their own consumption and a household public good such as expenditures for a child. Household decisions are Pareto efcient and depend on the relative decision-making power of each parent. Just as in
the unitary model, Pareto efciency raises the possibility that parents will engage in
offsetting behavior in the face of policy changes aimed at increasing their childrens
protection from health risks. In the collective model, Pareto efciency also implies
that the households marginal willingness to pay for a childs risk reduction may be
determined from either parents decisions.2
The noncooperative approach (e.g., Browning, Chiappori, and Lechene 2010) differs from the unitary and collective models in that household decisions are not
Pareto efcient. Thus, this model raises the issue that parents may provide too little
protection for children relative to socially efcient amounts. Noncooperative parents
make choices about expenditures for the child individually, rather than collectively.
The distribution of decision-making power between parents may or may not affect
household marginal willingness to pay for reduced childrens health risks.
Empirical estimates used to discriminate between the three models of household
decision making make use of data from a eld study of heart disease collected in
2011 from a nationally representative (US) online panel maintained by Knowledge
Networks, Inc.3 Heart disease is the leading cause of death in the United States and
has been linked to exposure to environmental pollution such as airborne particulates
(Brook et al. 2004). A unique feature of the data is that they include 432 matched
pairs of mothers and fathers (i.e., 864 parents) with biological children aged 616
years that live together in the same household. The analysis uses stated preference
methods to examine parents intentions to purchase hypothetical goods that reduce
their own and their childrens risk of heart disease. The approach taken has the advantage that characteristics of these goods, such as price and the extent to which health
risk is reduced, are experimentally controlled so that key parameters needed to test
hypotheses of interest can be estimated consistently. Special attention is paid to
minimizing the potential problem that respondents may misstate their true purchase
intentions.
2. The relationship between individual and household willingness to pay for a public good
has received considerable attention in environmental economics. For a recent treatment of this
topic, see Ebert (2013).
3. Bonke and Browning (2011) present survey evidence on household expenditures for
goods for children but do not estimate willingness to pay or test for Pareto efciency.
484
December 2014
Empirical results support four main ndings. First, intended purchases of a good
to reduce heart disease risks for parents and their children satisfy efciency conditions derived from the collective model. These conditions are: (1) for a given proportionate reduction in heart disease risk, the household is willing to pay the same
amount of money at the margin to protect the mother, father, and child and (2) parents choices about proportionate health risk reductions for their children are based
on household valuations, rather than their own individual valuations. This nding is
interpreted as evidence against the noncooperative model and contrasts with earlier
results of Bateman and Munro (2009) and Lindhjem and Navrud (2009). On the
other hand, Pareto efciency is consistent with both the unitary and collective models.
Second, results show no statistically signicant within-household differences between the marginal willingness to pay of mothers and fathers for an absolute (rather
than proportionate) heart disease risk reduction for their child. This nding reects
more than Pareto efciency, as it also requires the mother and father in each household to perceive that their child faces the same initial level of heart disease risk.
Estimates indicate that annual household marginal willingness to pay to reduce the
childs risk by 1 chance in 100 of being diagnosed with heart disease by the age of
75 is $5.62 (SE = 2.20) for mothers and $4.08 (SE = 1.49) for fathers. Nonetheless, parents perceptions of risk to their child for a different disease may not be in
close agreement. Thus, in general, even if Pareto efciency prevails, spouses may
give different values of marginal willingness to pay to reduce the childs health risk
by a given absolute amount even though both parents respond on behalf of the
household.
Third, a statistically signicant within-household difference is found between
mothers and fathers marginal willingness to pay to reduce their own heart disease
risk by an absolute amount. The annual marginal willingness to pay of mothers to
reduce their own risk of heart disease by 1 chance in 100 prior to the age of 75 is
$5.48 (SE = 1.97), whereas for fathers the corresponding estimate is $2.14 (SE =
1.00). Additionally, consistent with results of Hammitt and Haninger (2010) and
Alberini and Scasny (2011), parents marginal willingness to pay to reduce heart
disease risk by 1 chance in 100 for their children is larger than their marginal willingness to pay to reduce this risk for themselves, although these differences are not statistically signicant at conventional levels.
Fourth, evidence presented shows that a shift in intrahousehold decision-making
power between parents, resulting from a redistribution of the household budget, does
not signicantly affect the marginal willingness to pay for an absolute reduction in
heart disease risk of any household member. This outcome supports a central feature
of the unitary model and is also consistent with the collective model if each parents
marginal willingness to pay for absolute reductions in the childs risk is equally sensitive to a redistribution of household resources.
Adamowicz et al.
485
The paper is organized into four additional sections. Section 1 develops the theoretical model. Section 2 describes the data. Empirical estimates are presented in section 3. Conclusions and implications for future research on valuing health risk reductions are drawn out in section 4.
1. THEORETICAL FRAMEWOR K
This section adapts the collective model of household decision making by a couple
with children proposed by Blundell et al. (2005) to introduce household production
of perceived health risk.4 The household production feature is added to support empirical analysis using data from the eld study. This section also briey contrasts
implications from the collective model with those of the unitary and noncooperative
approaches to facilitate a more complete interpretation of the empirical analysis.
1.1. Model
The model envisions a household consisting of a mother (m), a father ( f ), and one
child (k). Because the couple has one child, decisions about fertility or resource allocation among multiple children are not considered. Parents allocate household resources during one period and exhibit safety-focused altruism toward their child. The
child has neither income nor bargaining power within the household. Parents divide
their time between leisure (L), health risk reduction activities (H) for themselves
and for their children, and work (T).5 Each parent faces a health risk (R) and is
concerned about a health risk facing the child. Parents treat reduced risk to the child
as a public good within the household. They may purchase a market good (G) to reduce their own and their childrens risk; H and G are choice variables, so health risk
is endogenous.
Each parents utility (U i) is determined by
U i = U i Ci ; Li ; Ri ; Rki ; i = m; f ;
where Ci denotes her or his consumption of private goods, Li denotes her or his
leisure time, Ri denotes the mothers or fathers perception of her or his own health
risk, and Rki denotes the mothers or fathers perception of health risk to the child.
486
December 2014
The central questions in this paper deal with resource allocation to the child, so
neither parent is assumed to care about her or his spouses health risk or consumption.6 The parents utility is strictly increasing and strongly concave in consumption
and leisure and strictly decreasing and strongly concave both in own and child health
risks.
Perceived risks are home produced in an efcient and cost-minimizing way (Chiappori 1997) according to the production functions
Ri = Ri Gi; H i; Ai ;
Rki = Rki Gk; H k; Aki i = m; f :
In these equations, Gi and Gk = g km g kf represent the parents and the childs use
of a market good (G) to reduce perceived health risk, where gki denotes the amount
of the good provided by parent i. Correspondingly, H i and H k = hkm hkf represent
the parents allocation of time to reduce perceived health risk to the parent and to the
child, where hki denotes the amount of time provided to the child by parent i. The terms
Ai and Aki represent exogenous indices of parent is attitudes and information concerning her or his health risk and the childs health risk.7 Parents can vary their purchases
of goods and their use of time to reduce health risk to themselves and to their children.
A couple may differ in attitudes and information Aki and may have different perception functions for the childs risk; thus Rkm will equal Rkf only as a special case. Perceived risks are diminishing and convex in G and H.8
The household money budget constraint is given by
Y wm T m wf T f = C m C f pGm G f G k ;
where Y denotes household nonlabor income, wiT i denotes labor earnings of parent
i (wi is the wage rate of parent i), the price of the private consumption good is equal
to unity, and p denotes the common price of risk-reducing goods for each household
6. A model in which each parent cares about the utility of the other yields efciency predictions that are consistent with those developed here. This outcome occurs because the efciency conditions tested are derived from rst-order conditions. As discussed by Bergstrom
(2006), marginal conditions necessary for efciency with benevolent preferences are the same
as those without caring. The model with caring, however, does not as readily support the perspective of decentralized decision making that is adopted below for consistency with the eld
study.
7. This formulation assumes that the risk-reducing good provided to the child by the
mother and father are perfect substitutes. Similarly, mothers time and fathers time are assumed to be perfect substitutes in reducing the childs health risk.
8. G and H are assumed not to be direct sources of utility. The possibility that riskreducing goods and activities directly affect parental utility is addressed in the econometric
methods discussed later.
Adamowicz et al.
487
member. Each parent is endowed with one unit of time and divides it among timeconsuming activities according to Li H i hki T i = 1. The household full income
budget constraint is
Y wm wf = C m C f pGm G f G k
i
wi Li H i hki :
m; f
subject to the full income budget constraint in equation (4), the health risk production functions in equation (2), and nonnegativity restrictions on all purchased goods
and time allocations. The Pareto weight , interpreted as the relative decision-making
power of the mother, is bounded by zero and one. It is a function of the price of riskreducing goods, nonlabor income, wage rates of both parents, and a set of exogenous
variables (z) referred to as distribution factors that can affect bargaining power in
the household without affecting preferences or the budget constraint (Browning and
Chiappori 1998).9 The function p; Y; wm ; wf ; z is assumed to be continuously differentiable in its arguments and homogeneous of degree zero in prices, wages, and nonlabor income (Browning, Chiappori, and Weiss 2014).
A Pareto-efcient allocation of resources involving public goods can be sustained
by decentralized decision making given a suitable distribution of household income
and an efcient choice of Lindahl shares (Cornes and Sandler 1996; Donni 2009).
Implications of decentralized decision making are emphasized here because in the
eld study spouses made independent decisions about whether to purchase goods to
reduce their own and their childrens health risks.
Decentralized decision making can be thought of as a two-stage process. In the
rst stage, parents agree on a sharing rule governing the distribution of household income and on Lindahl shares specifying each parents contribution to intrahousehold
public expenditures for the benet of the childs health. The full income share of
parent i, si, constrains her or his expenditures according to the individual time inclusive budget constraint si = Ci pGi pti Gk wi Li H i i H k , where sm sf =
Y wm wf . Also, ti = g ki =Gk and i = hki =Hk are Lindahl shares indicating the
relative contributions of Gk and Hk that each parent makes to reduce health risk for
9. See Browning et al. (2006) and Chiappori and Ekeland (2006) for a general discussion
on the key role played by distribution factors to identify the structure underlying the collective household model.
488
December 2014
the child. Values of si are determined by the same set of factors that determine . In
the second stage, each parent chooses Ci; Li; Gi; H i; g ki; hki to maximize her or his
own utility
U i Ci; Li; Ri Gi; H i; Ai ; Rki g km g kf; hkm hkf; Aki ;
subject to the individuals money and time budget constraints and nonnegativity restrictions on the choice variables Ci; Li; Hi; Gi; g ki; hki . To simplify discussion, we assume an interior solution for consumption, leisure time, and risk-reducing goods.
Empirical analysis presented in section 3 tests whether parents intended purchases of a good to reduce risk of heart disease satisfy the rst-order conditions of
the model. These conditions require that both for themselves and for their children,
each parents marginal willingness to pay to reduce risk equates to the marginal cost
of risk reduction.10
U i =Ri
p
=
;
U i =Ci Ri =Gi
U i =Rki
ti p
=
:
U i =Ci Rki =Gk
First-order conditions for efcient time use also are assumed to be satised, although
these conditions are not tested empirically.11
If the production of risk exhibits declining marginal productivity of G (see eq. [2]),
then the marginal willingness to pay for a one-unit reduction in risk increases as risk
declines. Likewise, each parents marginal willingness to pay to reduce her or his childs
risk by one unit equates to the marginal cost of doing so times the parents Lindahl
share. The product of the parents Lindahl share ti and p represents the parents individualized Lindahl price. Marginal willingness to pay for a one-unit reduction in the
childs risk also increases as risk declines. Outcomes in equation (7) do not depend on
the relative decision power of the mother (; see Blundell et al. 2005).
10. This outcome that marginal willingness to pay for risk reduction may be expressed in
terms of only the price of the risk-reducing good and its marginal product in reducing risk
(i.e., terms from the utility function are absent) follows from the assumption that riskreducing goods are not direct sources of utility (see Harrington and Portney [1987] for further discussion).
11. A more complete test of the model would examine efciency of time allocation. For example, as long as parents time spent reducing the childs risk has no direct effect on utility,
efciency requires that parents allocate time to child risk reduction according to intrahousehold
comparative advantage as determined by relative wages and relative marginal products of parents time inputs. Data used in this study, however, contain insufcient information to determine whether conditions for efciency in time allocations are met. Empirical implications of
this feature of the data are discussed in sections 3 and 4. In any case, to conserve space, rstorder conditions for time use are not presented.
Adamowicz et al.
489
Econometrically testing whether equation (7) holds is simplied by examining behavior in the face of proportionate health risk reductions, rather than absolute health
risk reductions. Thus, equation (7) is re-expressed as
U i =Ri Ri i
= p;
U i =Ci
U i =Rki Rki ki
= ti p;
W ki =
U i =Ci
Wi =
where i = Ri =Gi =Ri and ki = Rki =Gk =Rki denote proportionate health risk
changes. Equation (8) demonstrates that a parents marginal willingness to pay for
risk-reducing goods (W i and W ki) equals the parents marginal willingness to pay for
proportionate risk reductions U i =Ri Ri =U i =Ci and U i =Rki Rki =U i =Ci
weighted by i and ki .
Equation (8) implies a version of the standard Lindahl-Samuelson efciency condition for the public good Gk:
Wk =
oW
i
ki
o
i
U i =Rki Rki ki
= ptm tf = p i = m; f :
U i =Ci
Pareto efciency implies that household marginal willingness to pay for the riskreducing good for the child (W k), computed as the sum of parents individual marginal willingness to pay, equates to the price of the good. Furthermore, household
marginal willingness to pay equals the ratio of either parents individual marginal willingness to pay relative to her or his Lindahl share W k = p = W ki =ti .
Collecting results in equations (8) and (9) and considering an equal proportionate
reduction in all perceived risks = m = km = f = kf implies that the parents
allocate resources so that the household is willing to pay an equal amount for an
equal proportionate reduction in risk for any member as shown in equation (10).
U m =Rm Rm U m =Rkm Rkm U f =Rf Rf U f =Rkf Rkf p
=
= :
=
=
U m =C m
tm U m =C m
U f =C f
tf U f =C f
10
No information is available from the eld study to determine the individual contribution of a mother and a father to reducing health risk for their child, so the Lindahl
shares (ti) cannot be econometrically identied. Nonetheless, testing for Pareto efciency is not impeded because Lindahl shares sum to unity. As shown in equation (11), given that equation (10) is satised, this joint contribution equates to each
parents individual marginal willingness to pay to reduce her or his own health risk.
U m =Rm Rm U f =Rf Rf U m =Rkm Rkm U f =Rkf Rkf p
= :
=
=
U m =Cm
U f =C f
U m =C m
U f =C f
11
490
December 2014
Equation (11) provides the basis for the econometric tests for Pareto efciency
presented in section 3.
1.3. Effects of Redistribution of Household Resources
As previously indicated, equation (10) does not depend on the mothers relative
decision-making power ( ) in the household. The solution values for the choice variables Ci; Gi; Li; H i; g ki; hki in the parents utility maximization problem, however, do depend on because this parameter inuences the share of resources allocated to each parent. For instance, the solution value for Gi is determined according
to Gi p; Y; wm ; wf ; Ai ; Aki ; i ; ki ; z. The proportionate health risk changes (i and
ki ), treated in the eld study as experimentally assigned characteristics of the riskreducing goods, are included as arguments.
Inferences about how the distribution of household decision power affects children often rest on examination of effects of distribution factors (z) on a measure of
childrens health or on the demand for childrens goods. Theoretical results derived
by Blundell et al. (2005) imply that an increase in the mothers relative decision power
arising from a change in z will boost household consumption of the risk-reducing
good for the child if the mothers individual marginal willingness to pay for this good
is more sensitive to shifts in decision power than is the fathers. Section 3 presents a
test for whether changes in parents relative contributions to the household budget,
interpreted as a distribution factor, affect marginal willingness to pay to reduce childrens health risk.
1.4. Other Perspectives on Efciency and Distribution
Unitary and noncooperative models of household behavior provide alternatives to the
collective perspective on efciency and distribution within families. As previously indicated, the unitary model treats the household as the decision-making unit, and
interaction between parents is ignored. To set up the unitary model in a way comparable to the collective model just presented, assume that the household maximizes
the weighted sum of utilities given by equation (5). For values of the Pareto weight
strictly bounded by zero and one, resource allocation in the unitary household would
satisfy equation (11) just as in the collective model. In contrast to the collective model,
however, in the standard unitary model the Pareto weight is a constant rather than a
function of prices, wages, and distribution factors (Browning, Chiappori, and Lechene
2006). In consequence, solution values for the risk-reducing goods in the unitary
model do not depend on distribution factors, and changes in distribution factors do
not alter willingness to pay for risk reduction for any household member. Thus, the
unitary and collective models may be distinguished not by efciency, which occurs in
both models, but by the possible reaction to a change in a distribution factor. The
collective model allows the household equilibrium to change with changes in distribution factors, whereas the unitary model rules out this possibility.
Adamowicz et al.
491
The noncooperative model of Browning et al. (2010), on the other hand, yields
an alternative to the hypothesis of efciency. In this model, a noncooperative equilibrium occurs when each public good (except possibly one) is exclusively provided by
one parent. If parent i is the sole provider of the risk-reducing good for the child,
then
U i =Rki Rki ki
= p;
U i =Ci
U j =Rkj Rkj kj
< p; j i:
U j =C j
12
The eld study examined decentralized choices of goods that would reduce heart
disease risks for parents and children. The study also elicited parents perceptions of
their own and their childrens risks of developing heart disease and collected other
information. Data were obtained from Knowledge Networks, Inc.s national online
research panel.13 The panel is representative of the US population, and prior re-
12. Efcient and inefcient outcomes are not as easily distinguished, however, if noncooperative parents jointly contribute to one public good and if that good happens to be health
risk reduction for the child. In that case, both parents buy the risk-reducing good for the child
until meeting the equilibrium condition U i =Rki Rki =U i =Ci ki = p for i = m, f. The
sum of parents marginal valuations of the childs risk-reducing good would equate to twice
the price, and the households risk protection efforts for the child exceed the Pareto-efcient
amount. Although the resulting allocation is inefcient, it cannot be distinguished from the
efciency restriction in equation (9) without identication of the Lindahl shares.
13. Knowledge Networks recruits panel members by probability sampling from a sample
frame that covers about 97%99% of US households and includes cell-phone-only households,
households with listed and unlisted telephone numbers, nontelephone households, and households with and without Internet access. Recruited households are provided with Internet access and hardware if necessary. Panel members typically are asked to complete one survey
weekly. Members receive an e-mail invitation to participate in a survey and are provided with a
link to the Internet address of the survey questionnaire.
492
December 2014
search suggests that demands for health risk reductions estimated from samples
drawn from the panel are not biased by selection (Cameron and DeShazo 2010;
Yeager et al. 2011). Dickie and Gerking (2011) more fully document the eld study
and data. The survey instrument is available in an online appendix.
Panel members were eligible to participate in the study if they were parents ages
1855 years who had at least one biological child age 616 years living in the home
and had not previously been diagnosed with coronary artery disease or experienced
a heart attack. Parents with a prior history of heart disease were excluded to focus
on ex ante perception and valuation of risk. Older teenagers were excluded because
they are more likely than younger children to earn an income and make independent
consumption decisions.14 Children under age 6 years were excluded because in focus
groups (see below) conducted prior to administering the survey parents expressed difculty assessing and valuing heart disease risk for very young children. For the 74%
of parents with two or more children living at home, one child was randomly selected
and designated as the sample child. Roughly half (52%) of the sample children were
male, and the average age of sample children was 11 years.
A total of 2,554 parents living with their spouses or partners completed the survey during JanuaryMarch 2011. Among these parents, 864 observations are utilized
from matched pairs of spouses living together (i.e., 432 matched pairs).15 A key element of the sample design was to maximize the number of paired spouses in the
overall sample. The second parent in each matched pair received an e-mail invitation
to participate after the rst parent had completed the survey. On average, the second
parent in a pair completed the survey 29 days after the rst parent.16 Empirical analysis is based on the sample of matched parents because the focus of this paper is on
resource allocation within two-parent households.17 Each parent in a matched pair is
questioned about the same child, who was a biological child of both parents.
Table 1 presents a comparison of parents in the 432 matched pairs to demographic characteristics of married parents with spouse and own children present from
the Current Population Survey (CPS). Matched parents in the sample resemble married parents in the CPS in household size, age, employment status of husband, and
earnings differences between spouses. Also, mothers in the matched pairs are about
14. Dauphin et al. (2011) present evidence that teenagers 16 years of age and older are
decision makers in the household.
15. In the data, 434 matched pairs of parents are available. In two of these pairs, however,
one member did not provide information about willingness to pay to reduce heart disease risk.
16. The time gap between survey completion within pairs of parents had a minimum of 6,
a median of 28, and a maximum of 50 days. The second parent completed the survey within
2 weeks of the rst parent in 2.5% of cases and within 6 weeks in 90% of cases.
17. Potential differences between matched parents and married parents whose spouses
did not complete the survey are examined empirically in section 3.
57.4
Husband earns 20 to 30 more than wife
Husband is 1 to 2 years older than wife
80 to < 90
4.4
2.2
89.6
50 to < 60
10.8
43
51.8
81.6
3.5
7.6
88.8
40 to < 50
8.8
40 to 44
35.8
66.6
7.7
17.5
Husband
58.2
Husband earns 10 to < 30 more than wife
Husband is 2 to 3 years older than wife
75 to < 100b
4.3
1.9
B. Data on Household
64.5
10 to <20
38.0
41
55.5
83.6
2.1
5.5
Wife
67.8
15 to < 20
35.3
35 to 39
38.3
67.2
7.0
17.3
Wife
Employed (%)
Median earnings ($1,000)
Earn <$5,000 (%)
Median age (years)
Bachelors degree or more (%)
White, not Hispanic (%)
Black, not Hispanic (%)
Hispanic (%)
Husband
494
December 2014
equally likely as those in the CPS to be employed (65% vs. 68%) and about equally
likely to contribute less than $5,000 to household income (38% vs. 35%). Whites
and the college educated are overrepresented among the matched parents, and median earnings of men and total household income are about $10,000 higher in the
sample than in the CPS.
The nal version of the survey reected information obtained from 25 parents in
two focus groups and over 400 parents in two pretests. The introductory part of the
survey provided a brief description of coronary artery disease and explained that in
the survey, the terms heart disease and coronary artery disease would be used synonymously. The main body of the survey was divided into three parts to elicit parents (1) initial perceptions of the risk to themselves and to their children of being
diagnosed with coronary artery disease before the age of 75 years, (2) revised perceptions of these risks, and (3) willingness to pay to reduce coronary artery disease risk
to themselves and to their children. The median parent completed the survey in
27 minutes.
2.1. Initial Risk Perceptions
Parents estimated the risk of being diagnosed with coronary artery disease before age
75 years using an interactive scale similar to that used by Dickie and Gerking (2007).
The scale depicted 100 numbered squares arranged in 10 rows and 10 columns. All
100 squares initially were colored blue. Parents re-colored squares from blue to red
to represent risk levels (see g. 1). For example, a parent could use a computer mouse
to indicate a risk of 36 chances in 100 by selecting the square numbered 36 in the
scale, causing all the squares from 1 to 36 to turn red. Beneath the scale, the level of
risk was indicated by displaying the percentage of the 100 squares that were colored
red. By selecting different squares, parents could make as many changes to the scale
as desired before selecting the Continue button to record the nal answer.18
Parents practiced using the risk scale before estimating the risk of getting coronary artery disease. First, they were shown four examples of scales representing risk
levels of 25%, 50%, 75%, and 100% and were told the relationship between these
percentages and chances in 100. Second, parents used the risk scales to represent
the chances of experiencing an automobile accident for each of two hypothetical
people, Mr. A, a relatively careless driver who had 33 chances in 100 of an accident
and Ms. B, a relatively safe driver who had a 1% chance of an accident. Respondents
then were asked which of these two people had the lesser chance of an accident. The
11% of parents that answered incorrectly were provided with additional review of
the risk scales and then correctly identied the individual with lower risk.
18. Knowledge Networks used the JavaScript Dojo Toolkit to construct the custom graphics used in the survey.
Adamowicz et al.
495
Figure 1. How many chances in 100 do you think you have of getting coronary artery
disease before you reach age 75? Please mark the scale to show your answer. The respondent
would read, Risk level: 36% chance of heart disease. A color version of this gure is available
online.
After completing the risk tutorial, parents answered a few questions about familiarity with coronary artery disease. Most parents indicated that they were aware of
this disease; 92% said that they had heard or read about it, 75% knew someone
personally who had had it, 69% had thought about the possibility that they themselves might get it, and 32% had thought about the possibility that one of their chil-
496
December 2014
dren might get it. After answering these questions, parents used the risk scale to estimate chances of getting coronary artery disease before age 75, rst for themselves
and then for the sample child.
Parents initial assessments of heart disease risks are summarized in table 2 for
mothers, fathers, and their children. Five features of these subjective risk perceptions
are of interest. First, there is considerable variation in risk assessments, with standard deviations that are about 60%75% as large as corresponding means. Second,
the average mother indicated that she had 35 chances in 100, and the average father
indicated that he had 37 chances in 100, of getting coronary artery disease. The
average mother appears to have overestimated her risk, whereas the average fathers
assessment is relatively close to epidemiological estimates of this risk. Average assessments of childrens risks also closely match epidemiological estimates.19 Third, mean
risk assessments suggest that parents may recognize that risks are higher for men
than for women. The null hypothesis that mean risk assessments are equal between
mothers and fathers would be rejected at the 6% level in a matched samples test.
Fourth, parents believe that their own risk of heart disease exceeds the risk faced
by their children. The null hypothesis that mean risk assessments are equal for mothers and their children is rejected at the 1% level in a matched samples test. Likewise
the null hypothesis that mean risk assessments are equal for fathers and their children is rejected at 1% in a matched samples test.20 Finally, mothers and fathers
broadly agree on the risks faced by their children. The null hypothesis that mothers
and fathers make identical risk assessments for their children is not rejected in a
matched samples t-test ( p-value = .32).
2.2. Revised Risk Assessments
After making initial estimates of coronary artery disease risk, parents had the opportunity to revise their estimates after receiving information about the disease. In this
section of the survey, parents were rst told that the average person has about
27 chances in 100 of being diagnosed with coronary artery disease before age 75.
This average risk was illustrated using a risk scale showing the 27% risk level next to
19. Based on data from the Framingham Heart Study (Lloyd-Jones et al. 1999), the
average US person faces a 27% risk of diagnosis with coronary artery disease before age 75.
For females, this risk is 19% and for males it is 35%.
20. Additionally, a regression of the parents risk perception for the child on the parents
risk perception for himself or herself (and a constant) indicates a positive linear association
that is signicant at the 1% level. The null hypothesis that the linear association between the
risk that a mother perceives for her child and for herself does not differ by child gender is
not rejected at conventional levels. Likewise, the null hypothesis that the linear association
between the risk that a father perceives for his child and for himself does not differ by child
gender is not rejected at conventional levels.
Through
10
20
30
40
50
60
70
80
90
100
From
0
11
21
31
41
51
61
71
81
91
Mean
SD
Median
Mode
Chances in 100
.171
.166
.171
.127
.205
.053
.039
.058
.012
.000
35
21
30
50
Initial
Self
.104
.207
.281
.138
.157
.030
.037
.042
.005
.000
32
18
25
25
Revised
.263
.189
.180
.118
.171
.028
.021
.025
.005
.002
28
20
25
50
Initial
Child
.196
.297
.283
.111
.085
.014
.002
.007
.002
.002
24
14
21
20
Revised
.148
.118
.214
.120
.191
.065
.046
.081
.014
.005
37
22
33
50
Initial
Self
.092
.138
.286
.182
.136
.058
.042
.053
.009
.005
35
19
30
25
Revised
.254
.224
.224
.069
.159
.021
.018
.023
.009
.000
27
19
25
50
Initial
Revised
.201
.353
.283
.060
.069
.014
.009
.007
.005
.000
23
14
20
20
Child
Table 2. Relative Frequency Distribution of Matched Pairs of Parents Assessments of Risk of Coronary Artery Disease Diagnosis before Age 75
(N = 432)
498
December 2014
the risk scales that parents had marked for themselves and their children. Parents
then were told that they and their children would probably not have the same risk as
the average person, because chances of getting heart disease depend on six risk factors
that are different for everyone: gender, smoking, current health status, family history,
exercise, and diet. The survey elicited current information from parents about each of
these risk factors as they pertain to themselves and their children while also providing information about how the factors inuence risk. For example, parents smoking
status was assessed, and respondents were advised that coronary artery disease risks are
higher for the average smoker than for the average nonsmoker. Risk levels for smokers
and nonsmokers were illustrated using risk scales.
Parents were shown their initial assessments as previously marked on the risk
scales and were permitted to revise their assessments if desired. Revised risk assessments, which reect parents assessments of the heart disease risk factors described
above, are shown in table 2. Parents revised their own risk assessments about as frequently as they revised their assessments of their childrens risk. About 40% of fathers and 45% of mothers made revisions. Downward revisions predominated, with
parents on average reducing their own risk assessment by two to three percentage
points and reducing their assessment for their children by about four percentage
points. In separate matched samples tests, each of the hypotheses that (1) mean revised assessments are equal for mothers and their children and (2) mean revised assessments are equal for fathers and their children is rejected at the 1% level.21
Similar to the outcome when considering initial risk assessments, the null hypothesis that mothers and fathers made identical revised assessments of their childrens
risk of heart disease is not rejected in a matched samples t-test (p-value = .17). These
results indicate that the average pair of parents shares a common perception of the
level of risk faced by their child. Table 3 presents the frequency distribution for the
difference between parents revised risk assessments for the child. As shown, this
distribution has a median and mode of zero. About 13% of mother/father pairs made
identical revised risk perception estimates, about 27% of these pairs made revised risk
perception estimates that differed by three percentage points or less, and about 57%
of mother/father pairs made revised risk perception estimates that differed by less
than 10 percentage points or less.
2.3. Willingness to Pay
The nal section of the survey elicited willingness to pay to reduce risk of coronary
artery disease. Parents rst were told that chest pain, shortness of breath, activity
limits, and the need for more medical treatment and medication commonly followed
a diagnosis of coronary artery disease. Then, to describe the timing of heart disease
21. In contrast to the initial risk assessments, parents revised assessments of their childrens risk are about 2 percentage points higher for male than for female children, and this
difference is signicant at the 5% level in a two-tail test.
Adamowicz et al.
499
Absolute Frequency
Relative Frequency
20
21
19
28
54
27
57
37
52
36
22
41
18
432
.046
.049
.043
.065
.125
.063
.132
.086
.120
.083
.051
.095
.042
1
risks, the survey employed a graphical illustration of risk for all ages between the
present and age 75. The illustration (see g. 2) used a Gompertz hazard function to
approximate the empirical (Kaplan-Meier) hazards estimated by Lloyd-Jones et al.
(2006). Hazard functions showed the cumulative risk of contracting heart disease at
any age from the present until age 75. These functions were constructed based on
each parents revised risk assessments, so that at age 75 the cumulative hazard shown
on the graph was equal to the parents revised risk estimate. When the respondent
used the mouse to indicate a point on the hazard function corresponding to a given
age, a text box appeared on the screen stating the risk of heart disease between the
present and the selected age. Cumulative hazard graphs were displayed for the parent
and then for the child.
To value reductions in heart disease risks, parents were told that they would be
asked about their purchase intentions for each of two vaccines.22 One of the hypo22. Focus group participants expressed only limited concerns about potential side effects of
vaccines in general and the heart disease vaccine in particular. In consequence, survey respondents were told: If the vaccine becomes available, it would have rst been approved by
government authorities using the same strict approval process for all other medications. Thus,
it would be approved only after extensive testing in people over many years. Respondents also
were told: In previous trials with vaccines, some people experienced side-effects like soreness
in the arm, fatigue, or slight stomach upset. These side-effects generally disappeared within
12 days. More serious side-effects rarely occurred. Questions about possible respondent concerns about side effects of vaccines were not included in the survey.
500
December 2014
Figure 2. Display of hazard function for 42-year-old parents revised risk assessment of
34%, showing cumulative risk of 12% by age 60. The gure shows the display as it would appear to a respondent using the cursor to determine the cumulative risk at age 60 years. The
respondent would read: Your heart disease risk prole is shown in the chart below. The
height of the red curve shows your heart disease risk between now and any of the ages up to
75. To see how the chart works: Point to the red mark on the curve above the age of 75 to
see your total risk of 34% between now and age 75. Point to any red mark on the curve
above any other age to see your risk between now and that age. A color version of this gure
is available online.
thetical vaccines reduced risk for the parent and the other reduced risk for the child.
The two vaccines were presented one at a time in random order. Parents were told
that the vaccines would slow the build-up of fatty deposits in the arteries, would be
taken by injection annually, and would provide additional protection from coronary
artery disease beyond the benets that could be obtained from eating right and
getting enough exercise. As the vaccines were described, their effectiveness was varied at random. Parents were assigned risk reductions of either 10% or 70% of their
revised risk assessment, and children were assigned risk reductions of either 20% or
80%.23 Each parent in a matched pair was assigned the same percentage risk reduc-
23. Percentage rather than absolute risk reductions were assigned to facilitate testing for
Pareto efciency in health resource allocation. See section 1, equation (7) and accompanying
discussion.
Adamowicz et al.
501
tion for the child, which always was larger than that assigned to either parent. Parents were told that these risk reductions would materialize only for those who continued to receive the vaccinations annually through age 75. Parents were told that
risk reductions would be larger for children because the vaccination program produced greater benets if it was initiated earlier in life.
Each parent was asked to read the description of each vaccine and then was
shown the previously marked risk scales for herself or for her child, which now indicated: (1) their revised estimate of heart disease risk, (2) the risk reduction offered
by the vaccine, presented in chances in 100 based on the revised risk estimate, and
(3) the amount of risk that would remain if the vaccine was purchased. The experimentally determined reduction in risk offered by the vaccine is interpreted as its partial effect on risk holding constant all other risk factors and beliefs incorporated into
the revised risk estimate.24 Parents also were shown how the hazard function would
shift to reect lower heart disease risks over time, assuming annual vaccinations.
For the vaccine to reduce the childs risk, parents were asked, Would you be
willing to pay $p to put your child in the heart disease vaccination program for the
rst year? Parents were then told, As you think over your answer, please consider
two things: (1) If your child is in the program, you would have less money available
to pay for other family members to participate and to buy all of the other things that
your family needs, and (2) If your child is in the heart disease vaccination program
for the rst year, you may want to continue in future years to get the full heart
disease prevention benet. Of course, when your child becomes an adult it will be up
to [him/her] to decide whether to participate. Thus, in answering the valuation
question, parents were reminded of the budget constraint and the possibility of participating in the program in future years. The value of $p was randomly chosen from
the ve values $10, $20, $40, $80, $160. These values of p were selected on the
basis of focus group input and pretest results so as to expect the eld study to yield
at least some positive intentions to purchase at each combination of price and risk
change offered.
Parents who indicated that they were willing to pay $p were asked a follow-up
question about the certainty of their intention to purchase: You said that you
would pay $p for your child to be in the heart disease prevention program for the
24. This study focuses on willingness to pay, rather than on offsetting behavior, so total
effects of vaccines on heart disease risk are not measured. Total effects of vaccines would
include the net effects on risk from re-optimizing choices of all goods and time allocations.
For instance, Kaestner, Darden, and Lakdawalla (2014) present evidence that use of a drug
to reduce health risk is associated with changes in risk-related behaviors. Nonetheless, in the
present study, parents may consider alternative choices among other risk-related goods and
time allocations as they decide whether to express an intention to purchase the vaccine.
502
December 2014
rst year. If the program was actually available, how certain are you that you would
really do this? Three answer options allowed respondents to indicate whether they
were uncertain, or would probably or denitely pay. A parallel procedure was used
to elicit purchase intentions for the vaccine to reduce the parents risk. The prices
of the vaccines presented to mothers and fathers for their own vaccine and for their
childs vaccine were the same for each household: prices therefore varied only across
households.
Parents purchase intentions for vaccines to reduce heart disease risks are summarized in table 4. A majority of parents declined to purchase vaccines. Among those
who said they would purchase the vaccine, about 12% indicated on follow-up that
they were uncertain about their purchase intentions. Blumenschein et al. (2008) and
others (Champ et al. 1997; Blumenschein et al. 1998, 2001; Champ and Bishop
2001) have shown that hypothetical bias (misstatement of purchase intentions when
actual payment is not made) can be reduced by treating those that indicate uncertainty about purchase intentions as nonpurchasers. Thus, only parents who said that
they were willing to pay for a vaccine and that said that they probably or denitely would pay for it are treated as stating positive purchase intentions.25
Table 5 presents the proportion of parents who probably or denitely would
purchase a vaccine by price and by size of proportionate risk change. As shown,
parents were more likely to say that they would purchase vaccines that offered larger
risk reductions or that had lower prices. A positive fraction of respondents stated an
intention to purchase both the parent and child vaccine at all combinations of risk
change and price.
2.4. Other Variables: Distribution Factors
In the survey, distribution factors are measured by introducing hypothetical unexpected changes in parents income or expenses. Sensitivity of an expressed intention
to purchase the vaccine to reduced relative contribution to the household budget
was assessed by examining whether the parent would change this decision in the face
of an unexpected personal expense (such as for a medical procedure or for money
lost on an investment for which the parent was personally responsible) coupled with
an unexpected income increase for the spouse. The opposite scenario was used to
assess sensitivity of the decision to decline the purchase of the vaccine: the parent
was asked whether the decision would be reversed if she received an unexpected income increase and her spouse experienced an unexpected expense.26 In both cases,
25. Alternative treatments of parents purchase intentions are examined in the empirical
analysis of section 3.
26. The redistribution employed increases in expenses rather than decreases in income because some parents had no income.
Adamowicz et al.
503
Fathers
Decision For
Self
Child
Self
Child
.57
.43
.50
.50
.49
.51
.50
.50
.14
.51
.35
.14
.52
.34
.08
.55
.37
.13
.52
.35
432
432
This section uses the data described in section 2 to test hypotheses developed in
section 1. Section 3.1 describes results from a nonparametric test for Pareto efciency in allocation of household health resources. Section 3.2 describes methods
applied in a parametric test for Pareto efciency, and section 3.3 describes the test
outcome. Estimates of marginal willingness to pay for absolute rather than proportionate risk reductions are presented in section 3.4, and section 3.5 describes the
27. For example, the changes in expenditure and income were described as follows to parents who initially expressed an intention to purchase the vaccine for the child: Suppose that
you personally had a new expense. For example, suppose that you felt obligated to give nancial help to a relative on your side of the family, or that you had an expensive medical procedure, or that you lost money on an investment that you personally had made. Suppose that
the total cost to you is $X per year, for the next year. The value of $X was randomly assigned as either 2% or 10% of gross household income. At the same time, suppose that your
(spouse/partner) unexpectedly received an additional $Y of income per year for the next
year. The value of $Y was randomly assigned as either 50% smaller or 50% larger than $X.
Parents then were asked, If you had extra expense of $X per year and your spouse (partner)
had extra income of $Y per year, for the next year, would you be willing to pay $p for your
child to enroll in the prevention program for the rst year? A similar procedure was used for
parents who initially declined the vaccine for the child, except that the spouse was assigned
the increase in expense and the responding parent was assigned the increase in income.
504
December 2014
Table 5. Overall Proportion of Parents Who Would Buy Vaccine by Price and Risk
Change (N = 432)
Proportion Who Would Probably or Denitely Buy Vaccine
Risk Reduction (%)
Parent:
10
70
All
Child:
20
80
All
$10
$20
$40
$80
$160
All
.413
.585
.453
.385
.488
.410
.317
.508
.367
.267
.552
.328
.183
.304
.211
.315
.490
.356
.414
.531
.477
.337
.524
.422
.322
.416
.371
.242
.457
.341
.183
.315
.245
.299
.453
.374
outcome of the test for effects of redistribution of relative contribution to the household budget.
3.1. Nonparametric Test for Pareto Efciency
In the eld study, a matched pair of parents who behaved efciently would agree
on whether to purchase the vaccine for the child because both parents received the
same description of the vaccine (same percentage risk reduction and same price) and
this decision is made based on household rather than individual valuations of risk
reductions for the child (see eq. [9]). In a noncooperative household, however, parents make decisions based on individual rather than on household valuations of risk
reductions for the child (see eq. [12]), and one parent may be the sole provider of the
childs vaccine.
These observations motivate the cross-tabulation shown in table 6 of matched
parents stated purchase intentions for the vaccine that reduces heart disease risk for
their child. In 74% of pairs, spouses state the same purchase intentions despite the
mixture of yes/no responses reported in table 5 at each combination of price and risk
change.28 Additionally, a McNemar test (see Agresti 2002) can be used to test the
null hypothesis that, for a given pair of spouses, the probability that the mother will
indicate an intention to purchase the vaccine equates to the probability that the
father will indicate an intention to buy it. The McNemar test statistic, which can be
28. The extent of expected or intended agreement may be higher than actual agreement
reected in table 6. After stating their purchase intentions for the childs vaccine, parents
were asked whether they thought that their spouse would agree with their decision. About
95% of parents expected their spouses to agree.
Adamowicz et al.
505
No
Father
No
Yes
Total
188
58
246
54
132
186
Total
242
190
432
computed using information shown in table 6, is a chi square with one degree of
freedom under the null hypothesis and takes a value of 0.143 ( p-value = .71). Thus,
the null hypothesis is not rejected at conventional signicance levels, an outcome that
is consistent with efciency in the collective and unitary models, but not with the inefcient outcome of the noncooperative model.
3.2. Parametric Tests for Pareto Efciency: Econometric Procedures
Parametric tests for Pareto efciency apply the model developed in section 1 to
econometrically estimate determinants of parents willingness to pay for the hypothetical vaccine for themselves and for their children. Econometric methods applied build
on those developed in Dickie and Gerking (2007). Three key features of this approach involve (1) estimating a willingness to pay function, rather than an explicit
specication of a difference in random utility functions (see Cameron 1988), (2) treating the vaccine as a newly available private good that would provide an increment in
the amount of heart disease risk protection that was previously considered optimal,
and (3) as mentioned previously, interpreting experimentally assigned heart disease
risk reductions as the marginal products of vaccines, holding other risk-reducing activities constant. Randomly assigning both the price and the marginal product of vaccines means that the marginal cost of vaccines also is randomized (see eq. [7]).
The equations to be estimated are derived from the rst-order conditions in equation (8) and are rewritten in more compact notation as:
W i p = i i p;
W ki ti p = ki ki ti p
i = m; f :
13
In equation (13), W i and W ki denote the true (unobserved) annual willingness to pay
of parent i (i = m, f ) for vaccines to reduce heart disease risk, and ti represents the
ith parents Lindahl share of child vaccine expenditures. The variables p, i , and ki
respectively represent the vaccine price and the percentage-point risk reductions that
were randomly assigned to the parent and to the child in the eld study. Price is
measured in dollars per year and the risk change variables were coded 10 and 70 for
506
December 2014
the parent (i ) and 20 and 80 for the child (ki ). The parameters i and ki are interpreted as the annual marginal willingness to pay of parent i for one-percentagepoint reductions in heart disease risk for the parent (i ) and for the child (ki ). Equation (13) allows for differences in marginal willingness to pay by parents in the same
household and ensures that marginal willingness to pay for vaccines is proportional
to marginal willingness to pay for risk reductions (see also Hammitt and Graham
1999). Also, this equation applies to the ith parent in the hth household, but the
household observation index (h) is suppressed to economize on notation.
~ i and
Estimates of i and ki are based on stated willingness to pay for vaccines (W
i
ki
ki
~
W ) rather than on the true values of willingness to pay (W and W ). Stated willingness to pay is latent: parents only were asked whether they would be willing to
pay a randomly assigned price. To account for these features, the discrepancy between true willingness to pay and stated willingness to pay is modeled as a sum of
parent-specic effects as shown in equation (14),
~ i p = i i p i ;
W
~ ki ti p = ki ki ti p ki ;
W
i = m; f ;
14
where the random disturbances i and ki capture effects of household characteristics and unobserved heterogeneity among parents in determining the difference between true and stated willingness to pay for the vaccines. Some parents, in other
words, may be better able than others to state accurately their willingness to pay to
reduce their own risk and their childrens risk of heart disease.
The disturbances i , ki are distributed independently of i , ki , and p, because
the risk reductions and vaccine prices are randomly assigned experimental treatments.
Thus, Ei ji ; ki ; p = i and Eki ji ; ki ; p = ki . The constants i and ki may
not equal zero in part because of nonzero mean differences between stated and actual
willingness to pay for the vaccines. Therefore, let i i = i i and ki ki = ki ki
and re-express equation (14) as
~ i p = i i i p i i ;
W
~ ki ti p = ki ki ki ti p ki ki ; i = m; f ;
W
15
where i and ki are each assumed to be normally distributed with zero means and
unit variances, and 2i and 2ki represent the variance of stated willingness to pay for
the vaccine for the parent and the child, respectively. The possibility that i and ki
are correlated is considered momentarily.
A parent is assumed to state an intention to purchase a vaccine for herself or him~ i p and for the child if W
~ ki ti p. Thus, in the case where the parent
self if W
indicates a willingness to purchase a vaccine,
i i = i i = i i 1= i p;
ki ki = ki ki = ki ki ti = ki p:
16
Adamowicz et al.
507
Equation (16) indicates that there are two sets of two equations to estimate: (1) mother
and child and (2) father and child. These equations are estimated jointly by multivariate probit allowing for correlation of errors across equations.
The econometric design has ve important features. First, joint estimation of equation (16) may improve efciency and takes advantage of the fact that each pair of parents live together in a single household and provide responses on behalf of the same
child. Second, the coefcient of vaccine price in the parent equation (see eq. [16])
equals 1= i . This value can be used to recover the parents marginal willingness to
pay to reduce her or his risk (i ) from the normalized coefcient i = i (see Cameron
and James 1987).29
Third, as noted in section 1, insufcient information is available to estimate the
individualized contributions (Lindahl shares (ti)) of mothers and fathers to reduce
risk for the child. Nonetheless, the coefcient of price in the child equations equals
ti = ki . This coefcient can be used to recover an estimate of household marginal
willingness to pay for a one-percentage-point risk reduction for the child ki =ti from
each parent. The conceptual model in section 1 predicts that km =tm = kf =tf (see
eqs. [10] and [11]), so the inability to identify (ti) poses no difculties for testing the
hypothesis of Pareto efciency.
Fourth, methods applied facilitate testing the validity of equation (11) regarding
efciency in household allocations of health resources. Appropriate econometric tests
can be devised simply by testing coefcient equality constraints within and between
the four equations.30 Also, to test the validity of equation (11), it is unnecessary to
include covariates in these equations to account for other factors that may inuence
the purchase decision (e.g., distribution factors, wages, tastes for exercise or diet, nonlabor income, attitudes/information about heart disease risk, or other behaviors that
affect heart disease risk) because risk changes and vaccine prices are randomly assigned.31
29. If the risk-reducing goods are direct sources of utility, a possibility suggested earlier, the
correct cost value to use in equation (16) would be the randomly assigned cost variable presented to respondents net of monetized utility/disutility p = p . The term 1=
that would be added to equation (16) can be treated as an additional component of the error
already present in this equation. This term will affect the estimate of the constant term if it has
a nonzero mean but will not affect the point estimates of the coefcients of price and parent/
child risk reduction because these variables were randomly assigned. Also, if time spent on risk
reduction is a direct source of utility, the specication of equation (16) is unaffected because
the marginal utility of time spent reducing risk would not enter the rst-order conditions for
vaccines.
30. Bergstrom et al. (1988) present an alternative approach for using stated preference data
to test efciency of local (as opposed to intrahousehold) public goods provision.
31. Neither is it necessary to include variables for the child risk changes in the parent
equation or the parent risk change in the child equation.
508
December 2014
Fifth, the potential for errors in inference arising from parental misstatement of
purchase intentions is minimized in three ways. (1) As discussed previously, parents
expressing an intention to purchase are treated as purchasers only if they also indicated that they would denitely or probably make this decision if the vaccine was
actually available. (2) Inferences focus on marginal willingness to pay for heart disease risk reductions. Prior research indicates that stated and revealed preference methods yield similar estimates of marginal (as opposed to total) willingness to pay, particularly for private goods (e.g., Lusk and Schroeder 2004; Taylor, Morrison, and
Boyle 2010). (3) The econometric treatment of the possible discrepancy between
stated and true marginal willingness to pay implies that any systematic tendency for
respondents to misstate willingness to pay is conned to the constant terms that play
no role in estimating marginal willingness to pay for reduced risk of heart disease or
in testing for Pareto efciency.32 In light of these features and randomization of
prices and risk changes, the parameters i and ki =ti can be consistently estimated
from equation (16) provided that the functional form is correct.
3.3. Parametric Tests for Pareto Efciency
Table 7 presents estimates of normalized coefcients (e.g., i = i ) together with error
correlations in the four intended purchase equations for parent and child vaccines.33
Error correlations are positive and signicantly different from zero at the 1% level.
Estimates also show that parents were more likely to purchase a vaccine when it
produced a larger risk reduction or was offered at a lower price.34
32. Thus, in the specication applied (see eq. [16]), the experimentally assigned risk
changes and prices are assumed to be unrelated to the difference between true and stated willingness to pay. Another perspective on this specication is that it formalizes the Carson and
Groves (2007) theoretical result and empirical ndings in Taylor et al. (2010) and Lusk and
Schroeder (2004) that for private goods, the difference between true and stated responses, or
actual and hypothetical responses, is captured in the intercept of the willingness to pay model
and that marginal willingness to pay estimates are accurate (see Carson and Groves 2011). See
also Vossler, Doyon, and Rondeau (2012) for a discussion of this issue and conditions under
which it appears to hold for public goods.
33. Estimates were computed using the multivariate probit estimator in LIMDEP version 9.0. The software computes multivariate normal probabilities using the GHK (GewekeHajivassiliou-Keane) simulation-based quadrature method and computes derivatives numerically. The optimization method is BFGS (Broyden-Fletcher-Goldfarb-Shanno), a variation
on the Davidon-Fletcher-Powell method. The convergence criterion is 105 for the gradientweighted inverse Hessian. See Greene (2012).
34. Preliminary estimates of the table 7 equations also controlled for the randomized order of presentation of vaccines (child rst or parent rst), but no signicant effects of order
were found.
Adamowicz et al.
509
Father
Child
Equation
Parent
Equation
Child
Equation
Parent
Equation
.3715***
(.1209)
.0077***
(.0016)
.0037***
(.1209)
.3231***
(.1005)
.0100***
(.0019)
.0047***
(.0013)
.2347
(.1206)
.0060***
(.0016)
.0042***
(.0013)
.0585
(.1001)
.0055***
(.0019)
.0052***
(.0012)
1.0
.9193***
(.0212)
.6409***
(.0541)
.6008***
(.0579)
kf
f
kf
...
...
...
...
...
1.0
Log-likelihood
***
.6454***
1.0
(.0546)
.5675***
.9024***
(.0606)
(.0232)
839.4196
...
1.0
Coefcient estimates of the proportionate risk reduction variables are more easily
interpreted by obtaining estimates of unnormalized coefcients (i and ki =ti ). Point
estimates of these coefcients are computed by multiplying the normalized coefcients by estimates of i for parents and ki for children. This calculation suggests
that (1) mothers and fathers are willing to make annual payments of $2.14 (SE =
0.70) and $1.06 (SE = 0.44), respectively, to reduce their own heart disease risk by
age 75 by one percentage point and (2) parents estimates of household annual marginal willingness to pay to reduce their childs risk by one percentage point by age 75
are $2.08 (SE = 0.81) for mothers and $1.42 (SE = 0.56) for fathers.35
The unnormalized estimates can also be used to test the null hypothesis of
Pareto efciency in the collective model. The null hypothesis, which can be stated as
510
December 2014
36. Recall that in the noncooperative model, efcient and inefcient outcomes may be
difcult to distinguish. Thus, the nding of Pareto efciency may not be inconsistent with
implications of this model.
37. The null hypothesis of Pareto efciency is not rejected at conventional levels if (1) a
yes response to the initial vaccine purchase questions, regardless of certainty of intention, is
treated as an intention to purchase, as well as if (2) a positive purchase intention is indicated
only by a yes coupled with a denite response regarding certainty of intention (i.e., if
probably would purchase is treated as no).
38. For married parents who are not matched to a spouse who completed the survey, there
are two equations in the model, describing the parents purchase intentions (1) for the vaccine
that reduces his or her risk and (2) for the vaccine that reduces his or her childs risk. Bivariate
probit is used to estimate the equations, with percentage point risk reduction and price as
covariates (and a constant). The efciency hypothesis that the parent is willing to pay equal
amounts for percentage point risk reductions to herself or himself and to the child is not
rejected at conventional levels for mothers or for fathers. Also, the hypothesis that marginal
willingness to pay values for percentage point risk reduction for parents and children are equal
between mothers and fathers can be tested by comparing mothers and fathers between households, as opposed to the within-household test conducted in the sample of matched parents.
The hypothesis is not rejected at conventional levels. Thus, results in the sample of unmatched
married parents are consistent with those obtained in the sample of matched parents. Finally,
the two samples of married parents were pooled and bivariate probit was used to estimate
purchase intention equations with percentage point risk reduction and price as covariates (and
a constant). An indicator for membership in the matched pairs also was included, along with
interactions of this indicator with price and percentage risk change. At conventional signicance levels, a Wald test fails to detect differences between matched and unmatched parents in
marginal willingness to pay for percentage point risk reduction for parents and children.
Adamowicz et al.
511
Father
Child
Equation
Parent
Equation
Child
Equation
Parent
Equation
.2425**
(.1069)
.0209***
(.0044)
.0037***
(.0012)
.2766***
(.0980)
.0238***
(.0052)
.0043***
(.0013)
.1262
(.1039)
.0182***
(.0040)
.0044***
(.0013)
.0935
(.0977)
.0109**
(.0045)
.0051***
(.0012)
1.0
.9207***
(.0211)
.6419***
(.0535)
.6087***
(.0571)
kf
f
1.0
Log-likelihood
**
***
kf
...
...
...
...
...
.6444***
1.0
(.0550)
.5681***
.9067***
(.0614)
(.0226)
837.6961
...
1.0
512
December 2014
39. Consistent estimation of these coefcients can be demonstrated if (1) the randomly assigned prices and percentage reductions in heart disease risk are distributed independently of
both perceived risk and the disturbance term and (2) the fraction of respondents that are assigned a particular value of percentage risk reduction is xed and therefore independent of the
sample size. Both of these conditions are plausible in light of the way in which data for this
study were collected. In this case, both price and absolute risk change are uncorrelated with
the disturbance term as the sample becomes arbitrarily large, satisfying a key condition for consistency of maximum likelihood estimators.
40. Standard errors are computed using the delta method.
Adamowicz et al.
513
and Alberini and Scasny (2011). However, a joint test of the null hypothesis that
the marginal willingness to pay of mothers and fathers to reduce their own heart
disease risk by 1 chance in 100 equates to their marginal willingness to pay to reduce
heart disease risk to their child by 1 chance in 100 is not rejected at the 5% level of
signicance.
3.5. Effects of Redistributing Income/Expenses within the Household
The multivariate probit regressions in table 9 investigate whether parents marginal
willingness to pay to reduce heart disease risk for the child by 1 chance in 100 are
sensitive to a shift in the mothers relative contribution to the household budget.41
As previously indicated in section 1, the collective model allows for the possibility
that one parents marginal willingness to pay for risk reduction for the child is relatively more sensitive to shifts in income sources within the household, whereas the
unitary model does not. Estimates make use of data from the hypothetical redistribution of expenses/income described in section 2.4. Table 9 presents estimates from
four equations, two each for mothers and fathers. For mothers, for example, the dependent variable in the two equations reects the intended vaccine purchase decision
for the child before and after the hypothetical redistribution of income/expenses. All
equations include covariates measuring vaccine price and absolute risk change. The
after redistribution equations include interactions of absolute reduction in heart disease risk with (1) the percentage change in the mothers contribution to the household budget (coded positively if her income increased and negatively if her contribution to expenses increased) and (2) the percentage change in the overall household
budget (computed as the difference between the percentage change in income experienced by one parent and the percentage change in expense experienced by the other).
Thus, these percentage changes could be either positive or negative.
Estimates of the coefcients of absolute risk change and price variables are little
changed from their counterparts in table 8, and all error correlations are positive and
differ signicantly from zero at 1%. Coefcients of the interaction variables suggest
that for both mothers and fathers, a change in the mothers relative contribution to
the household budget leaves marginal willingness to pay to protect the child from
heart disease risk unaltered. The same outcome holds for a change in the overall household budget. This result supports a central feature of the unitary model that changes
in sources of household income have no effect on parents marginal willingness to pay
to reduce risk for the child. The collective model would predict this outcome if each
parents individual marginal willingness to pay to reduce the childs risk is equally
sensitive to a change in sources of household income.
41. The mothers wage relative to the fathers wage might be a better variable as it would
remove the inuence of hours of work. The eld study, however, did not collect information
on wages or on hours of work and some mothers are not employed outside the home.
***
**
Log-likelihood
kf
km
m
1.0
.8709***
(.0359)
.6290 ***
(.0569)
.5113 ***
(.0665)
861.2621
km
.0036***
(.0013)
...
Vaccine price
...
.1923
(.1164)
.0177 ***
(.0057)
Child Equation
before Redistribution
Constant
Covariate
Mother
.0046***
(.0013)
...
...
.0771
(.1091)
.0160 ***
(.0050)
Child Equation
before Redistribution
...
.5700***
(.0618)
.5131 ***
(.0656)
1.0
.9172***
(.0310)
1.0
...
...
kf
.0013
(.0009)
.0025 **
(.0012)
.0003
(.0005)
.1717
(.1115)
.0161 ***
(.0052)
Child Equation
after Redistribution
Father
1.0
...
...
...
.0011
(.0009)
.0032 **
(.0012)
.0003
(.0006)
.0024
(.1143)
.0138 **
(.0064)
Child Equation
after Redistribution
Table 9. Heart Disease Vaccine Purchase Intentions: Multivariate Probit Estimates, Effects of Redistributing Income/Expenses between Spouses,
Normalized Coefcients, 432 Matched Pairs of Parents
Adamowicz et al.
515
4. CONC L USIONS
This study uses a stated preference survey of matched mothers and fathers from
the United States to distinguish among alternative models of household decision
making for eliciting parents willingness to pay to reduce heart disease risk to themselves and to their children. Key ndings are that (1) the null hypothesis of Pareto
efciency in intrahousehold health resource allocations is not rejected and (2) parents marginal willingness to pay for health risk reductions for their child is not
affected by a redistribution of the household budget. These outcomes support both
the unitary and collective models of household decision making, but not the noncooperative model.
These ndings have important implications for health valuation research. First,
the outcome of Pareto efciency means that parents respond on behalf of the household when providing values of marginal willingness to pay for reductions in health
risk for their children; however, this does not mean that a couple will give the same
values. In a Pareto-efcient household, the marginal willingness to pay values for a
given percentage reduction in health risk for the child is the same for the mother as
for the father, but values provided by a mother and a father for a given absolute risk
reduction (e.g., by 1 chance in 100) may differ unless their initial perceptions of risk
faced by the child are the same. Second, two parents in a Pareto-efcient household
hold the same value of marginal willingness to pay for a given percentage reduction
in illness risk but hold different values to reduce this risk by 1 chance in 100 if they
each believe that they face different initial levels of risk. In the case of heart disease,
for example, results presented here suggest that men correctly believe that their risk
of contracting this disease exceeds that for women.
This study also points to three areas where further research would be worthwhile. First, as just indicated, results presented are consistent with the unitary model
of household decision making. This nding supports the use of prior estimates of
willingness to pay to reduce risks to childrens health in a policy context. Nonetheless, the unitary model has been decisively rejected in nonvaluation settings. As a
consequence, the nding obtained here that redistribution of the household budget
has no effect on household marginal willingness to pay for reducing the childs risk is
worth another look. In a health valuation context, it is crucial to know whether parents marginal valuations of their childrens risk reduction are equally sensitive to a
shift in household resources or more generally to a shift in decision-making power
within the household.
Second, further research might build on the estimation framework developed in
this paper to investigate (1) the role of both parent and child age in determining a
parents marginal willingness to pay to reduce health risk, (2) whether marginal willingness to pay to reduce health risk for sons and daughters differs by parent gender
and marital status, and (3) the role of baseline risk in determining marginal willingness to pay. More generally, additional research would be warranted to improve con-
516
December 2014
REFERENCES
Agresti, Alan 2002. Categorical data analysis. Hoboken, NJ: Wiley-Interscience.
Alberini, Anna, and Milan Scasny. 2011. Context and the VSL: Evidence from a stated preference
study in Italy and the Czech Republic. Environmental and Resource Economics 49:51138.
Bateman, Ian J., and Alistair Munro. 2009. Household versus individual valuation: Whats the difference. Environmental and Resource Economics 43:11935.
Becker, Gary S. 1993. A treatise on the family. Enlarged ed. Cambridge, MA: Harvard University Press.
Bergstrom, Theodore C. 2006. Benet-cost in a benevolent society. American Economic Review 96:33951.
Bergstrom, Theodore C., Judith A. Roberts, Daniel L. Rubenfeld, and Perry Shapiro. 1988. A test for
efciency in the supply of public education. Journal of Public Economics 35:289307.
Blumenschein, Karen, Glenn C. Blomquist, Magnus Johannesson, Nancy Horn, and Patricia R. Freeman. 2008. Eliciting willingness to pay without bias: Evidence from a eld experiment. Economic
Journal 118:11437.
Blumenschein, Karen, Magnus Johannesson, Glenn C. Blomquist, Bengt Liljas, and Richard M. OConor.
1998. Experimental results on expressed certainty and hypothetical bias in contingent valuation.
Southern Economic Journal 65:16977.
Blumenschein, Karen, Magnus Johannesson, Krista K. Yokoyama, and Patricia R. Freeman. 2001. Hypothetical versus real willingness to pay in the health care sector: Results from a eld experiment.
Journal of Health Economics 20:44157.
Blundell, Richard, Pierre-Andr Chiappori, Thierry Magnac, and Costas Meghir. 2007. Collective labour supply: Heterogeneity and non-participation. Review of Economic Studies 74:41745.
Blundell, Richard, Pierre-Andr Chiappori, and Costas Meghir. 2005. Collective labor supply with children. Journal of Political Economy 113:12771306.
Bolin, Kristian, Lena Jacobson, and Bjrn Lindgren. 2001. The household as health producer: When
spouses are Nash-bargainers. Journal of Health Economics 20:34962.
Adamowicz et al.
517
Bonke, Jens, and Martin Browning. 2011. Spending on children: Direct survey evidence. Economic
Journal 121: F123F143.
Brook, Robert D., Barry Franklin, Wayne Cascio, Yuling Hong, George Howard, Michael Lipsett, Russel
Luepken, Murray Mittleman, Jonathan Samet, Sidney C. Smith, and Ira Trager. 2004. Air pollution and
cardiovascular disease: A statement for healthcare professionals from the expert panel on population and
prevention science of the American Heart Association. Circulation 109:265571.
Browning, Martin, and Pierre-Andr Chiappori. 1998. Efcient intra-household allocations: A general
characterization and empirical test. Econometrica 66:124178.
Browning, Martin, Pierre-Andr Chiappori, and Valerie Lechene. 2006. Collective and unitary models:
A clarication. Review of Economics of the Household 4:514.
. 2010. Distributional effects in household models: Separate spheres and income pooling. Economic Journal 120:78699.
Browning, Martin, Pierre-Andr Chiappori, and Yoram Weiss. 2014. Economics of the family. Cambridge: Cambridge University Press.
Cameron, Trudy Ann. 1988. A new paradigm for valuing non-market goods using referendum data:
Maximum likelihood estimation by logistic regression. Journal of Environmental Economics and Management 15:35579.
Cameron, Trudy Ann, and J. R. DeShazo. 2010. Supplementary materials to accompany demand for
health risk reductions. University of Oregon, Department of Economics.
Cameron, Trudy Ann, and Michelle D. James. 1987. Efcient estimation methods for closed-ended
contingent valuation surveys. Review of Economics and Statistics 69:26976.
Carson, Richard T., and Theodore Groves. 2007. Incentive and informational properties of preference
questions. Environmental and Resource Economics 37:181210.
. 2011. Incentive and information properties of preference questions: Commentary and extensions. In International handbook of non-market environmental valuation, ed. Jeff Bennett. Northampton,
MA: Edward Elgar.
Champ, Patricia A., and Richard C. Bishop. 2001. Donation payment mechanisms and contingent
valuation: An empirical study of hypothetical bias. Environmental and Resource Economics 19:383402.
Champ, Patricia A., Richard C. Bishop, Thomas C. Brown, and Daniel W. McCollum. 1997. Using
donation mechanisms to value nonuse benets from public goods. Journal of Environmental Economics
and Management 33:15162.
Cherchye, Laurens, Bram De Rock, and Fredric Vermeulen. 2012. Married with children: A collective
labor supply model with detailed time use and intrahousehold expenditure information. American
Economic Review 102:33773405.
Chiappori, Pierre-Andr. 1997. Introducing household production in collective models of labor supply.
Journal of Political Economy 105:191209.
Chiappori, Pierre-Andr, and Ivar Ekeland. 2006. The micro economics of group behavior: General
characterization. Journal of Economic Theory 130:126.
Clinton, William J. 1997. Executive order 13045protection of children from environmental risks and
safety risks. Federal Register 62:19883.
Cornes, Richard, and Todd Sandler. 1996. The theory of externalities, public goods and club goods.
Cambridge: Cambridge University Press.
Dauphin, Anyck, Abdel-Rahmen El Lahga, Barnard Fortin, and Guy LaCroix. 2011. Are children
decision-makers within the household? Economic Journal 121:871903.
518
December 2014
Dickie, Mark, and Shelby Gerking. 2007. Altruism and environmental risks to health of parents and
their children. Journal of Environmental Economics and Management 53:32341.
. 2011. Morbidity valuation, benet transfer and household behavior. US Environmental Protection Agency, National Center for Environmental Economics, Washington, DC.
Dockins, Chris, Robin R. Jenkins, Nicole Owens, Nathalie B. Simon, and Lenelle B. Wiggins. 2002.
Valuation of childhood risk reduction: The importance of age, risk preferences, and perspective. Risk
Analysis 22:33546.
Donni, Oliver. 2003. Collective household labor supply: Nonparticipation and income taxation. Journal
of Public Economics 87:117998.
. 2009. A simple approach to investigate intra-household allocation of private and public goods.
Review of Economics and Statistics 91:61728.
Duo, Esther. 2000. Child health and household resources in South Africa: Evidence from the old age
pension program. American Economic Review 90:39398.
Ebert, Udo. 2013. The relationship between individual and household measures of WTP and WTA.
Social Choice and Welfare 40:36790.
Evans, Mary F., Christine Poulos, and V. Kerry Smith. 2011. Who counts in evaluating the effects of
air pollution policies on households? Non-market valuation in the presence of dependencies. Journal
of Environmental Economics and Management 62:6579.
Gerking, Shelby, and Mark Dickie. 2013. Valuing reductions in environmental risks to childrens health.
Annual Review of Resource Economics 5:24560.
Greene, William H. 2012. Econometric analysis. Boston: Prentice Hall.
Hammitt, James K., and John D. Graham. 1999. Willingness to pay for health protection: Inadequate
sensitivity to probability? Journal of Risk and Uncertainty 18:3362.
Hammitt, James K., and Kevin Haninger. 2010. Valuing fatal risks to children and adults: Effects of
disease, latency, and risk aversion. Journal of Risk and Uncertainty 40:5783.
Harbaugh, William T. 1999. Valuing childrens health and life: What does economic theory say
about including parental and societal willingness to pay? University of Oregon Working Paper.
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=445968.
Harrington, Winston, and Paul R. Portney. 1987. Valuing the benets of health and safety regulation.
Journal of Urban Economics 22:10112.
Jones-Lee, Michael W. 1991. Altruism and the value of other peoples safety. Journal of Risk and
Uncertainty 4:21319.
Kaestner, Robert, Michael Darden, and Darius Lakdawalla. 2014. Are investments in disease prevention
complements? The case of statins and health behaviors. Journal of Health Economics 36:15163.
Lindhjem, Henrik, and Stle Navrud. 2009. Asking for individual or household willingness to pay for
environmental goods. Environmental and Resource Economics 43:1129.
Lloyd-Jones, Donald M., Martin G. Larson, Alexa Beiser, and David Levy. 1999. Lifetime risk of
developing coronary heart disease. Lancet 353:8992.
Lloyd-Jones, Donald M., Eric P. Leip, Martin G. Larson, Ralph B. dAgostino, Alexa Beiser, Peter
W. F. Wilson, Philip A. Wolf, and David Levy. 2006. Prediction of lifetime risk for cardiovascular
disease by risk factor burden at 50 years of age. Circulation 113:79198.
Lundberg, Shelly A., Robert A. Pollak, and Terence J. Wales. 1997. Do husbands and wives pool their
resources? Evidence from the United Kingdom Child Benet. Journal of Human Resources 32:46380.
Adamowicz et al.
519
Lusk, Jayson L., and Ted C. Schroeder. 2004. Are choice experiments incentive compatible? A test with
quality differentiated beef steaks. American Journal of Agricultural Economics 86:46782.
Strand, Jon. 2005. Individual and household values of mortality reductions with intra-household bargaining. Journal of Risk and Uncertainty 32:21736.
Taylor, Laura O., Mark D. Morrison, and Kevin J. Boyle. 2010. Exchange rules and the incentive
compatibility of choice experiments. Environmental and Resource Economics 47:197220.
US Environmental Protection Agency. 2013. Americas children and the environment. 3rd ed. EPA 240R-13001. Washington, DC: Environmental Protection Agency.
Vossler, Christian A., Maurice Doyon, and Daniel Rondeau. 2012. Truth in consequentiality: Theory
and eld evidence on discrete choice experiments. American Economic Journal: Microeconomics
4:14571.
Yeager, David S., Jon A. Krosnick, LinChiat Chang, Harold S. Javitz, Matthew S. Levendusky, Alberto
Simpser, and Rui Wang. 2011. Comparing the accuracy of RDD telephone surveys and Internet surveys conducted with probability and non-probability samples. Public Opinion Quarterly 75:70947.