You are on page 1of 21

HEALTHECONOMICS, VOL.

5: 447-468 (1996)
ECONOMETRICS AND HEALTH ECONOMICS

WILLINGNESS TO PAY AND COST OF ILLNESS FOR


CHANGES IN HEALTH CAPITAL DEPRECIATION
WALTER RIED
Universio of Mannheim, Germany

SUMMARY
The paper investigates the relationship between the willingness to pay and the cost of illness approach with respect
to the evaluation of economic burden due to adverse health effects. The basic intertemporal framework is provided
by Grossman’s pure investment model, while effects on individual morbidity are taken to be generated by marginal
changes in the rate of health capital depreciation. More specifically, both the simple example of purely temporary
changes and the more general case of persistent variations in health capital depreciation are discussed.
The analysis generates two principal findings. First, for a class of identical individuals cost as measured by the
cost of illness approach is demonstrated to provide a lower bound on the true welfare cost to the individual, i.e. cost
as given by the willingness to pay approach. Moreover, the cost of illness is increasing in the size of the welfare
loss. Second, if one takes into account the possible heterogeneity of individuals, a clear relationship between the
cost values supplied by the two approaches no longer exists. As an example, the impact of variations in either
financial wealth or health capital endowment is discussed. Thus, diversity in individual type turns out to blur the
link between cost of illness and the true economic cost.

KEY WORDS- willingness to pay; cost of illness; Grossman pure investment model; comparative dynamic analysis

Among the major consequences of health pro- As is well-known, the literature up to now has
grams, the impact upon mortality and morbidity focused mainly upon the discussion of two appro-
occupies a prominent position. This statement aches, that is, the willingness to p a y approach on
applies equally to preventive services and to the one hand, and the human capital approach on
curative care. More generally, as research on the other. By working out the effects of a project
outcome measurement in health care has shown, on the present value of an individual’s earnings,
any project which is likely to affect individual the latter offers the advantage of yielding ‘hard
health will do so by way of influencing either data’ which can be obtained at a reasonably low
mortality or morbidity. To be sure, the latter is cost in the majority of cases. Conversely, the
taken to be understood in a wide sense, i.e. it refers willingness to pay approach is grounded firmly in
not only to the consequences of diseases but to any welfare economic theory in that it supplies, at least
change in individual well-being. at the conceptual level, the correct measure of
With respect to the costs and benefits of such either individual benefit or cost.
projects, it has long been recognized, of For the purpose of program evaluation, the
course, that the task of correctly capturing the human capital approach is often embedded in a
corresponding welfare effects of a change in somewhat wider approach which additionally takes
mortality or morbidity may pose specific problems. into account the impact on medical care resource

Address for correspondence: Walter Ried, Department of Economics, A 5 , B 247, University of Mannheirn, D-68131Mannheirn,
Germany.

CCC 1057-9230/96/050447-22 Received I November I995


0 1996 by John Wiley & Sons, Ltd. Accepted 23 April 1996
448 W. RIED
use. The latter represents the cost of illness appro- by the two approaches. Since no uncertainty is
ach as it is commonly used in applied work. In any taken into account, only the effects of changes in
situation where individual willingness to pay can morbidity will be discussed. In my view, however,
be taken to encompass the effect on medical care this restriction is more than offset by the fact that
resource use as well, cost of illness should be the model is rich enough to allow for labour supply
considered as the rival approach in program eval- reactions and for an impact on the intertemporal
uation. Of course, relying on the latter approach evolution of health capital.
will yield valuable information only to the extent As a cursory review of the literature reveals,
that a relation between cost of illness and willing- analysis of the impact of parametric changes in the
ness to pay can be established. While a Grossman model has tended to be less than com-
considerable body of literature investigates the prehensive. In fact, Grossman3 restricted his
possibility of such a relationship, it seems fair to discussion to the effects on gross investment in
conclude that, at present, this issue is still far from health and health capital, respective1 . Basically
being resolved. the same holds for Cropper’s Muurinen6
Consider, for example, two more recent papers rightly called her analysis ‘comparative static’
analysing willingness to pay for changes in since it was based on assumptions which will
environmental conditions which induce changes in generally not be satisfied in a comparative
health status. By relying on an essentially static dynamic analysis.6 In contrast, a more recent paper
model with stochastic mortality and morbidity, by Ehrlich and Chuma’ comes rather close to the
Berger et al.‘ explored the relationship between an approach adopted in this paper. Working with a
individual’s willingness to pay and the correspond- modified version of the full Grossman model in a
ing ‘cost of illness’. The latter concept, however, continuous time setting, they report a considerable
as the authors introduce it, really represents the number of comparative dynamic results. It is not
value given by the cost of illness approach as clear, however, whether their analysis is capable
described above augmented by the value of the of generating the information necessary to derive
change in consumption time. In a similar vein, the cost of illness measure as defined below.
albeit in a somewhat different model involving no The paper is organized as follows. The next
uncertainty, Harrington and Portney * investigated section presents Grossman’s pure investment
the same issue while employing a slightly different model. Particular care is taken to derive a complete
concept of cost of illness. In both cases, the set of conditions characterizing one type of opti-
authors were not able to establish a lower bound mal solution. Then follows an investigation of the
property for the cost of illness approach. comparative dynamic effects due to a parametric
While some of the papers dealing with the change in health capital depreciation. Next, I
subject do conduct an intertemporal analysis, none derive the two measures of interest, i.e. cost
of them incorporates a dynamic view of health according to the willingness to pay and the cost of
capital accumulation. This is surprising since the illness approach. Building upon this information,
application of human capital theory to the field of the issue of the relationship between the two
health economics surely must be counted among approaches is taken up. In particular, the useful-
the major theoretical developments over the last ness of cost of illness estimates for the purpose of
decades. The pioneering work, of course, is due to program evaluation comes under scrutiny. The
Gro~sman,~ who formalized the notion that an final section offers my conclusions as well as
individual, in each period of his life span, inherits several suggestions for future research.
a stock of health capital which is subject to depre-
ciation but can be augmented by means of
investment activities, the latter including medical THE MODEL
care as an input.
It is the main purpose of this paper to study the The purpose of this section is to derive the equa-
relationship between the willingness to pay and the tions governing the individual’s behaviour for the
cost of illness approach by relying on Grossman’s remainder of his lifetime. Given that the underly-
pure investment model. More precisely, I shall ing pure investment model is well the
analyse the effects of a change in the rate of depre- presentation will be rather brief, the principal aim
ciation of health capital due to, say, a variation in being to introduce the notation employed in this
environmental conditions, on the values supplied paper. In contrast, considerable space is devoted to
WILLINGNESS TO PAY AND COST OF ILLNESS 449
both the formulation and the solution of the The function I is linear homogeneous with respect
individual's basic intertemporal decision problem. to factor inputs, i.e. in any period, additional gross
Because a true comparative dynamic analysis investment is available at the constant marginal
requires comprehensive information on the trajec- cost n;.
tories of all endogenous variables, the emphasis Apart from spending healthy time on household
will lie on deriving an implicit representation for production, the individual may also supply labour
each period of the planning horizon. TW,.Relating labour income on the one hand and
Let the intertemporally separable utility function expenditure on both the market good and medical
care on the other, one obtains for the present value
change in the individual's financial wealth
position:

with
V,+I --- v - W , T W , - PYM,- P r X ,
(1 + r)'+'(1 + r)' (1 + r)'
u,= u(Z,) and u(Z,) > 0 2,> 0 (la) (5)
represent individual preferences, where u is a with V , denoting financial wealth, r the constant
strictly concave and twice continuously differenti- rate of interest, W, the wage rate, and P,"(P:) the
able function giving period t utility as generated price of medical care (the market good).
by the commodity Z,. The weights mr are deter- The stock of health capital is taken to provide a
mined by the individual's rate of time preference. healthy time service h ( H , ) which can be put to
Note that the specification of contemporaneous either of the three uses mentioned above:
utility by equation (la) denies any direct utility
effect of health capital while Grossman merely h ( H , )= rW,+ TH, + T, (6)
assumed the latter to provide no such effect at the
margin. Given that this paper investigates marginal with a h / a H , > 0; h, > 0 HH , > H,,,, where h is
parameter changes, this distinction does not mat- strictly concave and Hminmarking the boundary
ter. Thus, while it has no bearing on the results between life and death in the sense that whenever
derived below, equation (la) helps to avoid nota- his health capital is no higher than H,,, the indi-
tional clutter. vidual will be dead.
In a household production setting output Z, is It is important to note that part of the decision
produced by combining a market good, X,, with a problem under consideration consists of choosing
healthy time input, T, according to the optimal length of the planning horizon, i.e. it is
a free terminal time problem. More precisely, the
terminal period is determined endogenously by the
time path of health capital according to
The function Z will be taken to exhibit constant
returns to scale which implies a constant marginal n + 1 = min{t E NI H t s H m i n ) (7)
cost nf in any period.
In any period, the individual inherits a stock ofwhere N denotes the set of natural numbers. Hence,
health capital H , which is subject to depreciation n is the final period of the individual's life.
but can be augmented through gross investment I,: Although Grossman3is somewhat ambiguous on the
terminal condition for health capital, he actually
Hr+1 - H i = ] , - SPi (3) makes use of a slightly stronger condition, as I shall
argue below. Alternatively, one could suppose death
where 6, denotes the rate of depreciation and to take place once the stock of health capital falls
satisfies Oe 6,e 1. Gross investment in health strictly below the critical level Hmin.l0
represents the second production activity the indi- Now it is possible to formulate the individual's
vidual may engage in using a medical care input M, intertemporal decision problem using the theory of
and a healthy time input TH, according to discrete optimal control."*'2 Writing vt as a short-
hand for the vector (Z,, X,, T,,I,,M,,TH,, W,)of
control variables, the free terminal time problem
450 w.R1[ED
(P) is given as follows: existence of an optimal solution to (P), I shall
n assume the trajectories of exogenous variables to
(P) Choose n optimally and max m ,~ ( 2 , ) be such that for 1 s t b n health capital remains
up ,=, bounded away from H,, on an optimal path.
It is important to realize that, without further
assumptions, a great variety of optimal solutions
S.t. H,+ 1 - H I = I, - 6, H I ; may occur depending on the trajectories of all
Vr+I v, exogenous variables included in (P). In particular,
the possibility of comer solutions cannot be ruled
(1 + r),+' (1 + r), out for any of the control variables. On the other
- W,TW, - P:M, - P:X, hand, it would be rather convenient for compara-
- , tive dynamic analysis to be able to deal with just
(1 + r)' one type of optimal solution. Therefore, let me
Z ( X , , T I )- Z , = 0 ; I ( M , , TH,) - I , = 0; assume the optimal solution to the free terminal
time problem (P) not to involve comer solutions
h(H,) - TW, - TH, - T , = 0 ; U , 3 0 , for any of the control variables except possibly for
O s t s n ; H , > H ~ ,l s r s n ; gross investment in health in the final period. In
fact, this is in line with the literature exploring the
H , > H,, given; V, > 0 given; impact of parametric changes in the Grossman
model, a possible exce tion being the recent paper
H , , , s HIN.4 V n + l3 0 by Ehrlich and Chuma. P
Both the stock of health capital H , and the present The Lagrangean associated with problem (P)
value of financial wealth V,/ (1 + r ) represent state then reads
variables. Any sequence of the control and the
state variables satisfying all the restrictions
included in (P) will be called a feasible solution.
Thus, the *hdividual nee$ t? find feasible trajec-
tories { v, ] :-,
and { h', , V, :::1 such that the
=:,
associated sequence ( Z , 1 maximizes intertem-
poral utility. For the sake of notational conveni-
ence, in what follows optimal solutions will no
longer be marked by an asterisk. This should not
cause any confusion since the discussion below
will focus on optimal solutions.
The problem (P) contains both pure state and
pure control constraints as well as a mixed con-
straint involving health capital and the uses of
healthy time.13 It is worthwhile to examine the where ( y f , y;, B,, a:,a:, a,'") is a vector of non-
pure state constraint on health capital more closely. negative Lagrangean multipliers. An optimal
In any period t with 1 d f s n, this constraint solution satisfies the following necessary
defines an open set of admissible values for H,. It conditions: 1 1 * 1 2 * 1 4
is not difficult to see that this may cause optimal
solutions to (P) not to exist. This would be the
case, for example, if in some period optimality
were to require the individual to set health capital
arbitrarily close to H-. While this existence
problem can be circumvented in discrete time
models by resorting to the other terminal condition
on health capital mentioned above (which gives
rise to another difficulty, however), it is invariably " V " W aL,
present in any continuous time formulation. Yet as
far as I am aware, this has not received attention in
In order to guarantee the
the literature to date.3*4*6*7
WILLINGNESS TO PAY AND COST OF ILLNESS 45 I
as well as equations (2), (3), (4), (5) and (6). The time problem,13 the discrete time version fails to
transversality conditions are provide such an eq~ati0n.l~ Rather, the optimal
final period, i.e. n has to be determined through the
A:+lvn+I= 0; A+:, 30 ( 1 W analysis of a sequence of fixed terminal time
problems with terminal time varying over a plaus-
A,Hl(Hmin-H , + , ) = O ; A,H,,<O (12b) ible domain.
It is important to realise that the comparative
A solution satisfying these necessary conditions dynamic analysis undertaken below in no way
must be optimal. This holds because the Hamil- depends on exact knowledge of the optimal length
tonian associated with problem (P) is concave with of the planning horizon. Instead, it requires n to be
respect to the set of control and state variables for well-defined, i.e. finite, in addition to the unique-
every t. Therefore, the necessary conditions for an ness assumption introduced above. As the analysis
optimal solution are also sufficient.l2 will reveal, optimal health capital critically
Moving on to transversality conditions, consider depends on the marginal cost of gross investment
financial wealth first. From the necessary condition and the rate of depreciation. Thus, the requirement
(ll), the associated costate variable will be con- that n be finite places restrictions on the trajec-
stant over time. Assuming the individual to be tories of these exogenous variables. I shall follow
non-satiated with respect to the commodity, an the literature and take these restrictions to be
increase in initial financial wealth then ceteris satisfied.
paribus implies a higher level of intertemporal Now return to the set of necessary conditions.
utility. Hence, the costate must be strictly positive. After some algebraic manipulation, one obtains for
This implies, in turn, a terminal financial wealth any period t satisfying 0 s f s n - 2,
equal to zero, i.e. one has

Iy=Iz,v>o vt; vn+I=o (13)


With respect to the stock of health capital, proposi-
tion 1 in the Appendix states that the costate I
Jc, -- w,=o
relating to terminal health capital must be equal to aTH,
zero. Furthermore, I shall assume the optimal final
period to be uniquely determined. This implies that
sufficiently small perturbations of the trajectories
of exogenous variables will not alter the length of
the individual's planning horizon. In particular, as
the proof of proposition 2 in the Appendix demon-
strates, terminal health capital must be strictly
lower than Hmin:

Although Grossman3 in his seminal work at times


seems to imply that it should equal H - , it is clear
from his discussion of the impact of a marginal h ( H , ) = Tw,+ TH, + T, (22)
increase in gross investment that he, in fact, relies
on terminal health capital being strictly below endogenous variables:
Hmin. Note that a condition similar to equation (14) I,, Z,. M,,TH,, X,,T,, TW,, 1:
will generally fail to make sense in the continuous exogenous variables:
time case.
(HI),m,,r, ~ctz,4,~c:+~,W , W+l.d,, d,+l
Finally, since (P) is a free terminal time
problem, one may suspect that a condition for the At first sight, it may seem odd to include H , in
optimal length of the planning horizon is missing the set of exogenous variables since the stock of
in the set of necessary conditions considered health capital, as a state variable, cannot represent
above. However, unlike the analogous continuous an exogenous variable proper. Precisely for this
452 W. FUED

reason I have put H, in parentheses. Nevertheless, h ( H , ) = 7Wn+ T, (27)


when solving the optimal program for period t , the
individual inherits a stock of health capital which I , = M ,= T H , = 0 (28)
has already been determined in an optimal manner.
The only exception to this rule is, of course, the endogenous variables: H , , , , Z,, X,,T,, W,,Ax
initial period since the initial stock is truly
exogenous. exogenous variables: ( H , ) , m,,r. n:, W,, 6,
As for the structure of the system given by Whereas equation (25) determines terminal
equations (15) to (22), equation (18) determines health capital, equation (23) gives the optimal
H,+,the optimal stock of health capital for the next output of the commodity. Equations (24) and (26)
period since it contains no other endogenous provide the corresponding factor inputs, while
variable. Inserting the optimal value for H I + ,into inspection of equation (27) yields the optimal labour
equation (19) yields the optimal value for gross supply decision. Again, production of the commod-
investment I,. Next, equations (16) and (21) ity and labour supply are, at this stage, conditional
supply the solutions for the corresponding factor upon the value of the costate variable A:.
inputs of medical care M ,and healthy time TH,. It remains to indicate the determination of A,".
Equation (15) provides the optimal output of the Summing up the budget constraints given by
commodity Z , conditional upon the value of the equation (3,one obtains the intertemporal budget
costate variable A:. Using this information, equa- constraint
tions (17) and (20) determine the associated factor
inputs of the market good X , and healthy time T,, W,TW,- P y M , - P:X,
both conditional upon A: as well. Finally, equation = O (29)
(22) provides the optimal labour supply decision r=o (1 + r)'
as a function of A:, too. Postponing the case of A:
for a moment, this completes the description of the Inserting the optimal solutions for medical care M,
solution for a typical time period. as well as for the factor demand X,, and labour
With respect to period n - 1, the optimal sol- supply TW, described above, the latter two being
ution is again characterized by the system of conditional upon the costate variable A;, yields an
equations given by (15)-(22), with the exception equation in A: alone. Given that equation (29)
of equation (18), which needs to be modified to admits a solution, this is also the only one since, as
the analysis undertaken in the following section
will reveal, the left hand side represents an increas-
ing function of A:. Thus, equation (29) uniquely
determines the optimal value of the costate vari-
able A:. Relying on this information, it is possible
because the costate A,"+,equals zero. Hence, the to derive, at least in principle, the unconditional
rate 6, fails to influence health capital H,. This solutions for both production of the commodity
holds because a comer solution for final period and labour supply.
gross investment in health obtains. More precisely,
A,",,= 0 can be shown to imply a,' > 0.
Accordingly, the final period solution is charac- CHANGES IN THE RATE OF HEALTH
terized by CAPITAL DEPRECIATION

m,---
au z
n, =0 (23)
This section investigates the effects of changes in
az, (1 + r ) n health capital depreciation on the time paths of the
endogenous variables. In general, any determinant
of health capital depreciation could bring about such
a change. Since my prime intention is to analyse the
effects which ensue, there is no need to specify a
particular cause. There are a number of factors
which may reasonably be hypothesized to influence
the depreciation of health capital. Individual
lifestyle and environmental conditions should be
WILLINGNESS TO PAY AND COST OF ILLNESS 453
included among the more important determinant^.^^^ time path of health capital depreciation admissible
Although, in principle, any parametric influence on by equation (31), I shall at times focus on two
health capital could be modelled as working through rather special cases for the purpose of iIlustration.
either the gross investment function or the rate of Most simply, a purely temporary change prevails
depreciation, it seems more appropriate, following if, following the initial change in period k, no
Cropper, to focus on the latter. other rate varies subsequently, i.e. one has
Owing to space limitations, this section essen-
tially reports the results of the analysis of the
necessary conditions associated with (P). Readers
interested in a detailed derivation are advised to
consult a companion paper.g
For the endogenous variables other than A,: it is Apart from that, the case of equiproportional
useful to distinguish between two types of effects changes is also of interest. Indeed, this is the case
following a change in health capital depreciation. Grossman focused upon3 and it is characterized by
First, there may be a direct efect which operates
holding A: constant. In addition, there may also be
an indirect efect working through A:, if the
change in the rate of depreciation induces a change
in the value of this costate variable. The total efecr
of a change in the rate of depreciation in period j Consider first a change taking effect in the last
on an endogenous variable Y in period k other than period, i.e. k = n. As can be shown by analysing
A: is equal to the simple sum of both effects: the system of equations (23)-(28), the only
variable to be affected is terminal health capital
H, More precisely, a marginal rise in the rate 6,
+

will induce a marginal reduction in the stock H,+l.


All other endogenous variables remain as before.
In particular, there is no effect on the intertem oral
From the previous section, it is clear that there can 9
budget constraint (29), which implies that 1, will
be no indirect effect on either health capital or not be affected either.
gross investment since they are not influenced by Now suppose the change in health capital depre-
1:. Hence, for these four variables, the direct ciation to start off in any other period. Given that
effect of a change in the time path of the rate 6, the marginal alteration in the time path of d,, may
must be equal to the total effect. affect a possibly large number of rates, it makes
In the analysis to follow, I shall consider margi- sense to discuss the case of a purely temporary
nal changes in the rate of health capital change first. Since, in effect, any change which is
depreciation which may persist from some period, admissible by equation (31) represents a sequence
say k, onwards. The basic idea is that changes of purely temporary changes, this prepares the
occurring after period k, in which the change has ground for an investigation of more general
taken effect, are functionally related to the margi- variations,
nal change in 6,. More specifically, I shall take a In order to fix ideas, assume the purely tem-
change in health capital depreciation starting in porary change to take place in any period k with
some period k with 0 C k C n to be characterized by 1 s k g n - 1. From the system of equations
a set of derivatives as follows: (15)-(22), it is clear that the rate 6, affects the
endogenous variables directly both in period k and
in the preceding period. Consider the period k
effects first. Relying on equation (18) and on the
fact that there is no indirect effect, the total effect
on health capital is equal to
Thus, the analysis applies to any change in the time
path of 6, provided that the changes in periods t > k
are related to the initial change in period k by a
differentiable function and this function be known.
Out of the myriad of possible variations in the
454 W.NED
A rise in the rate of depreciation raises the cost of according to
acquiring new health capital. Hence, it is optimal
to choose a lower stock H,. Since the remaining
time path of health capital is not affected by the
change in dk, the individual needs to incur a higher (1 + r ) ' m , -
gross investment according to equation (19). a(zt)2
Combining this with the impact on the correspond-
ing factor inputs, one obtains
-2x1
- - z :2 c0 (35b)
an: z:,z:-z:,z:an;
-a-T, _ - Z azr c 0
-
1: (3%)
an: z:,z: -z:~z: an;
where Zl(Z:,) denotes the period t first (second)
partial derivative of the function Z with respect
to i ( i and j ) . Note that, at this stage, the
indirect effects cannot be determined in full
since there is no information on the impact of
where I,' ( I ; ) denotes the period k first (second) the rate dk on the costate A.: Finally, making
partial derivative of the function I with respect to i use of the healthy time budget constraint (6).
(i and j ) . The sign of equation (33a) must be the above results imply for the effects on labour
positive because the flow equation (3) implies supply
health capital to be non-negative. Otherwise, ie.
with a negative stock, the rate of depreciation ah dHk dTHk
aTwk =--.--- < 0 (36a)
would act to increase rather than reduce health
capital ceteris paribus. aHk ddk dd,
Proceeding in a similar manner, one obtains for
the total effects on gross investment in the previous
period, taking into account that health capital
remains as before,

(344
It remains to indicate the effect of the rate dk on
dMk- the value of the costate variable .A: Define net
-- 1
- dlk-l c 0 (34b) expenditures in period t as the excess of expendi-
1"-1 k-1-lk-1 k-I
ddk 22 II 21 12 ddk ture on medical care and the market good over
labour income. Accordingly, the present value of
net expenditures PVNE is given by
"
This completes the description of the impact of
changes in health capital depreciation on either
PVNE = 1P t X , + P Y +Mr)', - W ,TW,
t=O (1
(37)
health capital or gross investment since, in any
other period, the rate dk does not enter the determi- Employing the results derived above and mak-
nation of those variables. ing use of the first order optimality conditions, one
Now turn to the effects on the production of the can show for the direct effect of a change in the
commodity. Inspection of equations (15), (17). rate of health capital depreciation
and (20) reveals that all direct effects are equal to
zero. Owing to the influence of the costate A ,:
however, changes in health capital deprecia-
tion lead to indirect effects in every period
WILLINGNESS TO PAY AND COST OF ILLNESS 455
On the other hand, the impact of the costate is What happens if there is a change in the initial
unambiguously negative since a rise in A," in any rate do? In fact, the equations given above already
period acts to lower expenditure on the market cover this case if one bears in mind two additional
good while increasing labour supply: considerations relating to the evaluation of direct
effects. First, these are restricted to the initial
2x1
p : -- aTw' period since, for obvious reasons, there can be no
W' y impact on previous decisions. Second, owing to its
an;
--
apvNE
an; - 2
I=o (1 + r)'
exogeneity, initial health capital will not be
ano c 0 (39) affected, ie. dHo/dd0=O must hold. Then, direct
effects are obtained by specializing equation (33)
Clearly, the total effect on the present value of net and (36a) to the case k = 0 . Analogously, the
expenditure must be equal to zero. Combining associated indirect effects are covered by equations
equations (38) and (39), a purely temporary rise in (35), (36c) and (40). This is true because both
the rate of health capital depreciation implies a rise equations (38) and (40) can be shown to hold for
in n,V as well: changes in the initial rate do as well.
Following a marginal rise in do, the individual
reacts by undertaking more gross investment in the
dn," an," initial period in order to be able to maintain fully
-- -- > O ; 1 ~ k ~ n - (40) 1
d s k as, the stock of health capital over the planning hor-
izon. At the same time, consumption of the
This completes the description of effects. Fol- commodity will decrease in every period while
lowing a temporary rise in health capital labour supply will go up in any period other than
depreciation, the individual reacts by adjusting the first. Owing to a negative direct effect, the
both the output of the commodity and the corre- reaction of initial labour supply remains
sponding factor inputs downwards. Furthermore, ambiguous.
owing to a generally positive indirect effect, his The results for purely temporary changes in the
labour supply will increase in any period other time path of health capital depreciation generalize
than k. In period k, the sign of the total effect on in a straightforward manner to the case of other
labour supply remains ambiguous because of the variations compatible with equation (3 1). Consider
negative direct effect (Fig. 1). first direct effects. If there is a period prior to the

H,
H
G G
Y
M
m m

+I
ao"

0 5 t c k-1

Figure 1. Direct and indirect effects of a purely temporary change in health capital depreciation in period k
( k € (1, ...,n-1)).
456 W. RIED

change (i.e. k > l), the corresponding direct effects 6,. In the discussion of economic cost, this
can be read off from equations (34) and (36b). In assumption implies
any other period directly affected by the change in
the trajectory of b,, direct effects are determined

because a rise in the value of the costate A," lowers


- 1 (41) consumption of the commodity in every period
and, thus, imposes a true burden OR the individual.
As an example, consider the case of equipropor-
tional changes given by equation (31b), which
clearly induces a positive effect of the rate 6, on
n:.
dTW,
dsn 10"
_I
aTw,
as,
as, arw, as,,,
-
ask+KIA: ask
Note that assumption (43) involves no real loss of
generality. It excludes, in effect, any variation in
health capital depreciation beginning with a margi-
nal decrease which is eventually foilowed by
increases sufficiently high to induce a rise in the
costate A,". By appropriately modifying the
definitions of marginal cost introduced below, it is
where equations (42a) and (42b) hold for not difficult to show that the results of the analysis
0 s k S t d n - 1. Plugging these effects into the apply to such cases as well. This claim will be
intertemporal budget constraint and relying again substantiated below.
on first-order optimality conditions, one can show
that the overall direct impact on the present value
of net expenditures is given by TWO CONCEPTS OF COST
dPVNE Consider first the concept of cost according to the
dsk willingness to pay approach. As a measure of
welfare cost, I shall rely on a concept which would
(384 be equal to the negative of the compensating
variation if applied to discrete changes in health
Note thal the summation excludes the final capital depreciation. More precisely, I shall
period because the rate 6, fails to affect net investigate the minimum marginal amount of
expenditures. money the individual would be prepared to accept
Now turn to the evaluation of indirect effects. in order to prefer (weakly) the situation prevailing
In fact, equations (35) and (36c) apply again. The after the change. In her analysis of the impact of
only missing piece of information concerns the marginal changes in pollution, Cropper' employed
total impact of the change in the time path of 6, a closely related measure.
on the costate A:. While, as equation (40) shows, In fact, this measure of cost has already been
for purely temporary changes there is no ambigu- calculated in the previous section. The crucial obser-
ity, this does not translate to the case of more vation is that parametric variations in the rate of
general variations. As equation (38a) makes health capital depreciation complying with equation
clear, an initial rise in the rate of depreciation (43) will affect individual welfare only through their
may go along with a negative overall direct effect impact on the costate A:. This is true because inter-
on the present value of net expenditures, if it is temporal utility depends exclusively on the trajectory
followed by sufficiently high reductions in subse- of 2, which, in turn, is not directly affected by the
quent rates. For the sake of concreteness, time path of 6,. Therefore, if one were to enhance the
therefore, I shall focus on a subset of all changes individual's financial wealth so as to exactly offset the
in health capital depreciation admissible by welfare cost of the change in health capital deprecia-
equation (31) in order to avoid further notational tion, ttus would obviate the need for a correction of
complexity. More specifically, any such change A," and, hence, leave intertemporal utiLity constant.
will be taken to involve a marginal rise in the rate This motivates the following definition of cost
WILLINGNESS TO PAY AND COST OF ILLNESS 457
according to the willingness to pay approach: impact of a parametric change in health capital
depreciation, it is necessary to compute the result-
ing change in the costs of illness. This leads to the
following definition of the cost of illness value
associated with adverse changes in health capital
depreciation:
cCO/(dki *.*) dn-l)
Note that CWp will be positive if and only if a
change in the time path of the rate of health
capital depreciation decreases intertemporal
utility, i.e. imposes a burden on the individual.
Furthermore, Cwp exhibits the ordering property,
that is, for any two changes in health capital (45)
depreciation it assigns a higher value to the one
involving a bigger loss of individual welfare. In While changes in the demand for medical care
this sense, cost according to the willingness to enter C,, with a positive sign, changes in labour
pay approach measures the true welfare cost to supply cany a negative sign. This reflects the
the individual. principal difference between direct and indirect
The basic idea of the cost of illness approach is costs, i.e. whereas the former refer to increases in
to describe the impact of the workings of disease resource use, the latter relate to decreases in
on both resource use and the availability of resource availability. Note that all items included
resources. It is customary to denote any increase in Cco, are, at least in principle, observable since
in the use of resources as direct costs, while they represent total effects. In contrast, this is not
reductions in resource availability count as true for the cost according to the willingness to
indirect costs. Although the cost of illness appro- pay approach since Cwp only accounts for the
ach has been around for well over thirty years, it direct effect on labour supply which cannot be
seems fair to say that still no consensus on observed.
methodology has been reached as yet. 16.” Some
papers adopt a broad view of the range of costs to
be included in a cost of illness study. Thus, direct A LOWER BOUND PROPERTY
costs are taken to encompass not only health care
resource use (e.g. the cost of treatment), but also Consider now the relationship between C, and
non-medical costs such as the patient’s time input. C,,. Making use of the definition (37) of the
Similarly, the category of indirect costs includes present value of net expenditures, inspection of the
the value of leisure foregone in addition to earn- representations given by equations (44) and (45)
ings lost.I6*l8 in the previous section yields
In empirical studies, however, the dominant
approach has been to confine the analysis to an cCOI(dk, *.*, dn-1) = CWTP(dk, dn-1)

assessment of medical care cost and the value of


labour foregone due to disease,” the reason being
that access to data on these two cost items is far
easier to obtain than on the other items. On the
other hand, a narrow view of the costs of illness
has also been adopted at the theoretical level, as These equations provide a simple characterization
the example of a recent survey demonstrates.20 of the cost of illness in terms of the welfare cost as
Since my intention is to assess the validity of the measured by individual willingness to pay: cost of
approach as it is used in empirical studies, I shall illness equals the true welfare cost minus the
also take a narrow view of the costs of illness. present value of the value of indirect effects on
With respect to evaluation, the basic idea is to labour supply.
assess the effect on the cost of illness due to a By virtue of assumption (43), there is a positive
particular health programme or, for that matter, impact of variations in health capital depreciation
any other cause. Thus, in order to evaluate the beginning with a change in the rate dk on the costate
458 W.NED
variable I" Given that the effect of the costate critically depends upon assumption (43) since it
variable I,P; on labour supply is unambiguously has been used in their derivation. However, this is
positive as well in any period, a lower bound prop- not the case. To see why, consider a change in
erty can now be established for the cost of illness health capital depreciation not complying with
values. More precisely, whenever the change in equation (43) but nevertheless imposing a welfare
health capital depreciation acts to decrease intertem- cost on the individual. Clearly, such a change in
poral utility, i.e. there is a real cost to the individual, the time path of 6, must involve a marginal
the cost of illness approach can be shown strictly to decrease in the rate 6, while still forcing the
underestimate the true welfare cost. costate 1,"to rise. Thus,
Despite their discrepancy in general, both C,,,
and Cwp will always be of the same sign. This
follows from the impact of a change in health -d<I o (43")
capital depreciation on the intertemporal budget d6k
constraint (29) which is given by
must hold. Referring to the concepts of Cwp and
C,,, introduced in equations (44) and (49,
w--aTW, dLi respectively, it is not difficult to see that these have
been defined with respect to marginal rises in the
rate 6,. In order to be able to capture the cost due
to adverse changes involving an initial decrease in
the rate of health capital deprecktion, it is neces-
sary to work with Cwp and C-, instead, which
are defined as follows:

This equation holds because, after all effects have


been taken into account, the intertemporal budget
constraint necessarily must be satisfied again. Repeating the analysis carried out above, one
Rearranging terms and using equation (46), one arrives at exactly the same results. This proves
obtains their general validity, i.e. they hold for any adverse
change in health capital depreciation compatible
with equation (31).

HETEROGENEITY IN INDIVIDUAL TYPE

(48) A simple comparison of cost values supplied by


the willingness to pay and the cost of illness
Since the indirect effect on the demand for the approach represents but a first step on the way
market good is unambiguously negative in any towards a full understanding of the relationship
period, the cost of illness value of a change in between these two concepts. This holds because
health capital depreciation imposing a burden on the comparison as undertaken in the previous
the individual is always positive. section relates to a single individual. Hence,
Thus, for any change in health capital depre- it fails to provide information on how the discre-
ciation involving a lower level of welfare for the pancy between Cwp and CColvalues will vary
individual, cost as measured by the cost of among non-identical individuals. Since
illness approach provides a strict lower bound on individuals may safely be taken to differ with
the true economic cost. Moreover, the approach respect to possibly a host of characteristics,
correctly detects the presence of a cost, i.e. however, it is of paramount importance to know
whenever there is a welfare loss to the individ- about the impact of such differences. The present
ual, the corresponding cost of illness value will section addresses this issue by analysing the
be positive. influence of the financial wealth endowment and
It may appear as if the validity of these results initial health capital.
WILLINGNESS TO PAY AND COST OF ILLNESS 459
Consider first the example of the financial equation (46) with respect to V,. This yields
wealth endowment V,. Apart from the intertem-
poral budget constraint, this variable does not aTW, dn,"
appear in any equation of the model. Therefore,
w, - -
there are no direct effects on the endogenous
variables except for, of course, the costate 2;. - CCOI(.>]
~ [ ~ w T P ( . )
-
This implies that individuals differing only with
respect to Vo will build up the same stock of avo a VO
health capital and also take identical investment
decisions. As for the indirect effects relating to
a2Tw, dn," an,"
financial wealth, it is easy to see that A; must wt----- +w,-- aTw'
react inversely. This holds because a higher level
of financial wealth necessitates a higher present
value of net expenditures for the intertemporal
=T1-0
~ I ( A : ) ~ds, avo an; avo
(1 + r)'
budget constraint to be satisfied. With the other (50)
. .
optikality conditions as before, the only way to One would like to know, in particular, whether the
achieve this is through a lower value Of the last sum carries a definite sign. In order to check
costate variable A;. Thus, one has this, it remains to evaluate the second-order
an ;
-<o
derivatives. Consider the example of the second
partial of labour supply with respect to the costate
(49)
a vo A;. Employing equation (36c) yields
This completes the description of the influence of
V, on individual behaviour, which is also illus-
a 2 TW,
-- - z ;1 a2z,
trated by Fig. 2. Ceteris paribus, individuals a(n;)' z;,z: -zi12 ; a(lz;)z
endowed with a higher financial wealth will work
less and consume more of the commodity.
In order to see how the financial wealth endow-
a(
ZZ: - G 2: az,
) (51)
ment affects the relationship between Cwp and +
C , it is necessary to evaluate the derivative of an ; an 0"
Ho
b
Mo

nl,

20

xo
To
TWO

n2tT I

Figure 2. Effects of a variation in the financial wealth endowment.


460 W. RIED

The sign of the first term of the sum on the capital. As Fig. 3 illustrates, a variation in H ,
right-hand side depends on preferences in a initially induces a direct effect on gross investment
straightforward manner. It can be shown to equal and on labour supply but on no other endogenous
the sign of the third-order derivative of contem- variable apart from 2;. More precisely, an individ-
poranous utility with respect to 2,. On the other ual with a higher stock of initial health capital can
hand, without resorting to ad hoc assumptions, it is afford to undertake less investment in order to
impossible to obtain a definite sign for the second reach the optimal stock of period 1. In turn, this
term on the right-hand side of equation (51). implies the corresponding direct effect on initial
Hence, one is left with the negative result that the labour supply to be positive.
second-order derivative of the labour supply Along with these direct effects, a rise in initial
function with respect to Ax cannot be given a health capital must also produce indirect effects
definite sign. because its overall direct effect on the intertem-
By relying on the intertemporal budget con- poral budget constraint is to decrease the present
straint, it is possible to demonstrate a similarly value of net expenditures. Thus, one has
negative result for the other second-order deriva-
-an< ;
tive contained in equation (50) because its sign
depends on the second partials of labour supply o
and the demand for the market good with respect aH,
to the costate A:. Therefore, the impact of the
financial wealth endowment on the discrepancy Ceteris paribus, individuals enjoying a higher
between C , and C,,, remains unclear. initial health capital position will consume more of
Let me now turn to the impact of initial health the commodity and work less, the latter statement
capital. Consider first the influence upon individual holding from t = 1 onwards while the total effect
behaviour. Given that optimal health in any period on initial labour supply remains ambiguous.
depends only on a group of exogenous parameters Proceeding exactly as before in the case of
not including H,, individuals with different health financial wealth, the influence of the health capital
endowments will choose identical paths for health endowment on the discrepancy between the cost

n t t t 2

Figure 3. Effects of a variation in initial health capital.


WILLINGNESS TO PAY AND COST OF ILLNESS 46 1
values supplied by the two approaches under follows, I shall discuss this issue for the cases of
consideration is given by: identical and heterogenous individuals.
Consider two identical individuals, A and B,
and two programs, PI and P2. For a given marginal
benefit D, program P, entails an adverse change in
aH, health capital depreciation for A. This could be
achieved by, e.g., marginally freeing resources
devoted to pollution abatement in the vicinity of
individual A. Likewise, awarding the same margi-
n nal benefit to society, program P2 imposes an
=TL
unfavourable change in B's rate of health capital
r=o (1 + r)' depreciation. For the sake of notational simplicity,
(53) take C,(P,) and Cc,,(Pi) to denote the welfare
cost and cost of illness values, respectively, of
One may suspect that equation (53) contains less program P,.
than the full effect since initial labour supply also Now suppose the correct cost of illness values
depends directly on the health endowment. It can associated with programs PI and P2 to be at hand.
be shown, however, that the partial with respect to Furthermore, assume that program evaluation is
1; exhibits no such dependence. Comparing carried out according to a conventional cost-benifit
equation (53) with equation (50), the similarity criterion; i.e. a program is judged by the net benefit
should be obvious. The impact of initial health it provides. In this case, the cost of illness appro-
capital on the discrepancy between C , and C,, ach supplies useful information on the relative
values critically hinges upon the signs of the same desirability of the two programs under considera-
second-order derivatives. Thus, the negative tion. This holds because, as the size of the welfare
results obtained for the financial wealth endow- loss increases, the associated cost of illness value
ment fully apply to this case as well. Lacking goes up as well. More precisely, making use of
precise information on either individual prefer- equations (46) and (47), it is possible to establish
ences or household production technologies, and the following relation:
abstaining from the use of very specific ad hoc
assumptions, there is no way of knowing how the C,0/(6,, 6n-1)
* a * *

initial stock of health capital affects the discre- = b C W T P ( 6 k , - . . , d n -O


l )<; b < l (54)
pancy between the true welfare cost and its cost of
illness approximation. arw,
+ w, an;
L
,=O
-

(1 +r)'
ON THE USEFULNESS OF COST OF ILLNESS b= 1-
VALUES FOR PROGRAM EVALUATION ax,
aTW, - P : -
w, -
As explained above, a prime motivation for =?- an; an ;
L
employing cost of illness estimates relates to the 1=0 (1 + r)'
fact that these values are accessible for the
researcher. How useful, then, is the attempt to Thus, if in our example the cost of illness value
approximate the true welfare cost due to changes associated with the first program, C,,,(P,), turns
in the time path of 6, by resorting to the cost of out to be higher (lower) than C , ( P , ) , this
illness approach? Clearly, the usefulness of cost of implies the welfare cost of the second program to
illness values critically depends on the answer to be lower (higher). In this sense, for identical
the question of whether these values, if necessary individuals, the cost of illness approach offers
supplemented by information on individual type, valuable information for the purpose of program
provide information on the corresponding true evaluation.
welfare cost values. If one were able to give a In order to discuss the case of heterogenous
positive answer, this would certainly make a individuals, take individual A now to have a larger
strong case for the cost of illness approach. In what financial wealth endowment than B. With every-
462 w.Rl[ED
thing else unchanged, it is not difficult to see that the ranking of the true program cost can be any-
this affects the usefulness of cost of illness values. thing, e.g.,
First, observe that this type of heterogeneity fails
to produce any influence on the welfare loss due to
adverse changes in health capital depreciation.
This is true because, as Fig. 2 illustrates,
individuals with different financial wealth endow-
ments will choose identical paths for health capital
and gross investment. Hence, for a given change in
health capital depreciation, they will experience
~CO,(PI) > CCOl(P2) *
I
CWTP(PA

CWTP
' CWTP(P2)

CWTP(P1) = CWTP(P2)

(PI) < C W T P (PZ)

Thus, with individuals differing in their financial


(56)

the same direct effect on their present value of net wealth levels, the cost of illness approach fails to
expenditures. provide any useful information on the relative
Therefore, the relative desirability of programs desirability of the two programs in this simple
P I and P2 will not be affected by introducing example.
heterogeneity with respect to financial wealth. Finally, consider briefly the influence of initial
Consider now the impact on the cost of illness health capital. More specifically, take the better
values. In comparing the two programs, it is health capital endowment of A to be the only
necessary to account for the change in both the difference between the two individuals. Proceeding
size of the welfare loss and the financial wealth exactly as before in the case of the financial wealth
endowment. To see how this translates to the endowment, a similarly negative result obtains. In
respective cost of illness values, one has to particular, equation (55) can be used as well if one
evaluate the following derivative: substitutes H, for V,. Again, the impact of a
simultaneous variation in the size of the welfare
loss and initial health capital on the cost of illness
value remains unclear. Thus, a comparison of the
cost of illness values associated with P , and P2
supplies no information on the ranking of the true
program cost.
Thus, introducing type heterogeneity through
either the financial wealth endowment or initial
health capital acts to blur the relationship between
aTw,
w, y - P : - the ax, cost values produced by the cost of illness and

L
+
t=O
a84 0
(1 +r)'
an the willingness to pay approach. Moreover, the
discussion above has indicated a rather interesting
asymmetry between the two approaches. On the
one hand, cost of illness values arising from
avo changes in health capital depreciation are obser-
(55) vable whereas the corresponding welfare costs are
not. On the other, however, the impact of type
Carrying out the differentiation on the right heterogeneity on the welfare loss is clear whereas
hand-side, one obtains an expression involving its influence on the cost of illness remains
the second partials of the labour supply and the ambiguous.
market good demand function with respect to
the costate A:. Hence, drawing on the results
of the previous section, the sign of the CONCLUSION
right hand-side of equation ( 5 5 ) cannot be
established without further knowledge of both For the Grossman pure investment model dealing
preferences and household production with morbidity effects only, this paper has estab-
technologies. lished a lower bound property which cost of illness
It is important to see that this last result is values possess with respect to the true welfare cost
detrimental to the usefulness of the cost of illness of marginal changes in the rate of health capital
approach in program evaluation. It implies that, depreciation. In the literature on the cost of illness
for a given constellation of cost of illness values, approach, such a property has long been hypothes-
WILLINGNESS TO PAY AND COST OF ILLNESS 463
ized to hold on the grounds that empirically obser- On the other hand, the analysis in this paper has
vable estimates refer to the tangible cost of illness been confined to the impact of changes in health
only.'6p2o Hence, they fail to take into account the capital depreciation on a specific type of optimal
intangible cost of illness which, nevertheless, solution to the individual's decision problem. It
belongs to the overall economic burden due to would be interesting to know whether the results
morbidity. While this line of reasoning may sound derived above carry over to other types of optimal
plausible intuitively, it cannot apply to the model trajectories. A final suggestion concerning the
discussed in this paper. By neglecting any direct agenda for future research relates to the model
utility effect of health capital, the pure investment employed in this paper. Since the pure investment
model leaves no room for an intangible cost of model represents but a submodel, this raises the
morbidity. question of whether the results obtained in this
Instead, the lower bound property holds because paper continue to hold in the more general setting
of the labour supply reaction to changes in individ- of the full Grossman model.
ual welfare. Any change in intertemporal utility
must be brought about by a change in consumption
of the commodity which, in turn, implies corre- ACKNOWLEDGEMENTS
sponding changes in the respective factor inputs.
Thus, the indirect effect on labour supply counter- I am grateful to participants at the Fourth European
acts the change in individual welfare. In fact, a Workshop on Econometrics and Health Economics, Paris
result somewhat stronger than the lower bound 1995. particularly Martin Forster and Cynlle Piatecki, for
property of cost of illness values has been derived a number of helpful suggestions and fruitful discussions
for the present model. Ceteris paribus, the cost of on the full solution of the pure investment model. In
addition, I wish to thank two anonymous referees whose
illness value is increasing in the true welfare cost. criticisms gave me the opportunity to address the import-
Hence, for identical individuals it is possible ant issue of transversality conditions in greater depth. Of
validly to compare adverse morbidity effects course, I alone am responsible for any remaining errors.
associated with alternative health programmes by
relying on the cost of illness approach.
These positive results notwithstanding, the APPENDIX
significance of cost of illness estimates as a substi-
tute for the true welfare cost has been shown to be As explained in the main text, optimal solutions
rather low. As soon as one allows for diversity in will not be marked by an asterisk. Referring to the
individual type, program evaluation along the lines free terminal time problem (P), the optimal
suggested by the cost of illness approach offers no solution is given by sequences { v , ) : - ~ ,
guidance on the correct ranking of programmes. { H , , V,]::;, where v, denotes the vector of control
This has been demonstrated for the case of varia- variables as defined in the text. Hence, the
tions in financial wealth and health capital corresponding level of maximum intertemporal
endowments, respectively. Although at this stage utility T ( P )is equal to:
this is to some extent speculative, it seems highly
likely that similarly negative results will be
obtained for other parameters affecting variation in
individual type.
As far as the concepts of true welfare cost and
cost of illness are concerned, these have been Now consider a fired terminal time problem (Pk)
employed relating to marginal changes in health with k representing the final period of the
capital depreciation. While certainly not being individual's life and N G kd H. N constitutes a
uncommon in the literature on the subject,'*2*5 this lower bound defined by
represents but a first step on the way to an evalua-
tion of more general, i.e. non-marginal, changes in
the time path of the rate of health capital deprecia-
tion. Hence, a first suggestion for future research
would be to extend the analysis undertaken in this
paper to discrete changes in health capital i.e., N will be the final period if the individual
depreciation. never undertakes any gross investment in health.
464 W.N E D
Likewise, m
defines an upper bound on the Define the relation of weak (strong) dominance
individual’s length of life (which is equal to for any two of the intertemporal utility
w+ 11. Given that the ontimal length of life is maximization problems introduced above as
finite‘ by assumption, k can be chosen in follows. A problem weakly {strongly) dominates
accordance with another if its optimal solution offers a level of
n<Ai (A31 intertemporal utility no lower (higher) than any
optimal solution for the rival problem. Clearly,
More precisely, the problem (P,J is as follows: both relations refer to the function r, i.e. weak
I (strong) dominance implies the inequality for the
corresponding values of T to be weak (strict).

S.t. H I +1 - H , = I , - 6 , H , ; Lemma 1
The modified fixed terminal fine problem ( F k )
Vr+ I -~ V ,
weakly dominates the corresponding fixed terminal
(1 + r),+’ (1 + r ) , time problem ( P k ) ,i.e. one has:
- W ,T W , - P Y M , - P : X ,
- , T(Fk)3 r ( P , ) for any k (A6)
(1 + r)‘
Z ( X , , T , ) - Z , = 0 ; I ( M , , TH,) - I , = 0 ; Proof
h ( H , ) - TW, - TH, - T , = 0 ; 21, 3 0 , The set of restrictions included in (Fk)
0 s t c k ; H , > Hminr 1 d t c k ; constitutes a proper subset of the set of restrictions
H o > H,, given; V o > 0 given; contained in (PJ. It follows that any feasible
solution for the fixed terminal time problem is also
Hk+l Hmin; vk+l 0 feasible in the modified fixed terminal time
problem while the converse does not hold. This
The optimal tra’ectories solving (Pk) are given by implies (A6).
(d
sequences ( v, } k,=0, (H,‘”, V,‘”} ,’:f where the
superscript ( k ) has been added in order to indicate Lemma 2
the fixed terminal time problem under
consideration. This yields a level of intertemporal The optimal solution of the free terminal time
utility equal to problem ( P ) and the optimal solution of the fixed
terminal time problem ( P , ) coincide.
k

r ( P k ) = u((zy’}:=o)
= ~ m , U ( Z ~ (A4)
’)
r=O Proof
By assumption, n denotes an optimal final
Finally, define the modified fuced terminal time period. From the definition of both problems,
problem (pk) as the problem which does not take given n , the free terminal time problem (P)
into account the terminal constraint on health
reduces to the fixed terminal time problem (P,,).
capital but is identical with ( P k ) in all other In particular, lemma 2 implies (intertemporal)
respects. Note that transversality considerations
utility equivalence for the optimal solutions to
imply the terminal value of the corresponding problems (P) and (P,,),i.e. r ( P ) = T(P,,).
costate to be equal to zero since there is no
constraint on terminal health capital. Writing
v,‘k)}
W k ):=o,/ (RP’, f2,’- for the trajectories Lemma 3
optimal with respect to (Pp), the individual then
attains a level of intertemporal utility equal to: The optimal solution of the free terminal time
problem (P) weakly dominates the optimal
k solution of any fixed terminal time problem (Pk):
r(pk)=u((zf
-(k) }k, = o ) = C m r u ( Z I k ’ (A51
)
r=O u((Zr 1 :=o) 2 u((Z,‘k’}:=o) for any k (A71
WILLINGNESS TO PAY AND COST OF ILLNESS 465
financial wealth equal to zero, i.e. V z ) , = O must
Proof
hold. Now it is possible to extend the optimal
Relying again on the definitions of the free trajectories [ U,(")} :=o and [ R,("', V,("'):z,' in a
terminal time and the fixed terminal time problem, manner such that a feasible solution for a fixed
respectively, it is easy to see that any solution terminal time problem (Pk)with k > n obtains.
feasible in some fixed terminal time problem (Pk) Define f as follows:
is also feasible in the more general problem (P).
This establishes (A7). Note that, owing to lemma
2, strong dominance is ruled out.
Now it is possible to prove a result concerning
the transversality condition for the stock of health
capital. With no loss of generality, take f to be no higher
than m.
Clearly, n + f denotes the final period
which results if the individual decides, given a
Proposition 1 stock of health capital equal to RA:),> H,, in
In the free terminal time problem (P), the period n + 1, to undertake no more gross invest-
terminal value of the costate variable associated ment in health. Consider now the fixed terminal
with the stock of 'iealth capital will be equal to time roblem (f',,+d apd construct a solution
zero: { c j n + f } n +r =l O I ~ ~ = i n + l ) , v ~ n + ' ) ) ~ , ~as
: + follows.
l u p to
A:+, = 0 (A8) period n (period n + l), set the values of the
control variables (the state variables) equal to the
corresponding values of the optimal solution to the
Proof modified fixed terminal time problem (p,,):
The proof relies on the utility equivalence of the
-(n) ,
G i n + ] ) = 2),
solutions to problems (P) and (P,,) established by Oat4n
-
lemma 2. The basic idea is to show that A,",, < O Hl("fi)= R,("), 1 a t 4 n + 1 (A1la)
implies the existence of another fixed terminal V/n+LV,(n), latcn+l
time problem ( P J with k > n whose optimal
solution strongly dominates the optimal solution of
(P,) and, thus, of the free terminal time problem
(P). In view of lemma 3, this is a contradiction.
Consider the solution of the fixed terminal time
problem (P,,) which, owing to lemma 2, involves
no loss of generality. The weak inequality for
terminal health capital implies the transversality
condition '"'A,H,, $0. Now suppose ' " ' A ~ + , is
negative, implying a terminal stock of health
capital equal to Hmin.The fact that the costate
is lower than zero indicates that the
individual would benefit from a marginal increase
in terminal health capital above H,,,.While this is For the time paths of health capital and financial
impossible in the fixed terminal time problem (P,,) wealth, respectively, these choices imply:
it is clearly feasible in the modified problem (Fn). -
In the latter it is optimal for the individual to move H;:;"= H=y+i'(l- d,), n + 1 4 t 4 n + r
away from the optimal solution of (P,,) and, in ~ ~ : ~ " = O n, + l < t S n + l (Allc)
particular, to fix terminal health capital above
Hmm. Applying lemma 1, one has This completes the construction of a solution which
is clearly feasible, albeit not necessarily optimal,
r(F,,)3 r ( P n )= r(P) (A91 for the fixed terminal time problem (P,,.!). Note
that, given the definition of f in equation (AlO),
Due to the assumption of non-satiation, the health capital will be strictly above H,, in any
optimal solution of the modified fixed terminal period except the terminal one (i.e, t = n + I+ I).
time problem (F,,) will have a terminal level of Hence, in any period t with n + 1 s t a n + l , a
466 W.N E D
strictly positive amount of healthy time is available. time problem (F,) coincide. This implies utility
According to equation (A1 lb), this implies labour equivalence between both problems, i.e.
supply as well as both the time and the market good r(FJ= T ( P ) holds.
input to the production of the commodity to be
strictly positive. Thus, for n + 1 s rc n + 1, the
solution under consideration offers strictly positive
Proof
levels of the commodity. Consider the optimal solution to the fixed
Now compare this solution for (P,+i) with the terminal time problem ( P , ) which involves no
optimal solution of the modified fixed terminal time loss of generality due to lemma 2. Assume an
problem (F,) copidered- earlier. By construction, optimal solution to (F,) to be different from the
the sequence { Z,("+')]FJi provides the individual one to (F,,). This implies, in particular,
with the same amount of the commodity as the R',"!,> Hmk.As the proof of proposition 1 shows,
sequence { Z ~ n ) )up ~ =tooperiod n, while offering this implies, in turn, the existence of a fixed
strictly positive amounts for at least one additional terminal time problem (P,+r) with f a 1 which
period. It follows that the former must yield a strongly dominates (F,). Applying lemmas 1 and
strictly higher level of intertemporal utility: 2, the free terminal time problem will be weakly
dominated by (F,,).Thus, given R(,2,>Hmi,,
U ( ( q ( n ) } ; ==or(F,)
U({z,(n+i)};:i)> ) (A12) there exists a fixed terminal time problem which
strongly dominates the free terminal time
Moreover, the optimal solution of the fixed problem. This contradicts the weak dominance
terminal time problem (P"+')will at least weakly property of the problem (P) established by
dominate the feasible solution discussed above. lemma 3.
Hence, one has

r(Pn+i> L U ({ +I) 1YJ;) (A131 Proposition 2


Combining the inequalities (A9), (A12) and Assume the optimal final period in the free
(A13), one arrives at terminal time problem (P) to be unique. Then,
H,+lc H,, must hold, i.e. terminal health capital
r(pn+$r(P) (A141 will be strictly lower than Hmk
which contradicts the weak dominance property of
the free terminal time problem (P) established by Proof
lemma 3. Since the assumption that the terminal Again, the proof is indirect. It shows that the
value of the costate variable associated with health assumption H , = H,, implies non-uniqueness of
+

capital in (P) be negative leads to a contradiction, the optimal final period in the fFee terminal time
one must have A:+, equal to zero. problem ( P ) .
Note that a corresponding statement for fixed If the optimal final period is unique, this implies
terminal time problems will, in general, not be both problems (P) and (P,) to dominate strongly
true. In other words, when analysing the problem any other fixed terminal time problem (PJ:
(Pk),(k)Af+l < 0 cannof be excluded. With respect
to fixed terminal time problems, all the above r ( p )= r(P,) > r(P,) v k+ n (~15)
proposition asserts is that for the optimal solution
of the problem (P,), where n corresponds to an Owing to proposition 1, A,"+, = O while the
optimal choice of the final period, the costate terminal constraint on health capital implies
will be equal to zero. H,+ls H m h . Now suppose, contrary to the
The proof of proposition 1 shows that it is implication of proposition 2, Hn+, = H ~ "Relying
.
possible to supplement lemma 2 and to strengthen on lemmas 2 and 4, the optimal solutions to the
lemma 1 in the case k = n as follows: problems (P), (P.) and (F,) must coincide and, in
particular, have the same terminal stock of health
= R',"!,= H , + , which equals Hmk
capital: H',"!, by
Lemma 4
assumption.
The optimal solutions to the free terminal time Consider the impact of a small decrease in the
problem (P) and to the modified fixed terminal rate of health capital depreciation 6, on the
WILLINGNESS TO PAY AND COST OF ILLNESS 467
optimal solution to the modified fixed terminal occupational choice. Journal of Political Economy.
time problem (F,). Applying the results of this 1977; 85: 1273-1294.
paper, the only effect will be a small increase in 5. Cropper, M. L. Measuring the benefits from
terminal health capital. Thus, following the change reduced morbidity. American Economic Review.
in 6, one has H',=', > Hminwhile intertemporal 1981; 71: 235-240.
6. Muurinen, J.-M. Demand for health. A generalized
utility will remain unaffected. Grossman model. Journal of Health Economics.
Things are different for the fixed terminal time 1982; 1: 5-28.
problem (P,). Here, the optimal trajectory for 7. Ehrlich, I. and Chuma, H. A model of the
health capital up to period n must be modified in demand for longevity and the value of life exten-
such a manner that the terminal constraint will still sion. Journal of Political Economy. 1990; 98:
be satisfied after the change in the rate 6,. Hence, 761 -782.
the optimal solutions to (P,) and (F,) no longer 8. Erbsland, M., Ried, W. and Ulrich, V. Health,
coincide. Drawing on the proof of lemma 4, this health care, and the environment. Econometric
implies the existence of another fixed terminal evidence from German micro data. Health Econ-
time problem (P,+i)with 1 which strongly omics. 1995; 4: 169-182.
9. Ried, W. Willingness to Pay, Cost of Illness, and
dominates (F,) and thus, by lemma 1, the fixed the Human Capital Approach in Health Care-a
terminal time problem (P,). Comparative Study based on Grossman's Pure
Therefore, if terminal health capital is equal to Investment Model. Discussion Paper 521 -95.
H,, at the optimal solution to the free terminal time University of Mannheim.
problem (P), a small decrease in the rate 6, is 10. Van Doorslaer, E. K. A. Health, knowledge and the
sufficient to imply an increase in the optimal length demand for medical care. Van Gorcum, Assen/
of the planning horizon. Note that this argument Maastricht, 1988.
applies to arbitrarily s m a decreases in 6,. Making 11. Varaiya, P. P. Notes on optimization. Van Nostrand
use of the fact that Z,(,+')is continuous with respect Reinhold, New York, 1972.
to 6, (the comparative dynamic analysis undertaken 12. Berck, P. and Sydsaeter, K. Economists' mathemati-
cal manual. Second edition, Springer, Berlin, 1993.
in this paper relies, in fact, on the stronger 13. Seierstad, A. and Sydsaeter, K. Optimal control
assumption of continuous differentiability), T(P,+i) theory with economic applications. North-Holland,
will be a continuous function of the rate 6, as well. Amsterdam, 1987.
Given that arbitrarily small decreases in the rate 6, 14. Bensoussan, A., Hurst, E. G. and Nalund, R.
are sufficient to ensure that some other fixed Management applications of modern control
terminal time problem strongly dominates the theory. North-Holland, Amsterdam, 1974.
modified fixed terminal time problem (F,), 15. Canon, M. D., Cullum, C. D., and Polak. E. Theory
continuity then implies the existence of another of optimal control and mathematical programming.
fixed terminal time problem which is at least utility McGraw-Hill, New York, 1970.
equivalent to (F,) and, thus, to (P) without the 16. Hodgson, T. A. and Meiners, M. R. Cost-of-illness
methodology: a guide to current practices and
change in 6,. This however, contradicts the procedures. Milbank Memorial Fund Quarterly/
uniqueness assumption contained in proposition 2. Health andSociety. 1982; 60: 429-462.
17. Patrick, D. L. and Erickson, P. Health status and
health Policy. Quality of life in health care evalua-
tion and resource allocation. Oxford University
REFERENCES Press, Oxford, 1993.
18. Hodgson, T. A. The state of the art of cost-of-
Berger, M. C., Blomquist, G., Kenkel, D. and illness estimates. Advances in Health Economics
Tolley, G. S. Valuing changes in health risk: a and Health Services Research. 1983; 4: 129-164.
comparison of alternative measures. Southern 19. Decodes, J., Baker, T. D. and Schumann, D. The
Economic Journal 1986; 53: 967-984. hidden costs of Illness in developing countries. In:
Harrington, W. and Portney, P. R. Valuing the Sirageldin, I. and Sorkin, A. (eds) Research
benefits of health and safety regulation. Journal of in human capital and development. Vol. 5: Public
Urban Economics 1987; 22: 101-1 12. health and development. JAI Press, Greenwich,
Grossman, M. The demand for health: a theoretical CT,1988.
and empirical investigation. Occasional Paper 1 19. 20. Kenkel, D. Cost of illness approach. In: Tolley, G.,
National Bureau of Economic Research, New York Kenkel, D. and Fabian, R. (eds) Valuing health for
and London, 1972. policy. An economic approach. University of
Cropper, M. L. Health, investment in health, and Chicago Press, Chicago, 1994.

You might also like