Politics of social health insurance

Stéphane Rossignol

PII: S0176-2680(08)00018-9
DOI: doi: 10.1016/j.ejpoleco.2008.02.002
Reference: POLECO 1060

To appear in: European Journal of Political Economy

Received date: 7 April 2006

Revised date: 17 January 2008
Accepted date: 14 February 2008

Please cite this article as: Rossignol, Stéphane, Politics of social health insurance, Euro-
pean Journal of Political Economy (2008), doi: 10.1016/j.ejpoleco.2008.02.002

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Politics of social health insurance

PT
∗
Stéphane Rossignol

RI
January 17, 2008

SC
NU
MA
Abstract. This paper studies the political support for social health insurance
when a private alternative exists. Individuals differ only by their risk. For the more
realistic distributions of risk, a majority of agents do not want public insurance.
D

However, in a representative democracy, or in a direct democracy with altruistic

TE

agents, we show that social insurance can be adopted, particularly for treatments
which have the best cost-utility output. But if the low risk agents are more politi-
cally powerful than the high risk, the low cost treatments will not be refunded by
P

social insurance, even if their utility is high.

CE

JEL classification: D72; H51

Keywords: Health; Voting; Social insurance; Representative democracy
AC

∗
C.E.S., UMR 8174 CNRS Université Paris I and Université de Versailles-Saint-Quentin.
E-mail address: rossignol@math.uvsq, mail address: C.E.S. 106-112 Bd de l’Hôpital, 75647 Paris
Cedex 13, France. I would like to thank Hubert Kempf, Emmanuelle Taugourdeau and an anony-
mous referee for helpful comments.

1
 ACCEPTED MANUSCRIPT

1 Introduction
In this paper we study the political support for social health insurance when private
insurance is available. The agents are differentiated by their risk of sickness. In this
case, the median voter is against social insurance. However, we show that it can

PT
be adopted in a representative democracy or in a direct democracy with altruistic
agents, for the illnesses which have treatments with a good cost-utility output.
In many countries, the health systems are weakened by a continuous increase

RI
in health care spending. Health spending is now 10.5% of GDP in Germany, 9.8%
in France, with a record of 14% in the United States. This increase is higher than

SC
that of GDP and induces serious financing problems. It has three main causes:
first a demographic factor, the ageing of the population, since the need for medical
care is greater for older people. Secondly there is a technical factor, because with

NU
technological progress treatments are becoming more sophisticated and hence more
expensive. And finally a political factor: the increase of the coverage. Indeed most
OECD countries have evolved to offer wide coverage of their citizens.
MA
With respect to the importance of the public health insurance, any project for
reform the health system must be preceded by a thorough reflection on the respective
places for public and private insurance. There are several arguments of economic
efficiency in favor of public insurance: it has lower administrative costs, and it avoids
D

adverse selection, since it is universal. It is also redistributive, thus more equitable.

TE

However, with the public insurance, the agents are less free, and it weighs on public
taxation, which is distorsive and unpopular. In all countries the public and private
systems coexist, but their respective proportions vary. The Americans consider that
P

health is the individual’s responsibility and that of the market, the government must
CE

be involved only if necessary (as with Medicare or Medicaid). The Europeans attach
more importance to the public sector, private insurance being mainly a complement.
In this paper we adopt a political economy approach: the aim is to determine
AC

the political support for public insurance when there is a private alternative. More
precisely, we determine which treatments could be paid for by social insurance. It
will depend on the health utility of treatment, and on its cost.
Public health insurance has a solidarity function. Numerous political economy
models interpret this solidarity as a redistribution from the rich to the poor: Epple
and Romano (1996) and Gouveia (1997), using models where the agents are differen-
tiated by their incomes, establish the existence of an equilibrium where the majority
is in favor of public insurance. But this sort of redistribution can be achieved di-
rectly, without using health insurance, as shown by Meltzer-Richard (1985). The
median voter is indeed poorer than the average, hence in favor of redistribution. In
this case it is not a good way to justify the existence of public insurance. With a
similar model, Blomquist and Christiansen (1999) add a strong hypothesis: that the
qualification is private information, not available to the government.
It appears that to justify public health insurance, we must stress what distin-
guishes it from private insurance (with no asymmetric information): the mutuali-
sation of different risks. Consequently, we study here the solidarity between agents
differentiated by their risk of sickness, but with the same income. It allows us to

2
 ACCEPTED MANUSCRIPT

better understand the mutualisation of risks, and to distinguish it from the redis-
tribution between rich and poor. Hindriks (2001) and Hindriks-De Donder (2003)
have tackled this question, but by dropping the expected utility model. Our aim
is to establish the conditions of existence of public insurance (and the extent of its
coverage) in the usual expected utility model, with agents differentiated by their

PT
risk, and in particular to determine which treatments would be covered by public
insurance.

RI
Two systems are possible in our paper: purely public and purely private. For a
given treatment, no mixed insurance is possible. If public health insurance covers

SC
the treatment, we assume that it is compulsory and offers full coverage. Otherwise,
it is well known that, by an adverse selection mechanism, only high risk people
will take it, so public insurance then no longer has a solidarity function. There is

NU
here no moral hazard: if a person is sick, she has only one possible treatment (no
partial treatments). Without public insurance, the agents can subscribe to private
insurance, or choose not to take any treatment.
We show that the people having a lower than average risk (the majority here)
MA
would prefer a priori to reject public insurance. The other people want it if the
output of the treatment is sufficiently high1 , in a meaning specified later. We study
the conditions necessary to adopt a social system, even though it is contrary to the
D

interest of a majority.
The paper is organized as follows. The second section presents the model. The
TE

political economy of the choice of health insurance system is studied in the third
section. The fourth section is devoted to equity aspects. We conclude in section 5.
P
CE

2 The model
2.1 Health financing and agents’ welfare
AC

We consider an economy with n individuals, all with the same income y. They can
be healthy or sick. The probability p of falling ill depends on the agent. This risk
p is distributed in the population on the interval [0; 1] according to the cumulative
distribution function F (p).
A healthy person has a utility U(y), a non treated sick person has a loss of utility
V > 0, her utility is then U(y) − V . We assume that there is no loss of income due
to the illness. An agent of risk p (i.e. with probability p of falling sick) enjoys then
an expected utility without treatment:

W1 (p) = U(y) − pV (1)

U is an increasing and concave function.

We assume that the illness the agents can contract requires a treatment of cost
H > 0, and it leads to a complete recovery. The treatment is of the type ”all or
nothing”, it is pointless to take only a part of it. Our model is grounded on a
cost-utility evaluation of the treatment and its financing. It cannot be reduced to a
1
It will always be true for a fatal illness, since the utility gain is then V = +∞.

3
 ACCEPTED MANUSCRIPT

minimization of the cost, since we take into account the utility of the medical care.
V > 0 can be understood as the loss of utility due to the illness (without treatment)
with respect to a healthy state, but V is then also the utility gain of the treatment.
We will compare the welfare of the agents according to the way of financing the
treatment.

PT
- The illness risk can be covered by private insurance. We assume that the
insurers know perfectly the type of the agent, and then her risk. There are no

RI
administrative costs, the insurance contracts are actuarially fair. The contribution
rate θ is individual, it depends on the risk p. The expected utility is then W2 (p) =

SC
U (y(1 − θ(p))) with the budget constraint yθ(p) = pH, then θ(p) = p Hy . We obtain
the expected utility with a private insurance

NU
W2 (p) = U(y − pH) (2)

The agent can pay the insurance premium only if pH ≤ y. We will suppose it to be
MA
true henceforth.

- Note that if the agent herself pays for the treatment if she is ill, her expected
utility becomes
D

W3 (p) = (1 − p)U(y) + pU(y − H) (3)

TE

Moreover, since the agents are risk adverse, they prefer to pay for an actuarially
fair private insurance than to pay H. It means that W2 (p) > W3 (p) for any p > 0,
because U is strictly concave. If actuarially fair private insurance is available, the
P

solution of paying H directly will never be preferred by a rational agent.

CE

- If social insurance covers the treatment, it is compulsory and covers the whole
cost H. It is pointless to spend only a part of H, since the treatment is of the type
”all or nothing”.
AC

It is financed by a linear contribution rate t on income. The expected income is

Wsoc = U (y(1 − t)) since the agents are cured ifRsick.
1
The budget constraint of social insurance is 0 [yt − pH]dF (p) = 0, which gives
R1 R 1
yt = H 0 pdF (p) = pH, i.e. t = p Hy , where p = 0 pdF (p) is the average probabil-
ity of falling sick. We assume that social insurance is implementable, i.e. pH ≤ y.
We obtain finally the expected utility with social insurance:

Wsoc = U(y − pH) (4)

2.2 Who wants to be treated if there is no social insurance?

If there is social insurance, since it is compulsory and covers the whole cost H,
nobody needs private insurance. All the agents have the same welfare Wsoc =
U(y − pH).
We now want to examine the following question: what will the different agents
do if H is not covered by the social insurance? We have seen that it is always better
to pay for a private insurance than to pay H directly (since W2 (p) > W3 (p)). The

4
 ACCEPTED MANUSCRIPT

alternative here is then: to subscribe to private insurance or to give up on being

cured. The welfare without social insurance is then Wind (p) = max(W1 (p); W2 (p)).
An agent of risk p takes out a private insurance if and only if W2 (p) > W1 (p), which
is equivalent to U(y − pH) > U(y) − pV , i.e. V > p1 [U(y) − U(y − pH)]
We set

PT
1
V (p, H) = [U(y) − U(y − pH)] if p ∈]0; 1] (5)
p

RI
V (0, H) = lim+ V (p, H) = HU ′ (y)
p→0

SC
We now show that the low risk people will prefer to be better covered than those at
high risk.

NU
Proposition 1 An agent of risk p prefers to take out an insurance (rather than not
be cured) if and only if V > V (p, H), where V (p, H) is an increasing function of p
and H. MA
(i) If V > V (1, H), everybody wants to be treated.
(ii) If V ∈]V (0, H); V (1, H)[, an agent of risk p prefers to take out insurance
(rather than not be cured) if and only if p < p∗ , where p∗ = p∗ (H, V ) is such that
V = V (p∗ , H).
D

(iii) If V < V (0, H), nobody wants to be treated.

TE

Proof See Appendix.

P

Insert Figure1 here

CE

(i) If V > V (1, H), then everybody takes out insurance. For a given cost H , a
AC

serious illness is cured: everybody wants to get treated.

(ii) If V (1, H) > V > V (0, H), only the low risk agents take out insurance. An
agent of risk p takes out insurance only if V > V (p, H). The insurance premium
generates a too high loss of consumption for the high risk agents, so they prefer not
be cured. This result might seem counter intuitive. A priori we could think that it
is more useful to take out insurance if we have a higher risk. But a high risk means
a high insurance premium, which weighs on consumption.
(iii) If V < V (0, H), nobody takes out an insurance, since the gain V due to the
treatment is low compared to its cost.
Furthermore V (p, H) is an increasing function of H, which means that when the
cost of the treatment increases, fewer people take out insurance (for a given utility
V of the treatment).

2.3 How to define the output of a treatment?

In our model, a treatment is fully characterized by a couple (H, V ), where H is
its cost and V its utility. It is possible to compare the outputs of certain couples

5
 ACCEPTED MANUSCRIPT

(H, V ). First it is clear that (H2 , V2 ) is better than (H1 , V1 ) if it is less expensive and
provides a higher utility, i.e. if the two following conditions are fulfilled: H2 ≤ H1
and V2 ≥ V1 . If one of these inequalities is strict, then (H2 , V2 ) is strictly better
than (H1 , V1 ).
Of course, we cannot compare all couples of treatments (H1 , V1 ), (H2 , V2 ) in this

PT
way. For example if H1 < H2 and V1 < V2 , treatment 1 is less expensive but its
utility is lower. An additional criterion is necessary to compare the outputs of the

RI
two treatments. It will depend on the risk p of the agent. We can adopt V −V (p, H)
as a measure of the output of the treatment for an agent of risk p. Indeed V −V (p, H)

SC
is an increasing function of V and decreasing function of H, which is compatible
with our order relation for the couples (H, V ). Moreover, according to Proposition
1, an agent of risk p wants to be treated if and only if V > V (p, H), i.e. when the

NU
output of the treatment is positive (for her). If V < V (0, H), the treatment has a
bad output: it is negative for any risk p. If V > V (1, H), the treatment has a good
output: it is positive for any risk p.
MA
3 Choice of the system
3.1 Individual preferences
D

Our model works as a game with two stages:

TE

- first, a decision committing the whole society is taken: to adopt or not a public
health insurance system (for a given treatment (H, V )).
P

- second, the agents react individually to the preceding collective choice. It

means that if social insurance has not been adopted, they choose whether to take
CE

out private insurance or not. Proposition 1 deals with this second stage. If social
insurance has been adopted at the first stage, a private policy becomes of course
pointless.
AC

We will now study the first step of the game, taking into account the reaction
function of the agents, given in Proposition 1. Before studying the effective political
choice of the system, let us examine the individual preferences. Which system is
preferred by which people: public insurance or not? An agent of risk p prefers social
insurance if and only if Wsoc > Wind (p).
Since Wind (p) = max(W1 (p); W2 (p)), without social insurance, some people take
a private policy, the other people don’t. Our question is then split into two parts:
- For the agents who prefer a private policy to no medical care (i.e. such that
W2 (p) > W1 (p)), do they want social insurance or not? The answer is simple: they
want social insurance only if p > p. Indeed

Wsoc > W2 (p) ⇔ U(y − pH) > U(y − pH) (6)

which is equivalent to p > p. With medical care, social insurance is preferred if its
premium is less expensive, i.e. if p > p.
- The same question is valid for the agents who prefer no medical care to taking
out a private policy (i.e. such that W2 (p) < W1 (p)). They prefer social insurance

6
 ACCEPTED MANUSCRIPT

if Wsoc > W1 (p), i.e. if U(y − pH) > U(y) − pV , which means that V > p1 [U(y) −
U(y − pH)].
Finally, only the high risk agents can prefer social insurance, for the most bene-
ficial treatments. It is more precisely stated in the following proposition.

PT
Proposition 2 - If p < p, an agent of risk p always rejects the social insurance
system.

RI
- If p > p, an agent of risk p accepts the social insurance system if and only if
V > p1 [U(y) − U(y − pH)]

SC
(i) If V > p1 [U(y) − U(y − pH)], an agent of risk p wants social insurance iff
p>p  
U(y) − U(y − pH)
(ii) If V ∈ U(y) − U(y − pH); , an agent of risk p accepts

NU
p
the social insurance system if and only if p > V1 [U(y) − U(y − pH)]
(iii) If V < U(y) − U(y − pH), nobody wants a social insurance system.
MA
Proof Straightforward because: - if p < p, then Wsoc < W2 (p).
- if p > p, then Wsoc > W2 (p) but Wsoc > W1 (p) if and only if V > 1p [U(y) −
U(y − pH)].
D

Moreover H 7→ 1p [U(y) − U(y − pH)] and H 7→ U(y) − U(y − pH) are convex
functions since U is concave, which motivates the shapes of the curves in Fig 2.
TE

Insert Figure2 here

P
CE

- If p < p, the agent loses with redistribution: she always prefers the private
AC

insurance to the public one.

- If p > p, the agent wins with redistribution: here the public insurance is
preferred to the private one. However, if V is too low (for a given cost H > 0), even
the public system will be too expensive for the agent: she prefers to have no medical
care.

3.2 Political decision function.

We now study the health system chosen (social insurance or not) according to the
type of political decision procedure.
It is classically observed that a majority of health care spending is devoted to
the minority of the weakest people. We assume therefore that the low risk people
are the majority, i.e. p > pm , where pm is the median risk. It means that F (p) =
Rp
0
dF (p) > 21

7
 ACCEPTED MANUSCRIPT

Direct democracy with selfish agents. With direct voting, the median voter
decides: since pm < p, social insurance is rejected (according to Proposition 2), for
any values of V and H. It is not very realistic, because social health insurance
systems exist in most countries (even if they do not cover all the risks). And direct
democracy with selfish agents is a quite restrictive hypothesis. We now examine an

PT
alternative.

RI
Political decision function. We assume here that the choice is the result of the
maximization of a decision function D which is a weighted average of the welfare of

SC
the different agents. Z 1
D= W (p)dG(p) (7)
0

NU
W (p) is here the welfare of an agent of risk p, and G(p) measures the weight of
the
R 1 agents of risk p or lower than p in the political decision function. We have
0
dG(p) = 1. We can assume to simplify that the population is composed of k
MA
groups of homogenousPindividuals, differing by their probability of falling ill p1 ,
p2 ,...pk . We have n = ki=1 ni
Two different arguments can justify the use of a decision function of this type:
representative democracy, and direct democracy with altruistic agents.
D
TE

Representative democracy. With a representative democracy, the function D

can be obtained with a probabilistic voting model in an electoral competition. For
example in Coughlin et al. (1990) (and Persson Tabellini (2000) chapter 3.4), two
P

candidates (or two parties) A and B compete to win an election. They do not know
CE

exactly the preferences of the electors, so this uncertainty can be characterized by

a stochastic bias in favor of one candidate.
More precisely, each candidate announces a policy, and we assume that the
AC

winner will implement the policy announced. The agent j of group i votes for the
party B iff
WgA (pi ) + bi,j < WgB (pi )
where gA and gB are the policies announced by candidates A and B. The welfare
of any agent of group i is Wg (pi ) if the policy g is applied. The random variable bi,j
measures the bias (positive or negative) of agent j in favor of candidate A. For a
given group i, the random variables bi,j have the same law as a random variable bi ,
uniform on an interval [li , ri ].
The mathematical expectation of the numbers of votes for B is
XX
EV B (gA , gB ) = P (WgA (pi ) + bi,j < WgB (pi ))
i j
X
= ni P (WgA (pi ) + bi < WgB (pi ))
i
X
= ni Fi (WgB (pi ) − WgA (pi ))
i

8
 ACCEPTED MANUSCRIPT

where Fi is the cumulative distribution function of the uniform law on [li , ri ], i.e.
Fi (x) = rx−l i
i −li
on x ∈ [li , ri].
Then assuming that WgB (pi ) − WgA (pi ) ∈ [li , ri ] for any i :
X 1
EV B (gA , gB ) =

PT
ni (WgB (pi ) − WgA (pi ) − li )
i
li − ri

At thePequilibrium, candidate B maximizes EV B (gA , gB ), i.e. maximizes the

RI
1
function i ni αi WgB (pi ) where αi = ri −li
represents the influence P
of each agent of
group i in the decision function. Similarly candidate A maximizes i ni αi WgA (pi ).

SC
Let us now define a random variable Y of law given by: P (Y = pi ) = Pninαjiαj for
j
any i. We denote by G(p) the cumulativeR distribution function of Y . We have
1

NU
then found that each candidate maximizes 0 Wg (p)dG(p) with respect to g, where
g ∈ {social insurance; no social insurance}.
MA
Direct democracy with altruistic agents. If the agents are altruistic, in a
direct democracy, the electors will take into account the welfare of the other agents.
Altruism can be justified by social links: the agents care about the members of their
family or their friends who have health problems. If we denote by α the degree of
D

altruism, assumed to be the same for all agents, the utility of an agent of risk pj is
given by
TE

!
X
Vj = W (pj ) + α ni W (pi ) + (nj − 1)W (pj )
P

i6=j
CE

k
X
= (1 − α)W (pj ) + α ni W (pi )
i=1
AC

where α ∈ [0; 1]. The median voter is decisive here, his objective is to maximize
k
X
D = Vmj = (1 − α)W (pm ) + α ni W (pi ) (8)
i=1

Let Y be a random variable of law given by:

ni α (1 − α) + nm α
P (Y = pi ) = if i 6= m, and P (Y = pm ) = (9)
(1 − α) + nα (1 − α) + nα
R
Let G be the cumulative distribution function of the law of Y . Then D = W (p)dG(p),
i.e. D is a decision function of the type given in Equation (7).

3.3 Political choice of the system

The decision function D depends on G, H and V . We denote by Dsoc (G, H, V ) =
R1
0
Wsoc (p)dG(p) its value when there is social health insurance, and Dind (G, H, V ) =

9
 ACCEPTED MANUSCRIPT

R1
0
Wind (p)dG(p) when there is none. For any p, we have Wind (p) = max(W1 (p); W2 (p)).
Set
K(G, H, V ) = Dsoc (G, H, V ) − Dind (G, H, V ) (10)
Social health insurance is adopted if K(G, H, V ) > 0, rejected if K(G, H, V ) < 0.

PT
Let us denote by X (respectively Y ) a random
R 1 variable of cumulativeRdistribution
1
function F (respectively G). We have p = 0 pdF (p) = E(X) and 0 pdG(p) =
E(Y ).

RI
We distinguish three cases:
Hyp A: E(Y ) ≥ E(X).

SC
Hyp B1: E(Y ) < E(X) and E(Y 2 ) > E 2 (X)
Hyp B2: E(Y ) < E(X) and E(Y 2 ) ≤ E 2 (X)
Under Hyp A, the high risk agents have a higher political weight than the low

NU
risk agents (E(Y ) ≥ E(X)). Under Hyp B (i.e. E(Y ) < E(X)), the high risk agents
have a lower political weight than the low risk agents. We have two subcases: Hyp
B1 means that the high risk agents have a political weight softly lower than their
MA
demographic weight, Hyp B2 means than the high risk have a political weight much
lower than their demographic weight.
Proposition 3 specifies the choice taken, according to the values of H and V , for
any given cumulative function G.
D

For a given cost H > 0, we see that everything depends on the value of V > 0, in
TE

other words on the output of the treatment, i.e. the comparison between the utility
of the treatment, and its cost H.
P

Proposition 3 V 7→ K(G, H, V ) is a non decreasing function, and K(G1 , H, V ) ≤

K(G2 , H, V ) if G2 stochastically dominates G1 . Moreover, if U is a quadratic func-
CE

tion on [0; y], then there exist functions V ∗ (G, H) and H ∗ (G), such that:
1) Under Hyp A, we have
AC

K(G, H, V ) > 0 ⇔ V > V ∗ (G, H)

2) Under Hyp B1, we have

K(G, H, V ) > 0 ⇔ [V > V ∗ (G, H) and H > H ∗ (G)]

3) Under Hyp B2, we have K(G, H, V ) < 0 for any H, V > 0.

Proof See Appendix.

Let us examine the three cases of Proposition 3.
1) The first case (Hyp A) corresponds to a political decision function where the
political weight of the high risk people is at least equal to their demographic weight.
We find then that social insurance is adopted if the output of the treatment is suffi-
cient. It means that, for a given cost H > 0 , the treatment is covered by the public
health system if it provides a sufficient utility gain, higher than the critical value
V ∗ (G, H). Moreover, the higher the weight of the high risk agents, the wider the
coverage of the social insurance, since K(G1 , H, V ) ≤ K(G2 , H, V ) if G2 stochasti-
cally dominates G1 . However, the coverage will not be universal. Even if the high

10
 ACCEPTED MANUSCRIPT

risks decide alone, the bad output treatments will be rejected.

Insert Figure3 here

PT
RI
2) The second case (Hyp B1) corresponds to a political decision function where
the influence of the high risk agents (per capita) is lower than that of the low risk
agents, but not too small. For the treatments of cost H higher than H ∗ (G), the

SC
result is similar to the first case: a treatment is socially covered if the associated
utility gain V is high enough (greater than V ∗ (G, H)). For the low cost treatments
(H < H ∗ (G)), social insurance is rejected. Indeed, for any utility gain V of the

NU
treatment, its low cost can be easily paid for individually. This result may seem
counter intuitive, since it does not respect the ranking of the outputs of the treat-
ments. We could expect that the higher the output of a treatment, the better this
MA
one could be socially covered. But here under Hyp B1, a treatment (H1 , V1 ) can be
covered (if H1 > H ∗ (G) and V1 > V ∗ (G, H)) and another treatment (H2 , V2 ) of bet-
ter output not be covered (if H2 < H ∗ (G) and V2 ≥ V1 ). We can thus imagine that
D

a treatment (H1 , V1 ) initially covered by the social insurance could be not covered
anymore if its price decreases from H1 to H1′ , with H1′ < H ∗ (G) < H1 .
TE

It means that the health system is more liberal and less egalitarian than under
Hyp A: the agents must resort to a private policy, except for treatments which are at
the same time expensive and of high utility. The solidarity between people differing
P

by their risk becomes here a sort of minimal net covering only people who are poor
CE

and very sick.

AC

Insert Figure4 here

3) Under Hyp B2, the interest of low risk agents mainly matters. Social insur-
ance is always rejected, since it redistributes to the high risk agents, contrary the
private insurance. It gives the same result as direct democracy with selfish agents.

What does mean Proposition 3, in a direct democracy with altruistic agents?

Let Y a random variable
 of law givenP by Equation
 (9). Then we have:
E(Y ) = (1−α)+nα (1 − α)pm + α i=1 ni pi = (1−α)p
1 k m +nαp
(1−α)+nα
= p − (1−α)(p−p
(1−α)+nα
m)
,
P
where p = n1 ki=1 ni pi = E(X).
- Hyp A: E(Y ) ≥ E(X) is equivalent to −(1 − α) ≥ 0, i.e. α = 1.
It means that Hyp A is possible only if all agents are ”perfectly altruistic” i.e. they
worry as much for the other people than for themselves. If α = 1, the first case of

11
 ACCEPTED MANUSCRIPT

Proposition 3 applies. The treatment is covered by social insurance if and only if

V > V ∗ (G, H). It means that the low cost treatments are covered only if the agents
are perfectly altruistic.
- Hyp B1: E(Y ) < E(X) and E(Y 2 ) > E 2 (X), where: 
P
E(Y 2 ) − E 2 (X) = (1−α)+nα1
(1 − α)p2m + α ki=1 ni p2i − p2

PT
P
((1 − α) + nα) (E(Y 2 ) − E 2 (X)) = (1 − α)p2m + α ki=1 ni p2i − (1 − α)p2 − nαp2
= (1 − α)(p2m − p2 ) + nαE(X 2 ) − nαE 2 (X) = (1 − α)(p2m − p2 ) + nαV (X)

RI
= α (nV (X) + (p2 − p2m )) − (p2 − p2m )
(p2 −p2m )
i.e. E(Y 2 ) > E 2 (X) ⇔ α > nV (X)+(p

SC
2 −p2 )
m i h
(p2 −p2m )
It means that hyp B1 is true iff α ∈ nV (X)+(p2 −p2 ) ; 1 , i.e. if the degree of
m
altruism is sufficiently high (and lower than 1). In that case, a treatment is covered

NU
by the social insurance if its utility is high (V > V ∗ (G, H)) and its cost is not too
low (H > H ∗ (G)).
(p2 −p2m )
- If α < nV (X)+(p 2 −p2 ) , then the degree of altruism is too low: hyp B2 is true, no
m
MA
treatment will be covered by the social insurance.

3.4 An alternative specification

D

We have studied up to now a model with a fixed V and varying p. It is appropriate

TE

for diseases like heart attack, when the probability clearly depends on the person.
It can be interesting to discuss an alternative specification2 , i.e. with a fixed p and
varying V . This can be useful for diseases like influenza, when p is quite uniform in
P

the population, since having influenza does not depend very much on the individual.
CE

However V is not fixed, because the consequences of influenza can be much more
serious for old people than for young healthy people.
Let us examine our model with varying V and fixed p. The expected utility
AC

without treatment is

f1 (V ) = U(y) − pV
W
f2 (V ) = U (y(1 − θ(V )) with
With a private insurance, the expected utility is W
pH
the budget constraint yθ(V ) = pH, i.e. θ = y . Thus

f2 = U(y − pH)
W

does not depend on V .

fsoc = U (y(1 − t)) where yt = pH since p is
With a public insurance, we have W
a constant. Thus
fsoc = U(y − pH)
W
There is no difference between public and private insurance here.
Proposition 1 becomes then: W f2 > W f1 (V ) ⇔ V > 1 [U(y) − U(y − pH)], i.e.
p
people with a high V want an insurance.
2
I would like to thank an anonymous referee for suggesting this alternative specification.

12
 ACCEPTED MANUSCRIPT

Note that here high risk people (i.e. with high V ) want to be treated, since the
cost of the treatment does not depend on V . On the contrary, with a varying p and
fixed V , the high risk people (i.e. with high p) do not want to be treated, because
the premium depends on p.
Who is in favor of a social system here? An agentof utility loss V prefers

PT
social insurance if and only if W fsoc ≥ max W f2 ; W
f1 (V ) , where W
fsoc = W f2 . It
means that for V < 1p [U(y) − U(y − pH)], the agent prefers not to be treated, since

RI
f1 (V ) > W
W f2 = W fsoc . If V > 1 [U(y) − U(y − pH)], the agent wants to be treated
p
but is indifferent between social and private insurance. So some people are against

SC
the social system, the others are indifferent. Thus the social system cannot be
adopted if the agents are uniquely differenciated by V .

NU
4 Equity and political decision
MA
We want now to evaluate if a choice taken according to the political decision pro-
cedure as above, is compatible with the equity goals of public insurance. It means
comparing a positive result, the treatments effectively covered by social insurance,
with a normative analysis. We do not want to determine here if the political deci-
D

sion has been taken with the aim of being fair, but if its result can be considered
as fair or not. We will mention briefly two classical conceptions of economic justice:
TE

utilitarianism and egalitarian liberalism (Rawls).

- According to a strictly utilitarian conception, the fairest decision would max-
P

imizeR a social welfare function obtained by summing the welfare of all individuals,
1
i.e. 0 W (p)dF (p). The comparison public-private is reduced to the study of the
CE

sign of K(F, H, V ). It is then fair to choose public insurance only when K(F, H, V )
is positive, i.e. when V > V ∗ (F, H) (according to Proposition 3.1), with G = F and
Y = X). In other words, from a strictly utilitarian point of view, public insurance
AC

must cover the treatments of sufficiently good output. If the output is low, financing
public insurance imposes too many sacrifices on the low risk agents, with few gains
for the high risk agents. In the case of a lethal but curable disease, the utility loss
V is infinite, and H is still finite thus V ∗ (F, H) is finite: justice requires that any
lethal illness be covered by the public system. We note that under HypB1 or B2,
the low cost treatments (H < H ∗ (G)) are not covered, even those of high utility. It
is not fair from a utilitarian point of view: any treatment must be covered by the
social insurance, if it is sufficiently useful.
It is clear that if we compare
R 1 the utilitarian ethics with the result coming from
the function D(G, H, V ) = 0 W (p)dG(p), the nearer G is to F (i.e. any person
has almost the same weight in D), the fairer the decision taken. If F stochastically
dominates G, the high risk agents are too neglected, so public insurance is too rarely
adopted. If G stochastically dominates F , they are too favored, public insurance is
too often adopted, to the detriment of the consumption level of the low risk agents.
- With a Rawlsian approach, we must forget our real situation, to consider the
worst: the fair decision must maximize the welfare of the less favored. Here it
means caring only about the welfare of the agents of maximum risk pmax . According

13
 ACCEPTED MANUSCRIPT

to Proposition 3, the fair choice leads to adopting public insurance when V >
V ∗ (Fmax , H), where Fmax is the cumulative function of the distribution putting all
the weight on p = pmaxR 1 . If we compare the Rawlsian approach to the decision coming
from D(G, H, V ) = 0 W (p)dG(p), this one will be perfectly fair for G = Fmax . For
a lower G, public insurance is less often adopted, and the highest risk agents are

PT
harmed.
The Rawlsian approach leads more often to social insurance than the strictly

RI
utilitarian approach, because V ∗ (Fmax , H) < V ∗ (F, H).
For a given cost H, a treatment (H, V ) must be socially covered if its utility is:
- high: V > V ∗ (F, H) for the utilitarian ethics.

SC
- high or medium: V > V ∗ (Fmax , H) for the Rawlsian ethics.

NU
5 Conclusion
We have considered an economy of agents differentiated by their risk of sickness. We
MA
have studied the individual preferences concerning health insurance, with respect to
that risk. When the treatment of a disease is not covered by social health insurance,
the low risk agents take out a private insurance, and the high risk agents may prefer
not to be treated, to avoid a too dramatic decrease of their consumption (Proposition
D

1).
TE

If the treatment has a high cost-utility output, every agent with a risk above the
average wants social insurance, but the other agents reject it. If the output is less
good, only the very high risk agents want social insurance (Proposition 2).
P

With a majority of low risk people, social insurance is rejected in direct voting if
CE

the agents are selfish. However, in a representative democracy, or in direct democ-

racy if the agents are altruistic, social insurance can be adopted. It will be more
easily adopted if:
AC

- the output of the treatment is high

- the agents are very altruistic
- with a representative democracy, the political weight of the high risk agents is
great.
If now we wonder if the decision is fair, we can say that
- from a utilitarian point of view, the decision will be perfectly equitable (for all
the treatments possible) if every agent has the same political weight, or if the agents
are perfectly altruistic.
- with a Rawlsian criterion, the decision will be equitable if it is the one preferred
by the high risk agents.

References
Blomquist S., Christiansen V., 1999. The political economy of publicly provided
private goods. Journal of Public Economics 73, 31-54.
Coughlin P., Mueller D.C., Murell P. 1990., Electoral politics, interest groups and
the size of government. Economic Inquiry 28, 682-705.

14
 ACCEPTED MANUSCRIPT

Epple D., Romano R.E., 1996. Public provision of private goods Journal of Political
Economy, 104 (1), 57-84.
Epple D., Romano R.E. 1996. Ends against the middle: determining public service
provision when there are private alternatives Journal of Public Economics 62, 297-
325.

PT
Gouveia 1997 Majority rule and the public provision of a private good Public Choice
93, 221-244

RI
Hindriks J., 2001 Public versus private insurance with dual theory: a political econ-
omy argument. The Geneva Papers on Risk and Insurance Theory 26, 225-241.

SC
Hindriks J., De Donder P. . The politics of redistributive social insurance’. Journal
of Public Economics 87, 2639-2660.
Meltzer A. and Richards 1981, ‘A rational theory of the size of government’, Journal

NU
of Political Economy 89, 914-927.
Persson T., Tabellini G. 2000. Political economics, explaining economic policy. MIT
Press, Cambridge. MA
D
P TE
CE
AC

15
 ACCEPTED MANUSCRIPT

Appendix
A Proof of Proposition 1.

PT
We set
1
V (p, H) = [U(y) − U(y − pH)]
p

RI
We have just to show that V is an increasing function of p and H. We see that
V is clearly an increasing function of H. Let us show that it is also an increasing

SC
function of p.
∂V
∂p
= p12 [pHU ′ (y − pH) − U(y) + U(y − pH)]
Let µ(p) = pHU ′ (y − pH) − U(y) + U(y − pH)

NU
µ′ (p) = HU ′ (y − pH) − pH 2 U”(y − pH) − HU ′ (y − pH)
= −pH 2 U”(y − pH) > 0 because U is strictly concave.
µ(0) = 0 then µ(p) > 0, for any p > 0, thus ∂V > 0 if p > 0.
MA ∂p
V is then an increasing function of p.
∂2V
Moreover ∂H 2 = −pU”(y − pH) > 0, i.e. H 7→ V (p, H) defines a convex function.
D

B Proof of Proposition 3.
TE

a) First we show that V 7→ K(G, H, V ) is a non decreasing function.

We have
P

K(G, H, V ) = Dsoc (G, H, V ) − Dind (G, H, V )

CE

Z 1 Z 1
= Wsoc (p)dG(p) − Wind (p)dG(p)
0 0
Z 1
AC

= U(y − pH) − max Wj (p)dG(p)

0 1≤j≤2

Then we obtain three possible writing for K(G, H, V ) according to Proposition 1:

(i) if V > V (1, H), then
Z 1
K(G, H, V ) = U(y − pH) − W2 (p)dG(p)
0
Z 1
= U(y − pH) − U(y − pH)dG(p) (11)
0

K(G, H, V ) is independent of V.
(ii) if V ∈]V (0, H); V (1, H)[, then
Z p∗ Z 1
K(G, H, V ) = U(y − pH) − W2 (p)dG(p) − W1 (p)dG(p)
0 p∗
Z p∗ Z 1
= U(y − pH) − U(y − pH)dG(p) − [U(y) − pV ] dG(p)(12)
0 p∗

16
 ACCEPTED MANUSCRIPT

Let us show that K(G, H, V ) is an increasing function of V on ]V (0, H); V (1, H)[.
If we assume that (V1 , V2 ) ∈]V (0, H); V (1, H)[2 , with V1 < V2 , then:
Z p∗ (H,V2 )
K(G, H, V2 ) − K(G, H, V1 ) = − U(y − pH)dG(p)
0

PT
Z 1 Z p∗ (H,V1 ) Z 1
− [U(y) − pV2 ] dG(p)+ U(y−pH)dG(p)+ [U(y) − pV1 ] dG(p)
p∗ (H,V2 ) 0 p∗ (H,V1 )

RI
Z p∗ (H,V2 ) Z 1
= [U(y) − U(y − pH) − pV1 ] dG(p) + (V2 − V1 ) pdG(p)

SC
p∗ (H,V1 ) p∗ (H,V2 )

where [U(y) − U(y − pH) − pV1 ] = p [V (p, H) − V1 ] > 0 for p ∈ [p∗ (H, V1 ); p∗ (H, V2 )],
since V (p, H) is an increasing function of p. Thus K(G, H, V2 ) − K(G, H, V1 ) > 0,

NU
and K(G, H, V ) is an increasing function of V on ]V (0, H); V (1, H)[.
(iii) if V < V (0, H), then
Z 1
MA
K(G, H, V ) = U(y − pH) − max Wj (p)dG(p)
0 1≤j≤2
Z 1
= U(y − pH) − [U(y) − pV ] dG(p)
0
D

Z 1
= U(y − pH) − U(y) + V pdG(p) (13)
TE

0
∂K
R1
∂V
(G, H, V )= 0
pdG(p) > 0
P

We can then say that K(G, H, V ) is a non decreasing function of V on the three
CE

studied domains: V > V (1, H), V ∈]V (0, H); V (1, H)[, and V < V (0, H).
The function K is clearly continuous on these domains. Let us show that it is
continuous also at V = V (0, H) and V = V (1, H).
AC

R1
limV →V (1,H)− K(G, H, V ) = U(y − pH) − 0 U(y − pH)dG(p)
= limV →V (1,H)+ K(G, H, V )
then K is continuous at V = V (1, H).
R1
limV →V (0,H)+ K(G, H, V ) = U(y − pH) − 0 [U(y) − pV ] dG(p)
= limV →V (p1 ,H)− K(G, H, V )
then K is continuous at V = V (0, H).
K is then continuous everywhere, thus it is an increasing function of V (strictly on
V < V (1, H)).

b) Now we show that K(G1 , H, V ) ≤ K(G2 , H, V ) if G2 stochastically dominates

G1 , i.e. if G2 (p) ≤ G1 (p) for any p ∈ [0; 1].
(i) if V > V (1, H), then according to Equation (11), K(G1 , H, V ) ≤ K(G2 , H, V ),
because p 7→ −U(y − pH) is an increasing function.
(ii) if V ∈]V (0, H); V (1, H)[, then according to Equation (12), K(G1 , H, V ) ≤
K(G2 , H, V ), because p 7→ −U(y − pH) and p 7→ − [U(y) − pV )] are increasing
functions.

17
 ACCEPTED MANUSCRIPT

(ii) if V < V (0, H), then according to Equation (13), K(G1 , H, V ) ≤ K(G2 , H, V ),
because p 7→ − [U(y) − pV )] is an increasing function.

c) We study now the sign of K with respect to G.

We assume that U is an increasing, concave, quadratic function on ]0; y]. So we have

PT
for any x ∈ [0; y[, U(y − x) = a0 − a1 x − a2 x2 , where a0 = U(y), and a1 = U ′ (y) > 0
and a2 = −12
U”(y) > 0.

RI
• First we set V > V (1, H). R
1

SC
K(G, H, V ) = U(y − pH) − 0 U(y − pH)dG(p)
Recall that we denote by X (respectively Y ) a random variable of cumulative
R1
distribution function F (respectively G). We have p = 0 pdF (p) = E(X) and
R1

NU
0
pdG(p) = E(Y ).
2
R1
K(G, H,RV ) = a0 − a1 (pH)
 − a2 (pH)
R − 0
[a0 − a1 (pH) − a2 (pH)2 ] dG(p)
1 1
= a1 H 0 pdG(p) − p + a2 H 2 0 p2 dG(p) − p2
MA
= a1 H (E(Y ) − E(X)) + a2 H 2 (E(Y 2 ) − E 2 (X))
- Under hypothesis A, the high risks have a political weight which is greater than
their demographic weight, i.e. E(Y ) > E(X). This implies that E(Y 2 )−E 2 (X) > 0,
D

since E(Y 2 ) ≥ E 2 (Y ) > E 2 (X).

Thus K(G, H, V ) > 0 for any H > 0.
TE

- Under hypothesis B1, we have E(Y ) < E(X) but E(Y 2 ) > E 2 (X)
Then K(G, H, V ) > 0 ⇔ a2 H (E(Y 2 ) − E 2 (X)) > a1 (E(X) − E(Y ))
P

 
∗ ∗ a1 E(X)−E(Y )
i.e. K(G, H, V ) > 0 ⇔ H > H (G) where H (G) = a2 E(Y 2 )−E 2 (X) > 0
CE

- Under hypothesis B2, we have E(Y ) < E(X) and E(Y 2 ) ≤ E 2 (X)
Thus K(G, H, V ) < 0 for any H > 0
AC

• Now we set V < V (0, H).

R1
K(G, H, V ) = U(y − pH) − 0 [U(y) − pV ] dG(p)
So we have K(G, H, 0) = U(y − pH) − U(y) < 0
We can conclude then the proof of Proposition 3:

- Under Hyp A, K(G, H, V ) > 0 if V > V (1, H)

and K(G, H, 0) < 0. Moreover, V 7→ K(G, H, V ) is a continuous, increasing
function on V ∈ [0; V (1, H)], then there exists V ∗ (G, H) such that :
K(G, H, V ) > 0 if V > V ∗ (G, H)
K(G, H, V ) < 0 if V < V ∗ (G, H)

- Under Hyp B1,

- if H > H ∗ (G), then K(G, H, V ) > 0 for V > V (1, H)
and K(G, H, 0) < 0. Moreover, V 7→ K(G, H, V ) is a continuous, increasing
function on V ∈ [0; V (1, H)], then there exists V ∗ (G, H) such that :

18
 ACCEPTED MANUSCRIPT

K(G, H, V ) > 0 if V > V ∗ (G, H) and H > H ∗ (G)

K(G, H, V ) < 0 if V < V ∗ (G, H) and H > H ∗ (G)
- if H < H ∗ (G), then K(G, H, V ) < 0 for V > V (1, H)
thus K(G, H, V ) < 0 for every V if H < H ∗ (G).

PT
- Under Hyp B2,
K(G, H, V ) < 0 for V > V (1, H)

RI
thus K(G, H, V ) < 0 for every V

SC
NU
MA
D
P TE
CE
AC

19
 ACCEPTED MANUSCRIPT

Figure captions:
Figure 1: who takes out private insurance?

PT
Figure 2: who wants social insurance?

Figure 3: Hyp A

RI
Figure 4: Hyp B1

SC
NU
MA
D
P TE
CE
AC

20
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
ED
PT
CE
AC