. 5
( 11)


could have used years of formal schooling as a measure of g, but the results
are very similar.

Information Gauged by a question that asked people to declare their
degree of agreement with the following statement: “I feel that I have a pretty
good understanding of the important political issues facing our country.”
Respondents could indicate ¬ve levels of agreement: 1, strongly agree; 2,
agree; 3, neither agree nor disagree; 4, disagree; 5, strongly disagree. The
variable was reversed so that higher values measure more information.

Left“Right Position This variable is based on the classi¬cation of parties
from left to right developed by the International Social Survey Program to
facilitate comparison of party support across countries. Individual parties
are classi¬ed as follows (data on party support are not available for Italy).

National unemployment The standardized rate of unemployment at the
time of the national surveys (1996 unless noted otherwise below) minus the
OECD rate of unemployment at that time (the subtraction eliminates prob-
lems of multicollinearity while leaving the substantive results unaltered).

Left, Right,
Far Center Center, Conservative Far
left (1) Left (2) Liberal (3) (4) Right (5)
Greens Labour Democrats Liberal Party
Labour Liberal Conservatives
Communists NDP, Bloc PC, Liberal Reform Party
Quebecois, Party

Political Foundations of Social Policy

Left, Right,
Far Center Center, Conservative Far
left (1) Left (2) Liberal (3) (4) Right (5)
Communists, Socialist UDF RPR National Front
Far Left Party
PDS SPD, Greens FDP CDU/CSU Republicans
Worker™s Fianna Fail, Progressive
Party, Sinn Fine Gael, Party
Fein, Labour,
Democratic Greens
Red Alliance Labor, Christian Conservatives,
Socialist Left Democrats, Progress Party
Center Party,
Liberal Party
Alliance Labour New Zealand National Party
Labor, Center Party, Conservatives
Socialists Liberals,
Democrats Independent Republicans
Source: OECD (2000). Unemployment rates were: Australia, 8.5; Britain, 8.2; Canada, 9.6; France
(1997), 12.3; Germany, 8.9; Ireland, 11.6; Italy, 11.7; Norway, 4.9; New Zealand (1997), 6.7; Sweden,
9.6; United States, 5.4.

Appendix 3.D. Statistical Appendix
A problem arises in the use of s3 and s4 as explanatory variables. (Because
it is the same in both cases, it will simply be referred to as s.) Because the
question used as the basis for s was asked only in the 1997 survey, whereas
all the questions about spending were asked only in the 1996 survey, it
was necessary to “translate” the 1997 information on s so that it could be
used in the 1996 survey. For this purpose, averages for s were calculated
at the three-digit ISCO-88 level in the 1997, and then these values were
assigned to individuals in the 1996 survey based on their three-digit ISCO
Explaining Individual Social Policy Preferences

classi¬cation in that survey. It is shown here that the estimated coef¬cient
of b is consistent but has an approximate small sample bias that biases down
the estimated coef¬cient toward zero if b > 0, and biases it upward toward
zero if b < 0.
The structural model is

Ri, j = k + b · yi, j + c · s i, j + µi, j
96 96 96 96

where each observation is drawn from the 1996 survey and where i indexes
the ith individual in the j th ISCO three-digit level occupation group. There
96 96
are no data on s i, j . Assume s i, j is generated by the process

s i, j = s j + ·i, j
96 96

where s j is exogenous. s j itself is unobservable, but data are available from
the 1997 survey generated by the same process:

s i, j = s j + ·i, j
97 97

E·i, j = 0 ∀i, j, x

E·i, j · ·r, j = 0 = E·µ ∀i, r, j
96 97

E·2 = σ· ; Eµ 2 = σµ2

that is, ·ij and ·ij can be thought of as random drawings from the same

distribution. We now run the regression

Ri, j = k + b yi, j + c s j + µi, j + …i, j
96 96 96 96

s i, j
sj ≡ …i, j ≡ c s i, j ’ s j
96 96
where and (A3.14)
where Nj is the number of individuals in ISCO category j in the 1997
survey. From (A3.12) and (A3.13)
·i, j
…i, j = ·i, j ’
96 96
The exposition can be simpli¬ed considerably by assuming that there is no
correlation between s and y. This implies
Eb = a « 
s j · ·j
¬ ·
E cˆ = c 1 ’ E  (A3.16)

Political Foundations of Social Policy

By making the appropriate probability limit assumptions, it is not dif¬cult to
show that cˆ is a consistent estimator of c . We can get a better insight from
s .·/ s 2
the expectation of the exact ¬rst-order Taylor expansion of
around the expected values of numerator and denominator:

s· s· 1
≈E + · s· ’ E s·
s2 s2 s2

’ · s2 ’ E s2
E s

Let there be J ISCO categories and assume for convenience that Nj =
N ∀ j . Then this approximation produces
« 
¬ ·
E cˆ = c ¬1 ’ · (A3.17)
 2
s j
σ· + N ·
Since J is constant, (A3.17) tells us ¬rst that as N increases the approximate
bias goes to zero. Second and more importantly, it implies that for a small

c < E cˆ < 0 if c <0
c > E cˆ > 0 if c >0 (A3.18)

Finally, it implies that

If c = 0 ’ E cˆ = 0 (A3.19)

and the standard signi¬cance tests hold.

Explaining Individual Social Policy Preferences

Appendix 3.E. Multilevel Model
Write the level 2 observation on individuali in economy j as
Ri j = γ + ·yi j + µs i j + Xij δ + µi j
where Xij is the vector of controls, and de¬ne level 1 by the ¬xed ef-
fects model · = · + ·1 U j and µ = µ + µ1 U j . The single-stage regression
model is derived by substituting the level 1 model into level 2:
Ri j = γ + ·yi j + ·1 yi j U j + µs i j + µ1 s i j U j + X ij δ + µi j
Making the assumption that Uj s are exogenous and that this is a nonrandom
effects model implies that the single-stage multilevel equation conforms to
the standard ordinary least squares conditions.


Credible Commitment, Political
Institutions, and Social Protection—

In Chapter 3, peoples™ preferences for social protection were explained as
a function of the level and composition of human capital assets (i.e., as a
function of income and skill speci¬city). Any theory that seeks to understand
collective choice in democratic societies must begin with an account of
individual preferences. But we know from the seminal works of Arrow,
Olson, North, Shepsle, and Weingast that the aggregation of preferences
into public policy is anything but straightforward. Indeed, preferences may
never get translated into policies, even when a single (median) voter is
decisive in electoral competition.
One fundamental problem in the provision of social protection arises
because current pivotal voters choose policies that yield bene¬ts to them
only at some future point in time when these same voters are no longer
pivotal. This poses a problem because current voters can only commit the
government for one term at a time and because there is no way to bind future
voters to the policy preferences of current voters (for a similar logic, see
Franzese 2002, Ch. 2). This dilemma is referred to in this chapter as the
time-inconsistency problem in social policy provision, and it is shown that it can
lead to serious underprovision of social protection compared to the long-
term preferences of voters. This is particularly true in speci¬c skill systems
because the underlying demand for protection is higher.
The time-inconsistency problem is closely related to another incomplete
contracting problem in democratic politics. Political parties present policy
platforms in order to win elections, but they are simultaneously represent-
ing party-internal constituents who may not share the policy preferences

— This chapter builds on two unpublished papers with David Soskice (Iversen and Soskice
2002, 2004).

Credible Commitment, Political Institutions, and Social Protection

of the voters who are being courted in the election. This problem of repre-
sentation is particularly severe in majoritarian systems where the support of
the median voter is critical to win the election, but it is also severe where
parties™ core constituents can be expected to have preferences that are clearly
distinct from the median voter™s. Because the electoral platform is not an
enforceable contract, the median voter must worry about the actions that a
party may take after it wins the election. As we will see later in this chapter,
this affects the voting behavior of strategic voters and leads to outcomes
that are systematically different from the preferences of the median voter.
Two arguments are advanced in this chapter to show how these contract-
ing problems may be (partially) solved. The ¬rst is that political parties with
detailed policy programs and highly developed party organizations, espe-
cially links to unions, limit the ability of leaders to give in to short-term
electoral incentives and constrain the choice set of voters to alternatives
that are optimal in the long run (see also Franzese 2002, Ch. 2). These
mechanisms of commitments can work even when parties have the option
of adopting ¬‚exible organizations and even when third parties are allowed
to enter into the electoral competition with any platform or organization
they desire. However, in majoritarian systems, this solution magni¬es the
problem of representation because parties will deviate from the policy pref-
erences of the median voter. The result is that parties in majoritarian systems
have a strong incentive to adopt parties with strong leaders at the expense
of the ability to commit to long-term investment in social protection. This
incentive is weaker under PR where it is not critical for parties to win the
support of the median voter.
In addition to nurturing programmatic and responsible parties, PR elec-
toral systems also give centrist parties an incentive to ally with left parties
for purposes of redistribution. The reason is that the poor and the mid-
dle class have a common interest in taxing the rich and distributing the
revenues among themselves. As I have argued already, such redistribution
in turn serves insurance functions because those who are experiencing a
complete or partial unemployment of their assets as a result of adverse la-
bor market conditions will also bene¬t. Like wage protection through the
collective wage bargaining system, redistributive social spending serves as
a protection of income.
Majoritarian electoral systems are different because governments are
not (typically) chosen through coalition bargaining but directly by the elec-
torate. Whichever party gets a plurality wins, which, as we know well, places
the median voter in a very strong position. This is where the incomplete
Political Foundations of Social Policy

contracting problem in representation matters. If parties have a nonzero
probability of deviating from their electoral platforms after they are in
power, the median voter has an incentive to vote for the center-right party
because the median voter shares with the rich a common interest in limiting
redistribution to the poor. Hence, although both parties have an incentive
to cater to the median voter during elections, their incentive to deviate from
the platform after the election will make the center-right party more attrac-
tive to the median voter. This undermines spending for both redistributive
and insurance purposes.
This chapter is organized into three sections. The ¬rst section provides
a precise de¬nition of the time-inconsistency problem in the context of
social policies. A model of parties and voters that is then presented, which
shows the conditions under which the problem can be overcome. The next
section develops the electoral system argument, while the third explores
both arguments using micro-level electoral data as well as macro-level
data on institutions, government partisanship, and redistribution. In
equilibrium, electoral and party systems with high institutional capacity for
commitment (PR with programmatic parties) are always found in political
economies where speci¬c skills are important, whereas the opposite
(majoritarian systems with leadership-dominated parties) are always found
in general skills systems.
The broader ambition of this chapter is to tie economic institutions and
behavior to democratic institutions. Even though the varieties of capitalism
literature have explored the complementarities that link economic institu-
tions together, little theoretical or empirical energy has gone into exploring
the linkages between economic and political institutions. Here I focus
on the key institutions shaping democratic politics “ political parties and
electoral systems “ and show how they are linked to the production system.

4.1. The Time-Inconsistency Problem
Assume that all individuals have identical preferences and face the same
risk of a wide range of adverse events occurring, including ill health, unem-
ployment, loss of employment that uses workers™ speci¬c skills, and so on.
Against some of these events, individuals can insure privately; against oth-
ers, market failure (such as moral hazard) rules private insurance out. Under
what circumstances will a democratically elected government implement
the will of the majority to provide appropriate public income protection?
Credible Commitment, Political Institutions, and Social Protection

Elections take place at regular intervals with government terms lasting T
years. To simplify the argument, consider the following set of assumptions,
which expose the bare bones of the problem. At the start of each term, indi-
viduals have a probability p of some adverse event happening to them during
the T-year period. It is simple to think of this as a spell of unemployment,
although it can refer to any event that adversely affects the earnings capacity
of a worker, such as a skill becoming obsolete. In terms of the model in Cha-
pter 3, it refers to unemployment of any asset that the individual possesses.
After the election, it is revealed which individuals will be unemployed
during the electoral term. Finally, voters vote in a ¬rst-past-the-post-
election for one of two parties; each party™s platform is a tax t on those
who ¬nd themselves employed to ¬nance an unemployment bene¬t R =
(1 ’ p) · t/ p (given to those who have found themselves unemployed). The
goal of the parties is to win power, and the winning party is committed to
carrying out its platform. There is complete information on the side of vot-
ers and parties. This sequence (revelation of employment or unemployment
across individuals, followed by voting) is repeated at the start of each period.
Assume that p < .5, so that the median voter is employed. What level
of t will the median voter choose? If the employed individual receives a
pretax income of 1, then the expected utility of an individual at time „ = 0
before the individual knows whether or not he is unemployed is

δ „ [(1 ’ p) · u (1 ’ t„ ) + p · u(t„ · (1 ’ p)/ p)]
V0 ≡ (4.1)
„ =0

where δ ∈ [0, 1] is the discount factor and u(x) is the utility derived from
net receipts (income or unemployment bene¬t). De¬ne t —— as the value
of t , which maximizes (4.1) under the assumption that t„ has the same value
for all „ = 0, . . . , ∞ . Assume the conditions exist on u (·) to guarantee an
interior maximum. The example used in this chapter is u(x) = log x. In this
case, t —— = p.1 So the preferred level of taxation is directly proportional to
the risk of being unemployed.
Now consider a median voter at „ = 0 who knows that he or she is
employed. His or her expected utility is given by

δ „ [(1 ’ p) · u(1 ’ t„ ) + p · u(t„ · (1 ’ p)/ p]
V0 ≡ u(1 ’ t0 ) + (4.2)
„ =1

If t is constant for all „, V 0 = δ ’1 [(1 ’ p) log(1 ’ t) + p log((1 ’ p)t/ p))], which implies

‚ V0 1’ p 1’ p
t —— = p
’ = · or
‚t 1’t [(1 ’ p)·t]/ p p

Political Foundations of Social Policy

De¬ne t — as the value of t that maximizes (4.2) under the same assumptions
as before, and we ¬nd that t — = δp. Note that t — < t —— because there is
no possibility that the employed median voter requires an unemployment
bene¬t in „ = 0 . For example, if the probability of unemployment is, say,
5 percent, and the length of of¬ce is 4 years, δ (the discount factor) would
be around 0.8 and, hence, t — would be 80% of t —— .
Yet, the real problem arises because the median voter only chooses the
unemployment bene¬t for the current electoral period, „ = 0, that is t0 . For the
next period „ = 1, there will be a new median voter choosing t1 . If the
median voter in any one period is sincere, the vote in that period has no
effect on the votes of median voters in subsequent periods; current voters
cannot commit future voters. Hence, the median voter must take all future
tax rates as given. It can then be seen from (4.2) that the employed median
voter at „ = 0 chooses t0 to maximize

V0 ≡ log(1 ’ t0 ) (4.3)

implying an optimal choice of t0 = 0. And because each subsequent
median voter will also be employed, we see that the electoral system will
produce t„ = 0, for all „ ≥ 0, if parties are short-term vote maximizers or
there is free entry of parties. This is the fundamental time-inconsistency
problem of majoritarian democracy with periodic elections to which this
chapter draws attention. Although it would be optimal for the current
and for all future median voters to set t = t — = ‚ p (or t —— = p behind the
veil), the lack of any mechanism whereby the current median voter can
commit future median voters implies that each median voter will vote
for t = 0.

Proposition 1. The time-inconsistency problem. Given (i) the prefer-
ences of individuals de¬ned by (4.1), (4.2), and (4.3); (ii) sincere voting; and
(iii) free candidate entry and/or short-term parties: (a) the optimal ex-ante
tax rate is p, (b) the optimal tax rate at any point in time for the current and
all future median voters is δp, and (c) the equilibrium tax rate is 0.

The time-inconsistency problem as de¬ned here is related to overlap-
ping generations models of public goods provision. In a standard overlap-
ping generations setup, public goods in these models are provided in such
a manner that noncontributing older generations bene¬t from the contri-
butions of younger generations. This would be similar to a situation in
the current model where an employed median voter supported transfers
Credible Commitment, Political Institutions, and Social Protection

to those currently unemployed. Yet, all existing overlapping generations
models fail to offer solutions to the key time-inconsistency problem of
interest. In Browning (1975), median voters choose the level of social in-
surance, but because the insurance scheme is in effect for the rest of their
lives, time inconsistency is assumed away. Hu (1982) presents a three-period
overlapping generations model of social security where the current median
voter has some uncertainty about the future level of insurance. But because
uncertainty is a random variable with a mean that depends on past levels,
the median voter is in effect always making choices for both present and fu-
ture generations. Likewise, while Broadway and Wildasin (1989) correctly
note that the current median voter™s choice depends on the choice of future
median voters, they then go on to assume that the median voter chooses a
level of insurance that will remain in place for the rest of his or her lifetime.
Other ways around the problem have been to assume that different genera-
tions can write binding contracts with one another, as in Kotlikoff, Persson,
and Svensson (1988), or that governments weigh the preferences of each
generation regardless of which generation contains the median voter, as in
Grossman and Helpman (1998).
A more satisfactory class of models derives the solution endogenously
by allowing a possibility of punishment for noncooperation. In this setup,
members of a group in the overlapping generations model make continuous
decisions about whether to contribute to the public good or not, and this
creates opportunities for decentrally enforced cooperation in in¬nitely lived
groups (Dickson and Shepsle 2001; Rangel 2003). Yet, while this solution
makes sense in the context of some overlapping generations problems, it
makes little sense in our context. The reasons are that a voter is unlikely
to choose the level of insurance more than once in a lifetime and that
present and future median voters do not know the identity of one another.
Because of this, the decision of the median voter is equivalent to a member
deciding whether to cooperate in the last period of a ¬nitely lived group
in an overlapping generations model.2 We, thus, need an entirely different
solution. In the following, I focus on the key role played by political party
organizations and electoral systems.

2 A similar problem applies to the model proposed by Bawn (1999). Analogously to an over-
lapping generations model, she considers a situation where some members of a group are
bene¬ting from the effort of others, and where the game is repeated inde¬nitely. In this
game, cooperation by reciprocity is a possibility, but the solution is again highly implausible
for my purposes because it requires present and future median voters to know each other
and to transition repeatedly in and out of the median voter position.

Political Foundations of Social Policy

4.2. Two Quali¬cations
In reality, the problem is not as stark as stated in Proposition 1 for two rea-
sons. First, because the median voter at „ = 0 has a risk of unemployment
in the next electoral period, even if the voter knows that he or she is cur-
rently employed, that voter will demand insurance against that risk from the
currently elected government. That implies a positive tax rate. Yet, there
is still a time-inconsistency problem so long as p|E < p|U, where p|E is
the probability of unemployment in the next electoral period when currently
employed and p|U is the probability of unemployment in the next electoral
period when currently unemployed. The reason is that the optimal level of
insurance is chosen for a situation when the voter does not know whether
he or she is employed or unemployed (i.e., is behind the veil), which is
de¬ned by a probability somewhere between p|E and p|U. In general, the
difference between the risk before and after the veil is removed de¬nes the
severity of the time-inconsistency problem, and it will vary by policy area.
In the case of old-age insurance, for example, the median voter, who is
likely to be a middle-aged person, will know for certain that the condition
for collecting the bene¬ts will be zero in the following electoral period. For
most other types of social insurance, including unemployment insurance,
the risk is lower unveiled than behind the veil.
Second, and importantly, many bene¬ts are going to those who are both
employed and unemployed. Even when bene¬ts are targeted to the em-
ployed, to the extent that they reduce variability of earned income, what
I have referred to as income protection, they serve insurance purposes. Be-
hind the veil, spending on such insurance suffers from exactly the same
time-inconsistency problem as identi¬ed earlier. Unveiled, however, in-
come protection can be supported for redistributive reasons. The reason
is simple and goes back to the preference model in Chapter 3. If work-
ers discount insurance as a result of the time-inconsistency problem, the
demand for spending is equal to the demand for redistribution. In the
Meltzer-Richard model, this demand is represented by the median voter™s
preference for redistributive spending and is given by Equation (3.10). The
lower the relative income of the median voter is, the greater the demand
for spending is.
If the translation of the median voter preferences were direct, we
would know from the asset model in Chapter 3 that spending would
rise with higher risk of unemployment and more skill speci¬city. The
time-inconsistency problem cuts into this type of spending, while leaving
Credible Commitment, Political Institutions, and Social Protection

redistribution-motivated spending intact. The difference between the op-
timal level of spending desired by the median voter and the actual level
of spending depends on the probability of unemployment and skills of the
median voter. But as this discussion illustrates, there are two potential so-
lutions to the problem: One is to bind the hands of the government across
time, the other is to shift the locus of political power from high to lower
income groups. I consider each in turn.

4.3. Institutionalized Parties
Assume that there is majoritarian voting and that voters vote sincerely.
Furthermore, assume that voters have assets that are at risk of becoming
unemployed and for which they seek protection. As in previous chapters,
workers™ main asset is their skills, and voters are indexed by the speci¬city
of these skills, s , precisely as in Chapter 3. Because this section focuses
exclusively on the insurance motive and the intertemporal tradeoff to which
it gives rise, differences in income are ignored. In the next section, variability
in skills and the risk of unemployment is ignored, and only relative income,
and, hence, contemporaneous redistribution, is considered. This enables a
clear focus on each of the two problems described in the introduction.
Skill speci¬city is uniformly distributed across voters on the interval
[s min , s max ],3 with s m being the skill level of the median voter. Individual vot-
ers will be referred to by their skill level s . Voters are divided into Low and
High ( L and H ) risk where s ∈ L when s < s m and s ∈ H when s > s m . s is
known throughout the game and cannot be changed. As before, the prob-
ability that voters will be unemployed during the coming period of gov-
ernmental of¬ce is p; whether they are employed or unemployed is re-
vealed to them at the start of the period; and after it has been revealed,
they vote. Again, it is assumed that p < .5 so that the median voter is
There are two parties at the start of each period. If a third party can enter
and win, that will occur. With sincere voting, an implication of this is that
tax-¬nanced unemployment bene¬ts will always be zero if that is the only
policy dimension (as demonstrated in the previous section).

We can, thus, classify the average degree of skill speci¬city in a polity by sm =

(smax + smin /2.
4 Instead of speaking of workers being unemployed, one could more generally talk about assets
being unemployed. In terms of developing the logic of the argument, it does not matter.

Political Foundations of Social Policy

However, governments do not only provide protection against unem-
ployment of assets. They also seek to produce a variety of public goods “
including training for workers, wage restraint, and international compet-
itiveness “ that are neither guaranteed by the market nor produced by
simple tax and spend policies. For the provision of these public goods,
governments depend on the cooperation of private economic agents.5 It
is assumed, therefore, that the provision of a public good, a, can only be
effective if subsets of voters give the parties adequate support to do so.
The way this is accomplished, it is argued, is through a process that will be
referred to as party institutionalization.

4.3.1. Party Institutionalization
The key intuition behind the argument is that an institutionalized party has
a long-term relationship with some group in the electorate characterized by
a common set of interests. In exchange for contributions and other resources
from the group, which are critical for the production of public goods, the
group has an in¬‚uence on the policy platform of the party enshrined in
the party™s explicit or implicit constitution. The party also has a leader who
can, in principle, impose his or her will on the party™s platform. But in
compensation, the group has a say in the election and reelection of leaders.
The degree of institutionalization can, thus, be seen as the relative power of
leader and group. This is essential to the solution of the time-inconsistency
problem because, without power over platforms and party leaders, the time-
inconsistency problem would always tempt parties to offer lower tax rates.
This conception of party institutionalization is quite similar to the ex-
isting literature on parties and corresponds to what is usually referred to as
responsible and programmatic parties. For example, in a classic analysis of
party organizations, Schlesinger (1984) conceives of parties as voluntary or-
ganizations producing collective goods. To be electorally successful, parties
rely on the work of rank-and-¬le members, and these in turn must be “com-
pensated” through in¬‚uence over policy and leadership selection. Platform
control, in other words, is traded for rank-and-¬le support. In Schlesinger™s
conceptualization, however, the incentives of individual members to con-
tribute is problematic because policies represent collective goods. In the

5 There may be other types of public goods, such as a civil service or even a general educational
system, that do not require cooperation from private groups. The provision of these are
theoretically unproblematic and not considered here.

Credible Commitment, Political Institutions, and Social Protection

present model, the supporters are collective actors with the ability to use
selective incentives to solve internal collective action problems. This con-
ception of parties is very close to that in Aldrich (1995).
To provide greater precision to the argument, assume that institutional-
ized parties have a constitution. The constitution lays down that members
pay a fee in exchange for which they can nominate the party™s platform.
“Fee” is used here in the broad sense of any sacri¬ces made by a party™s
constituency to facilitate the provision of a public good in exchange for
policy in¬‚uence. The party leader is not bound to accept this platform, but
if it is rejected, the event , the leader™s probability of reselection, q , at the
end of the period is lower than the probability of reselection if the leader
accepts the platform, which is assumed to be 1. The difference 1 ’ q is
called the degree of institutionalization of the party because it measures the
degree to which the party constituency can control the leader “ or inversely
the leader™s degree of discretion over policy.
Assume that 1 ’ q is an increasing function of membership fees (again,
in the broad sense suggested earlier) and that these are determined by the
costs of the public goods provision, c (a). Because speci¬c skills require an
infrastructure of institutions to ensure the provision of such skills (in par-
ticular vocational training systems) as well as to cope with the hold-up
problems endemic to speci¬c asset investments (the wage bargaining sys-
tem), the choice of a i is a function of the s level of the median member of
group i = L, H.6 The simplest way to think about this is in terms of wage
restraint. Union leaders, on behalf of their members, may underexploit
their negotiation power in collective bargaining in order to encourage in-
vestment and employment. The greater the control over speci¬c assets is,
the greater is the potential collective good to be achieved. But if unions
have no guarantees that the government will keep its promise of future
insurance, the expected value of future earnings falls, and the attraction of
maximization in the present increases.
Stated slightly more formally, in order for the expected value of current
pay to be equal to the expected value of future pay, future pay must be
greater than the current pay. This follows trivially from discounting of
future consumption, but on top of that there must be a future pay premium

6 The implication is not that those with general skills have no interests in public goods but
that, insofar as this is the case “ as with general education “ they tend to be of a nature
that makes it possible for the government to provide it without the cooperation of private

Political Foundations of Social Policy

to compensate for the riskiness of future pay. Everything else being equal,
a worker with speci¬c assets (high s ) will face greater risks and, therefore,
value current over future pay more than a worker with general assets. This
difference disappears, however, as income protection reduces the riskiness
of future pay for high-s workers. For risk-averse workers, the expected
utility of future income rises as the protection of future income goes up,
keeping the expected value constant.
Using this logic, if R — is the optimal level of social protection, as de¬ned
previosuly, a government that can credibly make a promise to provide this
level of R will encourage current restraint compared to a government that
cannot credibly commit. In other words, a political party able to make
credible offers of future protection is in a better position to offer certain
collective goods such as wage restraint. Because collective goods provision
is a valuable electoral asset, there is an incentive for a party to yield power to
unions over platform and leadership selection in order to raise the credibility
of its social insurance commitments. Of course, this comes at the cost of
lower electoral ¬‚exibility and a more constrained party leader.

4.3.2. The Game between Parties, Voters, and Groups
The preferences of groups and voters are essentially variations of the utility
functions presented in the previous section, except the costs and bene¬ts
of public goods provision are taken into account. The key is whether deci-
sions are made behind the veil or not. The group choosing the party platform
makes its decisions behind the veil because choosing a platform has long-
term consequences if party leaders are made to accept the platform. The
employed median group member does not know whether he or she will be
unemployed in the future. The relative durability of party platforms mat-
ters here; if party platforms are chosen infrequently or if they are subject
to only incremental change, there is a clear difference between choosing a
platform and choosing a policy. Voters are not choosing platforms but only
a party to govern for the next electoral term. The veil behind which voters
are making decisions is, therefore, very thin: They can only project policies
into the next electoral period. This is a critical distinction.
More speci¬cally, the group™s utility function is analogous to Equa-
tion (4.1), except that there is a positive term representing the net utility
derived from the public good a (Appendix 4.A contains the details). The
preferences of s m , the median employed voter, after the veil has been raised
is analogous to Equation (4.3) (see Appendix 4.A for details). With these
Credible Commitment, Political Institutions, and Social Protection

assumptions, the party leader faces the following tradeoff. On the one
hand, there is an incentive to follow the nominated strategy because it
guarantees reselection. On the other hand, the leader would like to have
total policy ¬‚exibility because it makes it more likely that he or she can
win the next election and become head of government (or prime minis-
ter, PM). Also, it is more dif¬cult to control leaders as head of government
because prime ministers can make use of institutionalized resources, includ-
ing ample media access, that they do not enjoy to the same extent as party
To capture this logic, let B be the bene¬ts of of¬ce and L the bene¬ts of
leadership of the party. The more institutionalized the party is, the greater
are the constraints put on the leader and PM.
These constraints diminish the attractiveness of both leadership and PM
positions, especially as a increases, because a is exchanged for greater party
institutionalization. Because the preferred level of a is a positive function
of s , leaders of parties representing speci¬c assets are more constrained
than parties representing general assets, which also means that they are
constrained to accept a higher t. More speci¬cally, the period utility for the
leader can be written as
Upol = e · B(s ) + L(s ) (4.4)
where e is the probability of becoming a PM. To capture the idea that it is
easier to control leaders when they are not heads of governments, assume
‚ ln B ‚ ln L
‚ ln s ‚ ln s

4.3.3. Solving the Game
The chosen levels of t and a is the solution to the following stage game.
Stage i: The median voter in each of the two groups, s ML and s MH ,
chooses contribution levels and nominating platforms (behind the
Stage ii: Party leaders choose electoral platforms.
Stage iii: Voters vote (unveiled).
Stage iv: Leaders are reselected with a certain probability.
Because a new subgame that does not depend on the history up to that
point begins at the start of the next period, all we need to do is analyze the
Political Foundations of Social Policy

choices in the current period through backward induction. The solution is
characterized next, with the mathematical details explained in Appendix 4.A.

Stage iv: Reselection of Leaders. At the end of the period, the leaders are
reselected with certain contingent probabilities [1 if the leader accepted the
platform and (1 ’ q ) if the leader did not]. This involves no choices.

Stage iii: Voting. Assume that no new party has entered (an assumption
that is demonstrated to hold in Appendix 4.A) and that leaders have accepted
platforms that differentiate their party from the other. L will then vote
for L and H will vote for H because they vote sincerely and because L offers
an a closer to all voters in L than the a offered by H and similarly for mem-
bers in H. The median voter will choose the party with a combination
of t and a that comes closest to the median voter™s ideal combination of
policies: t = 0 and a = ±sm . If the two parties are equally attractive, then
the median voter will vote for each party with equal probability.

Stage ii: Party Leader™s Choice. The leader has to choose t and face two
opposing incentives. On the one hand, the leader would like to win the
election (and thereby become prime minister) by choosing a low t. On the
other hand, he or she wants to remain the leader of the party and produce a
high level of public goods (a), which requires the leader to pay attention to
the preferences of his or her constituent group (given by Equation (4.4)).
To model this choice, we assume that the base cannot monitor perfectly
what the leader does. In other words, there is a possibility that the leader
can “cheat” and get away with it. This may arise, for example, because an
unexpected recession makes a tax cut a prudent policy.
How high does 1 ’ q (the probability of non-reelection) have to be to
ensure leader i cooperates? Appendix 4.A derives the exact condition, but
the key variables affecting whether the condition is satis¬ed can be easily
summarized. Institutionalization (1 ’ q ) would need to be higher (i) the
greater is B (the bene¬ts of of¬ce), which raises the temptation to defect
for the leader; (ii) the lower is L (the bene¬ts of being party leader), which
again raises the temptation to defect; (iii) the higher is s (which raises the
group™s preferred tax level); (iv) the lower is the discount rate, »; and (v) the
higher is the probability that there are no shocks requiring a lower tax rate.
The group or party base would obviously want to choose a level of
institutionalization that ensures the cooperation of the leader. But whether
it can accomplish this cooperation depends on the resources that are at the
Credible Commitment, Political Institutions, and Social Protection

disposal of the group. Being able to extract a high fee, which is equivalent
to being able to in¬‚uence signi¬cantly the ability of the party to offer a
high level of the public good a, accords more group power over leadership
selection. Well-organized and encompassing groups will therefore tend to
be associated with more institutionalized parties, and effective organization
is in turn partly a function of the speci¬city of member assets, which confers
hold-up power.
But whether the leader cooperates also depends critically on his or her
temptation to defect. When winning government power carries a very high
reward (high B), it is harder for the party base to prevent leadership de-
fection. In an analogous manner, given a ¬xed prize of winning of¬ce, the
greater the effect of getting an additional vote on the probability of win-
ning is, the greater the temptation to cheat is. Both factors are positively
correlated with having a two-party majoritarian electoral system. Because a
single party wins the election outright, being the prime minister is likely to
be associated with considerably more power than in multiparty PR systems
with minority and/or multiparty governments. In addition, as argued by
Rogowski and Kayser (2002), majoritarian elections produce a very high
“vote-seat elasticity,” where winning a few more votes can have a big effect
on the prospect of winning of¬ce. Party leaders, therefore, have a relatively
greater incentive to pay attention to the median voter than to their base
compared to PR systems. Majoritarianinsm, in other words, makes it more
likely that the leader will fall for the “populist” temptation of cutting taxes.

Stage i: The Choice of Party Platforms. The groups sponsoring each
party choose the contribution levels and tax rates to maximize their welfare
function, which is given by Equation (A4.1) in the appendix [analogous to
Equation (4.1)]. Using a simple cost function, Appendix 4.A shows that the
optimal tax rate is
t — (s ) = p · (1 + γ (s )) (4.5)
where γ (s ) is a search cost for ¬nding a new job in the event of unemploy-
ment. It is assumed that it is harder for those with more speci¬c skills to
¬nd jobs that are suitable to their skills The optimal tax rate, t — (s ), is, thus,
positively related to the probability of unemployment p and also positively
related to search costs, γ (s ). Note that if there are no search costs, Equa-
tion (4.5) reduces to t — = p, which is the optimal long-term level of taxa-
tion preferred by the median voter as initially identi¬ed. Introducing s and
search costs, γ (s ) , implies policy differentiation. The optimal contribution
Political Foundations of Social Policy

level, a — , is also directly proportional to s (details are provided in Ap-
pendix 4.A).
Whether the contribution level is suf¬ciently high to induce leadership
cooperation cannot be determined a priori. To some extent, we have to treat
party organizations, and hence the ability of groups to in¬‚uence policies, as
historically given. The same is true for the incentives of leaders to cooperate.
However, the comparative statics is clear because as the speci¬city of assets
rises, so does the ability and willingness of groups to pay higher “fees,”
which in turn gives parties and their leaders a greater incentive to yield
in¬‚uence over party policies. Asset speci¬city and party institutionalization,
therefore, will tend to go hand in hand. In addition, because the prize of
winning a majority is very high in a two-party majoritarian systems, the
temptation of the party leader to present a short-sighted platform should
be expected to be greater than in multiparty PR systems where the marginal
effect of an additional vote on power tends to be lower.

4.4. Redistribution
As argued previously, there are two dimensions to the provision of social
insurance. One is an intertemporal dimension where current payment is ex-
changed for (potential) future receipts. The second is a contemporaneous
redistributive dimension where income is transferred from one group to
another. The two dimensions are intertwined because transfers motivated
by redistribution can serve insurance purposes, and transfers motived by
insurance can have redistributive consequences. Current unemployed may
in the past have supported unemployment bene¬ts for purely insurance rea-
sons, but they now have a redistributive motive to continue supporting it.
And if unemployment bene¬ts exist only because of political pressure from
the unemployed, that does not mean that those currently employed are not
bene¬ting from the insurance effects of such bene¬ts. Hence, if distributive
politics produces continuous pressure for redistribution, it can help over-
come the time-inconsistency problem. To understand the politics of social
insurance, we, therefore, need to understand the politics of redistribution.
This section seeks to provide such an understanding.
As noted in Chapter 1, most of the literature on redistribution focuses on
the effects of politics being dominated by the left or the right but does not
offer a convincing account of the source of such partisan dominance. Iden-
tifying this source is key to explaining redistribtuion. This section argues
that the source of partisan dominance is the electoral system.
Credible Commitment, Political Institutions, and Social Protection

The model is a special version of a very general model of redistribution
proposed in Iversen and Soskice (2004). That model assumes that (i) parties
represent groups, or classes; (ii) that parties maximize the distributive pref-
erences of their members (but subject to a time-inconsistency problem);
and (iii) that the net effect of government taxation and spending is nonre-
gressive in the sense that those with lower incomes need to bene¬t from
government policies as much as those with higher incomes. I present here
a simple version of this model that is easy to explain and relate to existing
work on the welfare state. The model makes speci¬c assumptions about the
structure of bene¬ts, but the conclusions hold for any tax-bene¬t structure
that is nonregressive and not strictly one-dimensional (the Meltzer-Richard
model is nonregressive but also unidimensional).
The model has two key results. First, in a two-party majoritarian sys-
tem, the center-right party has an electoral advantage whenever there is
a nonzero probability that the winning party will deviate from its elec-
toral platform once in power. The reason is that left party leaders under
majoritarianism need to compromise the ideal redistributive policies of
their members more than right party leaders and, therefore, face a greater
postelection incentive to adopt policies that are unattractive to the median
voter. In a multiparty PR system, by contrast, where each party represents a
distinct class and must ally with another party to govern, the typical pattern
is that the middle class (or center) party will ally with the lower class (or left)
party. The reason here is that the middle-class party can use taxes that fall
disproportionately on the rich to bargain with the lower-class party for a
level of social insurance (and hence taxation) that is closer to its ideal point.
The implications are that (i) center-left governments will be more frequent
under PR, (ii) center-right governments will be more frequent under ma-
joritarian rules, and (iii) redistribution will be greater under PR than under

4.4.1. The Politics of Redistribution
There is a huge empirical literature on the welfare state based on Esping-
Andersen™s (1990) classic study. Whatever the speci¬c aims of these stud-
ies, there is broad agreement that Esping-Andersen™s depiction of the dis-
tributive dimensions of social policy describes the policy space of advanced
democracies rather well. Most systems combine a universalistic or ¬‚at-
rate bene¬t with a means-tested bene¬t targeted at the poor, although the
relative weights of these bene¬ts vary across countries. Esping-Andersen
Political Foundations of Social Policy

also distinguishes earnings-related bene¬ts, but if these bene¬ts are di-
rectly proportional to income, they are equivalent for our analytical pur-
poses to people keeping their market income. Although such bene¬ts may
serve important insurance purposes, I deliberately ignore these in this
Assume that the ¬‚at-rate bene¬t, which is called f , is fully ¬nanced by a
proportional tax, t. This makes it equivalent to the Meltzer-Richard model.
But even though the Meltzer-Richard model is convenient, it is not descrip-
tively accurate and precludes us from understanding the multidimensional
politics of redistribution that give rise to coalitional politics. This is where
the means-tested bene¬t, g, enters. To make it maximally redistributive in
nature, I assume that it is ¬nanced by a progressive tax, which falls dispro-
portionately on those with higher incomes. Still, a small but nonnegligible
share of g, µ · g, is paid by the middle class (as is the case for most progressive
It is easy to see that these assumptions satisfy the nonregressivity con-
straint (i.e., the poor will not be made worse off ), and although they are
more restrictive than necessary, combining f and g, and their associated
taxes, yields a redistributive policy space that is ¬‚exible enough to describe
most actual social bene¬t systems. Again the key is nonregressivity, which
can be defended on many grounds. The marginal utility of money may
be higher among the poor than among the rich; or the costs of extracting
one tax dollar from a poor person may be higher than extracting a dollar
from a rich person. Most fundamentally, “reverse” redistribution may be
inconsistent with the underlying conditions that gave rise to democracy,
namely the need to attend to the distributive preferences of the poor and
the middle classes (Acemoglu and Robinson 2005). An argument along
these lines is spelled out in detail in Iversen and Soskice (2004). What-
ever the reason, data for advanced democracies, which are discussed later
in this chapter, show no case where taxes and transfer result in a more
dispersed distribution of income. To the extent that the democratic gov-
ernments redistribute money, it ¬‚ows from higher to lower incomes, and
no analysis of the welfare state implies otherwise. The two best-known
analyses, that of Meltzer-Richard and that of Esping-Andersen, are cases
in point.
More speci¬cally, the model follows Persson and Tabellini (1999) and
assumes that there are three equally sized income classes in the population:
L (low), M (middle), and H (high). Majoritarian systems are assumed to

Credible Commitment, Political Institutions, and Social Protection

be dominated by two parties (Duverger™s law), but under PR each group
is assumed to be represented by its own party.7 In both cases, parties rep-
resent income classes in the population (i.e., they are “class parties”). In
the majoritarian case, the middle class is split between a center-right and a
center-left party, making the median voter pivotal.
The transfer g (and corresponding tax) has a cost that includes expenses
for administration, rents to politicians who provide tax “loopholes” to the
rich, and the costs of paying lawyers to take advantage of these loopholes “
where the cost of g to H is ±g with ± > 0. In addition, it is assumed that
0 ¤ g ¤ g — , where g — is a constitutionally guaranteed upper limit that can
be thought of as a basic property right protection preventing complete
expropriation of property. For speci¬city, assume that this constitutional
protection can only be overturned by three quarters of the legislature. In
this case, H (assuming it has one third of the seats in the legislature) can
always block any attempt to raise g — , and H voters will have an incentive
to vote under PR regardless of whether they can anticipate H to be in
government or not. Loosely speaking, one can think of g — as measuring the
power of (high-income) veto players in the system.
It is possible to present the model with preferences over taxation that are
endogenously determined by the relative income of each group and the ef-
¬ciency costs of taxation.8 However, one can derive all the key comparative
statics from a simple indirect utility function model, in which each group
has preferences over t and g. The main difference is that, in the model with
endogenous policy preferences, relative income is an independent cause of
redistribution. But because relative income turns out not to have much of
an empirical effect, I omit the variable in the current presentation.
In this simple model, L is interested in maximizing g and t; H, in mini-
mizing both g and t; and M, in setting t as close as possible to some interme-
diate level of t, which is assumed to be 0.5, and in minimizing g. In terms
of t, this is the structure of preferences across income groups implied by
the Meltzer-Richard model. Also note that M would be equally distanced

7 Persson and Tabellini (1999) assume only two parties under PR, but they acknowledge that
it plainly does not make much sense: “We hold the party structure ¬xed, ignoring theoretical
arguments as well as empirical evidence for a larger number of parties under proportional
elections” (p. 706). They go on to say that “our excuse is pragmatic; we simply do not
know how to analyze multi-dimensional policy consequences of electoral competition in a
multi-party setting” (p. 706).
8 The model is available from the author upon request.

Political Foundations of Social Policy

from L and H if f (or t) were the only policy dimension. The preferences
over g follow trivially from the assumptions made. The goals of the three
groups, therefore, are
uL = g + t
u M = ’ |t ’ 0.5| ’ g ·(1 + ±)·µ
u H = ’t ’ g ·(1 + ±)·(1 ’ µ)

4.4.2. Majoritarian Elections
There are two parties, CL (center-left) and CR (center-right), which orga-
nize voters on either side of the median income. One party, thus, “repre-
sents” the center-left; the other, the center-right. Each will have different
ideal policies as a result. Given that the middle income group is a minority
in both parties, if these are characterized by the preferences of the median
constituent in each party, the center-left party will want {g, t} = {g — , 1}
while the center-right party will want {g, t} = {0, 0}. However, in a ma-
joritarian system, no party can affect policy without winning a majority
of the vote, so the platform presented in the election will clearly need
to deviate from the policy preferences of the median constituent in each
What is the vote-maximizing platform? It turns out that this is given by
a simple median voter result. Because there are two policy dimensions, it
is not obvious why this should be so, but Appendix 4.B proves that it is.
Essentially, if CR proposes taxes that are below 0.5, CL will offer a policy
of (0, 0.5); if CL proposes a g > 0, CR will offer a policy of (0.5, 0), and the
preferences of L and H are too misaligned to make it possible that both
would prefer a platform that is different than (0, 0.5).
Before proceeding, it is useful to characterize the median voter plat-
form brie¬‚y in left“right terms. Compared to the ideal policies of L and
H, the median voter platform is closer to the preferences of H than to
the preferences of L, and in that sense the median voter platform may be
thought of as right-of-center. The reason is that although the middle class
deviates from both the lower and upper class in terms of preferences over
the level of taxation and spending, it shares an interest with the latter in
restricting redistributive transfers to the poor. This is an old insight in the
welfare state literature, emphasized by Esping-Andersen in his discussion
of means-tested bene¬ts (1990, Chapter 1). It arises in the model because
of the two-dimensional nature of social spending.
Credible Commitment, Political Institutions, and Social Protection

Given that the two parties in a majoritarian system represent different
constituencies, is it realistic that they will converge on the median voter
platform? Most existing answers in the party literature suggest that even
though there is signi¬cant pressure on parties to present moderate platforms
in general elections, it is hard for party leaders to ignore the policy prefer-
ences of their core constituents completely. One argument is that leaders
need to mobilize their base in order to maximize voter turnout among their
prospective supporters. This involves appeals to the policy preferences of
the median activist and emphasis on policy differences with the other party
(Schlesinger 1984; Aldrich 1993, 1995, Chapter 6).
An alternative formulation with similar results is that parties cannot make
binding commitments to electoral platforms (Downs 1957; Persson and
Tabellini 1999). This is the generic time-inconsistency problem identi¬ed in
the previous section. Once in of¬ce, there is an incentive for both parties to
adopt policies that re¬‚ect the preferences of their median constituents. This
incentive is tempered by the concern for cultivating a reputation among
voters for reliability, but reputation is an imperfect commitment mechanism
in a world with short-sighted politicians. As a result, the median voter has
reason to worry that whoever wins the election will give in to the temptation
of pursuing policies that appeal to the party™s internal core constituents.
Thus, the temptation for the center-right party, if it wins, is to put the
policies {0, 0} into operation and for the center-left party to carry out the
policies {1, g — }. This affects the voting behavior of the median voter in a
subtle, but important way.
To understand this, assume that whether or not a party yields to the
temptation if elected depends on whether the costs outweigh the temptation
bene¬ts, TC L and TC R . These variables are straightforwardly calculated “
TC L = g — + 0.5, TC R = 0.5 “ in each case the gain from switching from
the median voter™s ideal point (0.5, 0) to (1, g — ) and (0, 0), respectively.
The cost of adopting more extreme policies is the loss of reputation.
The loss of reputation for trustworthiness matters to a government be-
cause, without such a reputation, governing is less effective because it
is harder for the government to make deals with other agents.9 This is
modeled by assuming that the loss of effectiveness is a cost, c C L and
c C R , respectively, which restricts government effectiveness if a defection
to more extreme policies takes place. Thus, the payoff to the left party from

9 An additional possibility is that voters punish defecting governments in future elections, but
this adds to the complexity of the model without altering its insights.

Political Foundations of Social Policy

defecting is TC L ’ c C L = g — + 0.5 ’ c C L and the payoff to the right party
is TC R ’ c C R = 0.5 ’ c C R .
Next, assume that c C L and c C R are random variables independently
drawn at each election from the same uniform distribution, normalized
for convenience to [0, 1], with 1 > max[TC L , TC R ]. Thus, in the election
campaign, the median voter forms an idea of how trustworthy each of
the party leaders are after they have set out their platforms; because this
trustworthiness can be valuably used by the executive if it carried out
the median voter policies, the loss of this attribute would be the cost
of yielding to the temptation of switching to left or right policies once
in power.
The median voter would be indifferent to which party he or she voted for
if TC L < c C L and TC R < c C R . But if TC L > c C L and TC R < c C R or if TC L >
c C L and TC R > c C R , the median voter would vote center-right; if TC L <
c C L and TC R > c C R , the median voter would vote center-left. Using the
joint cumulative distribution function of c C L and c C R , it is not dif¬cult to
see that the center-right would win a proportion
(1 ’ TL ) · (1 ’ TR )
πC R = TL · (1 ’ TR ) + TL · TR +
of elections against
(1 ’ TL ) · (1 ’ TR )
πC L = TR · (1 ’ TL ) +
won by the center-left. It follows that
πC R ’ πC L = TL ’ TR + TL TR > 0
In other words, the center-right party wins more of the time. The intuition
behind the result is simple and goes back to the observation that the median
voter shares an interest with the well-off to avoid means-tested transfers to
the poor. Even though both parties may fall to the temptation to adopt tax
policies that are unattractive to the median voter, it is only the center-left
party that has an incentive to adopt policies of means-tested transfers to
the poor. This makes the median voter more likely to vote for the center-
right party.
Whether a center-right party would also win against a center party de-
pends on the exact interpretation of what a center party is. This matters
only because some parties in the empirical analysis are classi¬ed as “center
parties,” speci¬cally, if the center party represents middle-class voters, then
it would be more attractive to the median voter than the center-right party.
Credible Commitment, Political Institutions, and Social Protection

But if the center party is really a center-left party with a platform mirroring
the preferences of the median voter, whereas the center-right party has a
platform that deviates to the right, then the prediction is ambiguous since
the center party is closer to the median voter, yet faces a greater incentive to
defect. Because no existing data clearly distinguish between these different
“types” of center parties, we cannot form any clear predictions about the
performance of center parties in majoritarian systems. The predictions for
center-left and center-right parties, however, are unambiguous: The latter
win more of the time.

4.4.3. Proportional Representation
For simplicity, assume here that there are three representative parties under
PR “ L, M, and H “ none of which have an absolute majority in the electorate.
It is furthermore assumed to be common knowledge that each party seeks to
promote the welfare of the class it represents. Because there is no imperative
under PR to win the median voter, a party does not have the incentive to
adopt a platform that is different from the optimal policies of its class.
If it did, it would not be credible. Indeed, this distinction between the
credible commitment of representative parties under PR and the dif¬culty
of such commitment under majoritarian arrangements is one of the central
differences between the two types of electoral systems. Recall from the
previous section that this difference was also important for the provision of
On the face of it, coalitions between M and H would seem as likely as
coalitions between L and M. When t is the only policy dimension, and if a
“split-the-difference” rule determines the policy a coalition will adopt, M
will indeed be indifferent between a coalition with H and a tax rate of 0.25
and a coalition with L and a tax rate of 0.75. Both imply utility of ’0.25
to M.
But this conclusion no longer holds when g is added. The reason is that M
can now offer concessions to L on g at a low cost that re¬‚ects the progressive
nature of the tax (i.e., most of the cost is paid by H ). In exchange for
such concessions, M can demand a tax rate that is closer to its preferred
rate. Speci¬cally, for a suitably low µ, the Rubinstein bargaining solution is
0.75 ’ g — /2 (see Appendix 4.C). Thus, a bargain with L will always be closer
to M™s preferred policy than 0.75. M has no such bargaining leverage over H,
and the outcome of that bargain would, therefore, be a simple split between
preferred tax rates (0.25). Consequently, M prefers to be in a center-left
Political Foundations of Social Policy

coalition. Appendix 4.C demonstrates this conclusion more formally and
addresses the objection that H can always break an LM coalition by offering
M a deal that is closer to M™s ideal policy. This cannot happen, it turns out, if
there is any cost of coalition breakup because that prevents H from making
a credible offer to M.
Combining the results for PR with those for majoritarian institutions,
the analysis has yielded an unambiguous and stark insight that has not been
articulated in any of the existing literature: Majoritarian electoral systems
tend to produce center-right governments, whereas proportional electoral
systems tend to produce center-left governments. The former will redis-
tribute less than the latter. The key to understanding redistribution is the
long-time political dominance of the left or right, and a key to understand-
ing long-term partisan dominance is the electoral rule.

4.5. Empirics
The theory is explored in two parts. The ¬rst examines the party institu-
tionalization argument using cross-sectional data for twenty-one countries.
Because time-series data are not available, the emphasis here is on estab-
lishing stable equilibrium couplings, although I will go as far as is possible
to establish causality using quantitative data. The second part explores the
redistribution argument. One section uses partisanship as the explanatory
variable to account for differences in the level of redistribution. Another
uses partisanship as the dependent variable to test the proposition that the
electoral system shapes coalition behavior and, therefore, the composition
of governments. The results of the analysis reinforce each other because
institutionalized parties are associated with more social insurance spend-
ing, and such parties are more likely to be found in PR systems, in which
redistribution tends to be greater.

4.5.1. Party Institutionalization
Insofar as party institutionalization varies systematically across countries,
the argument has two macro-level implications. The ¬rst is that we should
expect more young people to acquire speci¬c skills in countries where party
institutionalization is high. On the one hand, public goods provision, which
includes vocational training systems, is expected to be higher in countries
with highly institutionalized party systems. On the other hand, more young
Credible Commitment, Political Institutions, and Social Protection

people are likely to invest in speci¬c skills when such investments are backed
by credible institutional guarantees. Second, one would expect the skill
pro¬les of national labor markets to be associated with higher social trans-
fers. When a large portion of the electorate have highly speci¬c skills,
the preferred level of social insurance will be high. If parties are capable
of credible commitment, such preferences will get translated into actual
protection. In turn, protection encourages further investment in speci¬c
It is important to point out that this is an argument about institutional
equilibria. If the median voter is pivotal, for example, we would not expect
this voter to have skills that are inconsistent with the institutional capac-
ity of commitment or the level of social protection. If institutionalization
dropped below what is required to sustain con¬dence in the future viability
of public goods provision, and in the future of social protection, we should
see adjustments in peoples™ skill investments toward greater emphasis on
general skills until the point where institutionalization, skills, and social
protection are once again aligned in a steady-state equilibrium. When we
look at national systems over longer periods of time, we therefore expect to
observe a high degree of collinearity between these variables. Only if there
is a shock to one of the variables, or to the level of risks in the labor market,
would that variable become a causal agent triggering changes in the other
Chapter 5 examines what happens to social spending in different insti-
tutional settings when the system is exposed to exogenous shocks. Here I
simply examine if the observed covariation between variables across coun-
tries is consistent with the equilibria implied by the model. That is, I look at
whether heavy investment in speci¬c skills is in fact associated with high ca-
pacity of the party system for institutional commitment and whether these
variables in turn are linked to government spending. To this end, twenty-
one OECD countries are compared for the period 1980“95, which is the
period for which comparable ¬gures for the age cohort going through voca-
tional training exist.10 Vocational training activity serves as the macro-level
proxy for the importance of speci¬c skills in the labor force (corresponding
to average level of s in the theoretical model). As an indicator for the extent
to which the government engages in social spending “ the variable t in the

10 The countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France,
Germany, Ireland, Italy, Japan, the Netherlands, Norway, Portugal, Spain, Sweden,
Switzerland, the United Kingdom, and the the United States.

Political Foundations of Social Policy

theoretical model “ the sum of government consumption and transfers as
a share of GDP is used.11
The measure of the institutionalization of the party system is based on
two sources of data. One concerns the centralization and discipline of po-
litical party systems; the other, the extent to which interests are organized
in a corporatist manner. We need the ¬rst because a necessary condition
for parties to be able to credibly commit to a policy is that elected leaders
have individual incentives to run their campaigns on the policy platform of
the party rather than on personal resources and policy appeals. A necessary,
and potentially suf¬cient, condition is that such centralized control is cou-
pled with close ties to well-organized and centralized private groups, and a
measure of corporatism is used to capture this aspect. The combination of
these variables constitutes the, admittedly crude, variable for party system
The data are adapted from Carey and Shugart™s (1995) path-breaking
analysis of electoral systems in terms of the incentives they provide politi-
cians to either run on the party platform and toe the party line or to cultivate
their own personal following without regard to the preferences of the party.
The classi¬cation of party systems in this analysis is coupled with Siaroff™s
(1998) measure of corporatism, which measures both the organizational
centralization of private interests and their ties with the state and political
parties. The exact procedures for creating the combined index are explained
in Appendix 4.D.
Figure 4.1 shows the relationships between the three variables. Fig-
ure 4.1a suggests a fairly tight linkage between institutionalization of the
party system and vocational training activity, with a familiar clustering of
Anglo-Saxon countries at one end and the continental European countries
at the other (with France in a somewhat precarious intermediate position).
It is also notable that Japan is much closer to the Anglo-Saxon cluster than
to the continental European cluster. In addition to low institutional capacity
for commitment, the Japanese welfare state is underdeveloped, which would
imply underinvestment in speci¬c skills. In the Japanese system, however,
extensive training in highly ¬rm-speci¬c skills, which is not fully captured

11 Speci¬cally, the measure refers to all government payments to the civilian household sec-
tor, including social security transfers, government grants, public employee pensions, and
transfers to nonpro¬t institutions serving the household sector as a percent of GDP (1980“
95). Sources: Cusack (1991) and OECD, National Accounts, Part II: Detailed Tables (various

Credible Commitment, Political Institutions, and Social Protection

by the UNESCO educational categories, is made possible by very high job
security and protection of future earnings for skilled workers at the ¬rm
Figures 4.1.b and 4.1c show the relationship between government trans-
fers and vocational training and institutionalization. Again, the covariances
are relatively strong and in the expected direction. Countries with high lev-
els of vocational training also have high government spending, and spend-
ing is likewise rising with the degree of institutionalization of the party
system. Switzerland is a bit of an outlier because it is coded as having high

Figure 4.1 The relationship between institutionalization of the party system,
vocational training intensity, and government spending. Vocational training: The
share of an age cohort in either secondary or postsecondary (ISCED5) vocational
Source: UNESCO (1999). Government spending: Government civilian consumption plus gov-
ernment transfers in the form of payments to the civilian household sector (including social
security transfers, government grants, public employee pensions, and transfers to nonpro¬t
institutions serving the household sector) as a percent of GDP. Sources: Cusack (1991) and
OECD, National Accounts, Part II: Detailed Tables (various years). Institutionalization: Average
(after standardization) of Siaroff™s (1999) corporatism index and Carey and Shugart™s (1995)
classi¬cation of electoral systems. See Appendix 4.D for details.

Political Foundations of Social Policy

institutional capacity, yet it does not train or spend at a commensurable
level. The reasons, similar to those in Japan, could be the relatively high
protection built into the industrial relations system. It may also be that the
institutionalization variable gives an exaggerated measure of capacity for
commitment given that unions are much weaker in Switzerland than they
are in other corporatist countries.
Whatever the exact explanation for the somewhat aberrant positions
of Japan and Switzerland, the correlations between the variables are high
(ranging between .68 and .77). Again, this says nothing about causality, and
the three variables are best considered part of an institutional equilibrium
without a clear causal order. During particular historical periods, each can
be either a dependent or an independent variable. This complementarity
logic is one that has taken root in economics (see Aoki 1994; Cooper 1999),
and it is central to the varieties of capitalism. What this section has at-
tempted to show is that party institutionalization is an integral component
of the welfare production regimes that we observe.
However, we go beyond the identi¬cation of institutional complemen-
tarities in two ways. The ¬rst is to investigate the causal effects of partisan
governments on redistribution, as well as the effects of electoral systems on
partisanship. This is the task in the next section. Second, we want to inves-
tigate the extent to which countries with different institutional equilibria
respond differently to external shocks in terms of social spending. This task
is left for Chapter 5.

4.5.2. Redistribution
Most existing work on redistribution relies on indirect measures such as
government transfers, social spending, or some other indicator of welfare
state effort. Such measures are not entirely satisfactory because the data
come in a form that typically tell us very little about the extent of redistri-
bution as opposed to the level of spending.
Fortunately, relying on spending data to measure redistribution is no
longer necessary. During the past three decades, the Luxembourg Income
Study has been compiling a signi¬cant database on pre- and posttax and
transfer income inequality. The LIS data used for this study cover fourteen


. 5
( 11)