. 8
( 35 .)


proximate the presidential and legislative elections are, the stronger
the aggregation incentives will be, and the higher the probability
that the largest legislative party will also be the party of the president.
However, the effect of concurrent elections is conditional on the
number of viable presidential candidates. Where the presidential elec-
toral system regularly produces two strong presidential contenders,
candidates for the legislature are better able to distinguish between
front-running and trailing candidates and thus able to aggregate stra-
tegically. The number of viable presidential candidates, then, is key to
understanding aggregation incentives. Concurrent elections have their
strongest effect when there are two clear frontrunners for the presi-
dency. If distinguishing the frontrunners is dif¬cult, then the probability
that the largest legislative party will also capture the presidency is
greatly reduced, and so are the incentives to coordinate across districts.
This is true even when presidential and legislative elections are con-
current. A handful of recent studies ¬nd that a large number of presi-
dential candidates can completely undermine the effect of concurrent
elections on the number of parties (Amorim-Neto and Cox 1997;
Chhibber and Kollman 2004; Golder 2006).
What types of factors affect the number of viable presidential
candidates? The degree of social heterogeneity is one important factor “
greater ethnic fractionalization, for example, puts upward pressure on
the effective number of presidential candidates.20 However, just as the
social structure and electoral institutions interact to determine the
number of parties at the district level, so to do these variables interact to
shape the number of presidential candidates. Social heterogeneity only
increases the effective number of candidates when combined with a
permissive presidential electoral formula (Golder 2006). “Strong”

The effective number of presidential candidates is calculated by dividing 1 by the sum
of each candidate™s squared vote share: 1/Rvi2.
Building Party Systems in Developing Democracies

formulas such as the plurality rule generally produce fewer viable
candidates than more permissive arrangements (e.g., majority runoff),
even in the face of social heterogeneity. Indeed, several scholars have
found that more restrictive electoral rules produce fewer candidates than
more permissive rules (Cox 1997; Jones 1997; Jones 1999; Jones 2004;
Golder 2006). In short, consistent with Duverger™s law, restrictive rules
push the number of presidential candidates toward two, even where
there is a high degree of social heterogeneity (Golder 2006).21
A second, but oft overlooked factor that helps shape the number of
presidential candidates is the rule governing presidential reelection.
Systems with term limits, particularly those that ban any reelection,
should be associated with a higher number of viable presidential
candidates compared to those that do not limit reelection (Jones
1999). Why would limits on reelection be associated with more viable
presidential candidates? Assuming that presidential of¬ce has some
value and that there is open contestation for the of¬ce, there will
always be multiple possible presidential hopefuls. Yet often it is the
case that all but two of those hopefuls drop out of the contest along the
way, or never enter the race in the ¬rst place. Why? One factor, as
discussed previously, is certainly the restrictiveness of the electoral
formula. But another key factor that whittles down the ¬eld of can-
didates is the presence of an incumbent. In all democracies, the pres-
ence of an incumbent signi¬cantly raises the entry barrier for potential
challengers. Incumbents generally have better name recognition than
challengers and are able to bring the power, resources, and networks
associated with political of¬ce to bear on the campaign. In addition,
presidents (and presidential contenders) who are eligible for reelection
have strong incentives to invest in building and maintaining an
effective party organization and base of support “ assets that can be
mobilized for future presidential contests.22 The net effect of incum-
bency is to encourage an incumbent™s opponents to coordinate in

Evidence about the direct effect of social heterogeneity on the effective number of
presidential candidates is mixed (see, for example, Jones 1997; Jones 2004).
Note that candidates facing a single term limit might still have reasons to party-build
if the reelection ban applies only to consecutive reelection. The prospect of a future
run for of¬ce can be enough to induce lame-duck incumbents to take a longer-term
view. Lame-duck presidents may also have an incentive to party-build if their future
career prospects are tied to the party.
A Theory of Aggregation Incentives 43

support of a single standard-bearer for the opposition in order to
maximize their prospects at the ballot box.23
The lack of an incumbent, either by rule or the incumbent™s choice,
can dramatically reduce the entry barriers for presidential hopefuls.24
Why, though, would not a strong electoral formula eventually push
the effective number of presidential candidates down to 2, even if there
are multiple candidates to begin with? Indeed, that is what we would
expect given Duverger™s law and Cox™s M þ 1 rule. In single-seat
plurality contests, all but the two frontrunners should eventually
withdraw from the race (strategic entry). Where trailing candidates
fail to withdraw, voters should respond by voting strategically for
their most preferred of the two frontrunners (strategic voting). The
end result of this strategic behavior is an effective number of presi-
dential candidates near 2.
However, recall that in order for Duverger™s law and the M þ 1 rule to
work as predicted, certain assumptions must be met. One of these is that
voters, candidates, donors, and the like must be able to clearly identify
the two front-running candidates.25 Where they cannot do so, the
coordination on two candidates breaks down. Returning to the issue of
incumbency, I argue that without an incumbent in the race it will be more
dif¬cult to identify the two frontrunners, ceteris paribus. Imagine, for
example, a presidential contest in which there is one clear frontrunner
among all challengers. Where there is an incumbent in the race, identi-
fying the two frontrunners is relatively simple. If, however, there is not an
incumbent, actors have good information about one of the frontrunners,
but may have dif¬culty determining which of the remaining candidates is
the other frontrunner. A more concrete example is the case of the 1992
Philippine presidential election, which will be discussed in more detail in
Chapter 6. Due to a new prohibition on presidential reelection, the 1992
contest had no incumbent. Numerous challengers entered the race, but
unlike previous elections in the months, weeks, and days before the poll
no clear frontrunners ever emerged. The eventual winner, Fidel Ramos,
was victorious with only 23.6 percent of the vote. Only nine percentage

The strength of the incumbent will also in¬‚uence the opposition™s incentives to
coordinate on a single candidate.
Exactly how low that barrier is will depend in part on whether a clear, designated
successor exists that can inherit many of the advantages of incumbency.
See Chapter 4 for a discussion of each the assumptions.
Building Party Systems in Developing Democracies

points separated Ramos from the fourth place ¬nisher. Less than two
percentage points separated the second- and third-place candidates.
These ¬nal results re¬‚ected the fact that voters, candidates, party leaders,
and donors all had a dif¬cult time distinguishing the frontrunners from
the also-rans “ a task made much more dif¬cult without the presence of
an incumbent in the race.
In summary, the lack of an incumbent lowers the barriers to entry for
presidential contenders and undermines the incentives of presidents to
invest in party building. The result should be an increase in the number
of viable presidential candidates over elections where incumbents are
To conclude, the probability of capturing the prize of government
in presidential systems depends jointly on the proximity of legislative
and presidential elections and the effective number of presidential
candidates. The more proximate the presidential and legislative con-
tests are, the stronger the aggregation incentives will be. However, as
the effective number of presidential candidates rises, the relationship
between proximity and aggregation weakens. Aggregation incentives
are strongest when presidential and legislative elections are concur-
rent and there are a small number of presidential candidates. The
number of presidential candidates in turn is a function of social het-
erogeneity, the electoral formula, and the presence or lack of an

Parliamentary Systems
In pure parliamentary systems, the executive and legislative elections are
always concurrent, and the vote is always fused. As a result, the prob-
ability of capturing the aggregation payoff rests on the strength of the
prime ministerial selection method. A strong method of selection is one
in which the rules or norms of parliament are such that the leader of the
largest party always has the ¬rst opportunity to form a government. If
this is the case, and the leader usually succeeds, then the system looks like
a plurality election “ the leader of the party with the most support
becomes the head of the government (Cox 1997). If, on the other hand,
actors other than the leader of the largest party often form or get a chance
to form the government, then there are weaker incentives to try to
become the largest party. As a political entrepreneur with an eye on the
premiership, the actor™s willingness to go to the costly effort of
A Theory of Aggregation Incentives 45

attempting to organize a large party will be less if success will not
guarantee the premiership. As a candidate or faction leader, uncertainty
about whether a successful cross-district coordination effort will bring
with it the rewards of government is a strong disincentive. In short,
where there is a low probability that the leader of the largest party will
capture the aggregation payoff, aggregation incentives are weak, and
aggregation should be poor.
To summarize, aggregation incentives are a function of the payoff
to being the largest party at the national level and the probability
that the largest party will receive that payoff. Both are necessary to
produce maximum aggregation incentives. A low probability can
undermine cross-district coordination incentives, even if the potential
payoff is large. Likewise, a guarantee that the largest legislative party
will capture the reins of government will not produce strong aggre-
gation incentives if being in power at the national level is not worth
very much.

2.3 conclusion
In this chapter I have presented a theory of aggregation incentives. A
concentration of power within at the national-level of government may
indeed be an important determinant of aggregation incentives, but I
argue that by itself vertical centralization is not enough to produce
aggregation. Two other variables play key roles. First, the degree of
horizontal centralization “ the degree to which power is concentrated
within the national government “ in¬‚uences the incentives to aggregate
across districts. Horizontal centralization interacts with vertical cen-
tralization to determine the payoff to being the largest legislative party. I
argued that the presence of bicameralism, party factionalism, and reserve
domains increase horizontal decentralization and so decrease the size of
the aggregation payoff.
Second, the probability that the largest party will actually be able to
capture the aggregation payoff also shapes coordination incentives.
Where that probability is low, aggregation incentives will be weaker. In
parliamentary systems, the probability of capturing that prize is a
function of who typically becomes prime minister. In presidential sys-
tems, the effective number of presidential candidates and the proximity
Building Party Systems in Developing Democracies

of legislative and presidential candidates shape the probability that the
largest legislative party will also control the presidency.
In the next chapter, I use this theory to derive a series of testable
hypotheses. I then discuss various strategies for operationalizing my
dependent and explanatory variables and test my hypotheses on a large-
N dataset. In Chapters 4“6 I draw on the theory to explain the dynamics
of party system development and aggregation in Thailand and the

Testing the Theory

3.1 introduction
In Chapter 2 I developed a theory of aggregation incentives that stressed
the interaction of the size of the aggregation payoff (itself a product of
vertical and horizontal centralization) with the probability of capturing
that prize. In this chapter, I turn to the task of testing some of the theory™s
hypotheses using a dataset of 280 elections in 46 countries. In Chapters 5
and 6, I conduct further tests of the theory using data from Thailand and
the Philippines. The chapter proceeds as follows. In the next section, I
discuss the operationalization and measurement of the dependent vari-
able “ party system aggregation. I then devote a section each examining
the payoff to aggregation, the probability of capturing the payoff
in parliamentary systems, the probability in presidential systems, and
¬nally the effect of social heterogeneity on aggregation. In each of these
sections, I derive a set of hypotheses from the theory outlined in Chapter
2, discuss my strategy for operationalizing the various explanatory
(and control) variables of interest, describe the dataset used to test the
hypotheses, and ¬nally present the results of those tests. The ¬nal section

3.2 the dependent variable: aggregation
as in¬‚ation
As discussed in Chapter 1, the national party system is the product of
two types of coordination “ intra-district coordination and aggrega-
tion. Simply using the effective number of electoral parties in legislative

Building Party Systems in Developing Democracies

elections is insuf¬cient. The effective number of parties contains
information about both the district-level party systems and the extent
of coordination or aggregation across districts. To test the effect of
various factors on aggregation, we need a way to separate the district
and aggregation effects. One way to do this is to calculate the difference
between the effective number of electoral parties nationally (ENPnat)
and the average effective number of parties in each district (ENPavg)
and to use this as a measure of the extent of aggregation from the local
to national party system (see Chhibber and Kollman 1998). The larger
the difference is, the poorer the aggregation.

D ¼ ENPnat À ENPavg °1Þ

Cox uses this difference measure to calculate how much larger the
national party system is than the average district-level party system
in percentage terms. This measure, which he dubs the party system
in¬‚ation measure (I) is computed by dividing the difference between
ENPnat and ENPavg (D) by ENPnat and multiplying by 100 (Cox 1999,
17).1 Larger in¬‚ation scores correspond to poorer aggregation.


. 8
( 35 .)