. 9
( 35 .)


I ¼ ½°ENPnat À ENPavg Þ=ENPnat Š à 100 °2Þ

The interpretation of this measure is straightforward. If I is 10, this
suggests that only 10% of the size of the national party system can be
attributed to different parties garnering votes in different parts of the
country (poor aggregation), with the other 90% due to the average
number of parties at the district level (ibid.). In short, aggregation is very
good “ the same parties are generally the frontrunners in most districts
nationwide. On the other hand, if the in¬‚ation score is 60, we know that
poor aggregation deserves most of the credit for producing a large number
of parties nationally, while intra-district coordination can only account
for 40% of the national party system™s size. I use the in¬‚ation score as my
aggregation measure throughout this chapter and the remainder of
the book. Once again, higher in¬‚ation scores represent worse levels of
aggregation (or the more severe the cross-district coordination failures).
To create the in¬‚ation measure, I collected district and national level
election returns and calculated ENPnat and ENPavg for 280 elections in 46

Cox multiplies this by one hundred to convert the decimal into a percentage.
Testing the Theory 49

countries. Over half of these countries (25) are developing democracies. I
included only those elections where a country was minimally democratic
(de¬ned as a having a Polity2 score above 0).2 These election return data
were culled from various sources including Caramani (2000), Matt
Golder™s Democratic Electoral Systems Around the World dataset,3 Scott
Morgenstern™s District-Level Electoral Dataset,4 and the author™s own
work on elections in Southeast Asia (Hicken 2002).5 Calculating both
kinds of ENP can be complicated, particularly where party alliances are
common, election results include large “other” categories, or there are
large numbers of independent candidates. Fortunately, the percentage of
votes cast for parties in the “other” category, or for independents is
generally small, rarely more than 5% of total votes cast. Where I had
information about the number of parties in the “other” category, or the
number of independents I used this information to create an “average”
score for all other parties/independents by dividing the total vote share for
“others”/independents by the number of parties/candidates in that cate-
gory.6 Where I lacked this information, I was forced to treat “others” and
independents as if they were single parties.7 In the case of party alliances,
to calculate both ENPavg and ENPnat, I used as my basic unit of analysis
the entity for which voters actually cast their vote on election day. This
means that the alliance is counted as a single party if the alliance by itself
appears on the ballot and that is what voters vote for. If, on the other hand,
parties may enter into an electoral alliance but individual party names still
appear on the ballot separately, I count the votes for each individual party
in the alliance.
Table 3.1 presents the average in¬‚ation scores for the 46 countries I
use to test the theory. Since many of my key independent variables

The Polity2 score is a combination of the autocratic and democratic variables in the
Polity IV dataset (Marshall and Jaggers 2002).
http://homepages.nyu.edu/%7Emrg217/elections.html. See also Golder (2005).
Special thanks also to Gary Cox and Ken Kollman for sharing their district-level data
with me.
For example, (Vote share for “others”) / (# of parties in the “other” category).
The actual difference in the ENP score using the two methods is actually quite small as
long as the percentage of “others” or independents is not large (which is almost always
the case for the set of elections used here). For example, in the 1983 Thai election
independent candidates received 7.4% of the vote. If we divide that percentage by the
number of independents (in effect treating each as a party of one), then ENPnat is 5.9. If
we, instead, count all independents together as a single “party,” then ENPnat is 5.7.
Building Party Systems in Developing Democracies

table 3.1. Average In¬‚ation Scores

Country In¬‚ation
Costa Rica .00
Cyprus .00
Honduras .02
Austria .02
Chile .04
Denmark .04
Sweden .04
Jamaica .04
Dominican Republic .05
Greece .05
Norway .05
Ireland .05
Iceland .05
Netherlands .06
Mexico .07
Colombia .07
Mauritius .07
Venezuela .08
New Zealand .08
United States .10
Italy .10
Japan .10
Taiwan .12
Portugal .12
Botswana .13
Poland .14
Spain .15
United Kingdom .15
Argentina .16
Trinidad .17
Zambia .17
Australia .17
Kenya .18
Finland .19
Canada .20
Germany .21
South Korea .28
France .29
Brazil .29
Philippines .32
Malawi .32
Switzerland .38
India .47
Belgium .48
Thailand .50
Ecuador .55
Testing the Theory 51

range from 0 to 1, I convert the in¬‚ation score to a variable that also
ranges from 0 and 1.8 This eases the interpretation of the results. Scores
range from a low of .00 in Costa Rica to a high of .55 in Ecuador.

3.3 the aggregation payoff (size of the prize)
Chapter 2 discussed a variety of factors that I argue should affect
aggregation incentives. These can be broken down into two categories “
those that affect the size of the payoff for being the largest legislative
party and those that affect the probability that the largest party will
capture that payoff. Starting with the payoff, I argued that the size of
the prize is a function of both the degree of vertical centralization in the
polity as well as the degree of horizontal centralization.

Vertical Centralization
If Chhibber and Kollman (2004) are correct, the more power and control
of resources are devolved to subnational actors, the worse aggregation
will be. This is summarized in the following hypothesis. (In all of the
hypotheses that follow, I state the relationships in terms of the in¬‚ation
score. Recall that higher in¬‚ation scores equate to poorer aggregation.)

Vertical Centralization Hypothesis: The degree of vertical centrali-
zation is negatively related to in¬‚ation.

To estimate the effects of vertical centralization I use two newly
developed measures of ¬scal decentralization created by the World
Bank (World Bank n.d.). Subrevgdp measures subnational government
revenues as a share of GDP, while subexpengdp does the same with
subnational expenditures. The results are generally robust to the use of
either measure so for the sake of consistency I report only the revenues
measure in the following models, noting the few instances where the
choice of one or the other makes a substantive difference.9

This is done by simply taking the percentage difference in size between
the national and local party system, without multiplying the result by 100:
I ¼ (ENPnat À ENPavg)/ENPnat.
The World Bank also reports subnational revenues and spending as a percentage of
total governmental spending, but I prefer the percent of GDP ¬gures because they
simultaneously capture the degree of ¬scal (de)centralization along with the relative
importance of government spending vis--vis the economy as whole. However, the
results are not dependent on which measure I use.
Building Party Systems in Developing Democracies

Horizontal Centralization
In addition to vertical centralization the degree to which power is
concentrated within the national government (horizontal centraliza-
tion) also affects aggregation incentives. Speci¬cally, I argued that in
the presence of bicameralism, reserve domains, and party factionalism
aggregation incentives should be weaker. Together these variables
shape the degree of horizontal centralization. I also discussed a coun-
terargument that links horizontal decentralization to increased aggre-
gation. To summarize:

Bicameral Hypothesis: In¬‚ation will be higher in bicameral systems,
relative to unicameral systems
Reserve Domain Hypothesis: Where reserve domains exist in¬‚ation
will be higher.
Party Factionalism Hypothesis: The degree of party factionalism is
positively related to in¬‚ation.
Horizontal Centralization Hypothesis: The degree of horizontal
centralization is negatively related to in¬‚ation.
Alternative Hypothesis: The degree of horizontal centralization is
positively related to in¬‚ation.

In an effort to capture the extent of horizontal centralization, I use
several different measures created using data from the Database of
Political Institutions (DPI) (Beck et al. 2001) and Matt Golder˜s Demo-
cratic Electoral Systems Around the World dataset (Golder 2005).10 For
bicameralism, I created a variable called Senate that equals 1 if there is an
upper chamber in the legislature; 0 otherwise.11 As a proxy for the
presence of reserve domains, I use a measure of the military™s involvement

http://homepages.nyu.edu/%7Emrg217/elections.html. See also Golder (2005).
As an alternative, I also used Henisz™s measure of “an effective second legislative
chamber” (L2), which takes on a value of 1 if the second chamber that is elected
under a distinct electoral system has a substantial (i.e., not merely delaying) role in
¬scal policymaking (2000). I prefer my simpler measure of bicameralism for two
reasons. First, we know that upper chambers can affect policy even when they lack
formal veto authority (as the Thai case will demonstrate in Chapters 4 and 5). Second,
the Henisz measure excludes some important legislative powers that might constitute
a check on executive authority (e.g., appointment and con¬rmation authority). Using
L2 rather than Senate yields similar results, though L2 is always weaker than Senate
and often falls below traditional levels of signi¬cance. Other variables in the models
are substantively unaffected by the substitution of L2.
Testing the Theory 53

in politics. Military5 equals 1 if the chief executive has been a member of
the military in the last 5 years. The results are robust to expanding the
time frame from 5 to 10 years. These data come from the DPI.
As a strategy for operationalizing the degree of party cohesion/
factionalism, I rely on the coding scheme developed by Carey and Shugart
(1995) as adapted and extended by Wallack et al. (2003). This scheme is
designed to capture differences in the incentives to cultivate a personal
vote (versus a party vote) across different electoral systems. A large
number of scholars have argued that strong incentives to cultivate a
personal vote undermine party cohesion and promote party factionalism
(e.g., Katz 1986; Shugart and Carey 1992; Reed 1994; Lijphart 1994;
Hicken 2002, 2007). The personal vote undermines party cohesion and
promotes factionalism in at least three ways (Reed 1994). First, because in
personal vote systems candidates typically do not owe their election to the
party, they have less reason to be loyal to it once elected. Second, the
independent campaign organization needed to win in candidate-centered
elections gives politicians the means to stray from the party line without
fear of major repercussions or leave the party all together. Third, in
building an independent campaign base, candidates incur debts, make
compromises, and develop loyalties to constituencies that may be dif-
ferent from other candidates from within the same party (ibid.).
Carey and Shugart suggest three variables that they argue shape the
extent to which candidates running for of¬ce have an incentive to culti-
vate a personal vote. Each of the three variables, Ballot, Pool, and Vote,
are coded as 0, 1, or 2, where higher values denote greater incentives to
cultivate a personal vote.12 Using the Carey and Shugart coding scheme as
a template Wallack et al. (2003) collect data on incentives to cultivate a
personal vote for 158 countries for the years 1978“2001.13 They average
across Ballot, Pool, and Vote to create a variable called parindex, which I
use here as a proxy for party cohesion. Parindex ranges from 0 to 2 with
higher values being associated with stronger incentives to cultivate a

Ballot measures “the degree of control party leaders exercise over access to their
party™s label, and control over ballot rank in electoral list systems.” Pool captures
whether votes for one candidate affect the number of seats won in that district by the
party as a whole. Vote codes the nature of voters™ choice (for a party, candidate, or
multiple candidates). (Carey and Shugart 1995, 418“21).
Wallack et al. and Carey and Shugart differ in their treatment of single member
district systems. See Wallack et al. (2002, 7) for more details.
Building Party Systems in Developing Democracies

personal vote (and greater tendencies toward party factionalism).
Parindex is admittedly a very crude and indirect proxy for party
factionalism/cohesion. However, it has the major advantage of being
free of the endogeneity concerns that would accompany a more direct
measure of party cohesion. Still, as I discuss later, parindex may ulti-
mately be too crude a proxy to be useful.

Control Variables


. 9
( 35 .)