Professional Documents
Culture Documents
543–564
ABSTRACT
Introduction
This article introduces some improved measures that gauge the degree of
party linkage, defined as the extent to which parties are uniformly success-
ful in winning votes across districts. While this issue has been relatively
neglected in the literature on party politics until recently, it is of vital
importance if one is trying to understand the formation of national-level
party systems. The examples of the US and India illustrate this point. In
1354-0688[DOI: 10.1177/1354068804045387]
PA RT Y P O L I T I C S 1 0 ( 5 )
both cases, the system of election for the lower chamber is a single-member
district plurality system. At the district level, the effective number of parties
(the number of parties seriously contesting elections) for both countries is
around two. In other words, it appears that both countries have a two-
party system at the district level, on average. However, when we nation-
ally aggregate votes and calculate the effective number of parties at the
national level, a very different picture arises. While the US remains a two-
party system at the national level, India has a very fragmented national
party system in which the effective number of parties is about seven. This
difference stems from variation in the degree of linkage. India is an
example of poor linkage, in that parties across districts have very different
degrees of success in gaining votes. Thus, at the national level, the aggre-
gate size of the party system becomes highly inflated. The case of the US,
on the other hand, exhibits high linkage, in that the same set of parties
competes across districts with equal strength. In high linkage cases, there
is little inflation when local-level party systems are aggregated to the
national level.
Despite the importance of the issue, scholarly understanding about party
linkage, including how to measure it properly, is still very limited. Earlier
scholars who studied party politics often overlooked the issue altogether. As
a result, as Chhibber and Kollman (1998: 330) point out, there has been
‘confusion’ in theories of party systems at the district and national levels.
Even when scholars were aware of the issue (Duverger, 1954; Leys, 1959),
as Cox (1997: 182–5) observes, they assumed that the party systems in their
studies had almost perfect linkage as the result of nationalization of party
organizations. Such assumptions diverted scholarly attention away from
party linkage. Only recently have scholars begun to pay more serious atten-
tion to the issue of linkage (Chhibber and Kollman, 1998; Cox, 1997, 1999;
Hicken, 1999; Kasuya, 2001; Samuels, 2000). However, we have not yet
seen any measures that will allow us to appropriately gauge and compare
the degree of party linkage across countries and over time.
Responding to this concern, our article attempts to provide some
improved measures of linkage, building on Cox (1999), by (1) making it
more intuitive, (2) introducing an appropriate weighting scheme, and (3)
suggesting a subnational-level measurement of party linkage. We demon-
strate the properties and the usefulness of our measures by numerical simu-
lation and by empirical application to data from Italy, India, Germany, and
the United States.
The organization of the article is as follows. The first section discusses
the concept of linkage and provides an operationalizable definition of
linkage. In the second section, we first suggest some desirable criteria in
evaluating indices in general, and then analyze whether current measures of
linkage conform to the suggested criteria. The third section introduces our
suggested measures of linkage. The fourth section compares them with the
Cox index by numerical simulation, and the fifth section examines these
544
M O E N I U S A N D K A S U YA : M E A S U R I N G P A R T Y L I N K A G E
The expression ‘party linkage’ is taken from Cox’s study on electoral coordi-
nation (1997: ch. 10; 1999). He indirectly defines linkage as the ‘phenom-
enon of cross-district linkage of legislative candidates’ (1997: 181). In
particular, it refers to the outcome when ‘politicians seeking election to the
national legislature from different districts . . . run under a common party
label’ (1997: 186).1 Other studies that examine the issue of party linkage
(e.g. Hicken, 1999) use the term in a similar way. Party linkage in their defi-
nition refers to an organizational or labeling issue, which parties and their
candidates have to coordinate before or during the pre-election period. All
these studies, however, use measures based on electoral outcomes to indicate
the level of linkage. This is problematic, since the ex ante coordination
efforts and ex post voting outcomes will only be identical under perfect
information about voters’ preferences across districts and perfect foresight
about events that will happen until the election. In order to save the spirit
of the idea and at the same time ensure measurability, we suggest defining
the concept of party linkage as ‘the degree to which parties are uniformly
successful in winning votes across districts.’2 We believe that this definition,
on the one hand, captures the essence of the phenomenon that Cox has
identified, and on the other, can be operationalized and measured with
election results, as will be demonstrated below.
Party linkage is an issue involved in the process of national-level party
system formation. To isolate its involvement, we consider this process in
two steps, as displayed in Figure 1. Party system here refers to the outcome
of inter-party competition. At the first stage, voters cast their votes and the
district-level party system is formed in each district of a country. At the
second stage of party system formation, votes received by parties in various
districts are aggregated nationwide. The issue of linkage is of central import-
ance in the second step, namely during the aggregation of local party
systems into a single, national party system.
Party linkage varies in degree, and can be conceptualized as a continuum,
as illustrated in Figure 2. When linkage is at its lowest extreme, each district
has an entirely different set of parties, and there is no overlap of parties’
district-level vote shares when votes are aggregated at the national level. If
linkage is at its highest extreme, the set of parties running in each district
is identical, and each party receives the same proportion of votes across
districts. When there is high linkage, the national-level party system
perfectly mirrors the party system of each district. In the proceeding
analysis, we examine how to measure the degree of linkage in real-world
situations.
545
PA RT Y P O L I T I C S 1 0 ( 5 )
....
Figure 1. Conceptual scheme of national party system formation
) * + , ) *
) * + , ) * ) *
Figure 2. Degree of party linkage as a continuum
Cox’s measure of linkage starts with the Laakso and Taagepera (1979)
measure of ‘effective number of parties’ or ENP. ENP measures how many
‘serious’ parties are in the race, disregarding the parties without substantial
support. It is calculated as ENP = 1/(!v k2) where vk is the vote share of the
kth party. ENP can be calculated at any level of aggregation, such as
national, regional, and district levels. Cox (1999) combines the effective
number of parties at the national-level aggregate and at the district-level
average in constructing his measure of linkage. In his own writing, he named
it the ‘inflation measure’3 or I. However, as we will show below, it does not
exhibit the basic properties of an inflation measure. We therefore call it the
‘Cox measure,’ hereafter referred to as C. Cox (1999) defines his measure
as:
ENPnat - ENPavg
C== G $ 100
ENPnat
where:
ENPnat = the effective number of parties at the national level, and
ENPavg = the average effective number of parties at the district level.
Prior to Cox (1999), Chhibber and Kollman (1998) had measured the same
phenomenon4 by using the difference between the effective number of
parties in the national party system and the average effective number of
parties in the local party systems. In its formal expression, this measure
appears as D = ENPnat – ENPavg. This is an inverse measure of linkage: as
D becomes larger, linkage is poorer. Cox’s measure builds upon D. It divides
the difference between the national, and average local, party system size by
ENPnat, and multiplies it by 100. This gives a ‘measure of party system infla-
tion on a percentage basis’ (Cox, 1999: 155). In comparison to D, or the
simple difference between ENPnat and ENPavg, Cox’s measure can provide
us with a more fine-grained picture of the degree of linkage. Since the
measure suggested by Cox (1999) is the most sophisticated one thus far, we
focus on his measure as the point of comparison.5
In reference to the three criteria of evaluation suggested earlier, the
following remarks seem warranted. First, while C indeed is a valid measure
of linkage, as stated in the previous paragraph, it measures linkage in terms
of a share,6 but not as an inflation rate: if C is 10, according to Cox, a share
of about 10% of the overall size of the national party system is attributed
to different parties obtaining votes in different districts. However, C can
assume negative values (see Figures 5 and 6), which is hard to interpret in
terms of a share. Cox’s choice of the term ‘inflation’ itself already suggests
a mode to modify his measure so that it has the properties of an inflation
rate, which would also allow it to meet our first suggested criterion, intu-
itiveness.
Second, Cox’s measure does not allow adjustment for the variation in
district sizes. Here, what we mean by district size is the total number of
547
PA RT Y P O L I T I C S 1 0 ( 5 )
votes cast per district, but not the number of seats elected per district, which
is usually referred to as district magnitude.7 The variation in district size is
likely to be present in cross-country comparisons and may also occur over
time even within one country. As we will show in the simulation section,
this variation can introduce substantial measurement errors of inflation
rates.8 Since, in real-world situations, non-trivial differences exist in the size
of districts within countries, the absence of district size adjustments is likely
to produce biased results. Choosing an appropriate weighting scheme,
therefore, is necessary to meet our second criterion, the comparability across
units of observation.
Third, as we will demonstrate below in the case of Germany, situations
can occur where we observe a low inflation rate which should indicate high
linkage, but in fact parties are not uniformly successful across districts, so
linkage is poor. This happens when there are both districts with many
parties (as measured by ENP) and others with very few parties. The influ-
ences of these districts on ENPavg can offset each other. It follows that
ENPnat and ENPavg can be very similar in size, but parties are not uniformly
successful across districts. In these cases, all discussed measures, since they
only concern the average of district-level party system size, indicate high
linkage, but in fact linkage is poor. This suggests that additional information
is necessary to meet the third criterion, that the measure should gauge the
phenomenon of interest.
In summary, the above discussion guides us to develop three improve-
ments on the Cox measure. First, we attempt to provide a more intuitive
measure that has the properties of an inflation rate as used in economics.
Second, we introduce a weighting scheme that permits general comparison
across countries or over time. Third, we suggest a supplementary measure
that is necessary to gauge the phenomenon of interest accurately. The next
section addresses these three issues.
Pt + 1 = Pt $ d1 +
100 n
Im
Pt + 1 - Pt
Im= e o $ 100
Pt
where:
ENPnat = the effective number of parties at the national level, and
ENPavg = the average effective number of parties at the district level.
I is the simple, alternative inflation measure of party system linkage we
propose in this article. It should be noted that this expression differs in some
important aspects from an inflation rate of commodity prices. While the
commodity price inflation rate is calculated over time, the party system
inflation is constructed across districts. More specifically, the inflation in
commodity prices emerges through new information (new prices), while in
the case of party system formation, inflation is revealed through the differ-
ence in the distribution of parties’ vote shares at the district level versus the
national level.
Similar to the Cox measure, our measure I relates the size of the national-
level party system to the average size of the district-level party systems. If
the size of the national-level party system is larger than the average size of
party systems across districts, our measure indicates that there is inflation
of the party system from the district level to the national level. If the average
size of the party system across districts is larger than the size of the party
system on the national level, we observe party system deflation. In short,
the higher the inflation or the lower the deflation rate, the poorer the linkage
across district-level party systems.
Our measure I has the usual properties of an inflation rate in the sense
that it can exceed the value of 100, whereas C cannot. For example,
consider the case in which there are only three voting districts, with each
district having only one party (each party has a 100 percent vote share in
one district). In this case, there will be three parties on the aggregate level.
The inflation would be from one to three parties, and the measure I delivers
a numerical value of 200 percent. However, C yields the value of 66.6
percent. Therefore, the measure I is more intuitive as an inflation index than
the measure C, thereby improving on the first desirable property of a
measure, namely, the intuitiveness as an inflation index.
To be more precise, however, it should be noted that I is bounded both
from above and below,9 whereas the commodity price inflation rate does
not have an upper bound but its lower bound is –100 percent. This stems
from the fact that future prices are introduced as new information into the
formula. As we discussed earlier, since the cross-sectional inflation rates are
549
PA RT Y P O L I T I C S 1 0 ( 5 )
constructed from two different cuts of the same data, this is not the case in
the measurement of party system inflation. In other words, all the infor-
mation (number of votes parties receive) is revealed at the same time. Thus,
I is bounded both from above and below.10
ENPw-avg will converge to ENPavg when the district sizes become increas-
ingly equal to one another. Replacing ENPavg with ENPw-avg in our infla-
tion measure yields:
ENPnat - ENPw - avg
Iw = f p $ 100 =
ENPw - avg
J N J N
K ENPnat - !ENPi $ vot i O
n
vot K O
K nat O K vot nat $ ENPnat O $ 100
K n
i=1
O $ 100 = - 1
vot i K n ENP $ vot O
KK ! ENP i $ vot nat OO K ! i i O
L
i=1
P L i=1
P
where:
votnat = total number of votes cast at the national level,
voti = number of votes cast in district i, and
ENPw-avg = weighted average effective number of parties at the district level,
with each district’s number of votes cast relative to the national-level aggre-
gate number of votes cast used as weights.
Iw reflects the influence caused by the district size variation on the national-
level party system inflation more accurately than I, since both ENPnat and
550
M O E N I U S A N D K A S U YA : M E A S U R I N G P A R T Y L I N K A G E
ENPw-avg take the relative size of the votes per party into account, while
ENPavg does not. Therefore, using ENPavg may lead to distortions when the
district sizes vary.
552
M O E N I U S A N D K A S U YA : M E A S U R I N G P A R T Y L I N K A G E
P1 P2 P3 P1 P2 P3 P1 P2 P3
D1 1 0 0 D1 1 0 0 D1 0.33 0.33 0.33
D2 0 1 0 D2 0.125 0.75 0.125 D2 0.33 0.33 0.33
D3 0 0 1 D3 0.17 0.33 0.50 D3 0.33 0.33 0.33
Starting Point In-Between Point End Point
districts, as shown in the ‘End Point’ picture (Matrix 3). For each step of
the iteration procedure, we record I, Iw, and C at the national level. Figure
4 below plots this simulation exercise.
In Figure 4, the simulation is complete after 116 iterations. With each
iteration, competition within districts increases, as does the average effec-
tive number of parties at the district level (ENPavg). We record the average
degree of district-level party system competitiveness (ENPavg) on the hori-
zontal axis instead of the actual numbers of the iteration. On the vertical
axis, we record the degree of inflation as indicated by I, Iw, and C.
Figure 4 shows that Cox’s measure is bounded from above by 100%, and
always assumes smaller values than I. At the starting point of the iteration,
I and Iw report about 720 percent inflation.12 Their values then drop sharply
as the degree of competitiveness increases, until the end point of the itera-
tion where the inflation rate becomes zero. C captures this change in a
modest manner, within the range from close to 100 down to 0.
In Figure 4, I is always above Iw. This result is due to the following
mechanism. We start the iteration – that is, we introduce competition – from
the larger districts. This means that the effective number of parties in larger
districts is always greater than, or equal to, the number in smaller districts
until the last iteration, where all districts become fully competitive. If we
weight each district-level ENP with the district size, the ENP of larger
districts are more pronouncedly represented in ENPw-avg. Consequently,
ENPw-avg is always larger than ENPavg. Recall that I is derived by dividing
the difference of ENPnat and ENPavg by ENPavg. For Iw, we simply replaced
ENPavg with ENPw-avg. The size of ENPnat is identical for both I and Iw.
Since ENPw-avg is always larger than ENPavg, the gap between ENPnat and
ENPavg is smaller than the gap between ENPnat and ENPw-avg. Thus, in the
case of Iw, the denominator is larger and the numerator is smaller than in
800
700
Inflation in %
600
500 I
400 Iw
300 C
200
100
0
1.0 1.4 2.2 3.1 4.0 4.9 5.8 6.7 7.6 8.5 9.3 10.0
Degree of Competitiveness (=ENPavg)
the case of I. It follows that the value that results from the formula Iw is
always smaller than that from I.
The second simulation exercise illustrates the ‘concentrated monopoly’
case, in which one party has 100 percent of the votes in all districts at the
starting point. Along the simulation path we move to a perfectly competi-
tive situation. As in the previous case, the introduction of competition starts
in the larger districts. Figure 5 shows the result.
In Figure 5, we again record the average degree of competitiveness at the
district level (ENPavg) on the horizontal axis. We can see that C always
understates inflation and overstates deflation relative to I. Iw is again always
below I, for the same reason explained in the previous simulation.
This simulation result also helps us to understand cases in which we
observe negative inflation, that is, deflation of the national party system size
in comparison to the average size of district party systems. The figure shows
that all three measures first indicate party system deflation, then, in later
iterations, inflation. ENPavg first increases faster than ENPnat, before the
process is reversed. This suggests that when there are structural differences
in party competition across districts, with only a few parties dominating in
a number of districts and a fairly competitive structure in others, then we
are likely to see party system deflation.
The third simulation exercise also uses a ‘concentrated monopoly’ situ-
ation but in this case competition is introduced starting with the smaller
districts. The rest of the simulation procedure is the same as in the two cases
discussed above. The result is plotted in Figure 6.
Figure 6 reveals that Cox’s measure can drop below – 100 percent,
whereas a conventional inflation index is bounded by – 100 percent from
below. This, together with the fact that the Cox measure is bound from
above by 100 percent, confirms that the Cox measure does not have the
properties we usually expect from inflation indices.
We also observe in Figure 6 that, unlike in the previous two simulation
results, Iw is always above I. This is due to the fact that we introduced
30
20
10
Infl at ion i n %
0
1.0 1.4 2.2 3.1 4.0 4.9 5.8 6.7 7.6 8.5 9.4 10.0 I
–10 Iw
C
–20
–30
–40
–50
Degree of Competitiveness (=ENP avg )
-10 1.0 1.4 2.2 3.1 4.0 4.9 5.8 6.7 7.6 8.5 9.4 10.0
-60
inflation in %
I
Iw
-110
C
-160
-210
Degree of Competitiveness (=ENP avg )
V. Empirical Application
557
Table 3. Regional-level party system inflation rates and effective number of parties in Italy, 1994 election
Measure Rank
Region I Iw ENP ENPavg ENPavg-w I Iw ENP ENPavg ENPavg-w
National Aggregate 26.66 26.12 7.58 5.98 6.01 – – – – –
Trentino-Alto-Adige 9.34 8.40 5.15 4.71 4.75 1 1 18 19 19
Lombardia 8.96 4.09 6.37 5.85 6.12 2 4 12 12 12
Piemonte 8.45 3.48 6.59 6.08 6.37 3 7 8 9 11
Molise 6.06 2.42 7.97 7.51 7.78 4 11 1 1 1
Liguria 5.79 3.37 7.10 6.71 6.86 5 8 2 2 3
Veneto 5.70 5.94 6.46 6.11 6.17 6 2 11 11 10
Lazio 5.40 2.23 5.66 5.37 5.54 7 12 14 14 15
558
Basilicata 4.08 2.08 6.64 6.38 6.50 8 13 6 6 8
Sardegna 3.61 4.53 6.85 6.61 6.56 9 3 5 5 5
Calabria 3.44 1.77 6.90 6.67 6.78 10 15 4 4 4
Friuli-Venezia-Giulia 3.41 4.08 6.49 6.27 6.23 11 5 9 10 9
Campania 3.24 3.54 7.09 6.86 6.84 12 6 3 3 2
PA RT Y P O L I T I C S 1 0 ( 5 )
the highest inflation rates first. We also list the rankings of regional I and
Iw, as well as their regional ENP, ENPavg, and ENPw-avg.
The results shown in Table 3 suggest that Italy’s high inflation rate at the
national level actually comes from regional differences, rather than district-
level differences of inter-party competition. The individual regional-level
inflation rates are much lower than the national aggregate-level inflation
rate. This indicates that, within one region, the same parties compete with
similar strength in most districts, but cross-regionally the set of parties
competing is different. In other words, linkage is relatively high across
districts within one region, but the national overall linkage appears weak
due to the poor linkage across regions.
At the same time, Table 3 highlights the importance of using the weighted
inflation measure Iw rather than the unweighted I whenever appropriate.
The rank order of the regional inflation rates changes strikingly from I to
Iw. This change occurs despite the fact that the rankings of ENP, ENPavg,
and ENPw-avg are very similar to each other, but are uncorrelated with Iw,
and only slightly correlated with I. This suggests that, by not using the
weighted inflation measure, we may come up with an incorrect comparison
of regional, international, or across-time differences in the inflation rates.
Finally, we demonstrate the importance of considering both Iw and the
distribution of Ii for accurately measuring linkage. In Table 4, we present
Iw and the summary statistics of Ii, together with other important statistics
related to the district-level party system structure for Germany, India, Italy,
and the US.
The results shown in Table 4 suggest an important point: a low inflation
rate at the national aggregate level does not always result from high linkage,
but can result from the offsetting effect of having some districts contribut-
ing to the national-level party system inflation in an inflationary, and others
in a deflationary, manner. From the minimum value of Ii we can see that all
countries have some districts that exhibit negative values of Ii, which
suggests that these districts’ party systems are more fragmented than the
national-level party system. The existence of these districts contributes to
lower linkage. However, the national-level inflation rate Iw can be small,
due to the offsetting effects of positive and negative values of Ii. In this case,
a national-level low inflation rate is not the result of high linkage of parties
across districts.
To further highlight this point, we plot the histograms of Ii for the elec-
tions of the four countries in Figure 7. In order to make the cases
comparable, we standardized the distribution of Ii by subtracting the mean
of Ii from each variable and dividing it by the standard deviation of Ii. The
resulting variable Ii* is distributed with mean zero and variance one, which
allows us to make at least rough comparisons across countries:
Ii - Ii
I i* = ~ _ 0, 1i
v _Iii
559
PA RT Y P O L I T I C S 1 0 ( 5 )
In the histogram of the US case, we can see that the standardized local infla-
tion measures Ii* are quite narrowly distributed, with the median around
zero. The second peak on the right is due to the existence of districts where
one candidate received close to 100 percent of the votes. In contrast,
Germany’s distribution of Ii* covers a wide spectrum, indicating large differ-
ences in contributions of district-level party systems to the national-level
inflation rate. This can also be seen in Table 4, where we find a negative
median of Ii, indicating that more than 50 percent of the districts contribute
to the overall party system inflation rate in a negative manner. Thus we
conclude that although Germany has a very low national-level inflation rate
(Iw = 5.4), its party linkage is relatively poor. Italy’s case is similar to
Germany’s. It exhibits a wide distribution of Ii* while Iw is relatively small
(Iw = 14.2), suggesting that linkage is actually not so high in Italy. The
Indian case shows a wide distribution of Ii* combined with a very high Iw
(155.8). In this case, there are some offsetting effects but the presence of
districts having inflationary contribution is so large that the national aggre-
gate inflation index yields a very high value. What we can see from these
four graphs is that neither Iw nor Ii* alone are sufficient to determine degrees
of linkage. Only a small Iw combined with a narrow distribution of Ii*, such
as in the case of the US, indicates high linkage. It is therefore necessary to
look at both the national-level measure of inflation Iw and the distribution
of the local-level inflation measure Ii in order to identify the degree of
linkage accurately.
560
M O E N I U S A N D K A S U YA : M E A S U R I N G P A R T Y L I N K A G E
1 E
1
E
16%
14%
14%
12%
Frequency in %
12%
Frequency in %
10%
10%
8% 8%
6% 6%
4% 4%
2% 2%
0%
0%
.6
.2
.8
.4
.6
.2
-3
-1
6
1
3
0.
0.
1.
1.
2.
2.
-2
-2
-1
-1
-0
-0
.6
.2
.8
.4
.6
.2
-3
-1
6
1
3
0.
0.
1.
1.
2.
2.
-2
-2
-1
-1
-0
-0
Bin
Bin
1E
1
E
35%
14% 30%
Frequency in %
12% 25%
Frequency in %
10%
20%
8%
15%
6%
4% 10%
2% 5%
0%
0%
.6
.2
.8
.4
.6
.2
-3
-1
6
1
3
0.
0.
1.
1.
2.
2.
.6
.2
.8
.4
.6
.2
6
-3
-1
-2
-2
-1
-1
-0
-0
3
0.
0.
1.
1.
2.
2.
-2
-2
-1
-1
-0
-0
Bin
Figure 7. Histogram of distribution of Ii* for Germany, India, Italy, and the US
Conclusion
way, while others do so in a negative manner. This happens when the distri-
bution of local inflation measures is widely spread. Only when the national-
level party system inflation rate is low and when the distribution of local
measures is narrow can high linkage be identified. In sum, in analyzing party
linkage, we suggest consideration of the weighted version of our suggested
inflation measure together with the distribution of the local inflation
measure, especially for international or across-time comparisons.
Nevertheless, it should be noted that this article has not attempted to
analyze the causes of variation in the degree of linkage. Future studies could
aim to undertake this.
Notes
The authors would like to thank Frank Leibbrand, Arend Lijphart, Kelly Patterson,
and two anonymous reviewers for helpful comments, as well as Paola Sapienza for
making Italian data available to us. All remaining errors are ours.
1 In this passage, Cox seems to refer indirectly both to parties’ organizational effec-
tiveness in coordination across districts as well as to parties’ ability to uniformly
respond to voters’ preferences. We view these two factors as potential causes of
linkage, which cannot be measured directly by voting outcomes. We therefore
refer to linkage as the measurable concept defined below.
2 This definition is similar to what Jones and Mainwaring (2003) called ‘nation-
alization.’
3 We follow Cox’s choice of the term ‘inflation’ measure as a measurement of party
linkage, since our proposed measure formally and conceptually resembles an
inflation rate as used in economics.
4 Chhibber and Kollman refer to the phenomenon Cox called linkage as ‘party
aggregation’ (Chhibber and Kollman, 1998).
5 Jones and Mainwaring (2003) introduce an alternative measure based on the
Gini coefficient. We will compare their measure to our suggested measures at the
end of section III.
6 This can also be seen from the fact that Cox’s measure cannot exceed 100
percent, while an inflation rate can (see Figure 4).
7 As a way to adjust his index to real-world situations, Cox suggests that ENPavg
should be calculated as a weighted average, ‘where each district is weighted by
the share of assembly seats elected from within that district’ (Cox, 1999: 156
fn.1). In other words, he suggests weighting ENPavg by district magnitude. This
procedure is identical to weighting by district size (the total number of votes cast
per district) if the district size variation is proportional to the number of seats
per district (the district magnitude). However, if that is not the case, only
weighting each district by relative district size, but not by district magnitude,
addresses the variation in the number of votes cast per district.
8 The following example may illustrate the importance of weighting by district
size in the context of party linkage. Consider a country that has large voting
districts in cities and small ones in rural areas. With weighting, both the calcu-
lations of ENPavg and ENPnat take the variation of district sizes into account.
562
M O E N I U S A N D K A S U YA : M E A S U R I N G P A R T Y L I N K A G E
Without weighting, the large districts will get under-represented and small
districts over-represented in ENPavg, which implies that party linkage between
small districts receives the same level of attention as links between large
districts.
9 The same holds true for C, since I = 100 $ C
100 - C
10 In formal expression, I is bounded from above as: I ≤ (ENPnat – 1) · 100, and
bounded from below as: d 1 - 1n $ 100 < I
ENPavg
11 For an empirical application of this type of use of Ii see Kasuya (2001).
12 If all districts were of equal size, this number would of course be 900 percent.
13 In all cases, we analyze the recent election results for the lower chamber of the
national legislature; the Bundestag for Germany, Lok Sabha for India, Camera
dei Deputati for Italy, and the House of Representatives for the US. In the case
of the German Bundestag, we report the results of the single-seat constituency
electoral returns, and in the case of the Italian Camera we use the proportional
tier results. The sources of election data are as follows. The German data is from
the website of the Bundeswahlleiter (Federal Election Commissioner):
http://www.destatis.de/wahlen/bundestagswahl2002/deutsch/wahlvor98/index.h
tm (accessed July 2002). Data for Italy was obtained from the Ministero dell’
Interno (Ministry of Interior) courtesy of Paola Sapienza. Indian data is from the
Electoral Commission of India’s website: http://www.eci.gov.in/ge99/
index_ge1.htm (accessed July 2002). The US data is from the Inter-university
Consortium for Political and Social Research (1995).
References
Chhibber, Pradeep and Ken Kollman (1998) ‘Party Aggregation and the Number of
Parties in India and the United States’, American Political Science Review 92:
329–42.
Cox, Gary W. (1997) Making Votes Count: Strategic Coordination in the World’s
Electoral Systems. Cambridge: Cambridge University Press.
Cox, Gary W. (1999) ‘Electoral Rules and Electoral Coordination’, Annual Review
of Political Science 2: 145–61.
Duverger, Maurice (1954) Political Parties: Their Organization and Activity in the
Modern State. New York: Wiley.
Hicken, Allen D. (1999) ‘Political Parties and Linkage: Strategic Coordination in
Thailand’, paper prepared for delivery at the 1999 Annual Meeting of the
American Political Science Association, Atlanta, September 1999.
Inter-university Consortium for Political and Social Research (1995) ‘Candidate and
Constituency Statistics of Elections in the United States, 1788–1999.’ (computer
file) (Study #7757), 5th ICPR ed. Ann Arbor, MI: Inter-university Consortium for
Political and Social Research.
Jones, Mark P. and Scott Mainwaring (2003) ‘The Nationalization of Parties and
Party Systems: An Empirical Measure and an Application to the Americas’, Party
Politics 9–2: 139–66.
Kasuya, Yuko (2001) ‘Party System Linkage: Explaining its Variation in the Philip-
pine Case’, paper prepared for delivery at the 2001 Annual Meeting of the
American Political Science Association, San Francisco, August 2001.
563
PA RT Y P O L I T I C S 1 0 ( 5 )
Laakso, Marku, and Rein Taagepera (1979) ‘Effective Number of Parties: A Measure
with Application to West Europe’, Comparative Political Studies 12: 3–27.
Leys, Colin (1959) ‘Models, Theories and the Theory of Political Parties’, Political
Studies 7: 127–46.
Samuels, David J. (2000) ‘The Gubernatorial Coattails Effect: Federalism and
Congressional Elections in Brazil’, Journal of Politics 62–1: 240–53.
564