Party Aggregation and Political Consolidation in
the American States
James Hedrick & Jaclyn Kettler
Political Science Department
Rice University
March 15, 2011
1
Abstract
Electoral institutions and social heterogeneity have long garnered the most
attention from political scientists looking to explain the variation in party
systems across different countries. However, less attention has been paid to
explaining the overtime variation in the number of parties within political
jurisdictions, ignoring the longitudinal variation. Additionally, the study of
party systems in the United States has focused mainly on the failure of third
parties and the lack of a major American socialist party, neglecting the broader
party systems literature. This paper proposes to test the party aggregation
theory of Chhibber and Kollman which argues that fiscal centralization creates
incentives for parties and voters to coordinate that decreases the number of
effective parties within a jurisdiction, using several American states as a testing
ground. Our analysis provides qualified evidence for Chhibber and Kollman’s
party aggregation thesis, while incorporating more of the broader comparative
literature on party systems into the study of American state party systems.
1
Introduction
Electoral institutions and social heterogeneity have long garnered the most attention from political scientists looking to explain the variation in party systems across
different countries. However, less attention has been paid to explaining the overtime variation in the number of political parties within political jurisdictions. In
short, most research has addressed cross-national rather than longitudinal variation.
However, many countries have seen large fluctuations in the strength and number of
viable national parties over their history (Chhibber & Kollman 1998, Chhibber &
Kollman 2004). Institutions and social cleavages, however, rarely change and thus
do a poor job of explaining this overtime variation.
This paper tests Chhibber and Kollman’s central thesis that increasing fiscal cen-
2
tralization by a government incentivizes voters to choose broadly competitive parties
over local or regional parties and thus drives down the effective number of parties
(ENP). While Chhibber and Kollman have shown evidence that fiscal centralization
leads to fewer parties at the national level, their theory of party aggregation still
has not been extensively tested, and the mechanism has yet to be studied at the
subnational level, in this case the American states. This project proposes an empirical test of the theory on a selection of five American states to see if greater fiscal
centralization affects party aggregation at the state as well as the national level. In
short, does the theory of party aggregation by political centralization explain the
variation within the American states as well as at the national level?
2
Literature Review
As mentioned above, institutions and social cleavages have received the bulk of the
consideration from political scientists. Most modern institutional research on party
systems was kick-started by Duverger’s famous law that plurality electoral systems
and single-member districts (SMD) favor the emergence of two-party systems (1954).
Obviously, however, the empirical reality did not conform entirely to Duverger’s
expectations. Several states have displayed a stubborn tendency not to develop
nation-wide two party systems. India is an often cited example (Chhibber & Petrocik
1989, Riker 1982, Lijphart 1994). Other primarily English-based systems like Canada
and Britain have historically also seen divergence from the Duvergerian equilibrium
in their national legislatures, even though most of their legislative elections have used
3
SMD’s and plurality voting (Chhibber & Kollman 2004).
Since its publication, Duverger’s originally theory has been refined significantly. As
many scholars have noted, it is not a strictly determinative law (Ware 1996). The
most important refinement of Duverger is almost certainly Cox’s seminal work Making Votes Count (1997). Here Cox emphatically argues that Duverger’s law is an
explicitly district-level phenomenon. To illustrate, Cox argues that Duverger’s law
really only ensures that a country, or other district-divided political jurisdiction, will
have no more than twice the number of parties as electoral districts. The two-party
equilibrium is an institutional function of each individual district. In other words, if a
political jurisdiction has X districts, Duverger’s law only sets a theoretical maximum
of 2X parties for the jurisdiction overall.
Therefore, another process, something other than simple mechanical institutionalism,
is needed to go from district-level bipartisanism to jurisdiction-level bipartisanism.
Cox refers to this process as ‘linkage’ or ‘cross-district coordination’ and argues that
it is one of the most understudied areas in the party system literature (Cox 1999). In
fact, in his review of the electoral rules and coordination literature he only finds two
articles (Chhibber & Kollman 1998, Samuels 1998) that address this issue directly.
What is missing is some sort mechanism that drives the effective number of parties
at the national level towards the number of parties at the district level. In short, an
intervening variable or process is needed to bring the overall political jurisdiction’s
party system inline with the Duvergerian bipartisanism that develops at the districtlevel in plurality electoral systems.
4
A number of different theories have been proposed to explain linkage based on coordination of candidates in particular. Cox argues that the existence of an upper-tier
legislative chamber (such as the U.S. Senate which is geographically, rather than
district, based) or a presidential system can incentivize candidates to coordinate
in order secure control of the central government (Cox 1997, Cox 1999). Aldrich
presents a similar argument that elite coordination within the legislature helped establish national party labels in the early U.S. legislature (Aldrich 1995). In both
cases, parties were established by elites to overcome collective action problems so as
to better control public resources.
It’s important to note that this is an primarily bottom-up theory of party system formation, which argues implicitly that parties initially form at the district/local level
but some other incentive needs to exist to cause candidates and parties to coordinate
across districts. However, it is also important to point out that voters also need some
incentives to identify with a particular party brand name as well, and an even further
incentive to choose a nationally competitive party over a favored regionally competitive one (Chhibber & Kollman 2004). In short, explaining how parties aggregate (or
coordinate, or link) across districts to form broader-based political parties requires
theorizing about the incentives of both candidates to coordinate and voters to prefer
nationally competitive parties to local ones that might have policy preferences closer
to their ideal.
Chhibber and Kollman’s own theory, the obvious inspiration for this paper, argue
that fiscal and governmental centralization drives party aggregation by affecting the
incentives of both candidates and voters (1998, 2004). Each group must adjust its
5
strategic calculations to accommodate a more active and involved central government. Candidate cross-district coordination incentives can be a reaction to either
voter preferences or the desire to influence national policies. The later reason is fundamentally the same as the Cox and Aldrich arguments, but the former brings in the
desires of the citizenry. The important development in the Chhibber and Kollman
argument is that “under a system of central authority, voters see themselves as either benefiting or losing from national policies (1998, pg. 335).” In this cases, voters
have strategic reasons to choose an overall competitive party over a preferred local or
regionally competitive one. Basically, with a stronger centralized government, access
to public resources depends more on the actions of the national (or state) government
than the local one. In short, the stakes are too high for voters to continue to support
non-nationally competitive parties. This has the effect of driving down the number of parties. Empirical evidence for this phenomenon has been shown in the U.S.
and India (Chhibber & Kollman 1998) and Brazil, a multi-member district system
(Samuels 1998) as well as Canada and the Britain (Chhibber & Kollman 2004).
To be fair, it is necessary to briefly mention some research on the effect of social
cleavages on linkage or party aggregation. While the Chhibber and Kollman thesis
owes its intellectual heritage primarily to the institutional literature, there is some
evidence that cleavages can affect linkage as well, although primarily in a negative
sense. Simply put, social and ethnic heterogeneity can increase the “chances of
malcoordination (Cox 1999, pg. 159)” and therefore inflate the effective number of
parties in a jurisdiction. Social heterogeneity can intervene in the linkage process,
causing variations in the effectiveness of party aggregation. While research on the
6
effect of various types of social heterogeneity on party systems is extensive (Lijphart
1979, Kalyvas 1998, Lipset & Rokkan 1967), very little has been done on the effect
of social cleavages specifically on the process of linkage. Would voters of a broadly
diverse jurisdiction - ethnically, linguistically, racially, etc. - find it more difficult to
coordinate for political gain than voters in a more homogeneous jurisdiction when
faced with increased in political centralization?
In short, while many authors have argued that the outcome of the party system
is best shown as an interactive process between social heterogeneity and electoral
institutions (Cox 1999), our focus here is the effect of fiscal centralization on ENP,
and we ignore social heterogeneity. For one, heterogeneity is of less concern to our
overall empirical research question. Its proposed theoretical impact is primarily
believed to have a negative effect on party aggregation - meaning that heterogeneity
should inflate the effective number of parties - making it more difficult to find evidence
for our theory, therefore, it is unlikely to bias our results. We do intend to include a
measure of social heterogenity - likely the ‘birthplace index’ of ethnic fragmentation
(Costa & Kahn 2003, pg. 105) for both theoretical and data availability reasons - in
later versions of the analysis, but for the moment we ignore it.
Finally, while linkage has been only minimally studied, there has been a small, if
interesting, set of research on the electoral support for so-called third parties in the
United States (Hirano & Snyder 2007). However, this literature has been motivated
differently than the literature on linkage and party aggregation. Fundamentally, both
are interested in explaining the seeming stability of the U.S. two party system. However, the U.S. third-party literature has been motivated by the absence of a major
7
socialist party in the U.S. (Lipset 1977, Lipset & Marks 2001) rather than how the
overall party system itself emerged. Because of this, there is a major methodological
difference between the two literatures. While the party systems and elecotoral institutions research focuses on the number of parties as a measure of the party system,
the third-party literature typically uses electoral support rather than the effective
number of parties as a measure of the dependent variable.
Because of this, the American-focused third-party literature has developed a different
series of explanations for the stability of the U.S. party system. Coincidentally, some
authors have come back to an explanation of the ineffectiveness of third-parties based
primarily on the difficulties of overcoming SMD’s and plurality voting (Rosenstone,
Behr & Lazarus 1996). However, as mentioned above, this relatively constant institutional rule still does not explain longitudinal variation in the number of parties
nationally. Additionally, other institutional changes, specifically the adoption of the
direct primary and the Australian ballot, have been cited as explanations for the
overall lack of third-party voting in the United States (Epstein 1989, Ware 2002).
However, tests of these institutional variables on electoral support for third parties
have seen inconsistent results (Hirano & Snyder 2007, Crespin 2004).
Likely the most popular explanation for the failure of third parties in the United
States is co-optation, specifically by the Democratic party (Oestreicher 1988). This
theory argues that American third parties are primarily a left-wing phenomenon.
Third parties like the Populists and the Progressives account for most of the thirdparty movements, and the Democratic party has routinely adopted the platforms
of these parties as their own, most noticeably during the New Deal but also with
8
the free silver issue and the presidential nomination of William Jennings Bryan in
1896. However, the empirical evidence for this is incomplete and consists mainly of
evidence that the Democratic Party moved to the left ideologically, previously thirdparty affiliated candidates switched primarily to the Democrats, and counties that
had previously had high third party vote totals became more Democratic afterward
(Hirano & Snyder 2007).
Overall though, the American third-parties literature suffers from a lack of generality. First, it focuses on the party system in a single nation. Secondly, because it is
puzzled by the lack of a major socialist movement in the U.S., it is preoccupied with
explaining the actions of parties of the left and not the development of the U.S. party
system more generally. As we’ll show below, some of the longitudinal variation in
the effective number of parties within the U.S. states is related to malcoordination
on the right. Finally, there is nothing in the empirical evidence presented by Hirano
and Synder (2007) that is incompatible with the thesis of party aggregation by fiscal centralization or political consolidation. They acknowledge that the New Deal
spelled the death knell for third party voting, a finding equally compatible with the
party aggregation theory. In addition, party aggregation is likely to result in both
candidates and voters abandoning non-competitive regional parties for strategic reasons. Basically, while the mechanism, co-optation versus political consolidation, is
different between the two theories, the empirical evidence presented to support the
co-optation argument equally supports the party aggregation theory.
9
3
Research Design
In short, two main theories were presented above. The first, party aggregation by
political consolidation, predicts that the effective number of parties will decline as
the government becomes more centralized and begins to play a larger role in the lives
of its citizens. Alternately, the co-optation argument states that third party electoral
support declined as a result of the leftward ideological shift of the Democratic Party,
particularly after the New Deal. The primary difference between the two is the
mechanism of the theory, ideological movement or fiscal centralization. Additionally,
the co-optation argument is not necessarily incompatible with the fiscal centralization
argument but mistakes an effect for a cause. If the party aggregation theory of
Chhibber and Kollman is correct, political consolidation should precede the party
aggregation, and the ideological shift should be an effect of new party members
affecting the internal deliberations of the party, pushing them ideologically. Even a
strategic move meant to garner votes by party elites is only an effective co-optation
tactic if the voters have an incentive to shift to a party with less commitment to
their issues.
This design does not propose to offer a definitive test of the two theories. It simply
proposes to test the fiscal centralization argument at the state level using five American states. According to Chhibber and Kollman, “the logic for party aggregation
is independent of the level of government...these effects may occur at the state level
as well (Chhibber & Kollman 1998, pg. 340).” By using multiple states (instead
of a single country) as the units of analysis, we can test the strength of the party
10
aggregation theory in multiple jurisdictions, with considerable variation in the centralization of their governments and the expansion of the government service sector
both cross-sectionally and overtime. Also, according to the co-optation argument,
the decline of third parties was primarily a top-down, national phenomenon and a
result of elite strategic calculations. However, if the results from this design show a
variation in the decline of the ENP across different states at different times, related
to government centralization, then that will indicate that political consolidation and
strategic behavior on the part of both parties and voters are the underlying cause of
the stable two-party system in the United States.
4
Data & Methods
As mentioned above, the units of analysis for this design will be a selection of American states. Ideally, all the states would be included in the study and combined into
a time-series, cross-sectional design that would account for their time of entry into
the dataset based on when they joined the union. Unfortunately, there are a couple
of issues with this approach. First, there is the issue of the South and its political
history of de facto one-party government as well as how to account for the Civil
War period. Additionally, several states have used, and continue to use, some form
of multi-member districts to elect legislators.1 While evidence indicates that party
aggregation due to fiscal centralization may not be a phenomenon entirely limited to
1
See Maryland, Illinois pre-1980, Massachusetts, Ohio, etc. Some states have used non-partisan
systems as well, such as Minnesota between 1917 and 1976, which would also complicate the design.
For a brief overview, see Hamm & Moncrief (2007).
11
single-member district plurality systems (Samuels 1998), including these states would
unnecessarily complicate the design. In short, accounting for factors like region and
the effects of non-SMD districts is a subject for a separate paper. No to mention that
acquiring the district-level and state-legislative results for all states over an extremely
long time series would require a massive data collection effort.
Given the above, five states have been selected to include in the design: Pennsylvania,
New York, Virginia, Michigan, and California. The first three have been states since
1790 and can be analyzed over a time series similar to that used by Chhibber and
Kollman for the overall United States (1998). Michigan and California obviously were
admitted to the union later, 1837 and 1850 respectively, and offer the opportunity to
examine whether states joining the union later followed a different party aggregation
process. You’ll notice that one southern state, Virginia, was included, which will
hopefully shed some light on the party aggregation processes in the southern states.
However, the eventual results for Virginia should be viewed separately from the other
four states.
To test the party aggregation theory, it’s necessary to calculate two different measures
for each state. The first is the ENP at the state-level; the second is some measure
of state-level political consolidation. First, there are a number of ways to measure
ENP longitudinally. In their original statement of the party aggregation theory,
Chhibber and Kollman used a measure of average ENP per district combined with
a measure of state-level ENP which was itself based on an older measure (Laakso &
Taagepera 1979). Unfortunately, given the lack of available data on the popular vote
at the district level for state legislative seats, we cannot construct a similar measure
12
(currently) for the American states. Instead, we estimated the effective number of
parties at the state-level using data on the distribution of seats of each states’ lower
chamber.
The data for calculating these measures will come largely from a dataset available
from the ICPSR, Partisan Division of American State Government (Burnham 1993).
This data contains information on state legislative members from 1834-1985. While
not complete - some years are missing and some elections are missing party labels the data for the five states mentioned above is relatively complete and any missing
data should not affect the results of the statistical and graphical analysis. In the
future, we hope to collect data to extend the dataset back to 1790 for Virginia,
Pennsylvania, and New York. However, this will require extensive archival research
as well as direct contact with the secretary of state’s office for each state. This is yet
another reason we have limited this initial design to a select, manageable number
of states. Given the limitations of the fiscal data available, discussed in more detail
below, our time series for Pennsylvania, New York, Virginia stretches from 1834
to 1912, while Michigan and California are analyzed beginning in 1840 and 1850
respectively until 1912.
Finally, since our data did not, at times, fully describe the party affiliation of each
chamber member, we used an equation for ENP that allowed us to estimate a reliable
number from incomplete data (Taagepera 1997). Taagepera’s approach allows the
estimation of a reliable measure of ENP when some parties are simply categorized
as “Other.” In our case, the Burnham data only provided for up to four parties
within a legislature at a time. However, there were several instances where there
13
were more than four parties represented in a state legislature at one time, as well
as several instances in which multiple candidates had no particular party affiliation.
Taagepera’s method basically estimates a conservative measure of ENP based on the
number of parties that hold seats within the legislature (not the percentage of votes
garnered by each party). We believe this to be a reliable, valid, and conservative
measure of ENP at the state level overtime.
Before presenting our estimates of ENP, evaluating the party aggregation theory
requires some sort of measure of the political consolidation and fiscal centralization
of each state overtime. Unfortunately, the proportion of national to state/local
(nondefense) spending used by Chhibber and Kollman is unlikely to work at the
state level. Their measure combines state and local spending relative to national
spending. While this is fine for their design since they are concerned with the ENP
at the national level, obviously separating state and local spending is important for
this design. However, local spending data is not easily available or easily comparable
across jurisdictions for much of this time period. That said, there are a couple of ways
to address measuring state fiscal centralization, basically the primary explanatory
variable in the graphical and statistical analyses presented below.
First, it is important to determine exactly what type of spending would incentivize
voters and candidates to start coordinating across districts within states. In the
case of state governments, there are two broad types of spending that, in our opinion, might serve as acceptable proxies for overall fiscal centralization: social welfare
spending and education spending. Early in U.S. history, these two issues tended to
be handled at the local level by cities or private charity organizations. Then, be14
ginning during the mid-1800’s, states started to become seriously involved in social
welfare and education, drastically increasing their investment in these programs, extending their involvement in the day-to-day lives of their citizens, and centralizing
their control over spending and policymaking (Fishback & Thomasson 2006).
Ultimately though, spending in the admittedly broad category of ‘social welfare’
during much of the 19th and early 20th centuries was primarily dedicated to health
clinics, food, and other care primarily directed at the indigent. In contrast, public
education was broadly available and had an impact across all social classes. We would
argue that the education spending might show more penetration and therefore have
an even greater impact on party aggregation than other types of spending. In short,
voters are more likely to pay attention to state politics when the state is providing
funding and dictating policy to the schools that are educating their children. Since
the aggregation logic relates a reduction in the ENP to the benefits citizens receive
from a jurisdictions’ policymaking, we decided to focus on the relationship between
education and overall spending per capita on the ENP at the state level.
While not a relative measure of state versus local spending on education and welfare, which would be ideal, measuring total per capita state spending and education
spending should effectively operationalize increased state involvement in policymaking and the centralization of politics within the state. The implicit assumption in
using the per capita measures, apart from the assumption that these are the correct measures of spending to be using in the first place, is that the general trend
overtime is for a relative increase in state involvement, which we believe is defensible from the historical record (Fishback & Thomasson 2006). This measure will
15
allow the presentation of graphical analysis of the relationship between spending and
party aggregation in each state. The data to construct these measures comes from
ICPSR Study # 09728, Sources and Uses of Funds in State and Local Governments,
1790-1995 (Sylla, Legler & Wallis 1993).
In short, to analyze the relationship between political consolidation and party aggregation, we have constructed a data set of five states covering approximately 1834
- 1915. This data set allows us to estimate the effective number of parties for each
state in a given year. This measure of ENP is based on the distribution of the parties in the lower chamber and represents a conservative measure of overall effective
number of parties within the state. To this we have added fiscal data on overall state
spending and education spending per capita as a measure of political consolidation
at the state level. In the section below, we provide a graphical analysis of ENP and
state spending as well as a supplemental statistical analysis.
5
Results
In order to test our theory that state fiscal centralization decreases the effective
number of parties winning seats in state legislatures, we graphically analyze trends
in the effective number of parties (ENP) and fiscal centralization. We first discuss
the trends in the effective number of parties and then the levels of state total and
education spending per capita. After we compare these trends together, we present
a simple regression model that provides additional evidence that increases in state
education spending decreases the effective number of political parties.
16
[Figure 1 About Here]
The ENP rarely rises above two effective parties, but because our data is the number
of seats and not votes, we are more interested in whether the ENP decreases over
time. Without longitudinal variation in the ENP, the theory fails to clear an initial
hurdle. However, as shown in the figures below, the ENP fluctuates significantly
overtime, and generally decreases in all five states, although the pattern is clearer
in some states than others. In California, the ENP varies considerably initially but
levels off around 1870 and steadily decreases after 1880. The ENP in Michigan also
fluctuates significantly. However, the Michigan ENP trend line begins to decrease
after 1870 and considerably decreases around 1895. In Pennsylvania, the ENP appears to slightly increase until 1870 and begins declining around 1885. Virginia’s
ENP drops significantly around 1850 and mostly continues to decrease with a short
increase between 1880 and 1890. The trend line in New York shows little variation in
ENP. Although there are some fluctuations, the ENP in New York begins and ends
around 1.8 parties. Besides perhaps in New York, state ENP does vary significantly
overtime and generally decreases during the late 1800’s and the early 1900’s.
Having established that ENP fluctuates overtime in the states and generally declines,
we turn our attention to state fiscal centralization, operationalized as spending per
capita. Figures 2 and 3 below present state overall and educational spending per
capita from the early 1830s through the early 1910s. The total spending per capita
in this time period eventually increases for all five states; however, the timing of the
increase varies. In California, total state spending per capita begins a general upward
trend around the mid-1860s. State spending per capita remained relatively constant
17
in Michigan until a very large jump in spending after 1900. Total state spending per
capita in New York begins to increase in 1860, while total spending in Pennsylvania
does not consistently increase until after 1890. After a major drop in total state
spending per capita in Virginia after 1860 (perhaps related to the Civil War), the
total spending does not begin a steady upward trend until after 1890. Overall, the
general trend in overall state expenditures during this time period is upward, but
with considerable variation both within and across states.
[Figure 2 About Here]
However, the trend in education spending per capita is somewhat clearer. After
an initial jump in spending around 1870, education spending in California steadily
increases for the reminder of the sample. Despite some initial growth in education
spending in Michigan after 1880, the significant increases in education spending occur
after 1900, similar to total spending for the state. In both New York and Virginia,
education spending per capita begins to increase after 1870. Education spending
per capita also increases after 1870 in Pennsylvania; however, there is also a jump
in spending and faster growth after 1890. Overall though, education spending per
capita appears to increase in all five states included in our sample and increase
significantly, both in absolute terms and in spending per capita. Next, we visually
analyze the relationship between state spending per capita and state ENP.
[Figure 3 About Here]
In Figures 4 - 8 below, we present the graphs of state ENP and state total/education
spending per capita beside each other, making conclusions about the correspondence
18
of the trends feasible. The stall in growth around 1870 and the decline in California’s
ENP following the mid-1870s appears to correspond with increases the sharp rise in
state education spending in the preceding decade. There is an initial growth in ENP
immediately after the increase in education spending, indicating a potential lagged
impact of spending on ENP. The increase in total spending that occurs before 1870
also seems to correspond with the California ENP trend line and a lagged effect could
explain the initial increase in ENP. Therefore, decreases in Californian ENP seem to
coincide with increases in state spending per capita (both total and education).
[Figure 4 About Here]
The relationship between state spending per capita and ENP in Michigan is slightly
more difficult to visually assess than California. While ENP is inconsistent in its pattern for much of the sample, it appears to be decreasing in the decades before 1910.
Education spending per capita appears to be gradually increasing in the late 1800’s,
with a sharp increase before and after 1900, perhaps indicating some correspondence between Michigan’s decreasing ENP and increasing state education spending.
The large increase in Michigan’s total spending per capita also occurs around 1900,
so both spending figures offer some initial evidence that increases in spending and
decreases in ENP are related.
[Figure 5 About Here]
In New York, the ENP trend over time is relatively stable, although the actual ENP
varies considerably over a small range of values, making conclusions regarding spending and ENP more challenging than the other states. Additionally, a considerable
19
amount of missing data make it difficult to provide more accurate information on
New York state spending. There does not appear to be any relationship between
increases in state total spending per capita and New York ENP. However, the slight
decline in ENP beginning around 1880 follows the large increase in state education
spending per capita occurring after 1870.
[Figure 7 About Here]
The total state spending per capita in Pennsylvania varies considerably in the early
part of the 19th century, making it somewhat difficult to draw any conclusions from
those years. Despite variation in state total spending early in the 1800s, the decline
in Pennsylvania ENP from approximately 1875 onward seems to correspond to a
growth in total state spending per capita also beginning around 1885. Education
spending per capita, while still low, rises rapidly during this period as well, indicating
significant state investment in the provision of education. Significant increases in
education spending beginning around 1885 result in more substantive funding of
education and coincide with decreases in ENP also beginning near 1885 as well.
[Figure 6 About Here]
Virginia is a difficult state for which to draw conclusions, partly due to the possibility
that the Civil War impacted the state spending per capita. Both a decline in ENP
and an increase in total state spending per capita appear to occur after 1850, however,
there is no clear relationship again until after 1900 when increases in total spending
per capita again coincide with decreases in state ENP. The overall trend in state
ENP and educational spending also appear to be inversely related, with a general
20
downward trend for ENP and a general upward trend for educational spending. After
1870, Virginia state education spending per capita begins to increase. The state ENP
also begins to decrease after 1870. The inverse relationship between state education
spending per capita and the Virginian ENP appears to end after 1880. Overall,
both ENP and state education spending trend downward, particularly after 1850.
However, care should be taken from reading too much into these general trends,
given the legacy of political repression in the south.
[Figure 8 About Here]
Although there is considerable variation across out sample of states, increases in state
spending per capita do appear to correspond to decreases in state ENP overtime,
while the relationship between overall state spending and ENP is less clear. This is
theoretically expected, since increases in education spending would likely have more
impact on citizens’ daily lives, increasing their incentives to prefer more viable (i.e.
not third-parties) candidates. Even with a crude, conservative measure of ENP that
only accounts for parties that actually achieved legislative representation, there still
appears to be a recognizable inverse relationship between the effective number of
parties within a state and education spending per capita.
Finally, in an attempt to more fully test the inverse relationship between education
spending per capita and state ENP, we present a very basic statistical analysis of
the relationship between education spending and state-level ENP.2 Table 1 below
presents the results of a model that regresses state-level ENP by year on our measure
2
See Table 1, Chhibber and Kollman (1998) for a similar analysis.
21
of fiscal centralization (education spending per capita) as well as a one-year lag of
the dependent variable. The first column presents the results of the regression for all
states, while the second drops Virginia. As you can see, in both models, education
spending per capita is negatively related to the effective number of parties controlling
for a one-year lag in the ENP. In both cases, education spending achieves marginal
levels of significance (p < .10), is properly signed, and has a substantive impact on the
effective number of parties at the state level.3 Combined with the graphical analyses
above, this result further bolsters the argument that fiscal centralization, especially
educational spending, reduces the effective number of parties within a state.
[Table 1 About Here]
6
Conclusion
To conclude, we think this analysis adds further qualified support to the Chhibber
and Kollman thesis that political consolidation, as measured by fiscal centralization,
can drive party aggregation at the state level as well as the national level. State
ENP does decrease as state governments increased education spending per capita.
Previous explanations for the failure of third parties and socialist parties in the
United States have failed to incorporate the broader party systems literature and
test for a more general explanation of the U.S. case. Additionally, these results will
3
Excluded additional regressions substituting total spending per capita and social spending per
capita for education spending showed no significant relationship to ENP across either set of states.
This corresponds with the theory, since education spending would presumably have the most direct
fiscal impact on citizens.
22
hopefully advance research on linkage and party aggregation, helping to close the
gap between the institutional and social foundations of party systems and the final
empirical output.
In addition, we hope this analysis will spur further research on other questions related
to linkage. For example, what types of government spending are important for party
aggregation? Chhibber and Kollman only used a broad measure of non-defense
spending (1998). This analysis used total spending, which was not clearly related to
ENP, and educational spending, which showed a recognizable relationship with the
number of viable state-level parties. In our opinion, the relevant characteristic of the
spending should likely be penetration. In other words, how deeply and directly does
the spending impact the citizenry? How many people are directly impacted by the
spending and how much personal, identifiable benefit do they derive from it? This is
also another reason why the per capita measures used above are appealing, since the
measure captures, at least somewhat, how much benefit each citizen receives from
different types of spending.
For example, in the admittedly broad category of ‘social welfare’ spending, which
we also considered including in this analysis, most of the original spending was for
health clinics, food, and other care primarily directed at the indigent. In contrast,
public education was broadly available and had an impact across all social classes.
We would argue that the education spending shows more penetration and therefore
would theoretically have an even greater impact on party aggregation than other
types of state spending. In short, voters are more likely to pay attention to state
politics when the state is providing funding and dictating policy to the schools that
23
are educating their children.
Also, testing these theories out at the state-level allows for far more degrees of
freedom and the ability to include more control variables and overall get a clearer
picture of the process of linkage and party aggregation. Specifically, expanding this
pilot study to include all states, possibly excluding the South, would offer the opportunity to test the party aggregation theory in non-SMD electoral systems. As we
mentioned above, many states have used some form of MMD to elect their legislators
at various times throughout their history. This practice has declined over the last
several decades, largely due to accusations that the districts can be discriminatory,
but historical state elections data would offer an excellent opportunity to test the
effects of SMD’s vs. MMD’s on party aggregation.
In addition, we intend to expand our data collection effort to extend the analysis all
the way back to the early 1700’s. As you saw above, the effective number of parties
for New York was highly stable over the time period analyzed here. However, New
York had an over forty year old party system before it entered our data set and
an ENP highly stable around two, as would be expected. Therefore, it is possible
that we are missing some important party system dynamics during the early years
of the Republic in states like New York that we could present more fully with a
longer time series. Additionally, other analyses have stressed the importance of the
Great Depression on the American party system (Chhibber & Kollman 1998, Hirano
& Snyder 2007). As it stands now, our time series only extends to the mid-1910’s.
Expanding our dataset to the mid-20th century at least would also increase our
ability to examine the role of fiscal centralization on state party systems.
24
Finally, as mentioned above, we would like to include social heterogeneity in the
models as a control if nothing else. It is possible that heterogeneity mediates the
relationship between fiscal centralization and the party system, as well as having
independent effects on the party system. The cross-sectional and longitudinal variation in this variable within the American states make them an excellent laboratory
to examine this relationship. Historical Census data includes data on population
birthplace, race, and nativity that could be adapted to measure social heterogeneity.
In the end, institutions and social heterogeneity may be the foundation of party
systems, but party systems express themselves as they do because of intervening
variables that create linkage between broad influences and empirical outcomes. As
we hope this analysis has shown, political consolidation and fiscal centralization play
a major role in creating that linkage and motivating party aggregation. State party
systems tend to become stable two party systems because of the role of government
in the lives of its citizens. Increase that role, and voters will respond by turning to
parties that have the broad-based appeal necessary to govern, address those concerns,
and provide relevant representation.
25
Year
10
2
3
2.5
3.5
Virginia
19
1.5
2.5
New York
00
1
2
Year
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
1.5
Year
19
18
18
18
18
18
18
18
19
19
18
18
18
18
18
18
30
26
18
Year
Figure 1: Effective Number of Parties
Year
19
19
18
18
10
00
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
10
00
90
80
70
60
50
1
1
1
1.5
1.5
1.5
2
2
2
2.5
2.5
3
2.5
3
Michigan
18
18
18
18
18
19
19
18
18
18
18
18
18
19
19
18
18
18
18
18
California
Pennsylvania
Year
10
2
3
6
4
5
8
Virginia
19
1
4
New York
00
0
2
Year
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
0
Year
19
18
18
18
18
18
18
18
19
19
18
18
18
18
18
18
30
27
18
Year
Figure 2: Total Spending Per Capita
Year
19
19
18
18
10
00
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
10
00
90
80
70
60
50
0
0
2
2
2
4
4
4
6
6
6
8
8
8
10
10
Michigan
18
18
18
18
18
19
19
18
18
18
18
18
18
19
19
18
18
18
18
18
California
Pennsylvania
Year
10
1
1
Virginia
19
.5
.5
New York
00
0
0
Year
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
-.5
Year
19
18
18
18
18
18
18
18
19
19
18
18
18
18
18
18
30
28
18
Year
Figure 3: Education Spending Per Capita
Year
19
19
18
18
10
00
90
80
70
60
50
40
30
10
00
90
80
70
60
50
40
10
00
90
80
70
60
50
0
0
0
1
.5
1
2
2
1
3
3
1.5
4
4
Michigan
18
18
18
18
18
19
19
18
18
18
18
18
18
19
19
18
18
18
18
18
California
Pennsylvania
California: Education Spending
2
2
2.5
3
3
4
California: ENP
1
0
1.5
1
29
1850
1860
1870
1880 1890
Year
Effective Number of Parties
1900
1910
1850
1860
Fitted values
Figure 4: Education Spending & ENP
1870
1880 1890
Year
Education Spending Per Capita
1900
1910
Fitted values
Michigan: Education Spending
2
2
3
4
2.5
Michigan: ENP
0
1
1
1.5
30
1840 1850 1860 1870 1880 1890 1900 1910
Year
Effective Number of Parties
1840 1850 1860 1870 1880 1890 1900 1910
Year
Fitted values
Figure 5: Education Spending & ENP
Education Spending Per Capita
Fitted values
Pennsylvania: Education Spending
2
1
2.5
3
1.5
Pennsylvania: ENP
0
1
1.5
.5
31
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Effective Number of Parties
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Fitted values
Figure 6: Education Spending & ENP
Education Spending Per Capita
Fitted values
New York: Education Spending
2.5
.5
3
1
3.5
New York: ENP
-.5
1.5
2
0
32
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Effective Number of Parties
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Fitted values
Figure 7: Education Spending & ENP
Education Spending Per Capita
Fitted values
Virginia: Education Spending
0
1
1.5
33
.5
2
1
2.5
Virginia: ENP
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Effective Number of Parties
1830 1840 1850 1860 1870 1880 1890 1900 1910
Year
Fitted values
Figure 8: Education Spending & ENP
Education Spending Per Capita
Fitted values
Table 1: Effect of Education Spending on ENP
Variable
All States Non-Southern States
(Std. Err.)
Education Spending Per Capita
ENP (1yr Lag)
Intercept
†
-0.036
-0.0374†
(0.020)
(0.021)
∗∗
0.549
0.502∗∗
(0.046)
(0.052)
0.809∗∗
0.905∗∗
(0.085)
(0.098)
340
0.315
77.62
279
0.267
50.29
N
R2
F
Significance levels :
† : 10%
(Std. Err.)
∗ : 5%
∗∗ : 1%
34
References
Aldrich, John H. 1995. Why Parties? The Origin and Transformation of Political
Parties in America. Chicago: The University of Chicago Press.
Burnham, W. Dean. 1993. “PARTISAN DIVISION OF AMERICAN STATE GOVERNMENTS, 1834-1985.” [Computer file]. Conducted by Massachusetts Institute of Technology. ICPSR ed. Ann Arbor, MI: Inter-university Consortium for
Political and Social Research [producer and distributor], 1987.
URL: http: // www. icpsr. umich. edu/ icpsrweb/ ICPSR/ studies/ 16
Chhibber, P.K. & K. Kollman. 2004. The formation of national party systems: federalism and party competition in Canada, Great Britain, India, and the United
States. Princeton University Press.
Chhibber, Pradeep K. & John R. Petrocik. 1989. “The Puzzle of Indian Politics:
Social Cleavages and the Indian Party System.” British Journal of Political
Science 19(2):pp. 191–210.
URL: http://www.jstor.org/stable/193713
Chhibber, Pradeep & Ken Kollman. 1998. “Party Aggregation and the Number of
Parties in India and the United States.” The American Political Science Review
92(2):pp. 329–342.
URL: http://www.jstor.org/stable/2585667
Costa, Dora L. & Matthew E. Kahn. 2003. “Civic Engagement and Community Heterogeneity: An Economist’s Perspective.” Perspectives on Politics
1(01):103–111.
URL:
http: // journals. cambridge. org/ action/ displayAbstract?
fromPage= online&aid= 145059&fulltextType= RA&fileId=
S1537592703000082
Cox, Gary. 1999. “ELECTORAL RULES AND ELECTORAL COORDINATION.”
Annual Review of Political Science 2(1):145–161.
URL: http://www.annualreviews.org/doi/abs/10.1146/annurev.polisci.2.1.145
Cox, Gary W. 1997. Making votes count : strategic coordination in the world’s
electoral system. Political economy of institutions and decisions New York:
Cambridge University Press.
Crespin, Michael. 2004. Direct Primaries, Sore Losers, and the Openness of the TwoParty System, 1904-1910. In Paper presented at the annual meeting of the The
35
Midwest Political Science Association. Palmer House Hilton, Chicago, Illinois:
Midwest Political Science Association.
URL: http: // www. allacademic. com/ meta/ p83371_ index. html
Duverger, M. 1954. Political parties: Their organization and activity in the modern
state. Taylor & Francis.
Epstein, L.D. 1989. Political parties in the American mold. University of Wisconsin
Press.
Fishback, Price V. & Melissa A. Thomasson. 2006. Social Welfare: 1929 to the
Present. In Historical Statistics of the United States, Millennial Edition. Volume
Two: Work and Welfare., ed. S. Carter, S. Gartner, M. Haines, A. Olmstead,
R. Sutch & G. Wright. Cambridge University Press.
Hamm, Keith & Gary Moncrief. 2007. Legislative Politics in the States. In Politics
in the American States, Ninth Edition, ed. Virginia Gray & R. Hanson. CQ
Press.
Hirano, Shigeo & James M. Snyder. 2007. “The Decline of Third-Party Voting in
the United States.” The Journal of Politics 69(01):1–16.
URL: http://dx.doi.org/10.1017/S0022381600007751
Kalyvas, Stathis N. 1998. “From Pulpit to Party: Party Formation and the Christian
Democratic Phenomenon.” Comparative Politics 30(3):pp. 293–312.
URL: http://www.jstor.org/stable/421988
Laakso, M. & R. Taagepera. 1979. “Effective Number of Parties - Measure with
Application to West Europe.” Comparative Political Studies 12(1):3–27.
URL: ¡Go to ISI¿://A1979GT08700001
Lijphart, Arend. 1979. “Religious vs. Linguistic vs. Class Voting: The ”Crucial
Experiment” of Comparing Belgium, Canada, South Africa, and Switzerland.”
The American Political Science Review 73(2):pp. 442–458.
URL: http://www.jstor.org/stable/1954890
Lijphart, Arend. 1994. Electoral Systems and Party Systems: Study of Twenty-Seven
Democracies, 1945-90. Oxford University Press.
Lipset, Seymour M. & Stein Rokkan. 1967. Cleavage Structures, Party Systems, and
Voter Alignments: An Introduction. In Party Systems and Voter Alignments:
An Introduction, ed. Seymour M. Lipset & Stein Rokkan. New York: Free Press.
36
Lipset, Seymour Martin. 1977. Why No Socialism in the United States? In Sources
of Contemporary Radicalism I, ed. S. Bailer & S. Sluzar. Boulder, CO: Westview
Press.
Lipset, S.M. & G. Marks. 2001. It didn’t happen here: why socialism failed in the
United States. WW Norton & Company.
Oestreicher, Richard. 1988. “Urban Working-Class Political Behavior and Theories
of American Electoral Politics, 1870-1940.” The Journal of American History
74(4):pp. 1257–1286.
URL: http://www.jstor.org/stable/1894410
Riker, William H. 1982. “The Two-Party System and Duverger’s Law: An Essay
on the History of Political Science.” The American Political Science Review
76(4):pp. 753–766.
URL: http://www.jstor.org/stable/1962968
Rosenstone, S.J., R.L. Behr & E.H. Lazarus. 1996. Third parties in America: Citizen
response to major party failure. Princeton Univ Press.
Samuels, D.J. 1998. Careerism and its consequences: federalism, elections, and
policy-making in Brazil PhD thesis University of California, San Diego.
Sylla, Richard E., John B. Legler & John Wallis. 1993. “SOURCES AND USES OF
FUNDS IN STATE AND LOCAL GOVERNMENTS, 1790-1915.” [Computer
file]. ICPSR version. DeKalb, IL: Paul Kleppner, Northern Illinois University
[producer]. Ann Arbor, MI: Inter-university Consortium for Political and Social
Research [distributor], 1993.
URL:
http: // www. icpsr. umich. edu/ icpsrweb/ ICPSR/ studies/
09728/ detail.
Taagepera, Rein. 1997. “Effective number of parties for incomplete data.” Electoral
Studies 16(2):145 – 151.
URL:
http://www.sciencedirect.com/science/article/B6V9P-3SWXYXW1/2/0db755f4136619873f404c0a34dd59c5
Ware, Alan. 1996. Political parties and party systems. Oxford University Press.
Ware, Alan. 2002. “The American Direct Primary:: Party Institutionalization and
Transformation in the North.”.
37