The Fates of Challengers in U.S.

House Elections: The

Role of Extended Party Networks in Supporting
Candidates and Shaping Electoral Outcomes
Bruce A. Desmarais
Raymond J. La Raja
Michael S. Kowal

University of Massachusetts Amherst

University of Massachusetts Amherst
University of Massachusetts Amherst

Extended party network (EPN) theory characterizes political parties in the United States as dynamic networks of interest
groups that collaboratively support favored candidates for office. Electoral predictions derived from EPN theory have yet
to be tested on a large sample of races. We operationalize EPNs in the context of organized interest contributions to U.S.
House campaigns. We deduce that support by a partisan community of interests signals the ideological credibility and
appeal of a candidate. EPN integration overcomes voter ambiguity surrounding challengers ideological preferences, and
resources provided by these coordinating interest groups promote a consistent message about the candidate. Using data
from the 19942010 cycles, we apply network analysis to detect EPN support of challengers and find that EPN integration
substantially improves the electoral prospects of challengers. The effect of EPN integration is distinct from that of campaign
resources. The findings provide support for EPN theory, as applied to congressional elections.

The Extended Party Network and

Congressional Elections

rich scholarly tradition exists in studying the organizational forms of political parties and how
such characteristics potentially shape key political outcomes (see, e.g., Cotter 1989; Key 1949; Ostrogorski
1902; Schattschneider 1942).
That tradition is being renewed. Recent research
drawing on network theory conceptualizes political parties as dynamic, dispersed systems of interconnected interest groups, centered on traditional formal party organizations, which Koger, Masket, and Noel (2009) term

extended party networks (EPNs). According to this theoretical framework, the central functions of the political
party are to select and support candidates who are deemed
likely to advance the party coalitions policy agenda once
in office (Bawn et al. 2012; Cohen et al. 2008; Herrnson
2009; Koger, Masket, and Noel 2009; Masket 2009; Skinner, Masket, and Dulio 2012). This stands in contrast to
the dominant, politician-centered explanation for party
formation (Aldrich 1995), which attributes the origins
of parties to the need for legislators to compromise and
cooperate in passing legislation.
An important implication of the EPN theory is
that candidates, especially challengers, who are targeted
by partisan coalitions of interests are inherently more

Bruce A. Desmarais is Assistant Professor, Department of Political Science, University of Massachusetts Amherst, Thompson Hall, 200
Hicks Way, Amherst, MA 01003 (desmarais@polsci.umass.edu). Raymond J. La Raja is Associate Professor, Department of Political Science,
University of Massachusetts Amherst, Thompson Hall, 200 Hicks Way, Amherst, MA 01003 (laraja@polsci.umass.edu). Michael S. Kowal
is a Ph.D. Student, Department of Political Science, University of Massachusetts Amherst, Thompson Hall, 200 Hicks Way, Amherst, MA
01003 (mkowal@polsci.umass.edu).
We would like to thank the editor, the anonymous reviewers, Brian Schaffner, Maryann Barakso, Tatishe Nteta, Jesse Rhodes, David
Nickerson, Paul Herrnson, David Lazer, Scott McClurg, Rahsaan Maxwell and Dino Christenson for helpful feedback on this project.
Previous drafts of this article were presented at the UMass American Politics Research Working Group, the 2012 Political Networks
Conference at the University of Colorado Boulder, the 2012 Northeastern Political Science Association Annual Meeting in Boston, the
Connections Conference on Network Science at the MIT Media Lab and the Boston-Cambridge Colloquium on Complexity and Social
Networks at Northeastern University. This research was supported in part by a University of Massachusetts Amherst Department of Political
Science Graduate Research Initiative Assistantship. All mistakes are our own.
Data for replication are available on the AJPS Dataverse Archive (http://dvn.iq.harvard.edu/dvn/dv/ajps).
American Journal of Political Science, Vol. 00, No. 00, March 2014, Pp. 118

C

2014, Midwest Political Science Association

DOI: 10.1111/ajps.12106

2
appealing to the party, on a policy basis, than those who
do not receive the support of the network. By extension,
challengers who are backed by a significant contingent of
the EPN will gain support from attentive voters who take
cues from visible groups and activists in the party network
(i.e., partisan elites; Dominguez 2011; Lupia 1994; Zaller
1992) and should therefore have better prospects on Election Day than those who do not, regardless of campaign
expenditures by the candidates. The strength of this cue is
bolstered by the consistency of preferences in the network
of interests and the coherence of the ideological message
amplified by the groups resources.
Research addressing party networks has focused
mostly on identifying patterns of connectivity among
party organizations and interest groups (Grossman and
Dominguez 2009; Heaney et al. 2012; Herrnson 2009;
Koger, Masket, and Noel 2009; Kolodny and Dwyre 1998;
Skinner, Masket, and Dulio 2012). These studies reveal
partisan affinities among interest groups with respect to
information flows (Koger, Masket, and Noel 2009, 2010;
Skinner, Masket, and Dulio 2012) and electoral or legislative activity (Grossman and Dominguez 2009). To the
degree that research has examined the impact of these
networks on political outcomes, it has been confined to
looking at the link between endorsements of prominent
politicians and the likelihood that a candidate wins a
party nomination or general election (Cohen et al. 2008).
These studies rely on a relatively small sample of candidates in presidential nominations (Cohen et al. 2008)
or local legislative elections (Masket 2009). Such studies
provide preliminary evidence of an electoral influence of
EPNs. However, the electoral implications of EPN theory
have yet to be tested on a large and representative sample
of electoral contests.
What is missing in the literature is a clear blueprint
for linking the activity of networked interest groups with
overlapping political agendas (i.e., political parties as conceived by Cohen et al. 2008 and Bawn et al. 2012) to the
electoral fates of the candidates they target. Our first goal
in this article is to translate theories about extended party
networks into empirical expectations regarding how partisan interest groups rally around selected candidates. Our
second goal is to explore whether integration into this
network of partisan interests affects election outcomes.
To directly test interest group behavior and the electoral
consequences derived from EPN theory, we turn to campaign finance, which is a central mechanism by which
disparate interest groups attempt to influence electoral
outcomes. We focus on outcomes in U.S. House elections from 1994 to 2010. Political contributions by organizations provide a robust measure of the intensity of
support for candidates, while allowing us to distinguish

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

behaviors among a broad population of groups seeking

to influence federal elections. We draw on the framework
of network analysis to operationalize the EPN using data
on group contributions to, and independent spending
for, candidates for the U.S. House. We do this through a
community detection (Newman 2006) algorithm that is
informed by the structure of advocacy that underlies EPN
theory.
We find a consistent pattern of campaign finance
community structure that strongly coheres with EPN
theory. That is, challengers are integrated into either
highly partisan, well-resourced communities of interest group political action committees (PACs) exhibiting
dense, overlapping ties (i.e., the EPN) or a large bipartisan
catch-all community that is defined by meager support.
We find that integration of a challenger into the dense
partisan communities of PAC contributors substantially
increases the probability of electoral success, an effect that
goes well beyond campaign spending. Our findings provide support for a theory of political parties as interest
group coalitions of policy demanders. It also illustrates,
in practice, why political parties may fail to converge ideologically on the median voter due to the types of candidates they help elect.

The Extended Party Network

Despite the well-documented resurgence of congressional
political parties (Aldrich 1995; Heberlig and Larson 2012;
Herrnson 1988), contemporary research addressing congressional election outcomes offers little to say about the
impact of party organizations, or any coordinated political efforts beyond individual campaigns (see, e.g., Jones
2010; Lazarus 2008; Woon and Pope 2008). Of course, partisanship (e.g., of candidates, districts, the presidency, and
Congress) plays an essential role in every account explaining outcomes in congressional elections (Canes-Wrone,
Brady, and Cogan 2002; Jacobson 1989; Woon and Pope
2008). This dichotomythe prevalence of partisanship
and the absence of party organizationfits well with the
politician-centered understanding of parties. That is, parties arise in response to the needs of policy makers, mainly
legislators serving in the same body, to coordinate policy
agendas in order to win individual gains through mutual
support (Aldrich 1995). Thus, according to the dominant theoretical understanding, parties are organizations
born out of service to legislators, with the central organizational focus being the top-down enforcement by the
party leadership of adherence to a unified policy agenda
(Cox and McCubbins 1993).

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

The theory of the party as a coalition of policy demanders, which we equate with EPN theory (Bawn et al. 2012;
Cohen et al. 2008; Grossman and Dominguez 2009; Koger,
Masket, and Noel 2009), holds that parties arise from the
benefits that organized interests realize from aggregating
agendas and coordinating resources in pursuit of electoral
and policy goals.1 Specialized interests that are relatively
rich in resources (i.e., money, large memberships, or expertise) typically have very narrow policy objectives (e.g.,
support entitlement programs for retirees, zone more offshore space for oil drilling, oppose gay marriage). Substantial resources would be wasted if organized interests
pursued separate candidates corresponding to each individual groups agenda. This logic was expressed by Lee
Saunders, president of the American Federation of State,
County and Municipal Employees, in a recent statement
on the need for labor unions to align with other supportive interests (e.g., the NAACP; Trottman 2013): Labor
cant do it alone. Our density dictates that weve got to
have partners. To avoid this waste, policy demanders in
pursuit of separate agendas that are compatible, or at least
not contradictory, agree to pool resources in support of
candidates who will represent overlapping agendas. This
combination of agendas becomes the party platform, and
the separate interest communities become the coordinated base of support and activism for the party.
An archetypal example of the loose connections that
form among issue-oriented groups in partisan interest
networks is provided by the U.S. Senate election contribution activities of three PACs in the 2012 cycle: the
Gun Owners of America (Gun PAC), the Association of
Oil Pipe Lines (Oil PAC), and the Republican National
Coalition for Life (Abortion PAC). These three groups
are firmly aligned with central issues in the Republican
Partys policy platform, but they span the economic and
cultural issue dimensions.2 The Gun PAC contributed to
11 Senate candidates in 2012. The candidate to whom
they contributed the most was Ted Cruz (R-TX). Ted
Cruz was the second largest recipient of contributions
from the Oil PAC, which contributed the most to Lisa
Murkowski (R-AK). The Abortion PAC supported eight
Senate candidates, five of whom were also supported by
the Gun PAC. In this example, the Gun PAC is directly
connected to both the Oil PAC and the Abortion PAC,
which are indirectly connected through the Gun PAC.
The Abortion PAC indirectly helped the Oil PAC by freeing up resources for the Gun PAC to invest in Ted Cruz,
1

Cohen et al. (2008) and Bawn et al. (2012) do not discuss networks,
but the theory they elaborate clearly implicates the network concept
and inspired us to approach the study of parties in this manner.

The data for this example were gleaned from opensecrets.org.

which in turn helped the Oil PAC spread resources to Lisa

Murkowski and other potential supporters of pipeline
construction. This example highlights the loose network
of cross-supports that emerges from decentralized and
overlapping backing of candidates for Congress.
A common collective action motivation underlies
party formation in both the candidate-centered and EPN
theories of parties. That is, parties arise because there
are several political actors in pursuit of narrow policy
objectives, who will rarely achieve success if they go it
alone. The party forms as a collective action solution in
which each actor achieves more than would be possible
in solitary pursuit, in exchange for supporting the goals
of other party members. Despite pursuing somewhat disconnected policies, political actors choose to create a long
coalition that strives to stay together across time to increase the likelihood that each will achieve their particular
goals. The two theories diverge when it comes to (1) the
actors who require a collective action solution and (2) who
benefits from party activity. In the candidate-centered
theory, policy makers form a coalition (i.e., party) in
order to achieve policy objectives that appeal directly to
their constituents and therefore secure reelection (Aldrich
1995). In the EPN theory, active special interests benefit
from coordinating and standing behind candidates in the
form of a party, and thus they secure benefits for their
group members by electing politicians loyal to the integrated party platform (Bawn et al. 2012).

Extended Party Networks and

Challenger Success
Previous literature on challenger success has focused
largely on whether a candidate has held elective office.
Politically experienced challengers more frequently selfselect into races in which they will serve as an appealing
alternative to the incumbent (Carson et al. 2012; Lazarus
2008). The literature clearly demonstrates the role of candidate self-perception in choosing to run (e.g., Maisel
and Stone 1997). We build upon this work with theory
regarding the impact of organized interest perception and
mobilization in the context of the EPN.
The main challenge in achieving successful coordination among groups in the EPN is the identification
of candidates for office who (1) have strong electoral
prospects and (2) will deliver on the shared party agenda
(Bawn et al. 2012). Bawn et al. (2012) highlight that the
parties cannot simply commit resources to candidates
in the hopes of convincing them to adhere to a party
platform, since there is a principal-agent problem in

4
which the partys constituent groups cannot effectively
monitor legislators. This is why the party network needs
to focus its energies on bolstering the prospects of
candidates who would enter office with a priori credible
and agreeable policy stances (Cohen et al. 2008). From
this characterization of EPN activity, we deduce that EPN
support sends a strong signal regarding a candidates
interest in and capacity to deliver on the party agenda.
Because different communities of interests (e.g., gun
control advocates and environmentalist groups) converge
on shared, compromise candidates, the credibility signals
afforded by group support are concentrated on single
candidates. This helps challengers overcome the hurdle
(Bernhardt and Ingberman 1985) of convincing a
broader, nationally focused partisan constituency that
they will take appealing policy stances once in office.
The first boost to the candidacy will come from the
reputations of these groups. The fact that a coalition converges on a candidate sends a strong signal to attentive
voters. These voters may have ties with such groups and
comprise a significant portion of the electorate, particularly in primaries and low-turnout general elections.
Research shows that issue groups often contact members
directly to endorse favored candidates, and many broadcast their preferences to the wider public. Such issue-based
advocacy or endorsements serve as important heuristics
to attentive voters (Lau and Redlawsk 2001) and may
even help low-information voters make decisions (Lupia
1994), particularly when interest groups provide contextual information in campaigns to help voters understand
the ideological or partisan implications of group support
(Arceneaux and Kolodny 2009).
The potency of the signal sent by EPN support for a
candidate lies in the power of multiple consistent signals
to discriminate the underlying preferences of a candidate.
Since the ideology of interest groups can be inferred based
on the candidates they support (Bonica 2013), when many
like-minded groups in a policy domain back a candidate
(e.g., Sierra Club, League of Conservation Voters, and
National Resources Defense Council support the same
candidate), this provides multiplicative certainty in a candidates policy preferences. Suppose there is some probability p that a group interested in supporting a candidate
with position a component of the party agenda
supports a candidate who actually does not prefer position , and a probability q > p that a group supports a
candidate who actually does prefer position . Let there
be N groups that prefer , and k of them support the candidate. As k increases, the probability that the supported
candidate prefers position also increases.3 Specifically,
3

Note that a similar result could be reached by considering the

posterior variance of a continuous ideal point.

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

FIGURE 1 Posterior Probability That a

Candidate Prefers Given the
Number of Groups Preferring
That Contribute to the Candidate

Note: To parameterize these curves, the prior probability that

a candidate prefers is set at 0.5. The probability that a group
supports a candidate given that the candidate prefers (i.e.,
q) is set at an equal-spaced sequence of 20 values between
0.55 and 0.99, with darker lines representing higher q. Given
q, p is set to 1 q.

applying Bayes rule, P(|k) = P(k|)()/(P(k|)()

+ P(k|not )(1 ())), where P(k|) and P(k|not ) are
the binomial distributions parameterized with q and p,
respectively. The terms () and (not ) represent prior
beliefs regarding the prevalence of candidates who prefer position . Thus, the number of like-minded groups
supporting a candidate plays a key role in discerning a
candidates underlying preferences. Figure 1 illustrates
this relationship for a hypothetical situation in which N =
10. For each additional like-minded group that contributes
to a candidate, the probability of that candidates preferences aligning with the policies preferred by those groups
(i.e., ) increases by between 5 and 50%, depending upon
the number of groups already supporting the candidate
and the inherent accuracy of an individual group in vetting the policy positions of a candidate. In this way, support by a coherent community of interests sends a clear
policy signal about a candidate. This signal goes beyond
the resource and reputational effects of any one group.
Complementary to this dynamic is the formation of
a broader and looser connection of not necessarily likeminded communities, which converges around a single
candidate who satisfies the multiple interests of the partisan coalition without offending any particular group. By
integrating a coalition of interest communities, the EPN
draws together the support of otherwise disconnected

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

constituencies, raising the scale of candidate backing from

the interest community to the political party.
The second class of benefits that arise from
EPN integration derives from intentional coordination among interests. Interest groups provide electoral
resourcesbeyond campaign contributions and group
endorsementsin coordination with other groups in the
coalition that boost candidate prospects (Herrnson 2011;
Herrnson, Deering, and Wilcox 2012). These include collective efforts to mobilize voters (Magleby 2010), air campaign ads attacking the opposing candidate (Franz 2010),
and offer timely expertise on campaign matters (De Witt
1980; Herrnson 2011).
We are not arguing that EPN support is all that drives
electoral outcomesmany voters will directly evaluate
candidates and make their own decisions. However, as
Bawn et al. (2012, 57) note, Some voters who care nothing about the interests of the various groups are nonetheless attracted to their parties because of the values, such
as social order or equality, that they perceive in their
program. EPN support signals that a challenger credibly
represents the partys platform or significant components
thereof and is likely to deliver on those policy preferences
if elected. Candidates backed by the EPN have strong credentials on one or more elements of the party platform
and will give offense to no one [in the party] (Cohen
et al. 2008, 83). With this imprimatur, the chosen candidate is the beneficiary of the reputation, resources, and
expertise of members of the long coalition. Given low
attentiveness of many voters and the first-past-the-post
rules of American elections, the signaling and (noncontribution) resources of such coalitions will affect electoral
outcomes, beyond candidate spending.

Empirical Analysis
We focus on the electoral predictions regarding EPN
integration of challengers. Since incumbency offers access to party and other resources, we do not attempt
to differentiate between EPN support and the incumbency advantage (Abramowitz, Alexander, and Gunning 2006; Fenno 1978; Levitt and Wolfram 1997; Mayhew 1974). Rather, we study challengers and consider
whether the conception and analysis of extended party
networks can build upon our understanding of challenger
success.
There are two fundamental tasks before us in our
empirical analysis: (1) measurement of whether a challenger is integrated into a partisan network of organized
supporters and (2) estimation of the effect of EPN sup-

port on electoral outcomes. In the analysis that follows,

we show that the network analytic methodology referred
to as community detection (Newman 2006) is very well
suited to identifying groups of highly interconnected contributors and candidates. We then show that strong integration into a party network community substantially
increases the likelihood of a challengers success.

Contribution Network Data

Data on political action committee (PAC) and formal
party committee contributions to all candidates for the
U.S. House of Representatives and data on all House election outcomes from the 19942010 cycles form the bases
of our empirical analysis.4 From here on, we include the
formal party committees in our broad references to PACs.
We also gather data on additional district and candidate
attributes up to the 2008 elections.5
In order to identify candidates backed by EPNs, we
seek to operationalize our conceptual understanding of
the EPN. As noted above, we understand a party network
to be a collection of organized interests that concentrate
their support efforts around a core group of candidates.
Before detailing our approach to identifying these coalitions of interests, we must define the network that we
will analyze. We construct a bipartite network. A bipartite network is one in which there are two types
of nodes (i.e., actors), and ties can only connect two
nodes of different types (i.e., there are no intra-type
ties; Newman 2010). The two types of nodes in the networks we construct are PACs and candidates for the U.S.
House. The ties are weighted by the amount contributed
up to July 1 of the election year. We measure the network early in the election cycle to minimize the risk that
our inferences are driven by PACs giving to sure-thing
challengers.6
The network degree distributions in the bipartite
networks we construct (Newman 2010) are given in

Contribution data come from opensecrets.org. We focus on this

time period because we had significant data quality concerns, based
on our own checks, for the data predating 1994. Specifically, the
candidate type (e.g., incumbent, challenger) is riddled with error in
the 1992 data. We get around this issue in the Woon and Pope (2008)
replication presented below, as we use their challenger/incumbent
classification variable.

We use data from Jacobson (2009), which covers the 19942008

election cycles.

Our inferences are robust to measuring the network at any point

in the latter half of the election year. In the first two quarters of
the election year, there are not enough networked challengers to
provide a feasible sample size.

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

FIGURE 2 PAC and Candidate Cumulative Degree Distributions, 19942010

Note: Political action committee (PAC) and candidate cumulative degree distributions are shown in terms of total
amounts (in dollars) represented by ties and the total number of ties. The darker the line, the more recent the election
year. The solid lines depict distributions for incumbents, and the dashed lines depict distributions for challengers. Note
that the x-axis is on the log scale in the amount plots, such that an increment of 1 represents an order of magnitude
(i.e., a tenfold) increase.

Figure 2.7 The plots therein give the empirical cumulative distributions of the total amounts given by each PAC
over all candidates, the total amounts received by each
candidate over all PACs, the total number of candidates
to whom each PAC contributes, and the total number of
PACs that contribute to each candidate. Two main stories
emerge from these distributions. First, there is stark inequality in the contributions received by incumbents and
challengers. Over the entire period, the median amount
received by incumbents is approximately 100 times the
median amount received by challengers. This is a common finding, and it has led researchers to conclude that
7

Degree distributions are the distributions over vertices (i.e., PACs

and candidates) of the total connections to which those vertices
are incident (e.g., the distribution of the number of friends people
have, over people in a friendship network).

successful challengers owe their success to forces outside

of systematic organized party support (Herrnson 1995;
Jacobson 2009). Second, there is substantial inequality
across PACs and candidates. The best-funded incumbents
and challengers receive 10 times their median counterparts, respectively. The most active PACs contribute 100
to 1,000 times as much as their median counterparts. Regarding the second column of Figure 2, it is rare for a
challenger to attract a large number of contributors, as
the majority of challengers are supported by a handful
of PACs. Moreover, most PACs support just a few candidates, indicating that overlap among PACs should be an
informative signal.
We can explore the basic bivariate relationship between interest network integration and challenger success through the graphics in Figure 3. The plots depict

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

FIGURE 3 Contribution Network Visualizations

Note: Networks depict PAC nodes (invisible) connected to candidate nodes, which are
colored based on party and incumbency status. Coloration differentiates four groups:
Democratic incumbents (royal blue), Democratic nonincumbents (light blue), Republican incumbents (red), Republican nonincumbents (orange). Nodes are placed
based on the algorithm of Fruchterman and Reingold (1991), with edges weighted
by the amount contributed.

the structure of the network for three cycles (i.e., 2006,

2008, and 2010). Ties are anchored in the positions assigned to PACs in the node placement algorithm, but in
the interest of clarity, PACs are not drawn as nodes in
the network. Coloration indicates partisanship and incumbency status. The first column of graphics depicts all
candidates, and the second column displays only those

who win election.8 Figure 3 provides initial evidence that

interest network integration is associated with challenger
success. In all three cycles, nearly all of the nonincumbents
8

In these plots, candidates appearing close to each other receive

similar contribution amounts from a similar set of contributors.
We use the R package iGraph to produce these figures.

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

who are victorious are closer to the core of the network

than are nonincumbents who lose. Though compelling
visual evidence, this is not sufficient to identify an effect
of EPN support. For that, we need to control for other
explanations of challenger success and extract an explicit
measure of EPN support from the contribution network.

FIGURE 4 Illustration of Communities in an

Artificial Bipartite Network

Community Detection as Party Network

Extension Detection
In the previous main section and Figure 1, we presented a theoretical claim that when multiple like-minded
groups converge upon a candidate, a clear signal is established regarding the ideological position represented
by the candidate. To test this hypothesis, we must (1)
empirically identify collections of like-minded groups and
(2) identify the candidates on whom they converge. We
draw upon the network analytic method termed community detection to operationalize extended party networks of organized interests and the candidates they
support.
Community detection is the process of identifying
groups of nodes, within which ties occur with greater frequency and intensity than they do across communities
(Newman 2006). In our case, we have two node types:
PACs and candidates. The only ties in our data are those
from PACs to candidates. Thus, community detection
within the context of our data constitutes the identification of groups containing both PACs and candidates,
within which the PACs give more to the candidates in
that group than they do to other candidates, and the
candidates in that group receive more from the PACs in
that group than they do from other PACs. An illustrative
picture of a bipartite network with three communities is
given in Figure 4. There are still cross-community ties,
but ties within communities occur with much greater
intensity.
We characterize each PAC as drawing donations for
candidates in a given community from its communityspecific Poisson distribution.9 Thus, if there are k communities and m PACs, we infer k m Poisson distributions. We permit the number of communities (k) to vary
because we expect the network of groups supporting candidates to be characterized by several communities that
are densely connected within (i.e., focused on a coherent
9

We adopt a soft clustering approach, which means nodes are assigned to each community with some probability. Soft clustering methods are straightforward to express within a probabilistic
framework as mixture distributions (Imai and Tingley 2012). As
such, we can use the Bayesian information criterion (BIC) to determine the number of communities in the network (Fraley and
Raftery 1998).

Note: Brown circles represent one node type and navy squares
another (e.g., PACs and candidates). All edges connect one circle
and one square. The nodes in each community overlay a circle.

policy agenda) and loosely connected across (i.e., pursuing alternative aspects of a partisan agenda). The amount
contributed from PAC i to candidate j, denoted yi j , is
modeled as a mixture of Poisson distributions. A mixture
distribution is appropriate because, as central to the party
compromise underpinning EPN theory, we expect each
candidate will present a platform that appeals to overlapping interest communities. A total of k Poisson rate
parameters () are estimated for each PAC, and k community membership probabilities () are estimated for
each candidate. The rate parameters can be interpreted
as the intensity of an interest communitys support for a
candidate, and the community membership probabilities
represent the cross-community ties offered by candidates.
The probability of yi j is thus expressed as
P (yi j ) =

k


j t p(yi j, i t ),

t=1

where p() is the Poisson probability mass function.

In each election cycle, we estimate the model with k
from 5 to 50 and select the k that results in the lowest
BIC.10 For the purpose of the discrete analyses we perform below, we assign each candidate to the community
10
It might seem appropriate to begin at k = 2, simply capturing
the two political parties. However, the algorithm fits quite poorly
and exhibits convergence problems with k < 5. This is because,

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

FIGURE 5 Community Visualizations

Note: Each diamond corresponds to a community. Height is proportional to median contributions to a candidate in the
community, and width is proportional to the number of candidates in the community. Axes depict the partisanship and
incumbency characteristics of the candidates in the community.

of which he or she is most likely to be a member. This

facilitates the identification of candidates who are not integrated into the EPN. For each election, this results in
a partition of the set of the candidates into communities
that contain a mixture of candidates and PACs. These
memberships and attributes of these communities are
at minimum, there are five distinct connection patterns we see in
the data: (1) Some candidates (i.e., duck soup challengers) receive
meager or disparate PAC support, (2) some (i.e., strong Republican
challengers) receive primary support from conservative ideological groups, (3) some (i.e., strong Democratic challengers) receive
primary support from liberal ideological groups, (4) some (i.e., Republican incumbents) receive primary support from right-leaning
access-oriented donors, and (5) some (i.e., Democratic incumbents) receive primary support from left-leaning access-oriented
donors.

used to determine whether candidates are integrated into

EPNs.
The communities we identify, in four of the election
cycles, are illustrated in Figure 5. The regularity that we
seek to illuminate with this figure is the community that
appears as a nearly flat line in the lower half of each plot.
Based on the shape and position of this cluster, it can be
seen that (1) the median member is very poorly funded,
(2) the majority are nonincumbents, (3) there is little partisan bias to this community, and (4) it is by far the largest
community. In truth, this cluster might be better labeled
a noncommunity, capturing those candidates and contributors who cannot be classified into tight, coherent
networks. This is evidenced through the lack of a partisan bias in this community, the large membership, and

10

FIGURE 6 Community Partisan and

Incumbency Balance, 19942010

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

FIGURE 7 Month-to-Month Rate of Change in

Networked Challenger Classification
over the Election Year

Note: Each point corresponds to a community identified in one of

the election cycles. The y-axis gives the percentage of candidates
in one party in the community, and the x-axis gives the percentage
of candidates in the community who are incumbents.
Note: Plot gives the proportion of networked/non-networked classification of challengers that changed from the previous to the current month. Plot depicts the 19942008 congressional elections.
Darker lines represent more recent election cycles.

the paltry level of support. Though it might be possible

to derive more nuanced measures of networked support,
we see the classification of a candidate into this wide, flat
cluster as a clear, simple, and stark indication of a lack
of substantial support from any networked community
of interests. In our empirical analysis going forward, we
classify a challenger as having EPN support if he or she
is not assigned to the largest and most weakly funded
community (i.e., not in the wide, flat cluster). In the
supporting information, we show that this wide and flat
cluster appears in each cycle. The other clusters reflect
communities arising from highly dense PAC-candidate
linkages. And most importantly, the communities with a
large nonincumbent membership are highly partisan.
We have proposed community detection as an
approach to identifying the integration of candidates
into the EPNs. We have yet to show that the communities
into which challengers are integrated are, indeed,
partisan. It is conceivable, for instance, that bipartisan
communities exist with pro-business challengers from
both parties who receive support from business interests.
However, Figure 6 clearly illustrates that, insofar as
these communities of PACs are supporting challengers,
they are doing so on a partisan basis. We plot the
aggregate partisanship of candidates in each community
against the percent of community members who are
incumbents. Over all election cycles, nearly every community that contains any nonincumbents is decidedly
partisan. Over the nine election cycles, only a few small
communities emerge that are less than 90% in one party
and contain any nonincumbents. This supports our
claim that the community detection algorithm reliably

locates challengers in partisan networks of interest

groups.
Another concern we want to directly address empirically is the possibility that we are measuring instances
in which groups are piling on to challengers with a high
likelihood of success in a last-minute effort to gain access.
Though we can never perfectly discern a groups motives,
we can assess the aggregate dynamics in the campaign
under the assumption that the likelihood of challenger
success will be revealed as the campaign progresses. Figure 7 depicts the rate of change in the networked challenger classification throughout the election year. This is
the percentage of challengers for whom their networked
challenger status is not the same as it was in the previous month, where challenger status is 0 or 1 depending
on EPN support classification. The final spike in change,
which is the last-minute shuffle toward challengers winning the race, occurs at some point between August and
November. This is why we measure network integration
on July 1.

Electoral Success Analysis

We utilize two separate analytical approaches, applied to
two related but distinct data sets, to separate out the effect of EPN integration from other known predictors of
success in U.S. House elections. Our first analysis utilizes the matching framework to examine whether we see
the effects of EPN integration when we compare races in

11

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

which the EPN- and non-EPN-integrated challengers are

approximately the same on variables known to be strong
predictors of success in U.S. House races. The second analysis constitutes a replication and extension of a regression
model of partisan vote margins presented by Woon and
Pope (2008). Both analyses support the hypothesis that
EPN integration contributes to the success of challengers
in U.S. House elections.
Matching-Based Inference. Matching is an alternative
or complementary methodology for taking account of the
effects of potential confounding variables. The purpose
of matching is to derive balanced treatment and control
groups (Sekhon 2011; i.e., the treatment).11 When treatment and control groups are balanced, they are comparable on all potential confounding variables. Using the
integration of a challenger into a party network community as the treatment, we identify treatment and control
groups that differ only trivially on variables that have been
found to predict congressional election outcomes. We use
a mixture of exact and carefully tuned coarsened exact
one-to-one matching (Iacus, King, and Porro 2012). In
exact matching, the control case selected for each treated
case must have exactly the same values on potential confounding variables as the treated case. Coarsened exact
matching (CEM) is a process by which matches for each
treated case are selected to be within rounding error (i.e.,
within a bin) of the treated case.
The unit of analysis is the race, and the data span the
1994 to 2008 election cycles. We exactly match treatment
and control cases on election cycle, incumbent party, and
political experience.12 Matching on these variables controls for the macropolitical context and the traditional
measure of challenger quality, respectively. Coarsened exact matching is used to match races based on incumbent vote share in the prior election, challenger spending, and incumbent spending. These variables account
for the conventional measures of the competitiveness of
the district and resources, respectively. We start with a
bin size of one standard deviation in each variable and
iteratively reduce the bin size used for the CEM by incrementing the denominator of the standard deviation
(i.e., se/1, se/2, . . ., se/k) until hypothesis tests applied
to the noncoarsened treatment and control group values
indicate that there is no significant difference between
11
Though the terminology of causal inference and the potential
outcomes framework is commonly used in the development of
matching methods, we should be clear that matching is by no
means a case of design-based causal inference.
12
Exactly matching on election cycle and incumbent party controls
for a litany of partisan and economic contextual factors (e.g., party
of the president) that have been found to affect elections.

TABLE 1 Balance check for continuous variables

matched with coarsened exact
matching (CEM).
Challenger Incumbent Incumbent
Spending Spending Vote Share (t-1)
ATT
Mean (treated)
Mean (control)
KS test (p)
t-test (p)
ATE
Mean (treated)
Mean (control)
KS test (p)
t-test (p)
ATC
Mean (treated)
Mean (control)
KS test (p)
t-test (p)

1796270.48 2660867.38
1836609.71 2401184.62
0.96
0.16
0.82
0.22

54.63
54.70
0.75
0.70

1794847.61 2724234.73
1743907.41 2623581.18
0.39
0.10
0.65
0.65

57.60
57.59
0.41
0.95

1700922.49 2426617.67
1456979.00 2273322.29
0.52
0.33
0.18
0.63

55.22
55.38
0.78
0.50

Notes: The test statistic rows report the two-sided p-values associated with the test of the hypothesis that there is no difference
between the treated and controlled samples after filtering with
CEM. One- to-one CEM for estimating ATT, ATE and ATC results in matched samples of sizes 21, 26 and 22 (i.e., 42, 52 and 44
observations), respectively

treatment and control groups. Table 1 shows that our

treatment and control groups are well balanced, in that
there are no statistically significant differences between
treatment and control observations in the quantitative
variables.
Figure 8 gives our estimates of treatment effects. We
estimate the overall average treatment effect (ATE), the
ATE among control units (ATC), and the ATE among the
treated units (ATT; Morgan and Winship 2007).13 The
ATT and ATC constitute estimates of how much the party
network integration increased the probability of victory
among the treated units and how much it would have
increased the probability of victory among the control
units, respectively. The ATE constitutes an estimate of the
difference in the rate of victory between treated and control units, discarding units that are not comparable along
the control measures. The ATC is noticeably larger than
ATE and ATT, which indicates the need for our matching
approach in estimating ATEthe main effect of interest.
If treatment were truly unrelated to the confounding variables, then in expectation ATE = ATT = ATC (Konisky
13
The R package Matching (Sekhon 2011) is used to estimate the
effects and their standard errors.

12

FIGURE 8 Effects of Interest Community

Integration on the Probability of
Challenger Victory in U.S. House
Elections, 19942008

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

TABLE 2 Independent variables from Woon and

Pope (2008)
Variable

Description

incMID

Midpoint between incumbent Ideology and

challengers party mean
Heterogeneity of the party ideological signal
Ideological gap between incumbent and
challengers party
Proportion in the district voting for
Democratic president
Indicator of Democratic incumbent

incDHA
incGAP

Note: Our matching procedure for estimating ATT, ATE, and ATC
results in matched samples of sizes 21, 26, and 22 (i.e., 42, 52, and
44 observations), respectively.

and Reenock 2012). Focusing on ATE, we find that party

network integration increases the chance of challenger
success by approximately 12 percentage points, an effect
that is statistically significant at the 0.05 level (two-tailed).
This effect is substantively significant considering that a
challengers nave chance of success is, in nearly every election cycle, below 10% (Friedman and Holden 2009).14
Regression-Based Inference. Regression models constitute the predominant approach to studying the factors
that influence congressional election results (see, e.g.,
Abramowitz, Alexander, and Gunning 2006; Alexander
2005; Canes-Wrone, Brady, and Cogan 2002; Carson et
al. 2012; Jones 2010; Woon and Pope 2008). In this section, we extend a regression model from Woon and Pope
(2008), in which the dependent variable is the vote share
won by the Democratic Party candidate in U.S. House
elections, to include an indicator that measures whether
there is an EPN-supported challenger in the race. There
are two primary benefits of combining regression- and
matching-based inference. First, we can assess whether
our conclusion about the effect of EPN support is dependent upon the use of matching. Second, we are able to
incorporate many more potential confounding variables
in a regression model than can be accommodated in the
matching analysis.

14
Our conclusions are robust to the use of logistic regression with
the data from this section. Logit results are presented in the supporting information.

Presidential
Vote
Democratic
Incumbent
Dem. Quality
Challenger
Rep. Quality
Challenger
South
Midterm
Dem. Pres

Indicator of Democratic challenger who has

held elective office
Indicator of Republican challenger who has
held elective office
South region indicator
Midterm election
Indicator of Democratic president

Woon and Pope (WP; 2008) examine the relationship between the aggregate ideological characteristics of
political parties regarding the behavior of partisans in
Congress and the results of individual House elections.
Broadly speaking, they argue that party labels serve a
branding function for candidates in congressional elections, and that the behavior of party members in Congress
determines ideological information communicated by the
party brands. In order to evaluate the empirical accuracy
of their theory, WP study U.S. House election outcomes
between 1952 and 2000. They estimate a series of hierarchical regression models in which the dependent variable is the vote share of the Democratic candidate in a
district. They draw upon the literature to include several control variables and specify a number of additional
independent variables that operationalize concepts derived from their theory of party branding. The variables
included in their models are described in Table 2. The
interested reader should see the original article for indepth discussion regarding hypotheses related to these
variables.
Our data on campaign finance networks overlap
with WPs data in the election years between 1990 and
2000. We utilize the subset of WPs data that falls within
this range, which includes 1,939 races in total.15 The
model includes election-level random effects, which are

15

Uncontested seats are excluded from the sample.

13

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

TABLE 3 Replication of Woon and Pope (2008); Table 2, Model 1 (i.e., All)
No PAC Dummies
Variable
incMID
incDHA
incGAP (incDH A < 0)
incGAP (incDH A > 0)
Presidential Vote
Democratic Incumbent
Dem. Quality Challenger
Rep. Quality Challenger
South
Midterm Dem. Pres
Networked Rep. Challenger
Networked Dem. Challenger
Intercept
R2-within
R2-between
R2-overall
N

PAC Dummies

Coefficient

(Std. Err.)

Coefficient

(Std. Err.)

0.112
7.176
0.563
0.189
0.518
0.726
0.024
0.039
0.002
0.023
0.081
(0.006)
0.779

(0.018)
(0.921)
(0.059)
(0.031)
(0.018)
(0.103)
(0.006)
(0.005)
(0.004)
(0.004)
(0.005)
0.080
(0.072)

0.050
6.785
0.569
0.189
0.462
0.740
0.023
0.023
0.003
0.015
0.063
0.066
0.817

(0.018)
(1.641)
(0.108)
(0.058)
(0.020)
(0.186)
(0.006)
(0.005)
(0.004)
(0.007)
(0.006)
(0.007)
(0.131)

0.872
0.672
0.871
1,939

0.931
0.990
0.931
1,939

Notes: Unit of observation is the congressional district. Sample covers the 1990-2000 elections. Dependent variable is the Democratic Party
vote share in the district. Linear regression coefficients from model with election random effects are reported. All effects are statistically
significant at the 0.05 level (two-tailed) except for the South region indicator. The AC Dummiesodel includes 530 PAC/election yearparty
indicator variables

estimated by feasible generalized least squares.16 To test

the effect of EPN support, we include two additional
independent variablesNetworked Rep. Challenger and
Networked Dem. Challengerwhich are indicators of
whether there is a Republican challenger in the race with
EPN support and whether there is a Democratic challenger in the race with EPN support, respectively. We
expect that having an EPN-supported Republican challenger in the race will reduce Democratic vote share and
that having an EPN-supported Democratic challenger in
the race will increase Democratic vote share. In one of the
models, we also include a set of election year and partyspecific PAC dummies that indicate the support of highly
influential PACs. We do this to account for the benefits
of individual group support that go beyond campaign
contributions.17 The results are given in Table 3.
16
WP present several similar models of Democratic vote share. We
present just one replication. We replicate the model from Table 2,
since this one fits the data best. However, our conclusions regarding
the effect of networked challenger do not change if we use one of
the other specifications presented by WP.
17
We adopt a three-stage approach to incorporating influential PAC
dummies. A single PAC dummy indicates that group X gave to
the Republican/Democratic candidate in election year Y. They are

The results provide strong support for our hypothesis that EPN integration improves the electoral prospects
of challengers. The presence of a challenger in the race
with networked support statistically significantly reduces
the expected incumbent vote share by approximately 8
percentage points, an effect that is nearly identical for
Republican and Democratic challengers. To put the magnitude of this effect in perspective, the conventional measure of challenger qualitythe indicator of whether a
challenger has held elective officeresults in a 24 percentage point reduction in the expected incumbent vote
share. This analysis illustrates that our result is robust
to the choice between matching and regression to draw
PAC/election year/party-specific indicators. There are 26,138 potential PAC effects we could include in our model. We do not have
enough observations to identify that many effects. We use a combination of forward and backward search to arrive at our final set
of PAC indicators. In the first stage, we omit any indicators that are
not statistically significant in a simple linear regression in which
Democratic vote share is the dependent variable. This leaves 9.794
potential indicators. In the next stage, we use backward search with
the lassoa form of penalized regression that forces the coefficients
of variables that do not contribute substantially to the predictive
performance of the model (Tibshirani 1996)to further trim the
set of indicator variables. We use tenfold cross-validation to tune
the lasso (Friedman, Hastie, and Tibshirani 2010).

14

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

inferences. Despite the fact that the sample size is only

20% of that appearing in the original study, except for the
South regional indicator, the effects of all of the variables
are statistically significant and in the same direction as in
the original study.

Discussion
We demonstrate that the EPN is active and appears influential in congressional elections. In support of previous theoretical work about the role of interest groups as
policy demanders that shape the party, our analysis provides evidence of coordinated efforts by contingents of the
EPN to support and elect challengers of their choice. In
contrast to previous analyses that focused on relationships
among PACs (Dreiling and Darves 2011; Grossman and
Dominguez 2009), our community detection algorithm
was able to isolate unique subsets of PACs that converged
on distinctive subsets of challengers to finance their elections with contributions and independent expenditures.
We also provide evidencethe first of its kind with a
large sample of electionsthat such backing elevates the
chances of electoral success for selected challengers.
It is instructive to unpack the groups that comprise
the communities that we use to classify candidates as being integrated into the EPN. If these communities are
thought to represent EPNs, then their top supporters (1)
should be clearly on the same side of the partisan divide and (2) represent a diverse mix of specific interests.
Table 4 lists the top 10 contributors to the communities that supported the most challengers (i.e., the greatest
number of challengers were classified into their ranks
by the community detection), broken down by the partisanship of candidates in the communities for election
years 20042010. On the Democratic side, we observe that
the formal party organization, the Democratic Congressional Campaign Committee (DCCC), sits atop the list
each year. Right below the DCCC are interest groups that
are often associated strongly with the Democratic Party
coalition (Grossman 2012), including labor unions, trial
lawyers associations, environmental groups, and womens
organizations. On the Republican side, we observe antitax organizations, the National Rifle Association, pro-life
groups, and trade associations. In support of EPN theory, our study shows that these partisan and ideological
groups coordinate their efforts to support targeted candidates in ways that potentially shape the outcome of the
race.
Previous accounts document two primary ways
that partisan groups coordinate efforts: those driven by

formal party committees and those that arise through

loosely affiliated ideological groups. First, the formal
party committees like the DCCC and National Republican
Congressional Committee orchestrate partisan strategy with PACs (OConnor 2012), some of which have
leaders sitting on advisory boards of the party committee (Kolodny and Dwyre 1998). On the Republican side,
party leaders with strong policy preferences have created
new organizations dedicated to recruiting like-minded
challengers, providing them with seed money, and encouraging interest groups to support them (Peters 1999).
A recent example is the Young Guns PAC started in
2007 by Republican Party leaders Eric Cantor, Paul Ryan,
and Kevin McCarthy to recruit and support conservative
candidates for Congress (Burns 2010). A second coordinating strategy involves meetings among interest groups
to devise common campaign strategies. Campaign finance law prevents the party committees from coordinating with interest groups that use soft money. For this
reason, partisans form umbrella organizations that help
coordinate in the absence of a formal party organization.
In 2004, Democratic partisans used America Coming Together to convene meetings of labor, environmental, and
womens groups to coordinate voter mobilization efforts
(Skinner 2005). In the 2012 elections, a diverse collection
of conservative organizations took its lead from a prominent group called American Crossroads with the broad
electoral goal of helping the Republican Party take full
control of Congress (Confessore 2011). These coordination efforts have been documented in previous elections
(Magleby, Monson, and Patterson 2007).
These developments in congressional campaign activity provide evidence that the kind of coordination and
signaling among partisan elites in selecting presidential
candidates described by Cohen et al. (2008) also takes
place in legislative elections. EPN theory portrays the
party as a conglomerate of nonconflicting and cooperating interests who select and support candidates they
can trust to pursue the policy objectives of the party. This
implies that the integration of a candidate into the EPN
is a signal that they are credible and appealing on partisan
policies. Such signaling should pay electoral dividends
with a broader constituency of partisans. Moreover, the
presence of a community of like-minded PAC contributors may imply a broader electoral effort by the coalition
to campaign collectively for the candidate through direct
mobilization of voters and campaign advertising. Similar
to presidential elections (Cohen et al. 2008), participants
in the EPN have discovered ways to coordinate, but across
the many candidates and campaigns that constitute congressional elections.

15

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

TABLE 4 Top Ten donors to the community containing the highest number of challengers for each
party
Republican

Democrat

Republican

2004
NRCC
Natl. Assn of Realtors
NRA

DCCC
EMILYs List
Machinists/Aerospace
Workers Un.
Am. Medical Assn
Intl Brotherhood of Electrical
Workers
Am.s for a Rep. Majority
Am. Federation of Teachers
Natl. Assn of Realtors
Assn of Trial Lawyers of
America
Help Americas Leaders
United Food & Commercial
Workers Un.
Together for Our Majority
NEA
Keep Our Majority PAC
AmeriPAC
Rely on Your Beliefs
PAC to the Future
2008
NRCC
DCCC
RNC
AFSCME
NRA
SEIU
Rep. Campaign Cmte of NM Defenders of Wildlife
Club for Growth
Natl. Assn of Realtors
Credit Un. Natl. Assn
EMILYs List

NRCC
NRA
Am.s for Tax Reform
Rep. Party of Michigan
Natl. Rep. Trust PAC
Club for Growth

Am. Medical Assn

Freedom Project

Revere America
Freedom Project

Natl. Auto Dealers Assn

Every Rep. is Crucial PAC

AmeriPAC
Intl Brotherhood of Electrical
Workers
Am. Assn for Justice
Our Common Values PAC

Conclusion
We addressed two basic questions. First, whoin terms
of both supporters and candidatesconstitutes the political party via the extended party network, and second,
how are electoral outcomes shaped by the dynamics of
EPN support? We built upon emerging theories about
the organizational form of political parties and exploited
a new algorithm of community detection to reveal unique
patterns of behavior among different groups that embody
the conceptual definition of an EPN. Our study confirms
prior accounts regarding the existence of partisan networks, but it moves beyond prior work in two ways.
First, we integrate both interest groups and candidates
into communities using a bipartite network representation. This method enables us to identify the core of

Democrat
2006

NRCC
Club for Growth
RNC

DCCC
AFSCME
Moveon.org

Natl. Assn of Realtors

EMILYs List

Natl. Right to Life

Every Rep. is Crucial PAC

NEA
SEIU

Rely on Your Beliefs

Dem. Executive Cmte of

Florida
AFL-CIO
Dem. Party of NC
NEA
2010
DCCC
Am. Federation of Teachers
SEIU
Credit Un. Natl. Assn
Women Vote!
Intl Brotherhood of Electrical
Workers
AFSCME
Am. Assn for Justice

Freedom Project
Keep Our Majority PAC
NRA

Arizona Rep. Party

Natl. Fedn of Independent
Business

NEA
AmeriPAC

the party network and illustrate the tight links between

party resources and selected candidates in advancing electoral goals. Second, we demonstrate the stark difference in
electoral prospects for challengers within and outside the
EPN. Challengers supported by densely interconnected
partisan communities have a greater likelihood of winning compared to those with similar campaign resources
and political backgrounds, but without EPN backing.
We posit two related mechanisms that account for
the improved success among EPN-supported challengers.
First, a key challenge for nonincumbents is to surmount
the ambiguity that surrounds their candidacy. Given
that candidate emergence in congressional elections
tends to be highly entrepreneurial (Fowler and McClure
1989; Herrnson 2011), candidates must win an uphill
battle to present themselves in an ideologically clear and

16

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

compelling manner. When a challenger is backed by a

collection of like-minded interest groups, a clear signal
regarding the policy positions represented by the challenger is sent to the broader party network and attentive
members of the electorate. Second, many of these partisan networks arise from and engage in intentionally
coordinated actions in support of candidates. We assume
that EPN-supported candidates benefit from these coordinated support efforts. The relative magnitude of these
two mechanisms needs to be explored in future research,
but it is clear that EPN support plays a major role in accounting for challenger success in congressional elections.
The empirical findings from this analysis have
important implications for theories about contemporary
parties. By observing a unique set of groups that
coordinate actions to help targeted challengers, we have
revealed exactly the kind of electoral activity that scholars
attribute to strong parties (see, e.g., Schattschneider
1942). The patterns we find suggest that parties comprise
contending policy demanders in the electoral process
(Bawn et al. 2012). Our findings also present a challenge
to theories positing that officeholders in the legislature
shape the party to advance their electoral and policy goals
(Aldrich 1995; Cox and McCubbins 1993). If interest
groups with strong policy agendas shape who wins
office, then officeholders might have far less discretion
about the direction of party policy than previously
theorized (Masket 2009). Our findings also speak to
why candidates often fail to move toward the median
voter and instead take greater policy risks by adopting
more extreme positions (Fiorina, Abrams, and Pope
2005). An approach that draws on networked theories
of parties helps explain why party elites are converging
on an ideological point that is more extreme than voters
positions (Fiorina, Abrams, and Pope 2005).

References
Abramowitz, Alan I., Brad Alexander, and Matthew Gunning.
2006. Incumbency, Redistricting, and the Decline of Competition in U.S. House Elections. Journal of Politics 68(1):
7588.
Aldrich, John H. 1995. Why Parties? The Origin and Transformation of Political Parties in America. Chicago: University of
Chicago Press.
Alexander, Brad. 2005. Good Money and Bad Money: Do
Funding Sources Affect Electoral Outcomes? Political Research Quarterly 58(2): 35358.
Arceneaux, Kevin, and Robin Kolodny. 2009. Educating
the Least Informed: Group Endorsements in a Grassroots
Campaign. American Journal of Political Science 53(4):
75570.

Bawn, Kathleen, Martin Cohen, David Karol, Seth Masket, Hans

Noel, and John Zaller. 2012. A Theory of Political Parties:
Groups, Policy Demands and Nominations in American Politics. Perspectives on Politics 10(3): 57197.
Bernhardt, Daniel, and Daniel Ingberman. 1985. Candidate
Reputations and the Incumbency Effect. Journal of Public
Economics 27(1): 4767.
Bonica, Adam. 2013. Ideology and Interests in the Political
Marketplace. American Journal of Political Science 57(2):
294311.
Burns, Alexander. 2010. NRCC Adds Top 16 Targets. Politico, June 30. http://www.politico.com/
news/stories/0610/39212.html.
Canes-Wrone, Brandice, David W. Brady, and John F. Cogan.
2002. Out of Step, out of Office: Electoral Accountability
and House Members Voting. American Political Science
Review 96(1): 12740.
Carson, Jamie, Michael H. Crespin, Carrie P. Eaves, and Emily
O. Wanless. 2012. Constituency Congruency and Candidate
Competition in Primary Elections for the U.S. House. State
Politics & Policy Quarterly 12(2): 12745.
Cohen, Marty, David Karol, Hans Noel, and John Zaller. 2008.
The Party Decides: Presidential Nominations before and after
Reform. Chicago: University of Chicago Press.
Confessore, Nicholas. 2011. Outside Groups Eclipsing G.O.P.
as Hub of Campaigns. New York Times, October 29.
http://www.nytimes.com/2011/10/30/us/politics/outsidegroups-eclipsing-gop-as-hub-of-campaigns-next-year.html
Cotter, Cornelius P. 1989. Party Organizations in American Politics, Pitt series in policy and institutional studies. Pittsburgh,
Pa.: University of Pittsburgh Press.
Cox, Gary W., and Mathew D. McCubbins. 1993. Legislative
Leviathan: Party Government in the House. Berkeley: University of California Press.
David M. Konisky, and Christopher Reenock. 2012. Case
Selection in Public Management Research: Problems and
Solutions. J Public Adm Res Theory (2013) 23(2): 361
393.
De Witt, Karen. 1980. Candidate Grooming Isnt Left to
Chance: Advice and More Advice. New York Times, February 10,E2.
Dreiling, Michael, and Derek Darves. 2011. Corporate
Unity in American Trade Policy: A Network Analysis of
Corporate-Dyad Political Action. American Journal of Sociology 116: 151463.
Dominguez, Casey B. K. 2011. Does the Party Matter? Endorsements in Congressional Primaries. Political Research
Quarterly 64(3): 53444.
Fenno, Richard. 1978. Home Style: House Members in Their
Districts. Glenview, IL: Scott, Foresman.
Fiorina, Morris P., Samuel J. Abrams, and Jeremy Pope. 2005.
Culture War? The Myth of a Polarized America. New York:
Pearson Longman.
Fowler, Linda L., and Robert D. McClure. 1989. Political Ambition: Who Decides to Run for Congress. New Haven, CT: Yale
University Press.
Fraley, Chris, and Adrian E. Raftery. 1998. How Many Clusters? Which Clustering Method? Answers via Model-Based
Cluster Analysis. Computer Journal 41(8): 57888.

EXTENDED PARTY NETWORKS IN U.S. HOUSE ELECTIONS

Franz, Michael. 2010. The Citizens United Election? Or Same

as It Ever Was? The Forum: A Journal of Applied Research in
Contemporary Politics. 8(4): 15408884, DOI: 10.2202/15408884.1408, January 2011
Friedman, Jerome H., Trevor Hastie, and Rob Tibshirani. 2010.
Regularization Paths for Generalized Linear Models via Coordinate Descent. Journal of Statistical Software 33(1): 122.
Friedman, John N., and Richard T. Holden. 2009. The Rising
Incumbent Reelection Rate: Whats Gerrymandering Got to
Do with It? Journal of Politics 71(2): 593611.
Fruchterman, Thomas, and Edward Reingold. 1991. Graph
Drawing by Force-Directed Placement. Software: Practice
and Experience 21(11): 112964.
Grossmann, Matt, and Casey B. K. Dominguez. 2009. Party
Coalitions and Interest Group Networks. American Politics
Research 37: 767800.
Grossmann, Matthew. 2012. The Not-So-Special Interests: interest Groups, Public Representation, and American Governance.
Stanford, CA: Stanford University Press.
Heaney, Michael T., Seth E. Masket, Joanne M. Miller, and Dara
Z. Strolovitch. 2012. Polarized Networks: The Organizational Affiliations of National Party Convention Delegates.
American Behavioral Scientist 56(12): 165476.
Heberlig, Eric S., and Bruce A. Larson. 2012. Congressional Parties, Institutional Ambition, and the Financing of Majority
Control. Ann Arbor: University of Michigan Press.
Herrnson, Paul S. 1988. Party Campaigning in the 1980s. Cambridge, MA: Harvard University Press.
Herrnson, Paul S. 1995. Congressional Elections: Campaigning
at Home and in Washington. Washington, DC: CQ Press.
Herrnson, Paul S. 2009. The Roles of Party Organizations,
Party-Connected Committees, and Party Allies in Elections.
Journal of Politics 71: 120724.
Herrnson, Paul S. 2011. Congressional Elections: Campaigning
at Home and in Washington. 6th ed. Washington, DC: CQ
Press.
Herrnson, Paul S., Christopher J. Deering, and Clyde Wilcox,
eds. 2012. Interest Groups Unleashed. Washington, DC: CQ
Press.
Iacus, Stefano M., Gary King, and Giuseppe Porro. 2012.
Causal Inference without Balance Checking: Coarsened Exact Matching. Political Analysis 20(1): 124.
Imai, Kosuke, and Dustin Tingley. 2012. A Statistical Method
for Empirical Testing of Competing Theories. American
Journal of Political Science 56(1): 21836. Jacobson, Gary C.
1989. ", 1946-86."
Jacobson, Gary C. 1989. Strategic Politicians and the Dynamics
of U.S. House Elections, 194686. The American Political
Science Review no. 83(3):773793.
Jacobson, Gary C. 2009. The Politics of Congressional Elections.
New York: Pearson/Longman.
Jones, David R. 2010. Partisan Polarization and Congressional
Accountability in House Elections. American Journal of Political Science 54(2): 32337.
Key, V. O. 1949. Southern politics in State and Nation. [1st ed.
New York,: A. A. Knopf.
Koger, Gregory, Seth Masket, and Hans Noel. 2009. Partisan
Webs: Information Exchange and Party Networks. British
Journal of Political Science 39(3): 63353.

17
Koger, Gregory, Seth Masket, and Hans Noel. 2010. Cooperative Party Factions in American Politics. American Politics
Research 38: 3353.
Kolodny, Robin, and Diana Dwyre. 1998. Party-Orchestrated
Activities for Legislative Party Goals: Campaigns for Majorities in the US House of Representatives in the 1990s. Party
Politics 4: 27595.
Lau, Richard R., and David P. Redlawsk. 2001. Advantages and
Disadvantages of Cognitive Heuristics in Political Decision
Making. American Journal of Political Science 45(4): 95171.
Lazarus, Jeffrey. 2008. Incumbent Vulnerability and Challenger Entry in Statewide Elections. American Politics Research no. 36(1):108129. doi: 10.1177/1532673x07300908.
Levitt, Steven D., and Catherine D. Wolfram. 1997. Decomposing the Sources of Incumbency Advantage in the U. S.
House. Legislative Studies Quarterly no. 22(1):4560.
Lupia, Arthur. 1994. Shortcuts versus Encyclopedias: Information and Voting Behavior in California Insurance Reform Elections. American Political Science Review 88(1):
6376.
Magleby, David, ed. 2010. The Change Election: Money, Mobilization, and Persuasion in the 2008 Federal Elections.
Philadelphia, PA: Temple University Press.
Magleby, David B., J. Quin Monson, and Kelly D. Patterson.
2007. Dancing without Partners: How Candidates, Parties,
and Interest Groups Interact in the Presidential Campaign.
Lanham, MD: Rowman & Littlefield.
Maisel, L. Sandy, and Walter J. Stone. 1997. Determinants
of Candidate Emergence in U. S. House Elections: An Exploratory Study. Legislative Studies Quarterly 22: 7996.
Masket, Seth E. 2009. No Middle Ground: How Informal Party
Organizations Control Nominations and Polarize Legislatures.
Ann Arbor: University of Michigan Press.
Mayhew, David R. 1974. Congress: The Electoral Connection.
New Haven, CT: Yale University Press.
Morgan, Stephen L., and Christopher Winship. 2007. Counterfactuals and Causal Inference: Methods and Principles for
Social Research. Cambridge: Cambridge University Press.
Newman, M. E. J. 2006. Modularity and Community Structure
in Networks. Proceedings of the National Academy of Sciences
of the United States of America 103(23): 857782.
Newman, M. E. J. 2010. Networks: An Introduction. New York:
Oxford University Press.
OConnor, Patrick. 2012. Campaigns Drop Clues to
PACs. Wall Street Journal. http://online.wsj.com/article/
SB10001424052702303684004577509243899367240.html.
Ostrogorski, M., and Fredrich Clarke. 1902. Democracy and the
organization of political parties, by M. Ostrogorski, translated
from the French by Frederick Clarke, with a preface by the
Right Hon. James Bryce. 2 vols. New York: Macmillan Co.
Peters, Ronald M. Jr. 1999. Institutional Context and Leadership Style: The Case of Newt Gingrich. In New majority
or old minority? : the impact of Republicans on Congress,
edited by Nicol C. Rae and Colton C. Campbell, xi, 220 p.
Lanham, Md.: Rowman & Littlefield Publishers, pp. 4368.
Schattschneider, E. E. 1942. Party government, American government in action series. New York: Farrar and Rinehart.
Sekhon, Jasjeet S. 2011. Multivariate and Propensity Score
Matching Software with Automated Balance Optimization:

18
The Matching Package for R. Journal of Statistical Software
42(7): 152.
Skinner, Richard M. 2005. Do 527s Add Up to a Party?
Thinking about the Shadows of Politics. The Forum 3(3):
Article 5.
Skinner, Richard M., Seth E. Masket, and David A. Dulio. 2012.
527 Committees and the Political Party Network. American Politics Research 40(1): 6084.
Tibshirani, Robert. 1996. Regression Shrinkage and Selection
via the Lasso. Journal of the Royal Statistical Society, Series
B: Methodological 58(1): 26788.
Trottman, Melanie. 2013. AFL-CIO Members Back Link
to Outside Groups. Wall Street Journal, September
9. Taken online at: http://online.wsj.com/news/articles/
SB10001424127887323864604579065491962718868
Woon, Jonathan, and Jeremy C. Pope. 2008. Made in
Congress? Testing the Electoral Implications of Party
Ideological Brand Names. Journal of Politics 70(3):
82336.
Zaller, John. (1992). The Nature and Origins of Mass Opinion.
New York: Cambridge University Press.

BRUCE A. DESMARAIS, RAYMOND J. LA RAJA, AND MICHAEL S. KOWAL

Supporting Information
Additional Supporting Information may be found in the
online version of this article at the publishers website:
Figure S1: The analysis presented in this figure demonstrates that the high-membership, low-funding community appears in every election cycle in our data.
Figure S2: The analysis presented in this figure shows that
the inferences drawn in the matching analysis are robust
to the use of logistic regression on the full, unmatched
dataset.
Figure S3: The analysis presented in this figure shows
that the inferences drawn from matching are robust to
iteratively excluding each of the variables on which we
matched.
Table S1. The analysis presented in this table examines the
degree of association between EPN support and a variety
of challenger background variables.