Pinpointing the Powerful: Covoting Network Centrality as a Measure of Political Influence

• DOI: http://doi.org/10.1111/lsq.12129
• Date: 01 August 2016
• Published date: 01 August 2016
• Author: Nils Ringe,Steven L. Wilson

NILS RINGE

STEVEN L. WILSON

University of Wisconsin-Madison

Pinpointing the Powerful:

Covoting Network Centrality as a

Measure of Political Influence

This article introduces centrality in covoting networks as a measure of influence.

Based on a simple cueing dynamic, it conceptualizes those lawmakers as most central—

and thus as having the greatest signaling influence—who impact the greatest number of

colleagues’ voting decisions. A formal proof and an agent-based simulation show that

cue-providers are always more central than followers; hence, we can use real-world

voting data to identify the most influential legislators. To confirm the measure’s con-

struct validity, we predict covoting centrality in the European Parliament and find those

factors that are expected to impact legislators’ influence to predict their centrality.

Political actors are not created equal. Some are more inf‌luential

than others, be it due to the formal positions of power they hold, positive

personal traits like skills, experience, expertise, ambition, and determina-

tion, or less positive traits, like a lack of scruples. Yet, measuring political

power and inf‌luence is exceedingly diff‌icult. In this article, we introduce

a new measure of inf‌luence that draws from the study of power in social

networks: political actors’ centrality in covoting networks. Specif‌ically,

we use legislators’ voting patterns to measure their social connectedness

and, on this basis, examine the structural positions of individual law-

makers to determine their relative inf‌luence (Bailey and Sinclair 2008).

The basis for this conceptualization of voting as a relational activity is a

simple cueing dynamic in which some lawmakers (cue-receivers) follow

the lead of select colleagues (cue-providers) when voting in a legislature.

Those legislators whose cues inf‌luence the greatest number of colleagues’

voting decisions are most central in the legislative covoting network and

thus have the greatest signaling inf‌luence.

1

We maintain that cue-providers should always be more central

than their followers in covoting networks, which allows us to identify

those legislators who are most inf‌luential in the sense that their signals

affect the votes of the greatest number of colleagues. We make this case,

LEGISLATIVE STUDIES QUARTERLY, 41, 3, August 2016 739

DOI: 10.1111/lsq.12129

V

C2016 Washington University in St. Louis

f‌irst, by presenting a formal proof to show that cue-providers are always

more central than cue-followers. Second, we construct an agent-based

model that simulates legislatures of various sizes and parameterizations

to demonstrate the proof’s robustness to a variety of real-world scenar-

ios. In the empirical section, we calculate the covoting centrality scores

of members of the European Parliament (EP) and identify the factors

that determine them. Our analysis suggests the construct validity of our

measure (i.e., covoting centrality measures what we purport it measures)

in that factors that can be expected to impact individual legislators’

signaling inf‌luence are indeed predictive of the centrality scores of mem-

bers of the EP (MEPs).

The main innovation of this article lies in the use of social network

analysis to gain insights into legislators’ voting behavior. Social network

analysis focuses on the examination of social relationships (or ties)

between individuals (or nodes) and how these connections affect social

and political interactions, processes, and outcomes. The basic unit of

analysis is thus pairs (or dyads) of individual actors. Social network anal-

ysis also allows for the mapping of entire network structures, which are

composed, in the case at hand, of all dyadic covoting ties between a pop-

ulation of lawmakers. Legislative voting is more often than not

conceived of as an individual-level activity, and key to many analyses of

voting behavior is the assumption that voting decisions are independent

of each other.

2

However, in this article we think of voting, and voting

data, in broader terms, by considering how we can garner information

about individual-level inf‌luence by investigating the covoting network

structure as a whole. Treating covoting data as a relational measure

(where the nodes are individual legislators and the ties the extent to

which dyads of lawmakers both vote yea, both vote nay, or both abstain)

and using them to identify inf‌luential legislators constitutes a novel

approach to the study of legislative inf‌luence.

Importantly, the approach travels easily to other legislative arenas,

because roll-call votes are often readily available and because our model

is both parsimonious and general—most importantly, it does not make

any restrictive assumptions a priori about why legislators follow cues or

who serves as cue-provider or cue-receiver.

3

It is also less costly than

engaging in survey-based research because all that is needed to deter-

mine lawmakers’ signaling inf‌luence is a suff‌icient number of roll-call

votes. Our measure does not make assumptions about legislative inf‌lu-

ence by considering the formal positions of power legislators hold (such

as party or committee leader), their seniority, or their fund-raising prow-

ess; instead, it lets the data tell us who wields power by inf‌luencing the

votes of the greatest number of fellow legislators.

740 Nils Ringe and Steven L. Wilson