Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Electoral Systems and Fraud: Evidence from Ukraine's 2012
Parliamentary Election
Fredrik M Sjoberg†
New York University
&
Erik Herron‡
University of Kansas
Abstract
Institutional rules not only influence the legitimate activities political actors pursue to win elec-tions, but they also seem to influence illicit actions. We present a theory about electoral system design that focuses on how institutional rules relate to on-the-ground tactics of fraud. We test observable implications of the theory using a list experiment survey, crowdsourcing data, and fraud forensics on the October 2012 elections in Ukraine. We present evidence of vote manipula-tion, in the form of results falsification in the PR tier, and evidence of voter manipulation, in the form of vote buying in the SMD tier. We further document a preference of "boss" candidates to contest SMD elections, consistent with their incentives and capabilities to manipulate elections. The logic we present not only illuminates why the electoral system changed in Ukraine, but also in other semi-authoritarian states.
Keywords: Election Fraud, Proportional System, Plurality System, Mixed System, Crowdsourcing, List Experiments, and Fraud Forensics. JEL Classification: D72 (Political Processes and Rent-Seeking), D73 (Corruption). † Postdoctoral Visiting Scholar, Department of Politics, New York University, [email protected]. ‡ Professor, Department of Political Science, University of Kansas, [email protected] The authors would like to thank Maksym Palamarenko for help with CEC data, Jorge Soto at Citivox for sharing the ElectUA data, Natalia Zubar for sharing the Maidan Monitoring data, and Rakesh Sharma at IFES for the list exper-iment data. Names are displayed in reverse alphabetical order, but reflect equal authorship. An earlier version of the paper was presented at APSA 2013, Chicago.
2
Introduction
The scholarship on institutional design focuses on how political elites choose and modify rules to
support their vote-seeking, office-seeking, or policy-seeking goals (Müller and Strom 1999). An
underlying assumption of these explanations is that political actors behave as if they are con-
strained by the election rules and the rule of law. In societies where democratic rules are not
fully institutionalized, however, political actors may not behave as if they are constrained by the
rules. Indeed, the rules may provide cover for improper or illicit activities designed to influence
electoral outcomes. We propose that election rule reform may serve the needs of rulers by alter-
ing the environment in which they can manipulate elections.
Elections held in authoritarian regimes generally have a predetermined conclusion: pro-regime
outcomes prevail. The mechanisms leading to these outcomes can vary substantially, however.
Officials can manipulate the choice set facing voters, eliminating credible opposition candidates.
They can use control of the media to prevent the opposition from communicating with voters, to
demobilize opposition supporters, and to promote their own candidates. They can order or in-
centivize the implementation of intimidation and fraud to be perpetrated at different levels of
the process. Not only do the tools to engineer outcomes vary, but the inducements to ensure
compliance may also vary spatially and temporally. Regime proponents may employ centralized
and coordinated mechanisms, or they may rely on decentralized and uncoordinated mechanisms
(Sjoberg 2013a; Herron 2011). In short, while the dichotomous outcome (regime wins/regime
loses) does not tend to vary in authoritarian systems - the processes that produce the outcome
may vary substantially.
This paper uses data from two concurrent crowdsourcing initiatives, a list experiment, micro-
level official results data, and candidate biographical data to assess variation of electoral manip-
ulation in Ukraine across the two tiers, plurality (SMD) and proportional representation (PR).
We first outline our theoretical expectations, linking the incentives generated by election rules to
the opportunities available to commit illicit acts during the election. We continue by presenting
the case of Ukraine, indicating why it is an appropriate environment in which to evaluate hy-
3
potheses about election quality. Next, we present hypotheses that link institutional rules to evi-
dence of fraud, and we describe the four sources of data that we use to assess the hypotheses. In
the analytical section, we establish that allegations of fraud in Ukraine are plausible, and that
alleged methods of fraud vary. We further connect evidence of fraud to institutional rules and
behavior of political actors that flows from these incentives. We conclude by discussing implica-
tions of the findings.
Theoretical Expectations of Institutional Design and the Responses of Political Actors
The extensive literature on the design and consequences of election rules is often differentiated
by underlying assumptions about the direction of causation (Benoit 2007). Research that privi-
leges the work of Maurice Duverger generally asserts that electoral institutions affect the devel-
opment of party systems, measured most commonly by the effective number of parties
(Duverger 1959; Taagepera and Shugart 1989). The alternative approach asserts that pre-
existing partisan differences propel electoral system choice, with some scholars noting that
changes in the party system often precede electoral system re-design, suggesting that election
rules are chosen to accommodate existing partisan differences rather than generating them
(Colomer 2005).1
While the connection between electoral and party systems is the best established relationship,2
scholars have sought to connect electoral systems to other outcomes. Some research has suggest-
ed that the lines of accountability in PR increases the likelihood of corruption (Persson et al.
2003), especially as the linkage between citizens and politicians is less direct than in constituen-
cy based systems. Dishonest or corrupt politicians may be punished directly by voters in con-
stituency systems, whereas they may be insulated from accountability in PR if the party con-
1 Recent scholarship has noted that the dichotomy is false; Duverger recognized the importance of social conditions as well as institutional rules in structuring the party system (Clark 2006). 2 The literature on election rules and their effects on party systems has been regarded as an exemplar of the accumu-lation of knowledge in the field of political science (Riker 1982).
4
trols list construction. Constituency-based systems, however, create incentives for politicians to
accommodate local needs, a feature that could permit other forms of corrupt actions. Whether
or not voters hold politicians accountable depends on the nature of the relationship. In clien-
telistic societies the citizen-politician linkage does not revolve around democratic accountability,
but rather around the transfer of private material goods in exchange for political support
(Stokes et al. 2012; Kitschelt and Wilkinson 2007). In such a setting an electoral system that
promotes local-level linkages, like SMD, might well lead to more corruption.
Our underlying assumption is that all electoral systems provide incentives and disincentives for
the perpetration of fraud, but different rules may be more conducive to specific forms of corrupt
activities. Recent election reform in semi-authoritarian Ukraine, reinstating a mixed system fol-
lowing two relatively free and fair elections under PR, raises the question: What were the under-
lying motivations for the reintroduction of a plurality component in the electoral system? A
critical assumption in much of the electoral systems literature is that election rule designers,
politicians, candidates, and parties act in good faith, albeit in a self-interested way, when devel-
oping and implementing the rules. While these actors may craft rules to provide advantages to
themselves and their co-partisans, their behaviors are constrained by formal and informal rules.
In other words, any advantages that accrue from the rules come through fair play: politicians
follow the rules during the design phase, codify the rules with the intention that they will be
followed, and comply with the rules.
What if institutional designers and partisan actors do not intend to comply with the rules and
instead alter them to provide cover to illicit activities? If the assumption that actors follow the
rules is relaxed, how would this behavioral change affect the process of conducting elections, the
underlying observations that emerge from that process, and the outcomes as votes are translated
into seats?
As scholars have noted, political actors who intend to undermine democratic practices may in-
fluence the process at many levels. They may alter the authority of institutions, limiting elec-
toral choice or the power of elected officials, to preserve their control over decision-making.
5
They may influence the choice set facing voters, removing credible opposition candidates from
contestation. They may reduce access to information about certain competitors, or permit the
dissemination of biased information about competitors, to reduce the likelihood of opposition
success. They may increase barriers to entry for voters or candidates, especially targeting likely
supporters of the opposition. They may interfere with the vote-casting, counting, or aggregation
processes to ensure that their preferred candidates emerge victorious (Schedler 2002). Moreover,
efforts to undermine open competition may be dynamic as actors adapt to civil society or inter-
national efforts to undermine the effectiveness of the illicit activities (Simpser and Donno 2012;
Hyde 2011).
Underlying the argument about the electoral process is an implicit assumption that the locus of
influence and coordination is centralized. Indeed, some actions would require centralization to be
effective; modifications of national-level institutions and their authority, for example. Other as-
pects of manipulation could be perpetrated with less coordination, notably efforts to influence
the vote on election day. If political actors centrally coordinate efforts to engineer the casting,
counting, or aggregation process, spatial variation in regime support should be limited as the
regime has little incentive to suggest that opposition forces enjoy areas of concentrated support.
If actions are uncoordinated, election administrators may use a wider range of activities to se-
cure pro-regime outcomes (Herron 2011) .
Our theoretical expectations flow from these critical assumptions. First, the design of institu-
tional rules may be constrained by democratic norms, but may also be motivated by efforts to
manipulate processes to engineer preferred outcomes. Second, the implementation of election
rules can vary across a given country, providing differentiated opportunities to commit fraud.
Third, election rules may be designed to facilitate certain types of fraud, and the level of cen-
tralized control over efforts to falsify results may vary. The ability to manipulate processes is
contingent on the rules and on local capacity to conduct successful operations.
This paper takes advantage of the characteristics of Ukraine's mixed electoral system to assess
evidence of variation in the proportional representation and single-member constituency compo-
6
nents, notably in the evidence of activities that could be improper. In a departure from much of
the literature on electoral systems, we focus on adaptations designed to provide illicit advantage
to partisan actors. The next section provides additional details about our expectations, address-
ing the literature on election fraud and connecting the theoretical arguments to conditions in
Ukraine, the case under analysis.
Election Fraud
Early studies of election fraud were limited in number, emphasized historical, ethnographic, and
case study approaches, and focused on describing the phenomena associated with fraud
(Lehoucq 2003). Several events propelled research on election quality forward over the last dec-
ade. The contested United States presidential election in 2000 inspired several studies investigat-
ing how ballot design and voting technology affect outcomes (Wand 2001; Herron 2003; Knack
2003). Further, allegations of fraud served as a catalyst for electoral ‘revolutions’ all over the
globe (Tucker 2007) and researchers responded by advancing the tools available to assess allega-
tions of manipulation and extending the scope of election quality studies. Increased transparency
opened access to polling station level data in many countries, the ascendance of experimental
approaches provided tools to gather micro-level attitudinal and behavioral effects, and techno-
logical innovations permitted citizens to report on alleged violations in real-time via mobile de-
vices.
Three identifiable trends have emerged in the literature on election fraud: studies of election
forensics, focusing on the empirical assessment of election data to uncover anomalies attributa-
ble to fraud; studies of election integrity, focusing on theoretically-driven discussions about how
to evaluate election quality; and studies of incentives for perpetrators, addressing how, why, and
when actors use fraud. Among these strands of research, work on forensics has the most exten-
sive scholarship. While efforts to analyze election data for abnormal patterns predate the 2000s
(Sobyanin and Sukhovolskiy 1995), a diverse set of approaches has recently emerged. Research-
ers have explored the frequency of digits in naturally occurring data (Mebane 2006; Beber
7
2012),3 spatial and temporal distribution of turnout (Myagkov et al. 2009; Klimek 2012), evi-
dence of improper voter registration (Fukumoto and Horiuchi 2011), patterns of voting in spe-
cial polling stations (Herron and Johnson 2007), and audit data (Hood and Gillespie 2012) to
evaluate the quality of election administration practices.
Work on forensics tends to be empirically-driven rather than theoretically-driven. Research on
election integrity is oriented toward fundamental definitions of free and fair elections (Elklit
2005) and how these concepts might vary cross-nationally (Norris 2012). A key element of this
research is the effort to disentangle the potential effects on outcomes produced by different pro-
cesses (manipulation of institutional rules, influences on the formation and expression of citizen
preferences, and efforts to alter the voting activities) and different intentions of perpetrators
(Birch 2011; Vickery 2012). The latter specifically addresses the implications of ‘malpractice’
that is a product of inadequate training alongside intentional efforts to undermine the fair cast-
ing, counting, and compilation of votes. Scholarship in this area has also investigated how spe-
cific efforts to identify fraud may influence the behavior of officials and mitigate illicit activities
(Hyde 2007; Herron 2010; Kelley 2012; Ichino 2012; Sjoberg 2013b). Some research has bridged
forensics and integrity, wrestling with theoretical and empirical matters (Alvarez et al. 2008;
Alvarez et al. 2012).
Several scholars have explored another theoretical dimension of the study of election fraud: why
and under what conditions individual actors engage in illicit activities. Fraud has an immediate
benefit: it can help secure victory or the perception of a strong electoral mandate (Fearon 2011).
However, the primary impetus for fraud may be as a signal that benefits pro-regime actors out-
side the immediate electoral contest by demonstrating their strength (Magaloni 2006), and as an
observed activity (Simpser 2012) or hidden activity (Little 2012)4 that provides information
about opposition support.
3 The use of Benford’s Law and its assumptions about the distribution of digits for election data has been controver-sial. See Deckert, Myagkov and Ordeshook (2011); Mebane (2011); Beber and Scacco (2012). 4 The characterization of observed and hidden is from (Little 2012).
8
We integrate the three branches of election fraud literature by focusing on how institutional
variation may affect the incentives for perpetrators to pursue different forms of fraud. After em-
pirically establishing the presence of anomalies that could be attributed to fraud, we address the
incentives for perpetrators to use different forms of fraud under different conditions. By address-
ing sub-national variation in election quality, we add to the literature on election integrity. Our
analysis focuses on Ukraine, where institutional characteristics and access to varied forms of
data allow us to conduct the analysis while holding constant other contextual features.5
The Case of Ukraine
This section provides a brief introduction to Ukraine's election rules and political environment,
focusing on how it serves as an optimal case to assess questions about election fraud. Ukraine
has held six parliamentary elections since it gained independence following the Soviet Union's
collapse in late 1991. The elections have been held under three distinct electoral systems: major-
ity-runoff, mixed, and proportional representation, with important rule changes implemented
even when the overall system remained intact.
In the first election following independence, all 450 seats in Ukraine's parliament were selected
in constituencies using a majority-runoff formula. The rules were modified for the 1998 parlia-
mentary election, with seats divided evenly into a constituency tier using plurality rules and a
proportional representation tier with a 4 percent threshold. The system persisted in 2002, alt-
hough the ability to simultaneously contest seats in PR and the constituencies was removed.
The electoral system was altered once again prior to the 2006 election, with the mixed electoral
system replaced by national PR with a 3 percent threshold. The law was altered slightly for the
2007 snap elections. The mixed electoral system was reinstated for the 2012 election in a form
quite similar to the version used a decade earlier, but with a 5 percent threshold in the PR tier.
Here we focus on the 2012 elections and the rationale for re-introducing a SMD component.
5 Other countries that recently have reintroduced a SMD tier are Russia (2013), Moldova (2013), and Bulgaria (2009). In the latter two cases the SMD tier was later abandoned.
9
During the two decades of independence, external evaluations of election quality varied as well.
Early elections were criticized for political cronyism and the absence of a fair process. Following
the Orange Revolution in 2004, two parliamentary elections received more favorable evaluations
from international observers. The 2012 parliamentary elections featured a return to more nega-
tive assessments, similar to the 2002 elections, with concerns raised about media access, the be-
havior of election commissions, the deployment of administrative resources, and other efforts to
influence the election process. Ukrainian politicians have experienced variation in institutional
incentives as well as political environments more – and less – conducive to fraud.
Our theoretical argument suggests that the changes in electoral systems may be related to the
ways in which the institutional rules facilitate different forms of fraud. Indeed, one of the argu-
ments offered by the political opposition in the early 2000s in favor of adopting pure PR focused
on the perception that SMD facilitated fraud by pro-regime candidates. They suggested that
administrative resources could be more effectively used to secure victories in the districts, and
adoption of PR could mitigate these forms of fraud (Herron 2007).
While Ukraine's opposition suggested that PR was less conducive to fraud, we assert that varia-
tion in institutional rules permits the perpetration of different forms of fraud. That is, electoral
systems are always vulnerable to fraud, but the structure of rules and incentives they provide
may be more, or less, conducive to different illicit tactics. Moreover, institutional designers may
become cognizant of this variation and select rules to take advantage of these vulnerabilities. If
we interpret the return to a mixed electoral system as an effort not only to enhance electoral
prospects via legitimate means, but also through illegitimate ones, how does this narrative com-
port with the political dynamics taking place at the time of election rule change and the 2012
parliamentary election?
Having gained the presidency, but not yet controlling a majority in parliament, the Party of
Regions faced a considerable test in the 2012 parliamentary elections. Reports of sporadic fraud
in the 2011 local elections (Herron and Boyko 2012) raised concerns about the 2012 contest and
also garnered the attention of international organizations. Ukraine's upcoming term as chair of
10
the Organization for Security and Cooperation in Europe generated additional scrutiny and ren-
dered more challenging efforts to balance Western commitments with strong Russian ties.
As the 2012 elections approached, the ruling Party of Regions was faced with the task of max-
imizing seat acquisition without resorting to the egregious levels of fraud present during the
2004 presidential election that provoked the ‘Orange Revolution’. The regime successfully passed
election reforms that would reinstate the mixed electoral system, but also introduce some
measures to enhance transparency (e.g., installation of webcams in all polling stations). The
political opposition assented to election rule reform, but also vocalized concerns about the po-
tential for fraud.
From the perspective of parties and candidates, mixed systems permit hedging so that poor per-
formance under one method of seat allocation could be offset by good performance under the
other method. However, the adoption of a mixed electoral system could produce other benefits if
political actors intend to pursue sinister strategies to secure seats. The mixed system provides a
particularly flexible platform for large- and small-scale fraud, depending upon the level of scruti-
ny by domestic and international observers. If oversight is minimal, the choice set could be re-
stricted at the national level and votes could be manipulated at all levels of the administrative
process to produce the preferred outcome.
The mixed system platform provides the greatest amount of flexibility to actors who intend to
commit fraud but also wish to save face in the international community. When oversight is min-
imal, methods can be used to secure a pro-regime victory in both tiers. When oversight is
heightened, methods can be used to secure seats in the constituencies. This tactic could be espe-
cially fruitful if partisan actors are willing to intervene in any district where outcomes are uncer-
tain, rather than attempting to succeed in a specific district. The scale of the intervention could
be smaller, undermining detection, and the number of votes to manipulate and secure a victory
could be lower than the amount needed to acquire an additional seat in the PR tier.
11
In sum, we expect that institutional rules can affect behavior, even when the spirit of the rules
is violated. Institutional actors who intend to commit fraud, but are also mindful of detection,
have incentives to select systems that provide maximum flexibility for the methods that they
employ. In the case of Ukraine, we would expect to find variation in the forms of fraud em-
ployed in the PR and SMD components, as well as variation in the behaviors of actors who in-
tend to use fraud as a tool to gain elective office.
Hypotheses
Our theoretical expectations lead to two types of testable hypotheses. The first set of hypotheses
addresses how institutional incentives affect tactical decisions about electoral manipulation. We
differentiate between vote manipulation, or bureaucratic fraud, committed within polling sta-
tions by officials, and voter manipulation, acts of manipulation committed outside of polling
stations by partisan actors who are generally not election administrators (Sjoberg 2012). Bu-
reaucratic fraud encompasses actions taken by officials to stuff ballot boxes, fabricate results, or
otherwise interfere with the casting, counting, or aggregation of votes on election day. Voter
manipulation includes many activities such as vote buying and intimidation that may occur on
or before election day.
Dominant parties, such as Ukraine's Party of Regions, are generally well placed to control the
perpetration of bureaucratic fraud due to their authority over the bureaucracy, but other parties
may have localized access to these resources. Political parties nominate poll workers in Ukraine
and this characteristic of election administration incentivizes them to deliver the vote for their
own party, conditioned on local constraints such as the checks that other partisan poll workers
and election observers provide. Given the organizational prerequisites for committing bureau-
cratic fraud (i.e., control over the election administration apparatus), we anticipate that bureau-
cratic fraud is more commonly found under PR rules, where parties are the primary contestants.
While voter manipulation could be perpetrated by election officials, it is more likely to be asso-
ciated with candidate or party machines, and law enforcement officials. A common technique of
voter manipulation that was allegedly widespread in Ukraine is vote buying. Vote buying
12
schemes can be conducted in many ways, but inherent in the approach is the exchange of valua-
ble resources for the promise of a vote in favor of a particular candidate or party. Perpetrators
of this form of fraud have several tools at their disposal to check for compliance by voters, in-
cluding "voting carousel" techniques, as well as mobile phone images of completed ballots. Vote
buying does not require control over election administrators, but rather access to financial re-
sources and a network to implement the vote buying scheme. Given the resource and logistical
requirements, these manipulation tools are more likely to be implemented by local political en-
trepreneurs that have resources available for distributing material rewards. Because SMD incen-
tivizes local elites to take part in elections as candidates to consolidate influence over a specific
local territory, we anticipate that vote buying will be more prevalent in constituency races than
in PR. The following hypotheses flow from our expectations:
H1a: Elections in the PR tier are more likely to yield evidence of vote manipulation, or
bureaucratic fraud.
H1b: Elections in the SMD tier are more likely to yield evidence of voter manipulation.
The second set of hypotheses connects institutional incentives to the behavior of political actors.
We anticipate that certain types of candidates would be attracted to environments where their
comparative advantage can be acted upon. Specifically, locally powerful political machines
would be especially adept at voter manipulation like vote buying, leading political actors with
local power bases to be especially attracted to SMD.
H2a: "Boss"-type candidates should be more prevalent among the nominations in SMD
races than in PR, especially for large parties.
An increased frequency of "boss" candidates in SMD could be associated with the incentives of
the system and potential for manipulation, but it could also be due to alternate explanations.
Notably, candidates with this profile are more likely to be marginalized under PR due to ac-
countability concerns and the need for a balanced party list. While they may choose to contest
SMD races because of the greater likelihood of success, the tactics to secure victory may be more
13
benign. However, if these nominations are associated with reports of illicit behavior, then the
attraction to SMD may have been motivated by opportunities to commit certain types of elec-
toral manipulation. To investigate this possibility, we integrate candidate, election result, and
election monitoring data to investigate how the nomination of candidates more likely to be asso-
ciated with local machine politics affects reports of fraud, and other types of manipulation. Spe-
cifically,
H2b: Higher concentrations of boss candidates in districts are likely to be associated
with higher reports of fraud, and specifically higher reports of vote buying.
Our theoretical expectations differentiate the incentives that PR and SMD election rules provide
to political actors willing to engage in fraud, and are designed to probe the data for evidence in
outcomes. The hypotheses operationalize these expectations, assessing the evidence that fraud
reports vary between the tiers in two stages. In the first stage, we establish empirically that
differences exist. In the second stage, we investigate whether those differences are associated
with other features related to our theoretical argument, notably that certain types of candidates
would be attracted to SMD and that their activities would be associated with vote buying, and
that party strength would be associated with campaign violations.
Data
To investigate our research questions, we use data from several sources. In this section, we de-
scribe the characteristics of the data sources.
List Experiment
To explore voter manipulation, we need measures that capture the experiences of the voters
during the election period. One standard approach would be to conduct a survey with a repre-
sentative sample of all voters. However, direct questions can underreport vote buying, prompt-
ing us to implement a list experiment design to solicit more truthful responses about the extent
to which vote buying affected the vote choice (Gonzalez-Ocantos et al. 2012). Since our interest
is on the difference between the party list candidates and SMD ballots, we asked respondents to
14
answer questions separately for the two tiers. While this approach requires sophistication from
respondents, as individual voters were asked to distinguish between the campaign tactics that
were used by party and SMD candidates, and differentiate their experiences, it also permits us
to evaluate how tactics may vary across the tiers.
The survey was implemented by adding a battery of questions on electoral integrity to an exist-
ing post-electoral omnibus survey6 that included a representative sample of 2,048 individuals.7
For the list experiment, the sample was divided into three groups of equal size. One group got
the short list of non-sensitive items, another group got an additional item about vote buying,
and a third group got an item about intimidation. The question was: what determined your vote
in SMD or PR respectively. The groups were created at random, generating three groups with
the same characteristics except for the question format on the list experiment questions. To veri-
fy that there is no systematic difference the three groups, we present sample means (see Table
2). Results from a multinominal logit model that predicts treatment assignment as a function of
education, gender, age, self-reported voting, and region yields a non-significant likelihood ratio of
7.8 (p-value=0.846). In short, the three experimental groups are closely balanced in terms of
observable characteristics. This list experiment approach to uncovering evidence of variation in
fraud across different ballots is, to our knowledge, the first of its kind in the literature.
[Insert Table 1 about here]
Crowdsourcing Data
With the advent of social media and new information technologies, new horizons have opened to
explore voter experiences with the electoral process. Much of the literature on fraud has relied
on data from accredited election observers who spend election day inside polling stations (Hyde
2011; Kelley 2012). Crowdsourcing taps into the experiences of voters both inside and outside
polling stations. Crowdsourcing can be defined as "the act of taking a job traditionally per-
6 The survey was implemented by KIIS and the module commissioned by the International Foundation for Electoral Systems (IFES). 7 For the list experiment module we have 1,538 completed interviews.
15
formed by a designated agent and outsourcing it to an undefined, generally large group of people
in the form of an open call" (Howe 2006). The best-known early developer of crowd-sourced
election observation is the Kenyan based Ushahidi platform.8
An important distinction has been made between bounded and unbounded crowdsourcing, the
former relying on a set of trusted, pre-screened individuals, and the latter including the complete
set of potential reporters.9 In the context of election observation, the "bounded crowd" refers to
accredited domestic election observers who are trained and commissioned to submit reports on
election day from inside polling stations. The unbounded "crowd" is composed of the general
public who voluntarily submit reports. This distinction is not only important conceptually, but
also important in terms of fraud dynamics since accredited observers placed inside polling sta-
tions might displace fraud (Sjoberg 2012).
The data we collected from the 2012 Ukrainian elections come from two web-based election-
monitoring platforms, Maidan and ElectUA, that integrated regular citizen reports, local jour-
nalism, and accredited observer reports. While these data provide insight into on-the-ground
activities, they also present some challenges for analysis and interpretation. The most important
challenge is non-probability sampling. The voluntary nature of report submission, as well as its
technological requirements, renders reporters more likely to be young and urban, potentially
introducing selection bias into the data.10 Furthermore, the definition of the unbounded crowd
means that participants are not formally trained, and may vary in their evaluation of the severi-
ty of activities that they witness. Because any individual may submit reports, the provenance of
the report may be challenging to establish.
8 It was first deployed after the 2008 elections in Kenya, when information on post-electoral violence was collected by the use of mobile phones and visualized on a web-based map. The open-source platform has inspired efforts to engage citizens in observation all over the world, with thousands of Ushahidi websites created for a broad range of purposes: to monitor humanitarian crises, to gather information about corruption and governance, to track smoking bans, or to monitor elections. See http://www.ushahidi.com. 9 The distinction was first made by Patrick Meier in a blogpost, see ’Crowdsourcing in Crisis: A More Critical Reflec-tion’, Posted on March 31, 2009. 10 Because individual level data on the people who submitted reports are unavailable, we cannot evaluate this hypoth-esis, however.
16
Cognizant of these concerns, organizations that collect and visualize crowdsourcing data have
developed methods to verify reports. Reports are manually verified in a special purpose situation
room at the NGO HQ. Maidan, a domestic pro-democracy NGO, vetted reports prior to adding
them to the database by: (1) verifying the existence of the reporter; (2) reading the report and
determining if it is related to the elections; (3) checking the background and supporting supple-
mentary documentation (e.g., photos, videos, or audio recording); and (4) evaluating if the alle-
gation would qualify as a violation of the law. The report was discarded if it failed to meet any
of the four criteria.11
ElectUA, an organization sponsored by the Washington, D.C.-based media company Internews,
used similar protocols.12 In addition, ElectUA incorporated the bounded crowd in the form of
accredited non-governmental election observers, OPORA and ENEMO, and local journalists.
Reports submitted from the bounded crowd were automatically classified as verified and thus
not subject to the same vetting as reports from the unbounded crowd. These safeguards are
primarily designed to prevent "false positives" from entering the dataset. However, they could
exclude legitimate reports and do not address other concerns about the representativeness of the
data.
Incident reports consist of a field of text, which includes the substance of the report as well as
the location. Reports were submitted via SMS/MMS, Call Center, Twitter, Facebook App,
Online Web-form, or E-Mail. Maidan reported 1,643 individual incidents in its data and Elec-
tUA reported 1,724 incidents from across Ukraine (see Figure 1).13 There are more Maidan re-
11 Personal correspondence with Maidan Monitoring. The website, http://maidanua.org/vybory2012, was powered by Ushahidi. 12 ElectUA was set-up by a Mexican start-up company called CitiVox, and thus not operating on the Ushahidi plat-form. For more information, see http://innovation.internews.org/blogs/ukraine-parliamentary-elections-using-crowdsourcing-transparency-and-democracy. 13 The difference between the number of cases used in the analysis and the total number reported by the organizations is related to the quality of location data. If the reports did not include enough information to identify the district in which the alleged violation took place, it was excluded from our analysis. Also, Directed Denial of Service (DDoS) attacks on election day might have reduced the total number of submitted reports. The incident reports were scraped form the Maidan website, while the ElectUA data were obtained from CitiVox. We have the geo-coordinates for 1,270
17
ports from the South Eastern parts of the country, 45.9 percent, compared to 39.9 percent from
the Central regions, and only 15.1 percent from the West. The ElectUA reports are equally
spread out over Ukraine's territory.
In terms of bounded and unbounded crowd, we consider all of the 1,458 (88.7 percent) verified
Maidan reports to be genuine crowd reports since there was no formal collaboration with accred-
ited domestic observers.14 ElectUA separated out the bounded reports, 1,093, from the unbound-
ed crowd, 631. Overall, 40.4 percent of the ElectUA reports were verified according to the pro-
tocol, including both automatically verified bounded crowd reports and separately verified citi-
zen reports. Figure 1 illustrates the geographical distribution of all reports from both platforms.
[Insert Figure 1 about here]
The organizers of the two monitoring platforms manually coded latitude and longitude (geo-
referencing), and categorized the substantive text into different types of electoral malpractice.
We further code these categorizations into eight types of electoral malpractice, ranging from
outright vote-count fraud (counting) to administrative problems. One of the challenges is that a
single report might be categorized into many different malpractice categories. For instance a
report might have been categorized as ‘adminresurs, obman’ (administrative resources, fraud).
We use the primary malpractice category assigned to an incident report by the classification
team. In the example above this means that the main malpractice category would be adminis-
trative. Table 3 displays the relative frequency of each category.
[Insert Table 2 about here]
Bribing, counting, and observer obstruction are self-evident categories. Administrative violations
often involve officials failing to follow formal operating procedures. Campaign violations consist
Maidan reports and 1,449 ElectUA reports. The point-in-polygon operation was done in R using the maptools pack-age. Election district (n=225) shapefiles were obtained from Serhij Vasylchenko. 14 The analysis will be conducted both separately for verified and unverified reports in the next iteration. We use all reports in this paper.
18
of activities that violate rules about appropriate communication or mobilization, such as rallies
or meetings held on election day. Interference may include undue influence by a local official or
by a party proxy in the administrative process. Intimidation may include activities such as har-
assment by security officials to beatings by thugs. Voting violations include multiple voting or
problems with the secrecy of the ballot.
For the district-level analysis, we are especially interested in malpractice categories that tradi-
tional election observers find difficult to observe, such as vote buying. These techniques are
mainly used outside polling stations and observers often miss these activities because they are
stationed inside polling stations. Voters, however, are better placed to observe bribing and in-
timidation. Indeed, examining the type of malpractice reported by category, bounded or un-
bounded, we find that vote buying reports from ElectUa are more likely to come from the un-
bounded crowd.15
Electoral Returns Data
The Central Electoral Commission (CEC) in Ukraine makes polling station electoral returns
available online. Election protocols – the formal documents completed by election management
bodies at all levels of the hierarchy – establish basic information about turnout and the distribu-
tion of votes. The availability of these data at the polling-station level permits us to evaluate
election returns for the presence of anomalies that may be attributable to election fraud.
Ukraine had 33,762 polling stations spread out over 225 districts in the 2012 election; we use
data at different levels of aggregation in the analysis of our hypotheses.
Candidate Data
Parliamentary candidates in Ukraine are required to file information about their personal biog-
raphies and the CEC publishes this information online. This self-reported information includes
party affiliation, age, residency, education, occupation, and workplace. The positional data al-
lows us to extract information about the personal characteristics of candidates, permitting us to
15 χ2 (1, N=1,724) = 8.901, p = 0.003.
19
code candidates who have a ‘boss’-like profile. In addition to self-reported data, we also include
position on the party list for PR candidates and individual-level vote totals for SMD candidates.
Misreporting is a challenge because it may be correlated to key candidate attributes. For in-
stance, in earlier elections, some well-established candidates formally registered as ‘unemployed’
even if they were not. In the 2012 candidate data, only one candidate labels herself unemployed
under the field of ‘occupation’, alleviating the concerns about outright misreporting. However,
the largest gap in the data concerns the employment category, with 39.0 percent failing to dis-
close details about employment.16
For the analysis of candidate type and electoral system tier, we created a ‘boss’ score that ag-
gregates expected ‘boss-like’ attributes for each candidate. These attributes include whether or
not the occupation of the candidate includes the following descriptions: director, president, or
head (nachal’nik or golova), or if the employment field included: company, enterprise, or the
Ukrainian equivalent of LTD. The distribution of each attribute is displayed in the table below.
[Insert Table 3 about here]
Summing up all boss attributes results in an additive ‘boss’ score. Almost half of all candidates
in 2012, 2,660 or 46.1 percent, belong to one or more of the attributes listed above, which sug-
gest that this categorization is broad and probably includes candidates that would not have ac-
cess to the resources that we attribute to bosses. For the final ‘boss’ dummy we included all
candidates with a boss characteristic in both the occupation and the employment field. We also
filtered by age and gender. While young entrepreneurs can develop patronage networks, experi-
ence and time facilitate the development of a political machine. Further, while women could, in
principle, serve in this capacity, men in Ukraine dominate political competition. Consequently,
16 To evaluate whether failure to report is systematically associated with candidate strength we can regress party list position for the PR candidates and vote share for the SMD candidates on a dummy for failure to report employment. There is no relationship in terms of party list position (β=2.73, p-value=0.299), but there is a small negative associa-tion with individual vote share (β=-.0117, p-value=0.019). This finding suggests that there does not seem to be a pattern of powerful candidates failing to report their employment.
20
the final ‘boss’ dummy excludes candidates under the age of 35 and women.17 The ruling Party
of Regions has more boss candidates than in other parties, with 56.6 percent of the candidates in
the ruling party coded as a bosses compared to an average of 29.0 percent for all the other par-
ties.
For the district-level analysis we collapsed the candidate data to create a ‘boss’ score summing
the number of boss-type candidates for each single-member district. This operation could only
be conducted for the SMD candidates since the PR candidates run in one single nation-wide
constituency.
Analysis
Our analysis encompasses two stages. In the first stage, we establish that electoral malpractice
was reported in Ukraine's 2012 election, and that the types of fraud reported vary. In the second
stage, we directly address the hypotheses detailed in the previous section. We assess variation in
fraud across the two ballot types, using election forensics and list experiment data to investigate
evidence of illicit activities taking place inside and outside polling stations. We subsequently
present analysis of biographical candidate data to assess variation in candidate types across the
ballots and the consequences of this variation.
Descriptive: Variation in Types of Fraud
Monitoring data provides initial evidence in the presence of and variation in forms of illicit elec-
toral behavior reported during the 2012 parliamentary elections. Table 3 presents data from the
monitoring projects managed by Maidan Monitoring and ElectUA where the location of the al-
leged violation could be identified.
The proportion of reports identifying campaign irregularities, bribery, and improprieties in the
vote count was similar for both organizations; these alleged violations constituted 60-65 percent
17 We developed several alternate conceptualizations of ‘bosses’ to explore how different categorizations by age, gen-der, and type of enterprise affected outcomes. In general, the models displayed in this paper are consistent with alter-natives.
21
of the reports. Improper activities by electoral management bodies made up 22 percent of Mai-
dan's data and around 10 percent of ElectUA's. Interference and intimidation accounted for 2
percent of Maidan's reports, but 7 percent of ElectUA's. Only ElectUA reported observer ob-
struction and voting as separate entries. While these reports may not fully represent the range
and scope of illicit activities that took place, they provide insight into variation. Some activities
that we characterize as bureaucratic fraud, involving electoral management bodies and other
officials, are evident alongside activities taking place outside of polling stations (e.g., bribery).
Confirming the reports of citizens, official observers found evidence of manipulation and fraud in
the election process. The Organization for Security and Cooperation in Europe's (OSCE) final
report noted ‘…cases of harassment, intimidation and abuse of administrative resources were
observed in a significant number of electoral districts, which negatively affected the ability of
candidates to get their messages to voters and to compete under equal conditions…’ While the
OSCE did not provide details about the frequency or spatial distribution of claims, the report
language suggests that campaign violations were widespread. On election day, OSCE observer
reports indicated that the ‘vote count was assessed negatively in 11 per cent of polling stations
observed.’ (OSCE 2013, 2). OSCE sampling of polling stations is not fully randomized, but the
organization strives to approximate random selection. If we treat the reports as generally repre-
sentative of the population of polling stations, then we could conclude that voter manipulation
was more prevalent than vote manipulation. This conclusion would be consistent with the in-
terpretation based on monitoring data.
Relying on reports from official and unofficial observers, we have established in this section that
various forms of fraud were likely to have been perpetrated during the parliamentary election
and that voter manipulation was likely more prevalent than vote manipulation.
Variation in Fraud Across PR and SMD
We are primarily interested in evaluating how different forms of fraud are associated with incen-
tives, particularly how fraud varies under different institutional rules (H1a and H1b). In this
section, we present tests of digit frequency – notably last digits – in the polling-station level
22
returns (Beber 2012). Deviation from the expected distribution of digits suggests human inter-
vention; in the most sinister interpretation, election fraud. Figure 2 illustrates the distribution of
data for the PR and SMD tiers separately.
[Insert Figure 2 about here]
Deviations from the clean-count distribution of digits are evident in both PR and SMD returns.
However, the scale of the deviation is greater in the PR data, indicated by the chi-squared sta-
tistic that is almost twice as large as for the SMD tier. Moreover, the number of zeroes reported
in the PR data significantly higher than expected, providing additional evidence for human ma-
nipulation of electoral returns.18 Digit forensics therefore suggests that there is more vote manip-
ulation in PR. Since poll workers in Ukraine are nominated by parties it is not surprising that
they would benefit their own parties at the counting stage (Sjoberg 2013a).
An additional forensics test evaluates turnout data (Myagkov et al. 2009). The underlying logic
is that the distributions should be similar across the two tiers. However, Figure 3 illustrates that
the PR tier shows a higher proportion of polling stations with high levels of turnout. The differ-
ences in the distribution between tiers could be due to undervotes; voters may not cast SMD
ballots as consistently as PR ballots because candidates are less well known than parties. While
this explanation would address differences in the mean, it does not fully explain the ‘hump’ on
the right hand tail that indicates a substantial proportion of high-turnout polling stations in PR.
[Insert Figure 3 about here]
The distribution of turnout suggests the presence of anomalies that may be associated with
fraud. However, the data do not provide insights into the causal mechanisms. That is, are these
anomalies associated with activities inside or outside the polling stations, with vote or voter
manipulation? The subsequent sections use experimental and observational data to address the-
se issues.
18 Given that humans have been shown to have a preference for lower digits (Boland and Hutchinson 2000).
23
List Experiment
Hypothesis H1b stipulates that we should find more evidence of vote buying in the SMD tier
than in PR. Using a list experiment design, we find that 9.5 percent (two-sided p-value 0.084) of
the respondents admit that vote buying had an impact on their vote choice in the Single-
Member District elections (see Table 4). In the party list elections, the equivalent number is a
non-significant 5.2 percent. Unfortunately for the purposes of hypothesis testing, the difference
of 4.3 percent is not in itself statistically significant.19 As we note in the data section, asking
voters to distinguish between their vote choices in the two tiers requires is challenging. The fact
that the results of the list experiment point in the hypothesized direction, i.e. more vote buying
in the SMD elections, adds to the pool of evidence we present throughout this paper.
[Insert Table 4 and Figure 4 about here]
In addition, if we divide up the survey data according to the geographical location of the re-
spondents, we find that vote buying in the SMD contests seems more prevalent in the West,
where 16.0 percent (p-value = .044) of respondents indicate that vote buying was a factor in
their vote choice in the SMD contest. The difference between vote buying in the West and the
East is close to the ten percent significance threshold in a two-sided test, p-value 0.1150. The
difference between the regions is especially interesting since fraud forensics suggested that there
is more bureaucratic fraud in the East and the Central parts of the country than in the West.20
Candidate Characteristics
The previous section provided some evidence that the types of fraud perpetrated seem to vary
across the two tiers, as anticipated. This section pursues the issue further, assessing how the
races in the two tiers are associated with different forms of fraud.
19 Two-tailed difference of proportions tests (z-test). 20 A last digit test for the vote counts from the East and the Center of the country indicates that there is vote-count fraud in these two regions, while not in the West.
24
The first step is to examine whether candidate characteristics co-vary with differences in institu-
tional rules. We cannot discriminate among candidates by their reputation for involvement in
improper activities, but biographical data permit us to account for candidates who self-identify
as business owners or directors (see the data section). As outlined in hypothesis H2a, we expect
the incentives of SMD rules to attract ‘boss’-like candidates.
[Insert Table 5 or Figure 5 about here]
Table 5 confirms these expectations. Looking across the row for SMD, we see that 54.8 percent
of the total sample of non-incumbent candidates take part in the SMD contest, but among
‘bosses’ this number if significantly higher, 58.3 percent. The difference is even larger if we only
examine the ruling party candidates, with 70.5 percent of the SMD candidates being bosses
compared to only 53.5 percent of non-bosses.21 Applying the same methodology to the most re-
cent preceding mixed system parliamentary elections, in 2002, we see that the pattern is the
same. 22 Boss type candidates prefer to run in SMD and not in PR. Interestingly, the proportion
of boss candidates went down dramatically in the two PR-only elections held in 2006 and 2007.
[Insert Table 6]
The pattern suggests that bosses were disincentivized to run in PR-only elections and that they
returned to competition after the re-introduction of the mixed electoral system in 2012. The
evidence fits the overall expectations presented in this paper.
SMD-Level Analysis
If bosses prefer SMD and the type of fraud used in SMD is more likely to be in the form of voter
manipulation, then we should expect increased evidence of vote buying in districts where more
boss candidates participate (H2b). As the dependent variable, we use crowdsourcing data aggre-
21 Table not included in presentation. 22 χ2 (1, N=1,724) = 10.070 p = 0.002 for the 2002 elections.
25
gated to the level of the election districts (N=225).23 We report results from the overall count of
illicit activities as well as counts of alleged vote buying and alleged campaign violations. We
further anticipate that campaign violations are more likely to be reported outside of a party's
areas of core support (H2c).
[Insert Figure 6]
For the analysis of vote buying, the primary independent variable of interest is the presence of
local ‘boss’ candidates (see Figure above).24 However, other explanatory factors may affect alle-
gations of vote buying. Similar to bosses, local incumbents may have access to political machines
that could perpetrate fraud. The variable is a count of current MPs campaigning in the district.
We also include the performance of the ruling Party of Regions in PR, aggregated to the SMD
district, as an explanatory variable. Higher levels of PR performance suggest that a larger core
of mobilized supporters is present in the district. The variable uses Party of Regions perfor-
mance because this party possesses the most extensive resources to commit fraud. We anticipate
that parties are more likely to engage in campaign violations outside of their core areas, where
illicit campaign tactics might be more effective in mobilizing voters to support them (or de-
mobilizing the opposition).
We control for a number of additional factors in the analysis, including turnout, non-party affil-
iated victory in SMD, and region. Turnout reflects the proportion of valid votes cast in the dis-
trict. High levels of turnout have often been associated with fraud, especially ballot box stuffing
(although voter suppression can influence turnout in the opposite direction). We anticipate that
higher levels of turnout will be associated with more reports of improper activities. We identify
districts where independent candidates were victorious as an additional control for a strong local
‘boss.’ Lastly, we control for region, using an eight-region definition (Barrington and Herron
2004). Reports of fraud have varied regionally, due in part to variation in local control.
23 We have no violation data reported from seven districts. 24 Note that the analysis here is restricted to only the SMD tier results.
26
[Insert Table 7 about here]
We used a negative binomial regression to assess the models because the dependent variables
reflect a count of reported violations. In the model of vote buying allegations, the number of
boss-type candidates is positively associated with allegations. Other features, independent victo-
ry and the eastern region, are also significant. The findings of both models are consistent with
our expectations. In the model of campaign violations, the performance of the Party of Regions
is negatively associated with allegations. In addition, two of the regional variables are signifi-
cant, suggesting that these violations have a spatial component.
[Insert Figure 7 about here]
Figure 7 further illustrates how the expected number of reported cases of vote buying varies
based on the number of boss candidates present in a district (H2b). To generate the figure, we
used the crowdsourcing data on campaign violations, setting the location as East Ukraine, turn-
out at its mean, and an independent winner in the SMD race. The figure shows how the predict-
ed number of vote buying allegations varies with the number of boss candidates, and 99 percent
confidence intervals. The predicted number of vote buying allegations increases along with the
number of boss candidates, supporting our expectations.
27
Conclusions
We investigated how the incentives provided by institutional rules are associated with variation
in the forms of electoral manipulation perpetrated during a national election. Our paper was
motivated by the observation that political actors in authoritarian and semi-authoritarian re-
gimes of the post-Soviet region have re-adopted mixed electoral systems after experimenting
with nationwide proportional representation, most recently Russia and Ukraine. We speculated
that the flexibility provided by mixed systems would be conducive to certain types of electoral
manipulation, suggesting that institutional designers may be motivated not only by the ad-
vantages presented by election rules in fair competition, but by the opportunities election rules
provide for illicit behavior.
We focused on the case of Ukraine, a semi-authoritarian regime that recently returned to a
mixed system after holding two elections under national-level proportional representation.
Ukraine's elections have been closely scrutinized by international observers, and its leadership
has demonstrated concerns about international perceptions of its practices. But, pro-regime par-
tisans also have strong incentives to retain power, especially because they lost the presidency to
the opposition, albeit briefly. Mixed electoral systems provide the flexibility and incentives to
perpetrate different types of manipulation that are likely to vary based on ballot type.
The paper makes two main contributions to the study of election quality and political accounta-
bility. First, it advances our theoretical understanding of the micro-level foundations of illicit
behaviors. Just as political institutions create incentives for political actors to pursue different
types of legitimate strategies and tactics to win elections, they also create incentives for different
types of illicit behaviors. We divided the illicit behaviors into those that rely on vote manipula-
tion and those that rely on voter manipulation, arguing that PR election rules are especially
conducive to the former and constituency rules are especially conducive to the latter. Election
forensics revealed stronger evidence of vote manipulation in PR, consistent with our expecta-
tions. The list experiment suggested elevated reports of vote buying in SMD, although we could
not fully substantiate the differences among tiers. Candidate-level analysis shows that boss-type
28
candidates prefer the SMD tier over party lists. Finally, the district-level analysis demonstrated
that more "boss type" candidates are associated with more reports of vote buying, another piece
of evidence consistent with our expectations. In sum, the evidence suggests that on-the-ground
tactics vary, and that institutional rule variation is associated with tactical considerations.
Second, the paper uniquely combines traditional sources of analysis (election returns) with new
sources of data, notably crowdsourcing reports and a list experiment. The study of forensics,
including election forensics, does not rely on a single test to assess data quality. Perpetrators of
fraud attempt to hide their work, and various tools of fraud may be more evident with different
analytical tools and data sources. Most importantly crowdsourcing data and innovative survey
approaches allows us to tap into the experiences of voters, and not only focus on what happens
inside polling stations. While the tests in this analysis are not individually conclusive, combined
they provide evidence that institutional rules affect fraud tactics.
While preliminary, these findings speak to several questions in the scholarly literature and
among democracy-promotion practitioners. They suggest that research on election quality
should adopt a more nuanced approach to fraud and manipulation; actors respond to incentives
when selecting the optimal tools to use. This observation is also important for practitioners as
efforts to mitigate fraud must be responsive to the incentives inherent in the rules. Further, they
suggest that literature on institutional design must address the potential effects of both legal
and illegal behaviors as designers may intend to benefit from the illicit incentives that systems
provide rather than playing fair.
29
Tables and Figures
Table 1: Self-reported Parliamentary Candidate Characteristics, November 2012
Boss-like Attribute Freq. % all candidates
Incumbent MP (occupation) 345 6.4 % Director (occupation) 1,273 23.5 % President (occupation) 96 1.8 % Nachal’nik (occupation) 316 5.8 % Golova (occupation) 394 7.3 % Company (employment) 183 3.4% LTD (employment) 1069 19.7% Enterprise (employment) 305 5.6% Subsidiary (employment) 186 3.4% Empty (employment) 2,399 44.2%
* Note: Separate dummies for each boss attribute. A candidate can have a positive value on multiple boss attributes.
30
Table 2: List Experiment Sample Description, November 2012
Form Education Male Age Voted Western re-gion
A 5.5 37.2 % 49.9 77.7 % 51.8 % B 5.6 36.6 % 48.9 73.8 % 51.8 % C 5.7 39.6 % 48.7 73.7 % 51.8 % Total 5.6 37.8 % 49.2 75.1 % 51.8 %
* Note: The education scale ranges from '1' primary education (complete or not) to '8' completed higher education.
31
Figure 1: Incident Report Location, Ukraine 2012
32
Table 3: Monitoring Reports of Alleged Violations
Maidan ElectUA Malpractice category Freq % Freq % Administrative 348 22.28 172 9.98 Bribing 235 15.04 187 10.85 Campaigning 816 52.24 867 50.29 Counting 14 0.9 19 1.1 Interference 1 0.06 109 6.32 Intimidation 34 2.18 20 1.16 Observer Obstruction 18 1.04 Voting 107 6.21 Other 114 7.3 225 13.05
Total categorized 1,562 1,724 Not categorized 81 Total reports 1,643 1,724
Note: Some reports are not categorized by the organizers and thus not included in our data.
33
Figure 2: Last Digit Forensics Tests, Parliamentary Elections Ukraine 2012
* Note: The horizontal line at 10 percent indicates the expected uniform distribution under conditions of a clean vote count and the capped spikes indicate the 95-percent confidence interval for each individual digit (point-wise). Data: all three digit vote-counts for all parties.
9
9
910
10
1011
11
11Percent
Perc
ent
Percent0
0
01
1
12
2
23
3
34
4
45
5
56
6
67
7
78
8
89
9
9Digits
Digits
DigitsN=62,788, x2=41.59, p=0.000
N=62,788, x2=41.59, p=0.000
N=62,788, x2=41.59, p=0.000PR
PR
PR9
9
910
10
1011
11
11Percent
Perc
ent
Percent0
0
01
1
12
2
23
3
34
4
45
5
56
6
67
7
78
8
89
9
9Digits
Digits
DigitsN=56,529, x2=22.46, p=.008
N=56,529, x2=22.46, p=.008
N=56,529, x2=22.46, p=.008SMD
SMD
SMD
34
Figure 3. Turnout Distribution Anomalies, Parliamentary Elections Ukraine 2012
0
0
01
1
12
2
23
3
34
4
4Density
Dens
ity
Density0
0
0.2
.2
.2.4
.4
.4.6
.6
.6.8
.8
.81
1
1turnout
turnout
turnoutTurnout PR
Turnout PR
Turnout PRTurnout SMD
Turnout SMD
Turnout SMDkernel = epanechnikov, bandwidth = 0.0131
kernel = epanechnikov, bandwidth = 0.0131
kernel = epanechnikov, bandwidth = 0.0131Kernel density estimateKernel density estimate
Kernel density estimate
35
Table 4. Vote Buying and Coercion by Electoral System, List Experiment Survey
Party List SMD Difference Mean N/s.e. Mean N/s.e.
Control 1.909 507 1.81 499 Treatment - Money 1.948 483 1.89 480 Estimated % Money 5.17% 6.72% 9.46% 5.26% 4.29% Treatment - Coercion 1.949 491 1.88 486 Estimated % Coerced 2.68% 5.72% 4.63% 4.11% 1.95% N 1,481 1,465
* Note: List-experiment control and treatment values are the mean number of items identified by re-spondents. The numbers of subjects in each condition are in the second column. Linearized standard er-rors adjusted for the survey design.
36
Figure 4. Difference in Vote Buying Reports, List Experiment Survey
PR vote buying
PR vote buying
PR vote buyingSMD vote buying
SMD vote buying
SMD vote buyingParameter name
Para
met
er n
ame
Parameter name-.1
-.1
-.10
0
0.1
.1
.1.2
.2
.2Parameter estimate
Parameter estimate
Parameter estimate95% Conf. Interval
95% Conf. Interval
95% Conf. Interval
37
Table 5. Cross-Tabulation of ‘Boss’ Dummy and Electoral System Tier, Ukraine 2012
Non-Boss Boss Total PR 1,486 965 2,451
47.8 % 41.7 % 45.2 % SMD 1,626 1,350 2,976
52.3 % 58.3 % 54.8 % Total 3,112 2,315 5,427
100.0 % 100.0 % 100.0 % * Note: Pearson chi2(1) = 19.7231 Pr = 0.000
Figure 5. Boss Candidates by Electoral System Tier, Ukraine 2012
1,626
1,626
1,6261,350
1,350
1,3501,350
1,350
1,3501,486
1,486
1,486965
965
965965
965
965SMD
SMD
SMDPR
PR
PRBy E-System Tier
By E
-Sys
tem
Tie
r
By E-System Tier0
0
01
1
1By 'Boss' Dummy
By 'Boss' Dummy
By 'Boss' DummyN=5,427, x2=19.72, p=0.000
N=5,427, x2=19.72, p=0.000
N=5,427, x2=19.72, p=0.000
38
Table 6. Bosses and Non-Bosses in Parliamentary Elections in Ukraine 2002-2012
2002 2006 2007 2012 No-Boss (n) 2,539 4,075 2,001 2,751
Boss (n) 2,643 4,028 1,719 3,021 No-Boss (%) 49.0% 50.3% 53.8% 47.7% Boss (%) 51.0% 49.7% 46.2% 52.3%
* Note: The exact same methodology is used across all four elections. Publicly available self-reported data is available from the CEC.
39
Table 7. SMD-Level Results of Negative Binomial Regression on Monitoring Data
Campaign Vote Buying PoR PR -0.023**
(0.011) -0.964
(1.450) # Bosses 0.015
(0.034) 0.110**
(0.045) MP 0.028
(0.072) 0.044
(0.097) Turnout 0.709
(1.554) -1.244
(1.862) Ind. Win 0.073
(0.165) 0.572**
(0.208) East 0.747
(0.508) 0.753
(0.659) Eastcentral -0.304
(0.327) -0.346
(0.434) Crimea 0.773
(0.500) -0.207 (0.672)
South 1.041** (0.363)
0.125 (0.453)
Westcentral 0.242 (0.215)
-0.022 (0.285)
West 0.168 (0.294)
-0.132 (0.410)
Southwest 0.172 (0.329)
0.003 (0.476)
Constant 1.711* (0.921)
0.891 (1.138)
N 218 218 LRX2 46.86** 26.62**
Note: Standard errors in parentheses. North-central is the excluded category for regions. ** significant at the .05 level or above. * significant at the .10 level.
40
Figure 6. Crowdsourced Vote Buying Reports per SMD with ‘Boss’ Candidate Count Visual-
ized, Ukraine 2012
* Note: These are only the reports from the unbounded crowd.
41
Figure 7. Predicted Number of Vote Buying Allegations, SMD-Level Analysis
42
References
Alvarez, R Michael, Lonna Rae Atkeson, and Thad E Hall. 2012. Evaluating Elections: A Handbook of Methods and Standards: Cambridge University Press.
Alvarez, RM, TE Hall, and SD Hyde. 2008. Election fraud: detecting and deterring electoral manipulation: Brookings Inst Pr.
Beber, Bernd and Alexandra Scacco. 2012. "What the Numbers Say: A Digit-Based Test for Election Fraud." Political Analysis 20 (2):211-34.
Benoit, Kenneth. 2007. "Electoral laws as political consequences: explaining the origins and change of electoral institutions." Political Science 10 (1):363.
Birch, Sarah. 2011. Electoral malpractice. Oxford ; New York: Oxford University Press. Boland, P.J., and K. Hutchinson. 2000. "Student selection of random digits." Journal of the
Royal Statistical Society: Series D (The Statistician) 49 (4):519-29. Clark, William R. and Matt Golder. 2006. "Rehabilitating Duverger’s Theory: Testing the
Mechanical and Strategic Modifying Effects of Electoral Laws." Comparative Political Studies 39 (6):679-708.
Colomer, Josep. 2005. "It’s the Parties that Choose Electoral Systems (or Duverger’s Laws Upside Down)." Political Studies 53 (1):1-21.
Duverger, M. 1959. Political Parties. Translated by R. N. a. B. North: New York: John Wiley. Elklit, Jorgen and Andrew Reynolds. 2005. "A Framework for the Systematic Study of Election
Quality." Democratization 12 (2):147-62. Fearon, J.D. 2011. "Self-enforcing democracy." The Quarterly Journal of Economics 126
(4):1661-708. Fukumoto, K., and Y. Horiuchi. 2011. "Making Outsiders' Votes Count: Detecting Electoral
Fraud Through a Natural Experiment." American Political Science Review 105 (03):586-603.
Gonzalez-Ocantos, Ezequiel, Chad Kiewiet de Jonge, Carlos Melendez, Javier Osorio, and David W Nickerson. 2012. "Vote buying and social desirability bias: Experimental evidence from Nicaragua." American journal of political science 56 (1):202-17.
Herron, Erik. 2010. "The effect of passive observation methods on Azerbaijan's 2008 presidential election and 2009 referendum." Electoral Studies.
Herron, Erik, and Nazar Boyko. 2012. "Reversing Democratic Revolutions: The Implications of Ukraine's 2010 Local Elections." Journal of East European and Asian Studies 3 (1):79-100.
Herron, Erik S, and Paul E Johnson. 2007. "Fraud before the 'revolution': special precincts in Ukraine's 2002 parliamentary election." In Aspects of the Orange Revolution III: The Context and Dynamics of the 2004 Ukrainian Presidential Elections, ed. I. Bredies, A. Umland and V. Yakushik.
43
Herron, Erik S. 2007. "State Institutions, Political Context and Parliamentary Election Legislation in Ukraine, 2000–2006." Journal of Communist Studies and Transition Politics 23 (1):57-76.
———. 2011. "Measuring Dissent in Electoral Authoritarian Societies: Lessons From Azerbaijan’s 2008 Presidential Election and 2009 Referendum." Comparative Political Studies 44 (11):1557-83.
Herron, Michael C. and Jasjeet S. Sekhon. 2003. "Overvoting and Representation: An Examination of Overvoted Presidential Ballots in Broward and Miami-Dade Counties." Electoral Studies 22 (1):21-47.
Hood, MV, and William Gillespie. 2012. "They Just Do Not Vote Like They Used To: A Methodology to Empirically Assess Election Fraud*." Social Science Quarterly 93 (1):76-94.
Howe, Jeff. 2006. "The rise of crowdsourcing." Wired magazine 14 (6):1-4. Hyde, S. D. 2007. "The observer effect in international politics: Evidence from a natural
experiment." World Politics 60 (1):37. Hyde, S.D. 2011. The Pseudo-Democrat's Dilemma: Why Election Monitoring Became an
International Norm: Cornell University Press. Ichino, Nahomi and Matthias Schundeln. 2012. "Deterring or Displacing Electoral Irregularities?
Spillover Effects of Observers in a Randomized Field Experiment in Ghana." Journal of Politics 74 (1):292-307.
Kelley, J.G. 2012. Monitoring Democracy: When International Election Observation Works, and Why it Often Fails: Princeton University Press.
Kitschelt, Herbert, and Steven Wilkinson. 2007. Patrons, clients, and policies: patterns of democratic accountability and political competition. New York: Cambridge University Press.
Klimek, Peter, Yuri Yegorov, Rudolf Hanel, and Stefan Thurner. 2012. "Statistical Detection of Systematic Election Irregularities." Proceedings of the National Academy of Sciences 109 (41):16469-73.
Knack, Stephen and Martha Kropf. 2003. "Voided Ballots in the 1996 Presidential Election: A County-Level Analysis." Journal of Politics 65 (3):881-97.
Lehoucq, Fabrice. 2003. "Electoral Fraud: Causes, Types, and Consequences." Annual Review of Political Science 6:233-56.
Little, Andrew. 2012. "Elections, Fraud, and Election Monitoring in the Shadow of Revolution." Quarterly Journal of Political Science 7:249-83.
Magaloni, B. 2006. Voting for autocracy: Hegemonic party survival and its demise in Mexico. New York: Cambridge University Press.
Mebane, Walter. 2006. "Election Forensics: The Second-digit Benford’s Law Test and Recent American Presidential Elections." Election Fraud Conference.
Müller, Wolfgang C., and Kaare Strom. 1999. Policy, office, or votes? : how political parties in Western Europe make hard decisions. Cambridge England ; New York: Cambridge University Press.
44
Myagkov, M.G., P.C. Ordeshook, and D. Shakin. 2009. The Forensics of Election Fraud: Russia and Ukraine. Vol. 100: Cambridge University Press.
Norris, Pippa. 2012. Making Democratic Governance Work: How Regimes Shape Prosperity, Welfare, and Peace: Cambridge University Press.
Persson, T., G. Tabellini, and F. Trebbi. 2003. "Electoral rules and corruption." Journal of the European Economic Association 1 (4):958-89.
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):753-66.
Schedler, A. 2002. "The menu of manipulation." Journal of Democracy 13 (2):36-50. Simpser, Alberto. 2012. More than Winning: Why Governments and Parties Manipulate
Elections. Simpser, Alberto, and Daniela Donno. 2012. "Can International Election Monitoring Harm
Governance?" The Journal of Politics. Sjoberg, F M. 2013a. "Political Parties and Election Fraud." New York: Columbia University. Sjoberg, F.M. 2012. "Making Voters Count: Evidence from Field Experiments about the
Efficacy of Domestic Election Observation." Harriman Institute Working Paper Series 1 (1).
———. 2013b. "Autocratic Adaptation: The Strategic Use of Transparency and The Persistence of Election Fraud." Electoral Studies (forthcoming).
Sobyanin, Alexander, and V. G. Sukhovolskiy. 1995. "Demokratiya, Ogranichennaya Falsifikatsiyami: Vybory I referendumy v Rossii v 1991–1993." Moscow: Planning Group for Human Rights.
Stokes, S.C., T. Dunning, M. Nazareno, and V. Brusco. 2012. "Brokers, voters, and clientelism." Unpublished manuscript, Yale University and Universidad Nacional de Córdoba.
Taagepera, Rein, and Matthew Soberg Shugart. 1989. Seats and votes: the effects and determinants of electoral systems. New Haven: Yale University Press.
Tucker, JA. 2007. "Enough! electoral fraud, collective action problems, and post-communist colored revolutions." Perspectives on Politics 5 (03):535-51.
Vickery, Chad and Erica Shein. 2012. "Assessing Electoral Fraud in New Democracies: Refining the Vocabulary." Washington DC: IFES Working Paper.
Wand, Jonathan N., Kenneth W. Shotts, Jasjeet S. Sekhon, Walter R. Mebane, Jr., Michael C. Herron, and Henry E. Brady. 2001. "The Butterfly Did It: The Aberrant Vote for Buchanan in Palm Beach County, Florida." American Political Science Review 95 (4):793-810.