Election Ink and Turnout in a Fragile Democracy
Karen E. Ferree
University of California, San Diego
[email protected]
Robert Dowd
University of Notre Dame
[email protected]
Danielle F. Jung
Emory University
[email protected]
Clark C. Gibson
University of California, San Diego
[email protected]
December 5, 2015
Abstract
Inking voters’ fingers after they cast their ballots is standard practice in elections
in emerging democracies. Because of unreliable voting records and weak administrative capacity, electoral procedures rely on inking to prevent double voting. While
the putative e↵ect of inking is to decrease electoral fraud, it may unintentionally facilitate turnout manipulation strategies. We explore the e↵ects of inking on turnout
through a GOTV experiment we conducted in Kampala during the February 2011
Ugandan general election. Our results suggest that the e↵ects of practices like inking that publicize turnout decisions are not always benign. In Kampala — an area
of mixed political allegiance within a coercive semi-authoritarian regime — some
potential voters responded to our relatively mild informational treatment about the
use of ink by deciding not to vote.
Acknowledgements: We gratefully acknowledge Barak Ho↵man, Craig McIntosh,
and especially James Long, who contributed in essential ways to design and/or
implementation stages of this project. We also thank David Nickerson, James
Fowler, Zoli Hajnal, Eddy Malesky, Jesse Driscoll, Megumi Naoi, Claire Adida,
Don Green, Guy Grossman, William Evans, Scott Mainwaring, Guillermo Trejo,
Jaimie Bleck, Sarah Daly, Sean McGraw, Anke Hoe✏er and the participants of
seminars at University of Notre Dame, Stanford University, Oxford University, and
Yale University for their comments. Doreen Lwanga provided expert advice in the
field. We are grateful for the excellent research assistance of Dennis Inbok Rhee,
Drew Wagsta↵, and Kai Ostwald. This research was funded by the Kellogg Institute
for International Studies at the University of Notre Dame.
Introduction
Inking voters fingers after they cast their ballots is standard practice in emerging
democracies. From Afghanistan to Zimbabwe, an inked finger symbolizes a citizen’s participation in the electoral process. Ink o↵ers a relatively low cost, easy to
implement solution to the challenging problem of double voting in countries with
unreliable voting records and weak administrative capacity. However, while the
putative goal of inking is to decrease electoral fraud, we argue that it may unintentionally undermine free and fair elections by enabling politicians and other actors
to manipulate turnout.
Inking makes the decision to vote visible. An inked finger identifies a voter
not just when she emerges from the polling station, but until the ink on her finger
wears o↵, a process that can take days or weeks. During this time, her decision
to vote is public: easily seen by family and friends — as well as politicians, party
supporters, security personnel, and community and religious leaders. By rendering
the decision to vote visible, inking enables politicians and other actors to condition
rewards and punishments on the presence of an inked (or un-inked) finger, thereby
facilitating turnout manipulation strategies. Depending on the electoral context,
such strategies may aim to either encourage or suppress turnout.
Evaluating the relationship between electoral ink and turnout presents challenges. Depending on the mixture of electoral strategies in a particular election,
the e↵ects of ink vary, increasing turnout in some instances and depressing it in
others. Studies ignoring this heterogeneity may underestimate the impact of inking
on outcomes. The virtually universal adoption of ink further complicates evaluation
e↵orts. Not only is there little variation in observational data, but randomization of
the inking practice itself is infeasible. Finally, few credible measures of turnout exist
in emerging democracies. Standard practice involves asking people if they voted, an
approach that produces significant distortions if people misrepresent their behavior.
To address these challenges, we pursue a novel strategy to evaluate the relationship between ink and turnout. In a Get Out the Vote (GOTV) experiment
conducted in Uganda’s capital city Kampala during the country’s February 2011
national elections, we manipulated information about the use of ink in the elec-
1
tion. We randomly assigned registered voters to one of three groups: a control group
given a survey about mobile phone use; a treatment group that received the mobile
phone survey plus a reminder to vote; and a treatment group that received the mobile phone survey, the reminder to vote, and an informational message explaining
that ink would be used in the election allowing “everyone to know” whether or not a
person had voted. Ugandan enumerators delivered these messages a few days prior
to the election to registered voters; they returned to inspect fingers for ink during
the days immediately after the election. We find the informational treatment about
inking reduced the probability of having an inked finger, especially for younger and
less educated voters — precisely the set of voters least likely to be familiar with
inking practices and therefore most likely to respond to an informational treatment
about them.
These results indicate the plausibility of a link between the use of ink and
voter turnout. They suggest that the e↵ect of practices like inking that publicize
turnout may not be benign, in contrast to experiments in the United States that
routinely find increasing turnout’s visibility raises participation (Gerber, Green
and Larimer, 2008; Davenport et al., 2010; Mann, 2010; Sinclair, 2012). They also
reveal that practices adopted to improve the quality of elections in one domain may
have unintended consequences in others. We thus echo recent findings by Driscoll
and Hidalgo (2014) suggesting that relatively innocuous interventions aimed at
promoting electoral participation may backfire in some contexts. Studies of turnout
behavior in emerging democracies might benefit from these insights, specifically
examining institutional features of elections in addition to individual level factors
like age, education, and partisanship emphasized in most extant work (Bratton,
1999; Bratton, Mattes and Gyimah-Boadi, 2005; Bratton, 2008; Isaksson, 2010;
Kuenzi and Lambright, 2005, 2007). Only by placing voters in a specific political
context can we make sense of their behavior.
2
Inking as a Standard Electoral Practice of Emerging Democracies
The first use of election ink dates to the early 1980s, when election officials began
marking voters with ink detectable only with ultraviolet light. However, difficulties
with administration (in particular, the need to equip polling stations with the
special lights) promoted the adoption of visible, indelible ink as the standard by the
mid-1980s (Cody, 1985; Meislin, 1982; Taubman, 1982; Wattenberg, 1986). Inking
has since become an international norm in election administration. Virtually all
emerging democracies use it to identify voters who have finished the process of
casting a ballot.
The attractions of the practice are straightforward. Establishing accurate voter
registries is costly, difficult, and often controversial (Organization for Security and
Cooperation in Europe (OSCE), 2010). Would-be voters in newer and poorer
democracies frequently lack valid forms of identification to use for voter registration or voting. Given these limitations, multiple voting presents a serious threat to
election quality. If election administrators attempt to eliminate multiple voting by
placing stringent criteria on registering and voting, however, they risk disenfranchising large (typically poorer and less educated) parts of the population. Ink o↵ers
a low-technology, relatively inexpensive solution. Polling station sta↵ have only to
mark an individual’s finger (or other part of the hand) with indelible ink after they
vote. The ink can last from days to weeks, preventing voters from repeat voting.
It can be delivered to polling stations along with election materials, requires no
electricity to operate, and does not involve extensive training.
Although used pervasively, controversies periodically surface regarding ink use.
Several recent Malaysian elections, for example, have seen extensive discussions
of inking in the popular press as well as policy reversals by the government. In
response to public concern about multiple voting, the government purchased nearly
50,000 bottles of indelible ink from India for use in its 2008 General Election.
Yet just days before the election, the Election Commission reversed its decision
and abandoned the ink, citing claims that the ink might be used illicitly to mark
3
unsuspecting citizens prior to election day, preventing them from casting votes.
It later emerged that the order to abandon the ink plans came from the Prime
Minister’s cabinet, despite the Commission’s nominal independence (Weiss, 2009).
Malaysia’s ink woes continued in its 2013 General Election, when the “indelible”
ink proved to be removable with soap and water (Ostwald, 2013; Welsh, 2013). In
nearby Philippines, political operatives gave rewards to potential opposition voters
who allowed them to ink their fingers and so disqualify them from voting (Schedler,
2002).
Ink controversies have also dogged elections in Afghanistan. In the 2004 Afghan
elections, sta↵ at some voting stations apparently used non-permanent ink instead
of indelible ink, leading to extensive allegations of ink removal and double-voting
(Raman, 2004; Baldauf, 2004). During later elections in Afghanistan, inking controversies continued, with questions about the quality of the ink and inconsistent
application (Democracy International, 2010).
In spite of such controversies, inking is universally considered a fair and e↵ective
means to prevent double-voting. We could not find any systematic studies of inking
in either the policy or academic literature.
Inking and Turnout
In this section, we flesh out the logic underlying our intuition that inking may have
unintended consequences for voting behavior. This logic works through the causal
mechanism of turnout manipulation strategies, which can be initiated by a variety
of actors. As the goals of turnout manipulation strategies vary — some seek to
encourage turnout, others to suppress it — the e↵ects of inking on voting behavior likely vary with local context. Where turnout encouragement is the dominant
strategy in an election, inking should increase turnout; where discouragement dominates, inking likely has the opposite e↵ect. Hence, we hypothesize a relationship
between inking and turnout, but one that is context-specific and depends on the
nature of electoral strategies in a particular election.
Electoral ink prevents double-voting by marking voters in a way that is visible
4
to polling station officials and difficult to remove. In the process, ink also makes
visible the decision to vote outside of the polling station, typically for days or even
weeks. During this time, a range of actors can observe whether an individual has a
marked or unmarked finger and from this make inferences about his or her decision
to vote. A husband can check his wife for an inked finger. A boss can examine his
workers. A religious leader can observe congregants at religious services. A local
party leader can check the fingers of the voters in his precinct. Soldiers or police
can examine people passing through at a checkpoint. In short, ink substantially
reduces the cost and raises the e↵ectiveness of turnout monitoring for a wide variety
of actors.
By facilitating turnout monitoring, ink makes it easier to condition rewards or
punishments on the presence or absence of ink, thus enabling strategies of turnout
manipulation. Rewards and punishments for voting can assume many forms, from
the subtle (a raised eyebrow, a verbal rebuke, shunning) to the direct (cash payments, physical injury, loss of employment, destruction of property). They range
from legal strategies like providing voters with “I voted” stickers to illegal strategies involving illicit transfers of money or gifts, and they may or may not involve
coercion and threats of violence.1
External actors may have a range of di↵erent motivations when they attempt
to manipulate turnout. We propose a typology to capture these variations in Table 1, which di↵erentiates four strategies based on whether they aim to increase
or decrease turnout and whether their goals are narrowly electoral or focused on
enforcing social norms about political participation.
An electoral encouragement strategy tries to increase the turnout of a particular voter or group of voters for the purposes of winning an election. Parties
frequently engage in electoral encouragement strategies vis-à-vis their own known
partisans, especially in close elections, as doing so enhances their chances of win1
In keeping with prominent work on clientelism like Stokes (2005) and Nichter (2008), our perspective
assumes that political strategies focus on rewarding and punishing individuals rather than collectives like
villages or polling stations. This is a reasonable assumption in areas, like the one in this study, where
mixed political allegiances make group targeting inefficient. In politically homogeneous areas where
targeting focuses on collectives rather than individuals, ink may play a less significant role.
5
Table 1: Turnout Manipulation Strategies
Motivation of External Actor
Enforcement of Norms
Electoral Result
Lower Turnout
Normative Discouragement
Electoral Discouragement
Higher Turnout
Normative Encouragement
Electoral Encouragement
Desired Outcome
ning. The Obama campaigns of 2008 and 2012 in the United States, for example,
made targeted attempts to raise turnout of African American voters, students, and
other groups strongly predisposed to vote Democrat. (Nichter, 2008) refers to such
strategies as “turnout buying” when they involve an actual exchange of benefits for
turning out to vote. Parties may also engage in electoral discouragement strategies. Cox and Kousser (1981) document examples of parties paying partisans of the
opposition to stay home in late twentieth century New York State elections. The
Democratic Party used a variety of coercive methods to discourage African American voting in the pre-Voting Rights South (Kousser, 1974). Collier and Vicente
(2008) and Bratton (2008) discuss e↵orts by Nigerian parties in the 2007 election
to discourage turnout of opposition voters through intimidation.
The goals of actors in manipulating turnout may not always be narrowly electoral. In some contexts, the intention may instead be to enforce community norms
about political participation.2 During widespread election boycotts, for example,
community actors may seek to suppress turnout to deny legitimacy to a ruling
party or government (Beaulieu, 2014). Community, religious, and even family actors may seek to suppress female participation where such voting is seen as violating
social norms, a dynamic at play in recent Afghan elections.3 We classify these as
normative discouragement strategies. Alternatively, where community norms value
participation, normative encouragement strategies may predominate. Campaigns
emphasizing voting as a central duty of all citizens are an example of this type of
strategy and are common in the elections of many countries.
2
See citation redacted nd on Social Sanctioning and Turnout
For a recent example, see US Election Observation Mission in Afghanistan, Presidential and Provincial Council Elections 2009 Final Report, pp. 30–31.
3
6
Anecdotal evidence suggests inking plays a role in both turnout discouragement
and encouragement strategies. During the 2009 and 2010 Afghan elections, the Taliban threatened anyone found with an inked finger with its amputation (or worse);
in some instances, they carried out their grisly threat. Election observers noted
these intimidation e↵orts kept many voters at home, while other voters begged
polling station sta↵ not to ink their fingers after voting (Democracy International,
2010, 2011). Similar reports come from the 2002 Indian elections in violent Kashmir, where one journalist noted that ink which is a “nuisance” in most parts of
India was “virtually a passport to death in Kashmir” because militants threatened
retribution to voters found with inked fingers (Jha, 2002). Militias supporting President Mugabe used ink to encourage turnout in the second round of Zimbabwe’s
2008 presidential election by going door-to-door in Harare townships and promising “the most ferocious retribution” for anyone lacking an inked finger (BBC News,
2008).
GOTV experiments conducted in the United States provide evidence consistent with normative encouragement strategies. Gerber, Green and Larimer (2008),
for example, find that letters publicizing past turnout decisions of neighbors and
promising to reveal future ones increased the probability of voting in a Michigan primary election by 8 percentage points. They speculate that citizens did not want to
be seen as violating the community norm of civic participation. Additional GOTV
experiments generally reinforce these findings (Davenport et al., 2010; Mann, 2010;
Sinclair, 2012) and compliment research across social science disciplines demonstrating that making behavior public can alter it (e.g., Ariely and Meier (2009);
Becker (1968); Becker and Stigler (1974); Cialdini and Goldstien (2004); Cialdini
and Trost (1998); Habyarimana et al. (2007, 2009); Posner and Rasmusen (1999);
Rind and Benjamin (1994); Schultz (1999); Schultz et al. (2007); Whatley et al.
(1999)). Election ink reveals turnout decisions just as surely as Gerber, Green,
and Larimer’s experimental letter treatment, but in a far broader and more public
fashion. If letters influence turnout decisions, it seems plausible ink could do the
same.
In sum, we argue that the use of ink in elections may lead to changes in voter
7
turnout by facilitating turnout manipulation strategies. Turnout manipulation may
attempt to raise or lower turnout and may have normative or strategic motivations.
Inking lowers the cost of all such strategies by rendering the voting decision public,
making it easier for outside actors to incentivize desired voting behaviors. Anecdotal evidence, as well as results from American GOTV experiments, supports these
claims.
Evaluating Inking
Rigorous empirical evaluation of the practice of electoral inking presents challenges.
First, as suggested in the previous section, inking likely has context specific e↵ects
that depend on the mixture of turnout manipulation strategies present in a particular place. Because such strategies are often illicit, identifying and controlling
for them directly may be impossible. On the other hand, ignoring this heterogeneity may wash out treatment e↵ects. Second, the practice of inking voters has been
adopted so universally that variation in treatment status is rare. Although electoral
bodies may fail to ink voters at times because of political controversy or insufficient
ink, these instances are almost certainly not random, leading to standard problems
with identifying treatment e↵ects. Nearly universal adoption of inking and the importance placed on the practice by election officials would also seem to rule out
direct randomization of inking itself.
Measuring turnout generates its own set of problems. Official voting records
detailing individual turnout (as in the US) rarely exist in developing countries.
Moreover, social norms strongly favor voting in many places, leading to potential
social desirability bias in self-reports of voting. Bratton, Mattes and GyimahBoadi (2005, pp. 145 - 46) show, for example, that self-reported turnout levels for
respondents in the Afrobarometer are, on average, 13 percentage points higher than
official turnout levels.
We circumvent these challenges through a unique experiment we conducted during the February 2011 Ugandan election in Kawempe South, a constituency located
in central Kampala. In the experiment, we randomly exposed study participants
8
to information about the use of electoral ink in the election as part of a face-to-face
survey delivered just before the election. We then evaluated the e↵ect of this intervention on turnout by returning a few days after the election to check fingers for
telltale election ink (see Giné and Mansuri (N.d.) and Aker, Collier and Vicente
(2013) for similar measurement strategies). Initially conceived of as an extension of
American GOTV experiments to a new and di↵erent setting, the experiment also
provided a unique opportunity to evaluate the practice of electoral inking.
Our design had several advantages. By restricting the experiment to a single
constituency, we reduced the problem of strategy heterogeneity, albeit at the cost of
lowering external validity. By measuring turnout through inked fingers, we avoided
the problem of response bias. And by pursuing an experimental design, we mitigated standard problems with observational data. However, one drawback to the
design is that the informational ink treatment is weaker than a treatment directly
manipulating the inking practice itself. Informational treatments a↵ect only those
participants who previously lacked information or awareness about the practice.
Hence our results will likely understate the true e↵ects of inking. Nonetheless,
given the challenges of non-experimental approaches and the difficulty of manipulating the use of ink directly, the ink treatment o↵ers an alternative, albeit indirect,
way to investigate the e↵ects of inking on turnout.
In the following sections, we first discuss the setting of the experiment; we then
provide details on the experimental treatments, sampling strategy, and randomization protocol.
Setting
We conducted the experiment in Kawempe South, a constituency in the heart of
Kampala, the capital city of Uganda. Kawempe South is a large, socially, economically, and ethnically diverse area. Built over a flood plain and settled by waves
of migrants, it contains dense neighborhoods of closely spaced small one or two
room structures, many informal, with some larger homes interspersed. Markets,
schools, churches, mosques, clinics, farms, and small businesses coexist with residences. Small streets and paths meander and branch o↵ main roads clogged with
9
buses, motorcycles, pedestrians, cars, and animals. Kawempe is also home to Makerere University, Uganda’s premiere post-secondary educational institution. It is
neither particularly poor nor rich relative to other urban areas of Uganda.
Kawempe has a history of oppositional politics, di↵erentiating it from most of
Uganda. The ruling National Resistance Movement (NRM) has dominated Ugandan politics since the mid 1980s when its leader, Yoweri Museveni, seized power
during a civil war. After long resisting pressures to democratize, Museveni removed
a ban on multiparty politics in 2005. He won the presidency in the country’s first
multiparty election in many decades in 2006, soundly defeating his perennial rival, Kizza Besigye of the Forum for Democratic Change (FDC) 59.3 percent to
37.4 percent. Ironically, political repression and the consolidation of power around
Museveni appear to have increased in the period after the introduction of multiparty competition (Barkan, 2011; Makara, Rakner and Svåsand, 2009; Mwenda,
2007; Tripp, 2010). Museveni bolsters his political dominance through a “carrot
and stick” mixture (Conroy-Krutz and Logan, 2012) of good economic performance,
control over state resources and service delivery, constitutional manipulation, lavish
campaign spending, election fraud, and intimidation of opposition candidates and
supporters in the context of a highly militarized state (Juma, 2011; Barkan, 2011;
Tripp, 2010; Mwenda, 2007).
Fractured, weak, and underfunded, the Ugandan opposition has had few means
to challenge the NRM in most of Uganda (Tangri and Mwenda, 2010). Rural areas
in particular remain firmly tied to the incumbent (Barkan, 2011). In Kawempe,
however, the opposition won a narrow majority in the 2006 presidential election,
supporting Besigye with 55.5 percent of the vote to Museveni’s 41.4 percent.4 In
2011, Museveni once again faced an opposition that was weak and divided throughout most of the country, but was relatively strong in the capital city (Conroy-Krutz
and Logan, 2012). Kawempe thus represents an opposition leaning area in a single
party dominant, semi-authoritarian state with a long history of political repression.
As Barkan (2011) writes: “Kampala serves as something of a ‘Potemkin Village of
Democracy’ — impressive at first encounter, but not representative of the country
4
Electoral data available at http://www.ec.or.ug/Elec results/2006 pres polling.pdf, accessed June
12, 2014.
10
or the regime.” We therefore stress that our results may be particular to Kawempe
and not generalize to Uganda as a whole. Rural, politically homogenous areas in
particular, might exhibit quite di↵erent dynamics.
Treatments
The experiment had three treatments: a non-political “control” treatment, a “reminder to vote” treatment, and an “ink prime” treatment. Table 2 summarizes our
treatment groups. All were delivered face-to-face through a household survey. The
control consisted of a short, benign set of questions on demographic factors (age,
education, etc.) and mobile phone usage and provider. The “reminder to vote’
treatment included the mobile phone survey, plus a generic message reminding the
recipient that there would be presidential and legislative elections on February 18th .
It also stated that “all registered voters will have the opportunity to vote in this
election and voting is very important. The purpose of the reminder to vote treatment was to evaluate the generalizability of American GOTV experiments, which
have typically found that personal delivery of a pro-turnout message raises turnout
levels (Adams and Smith, 1980; Eldersveld, 1956; Gerber and Green, 2000, 2001;
Green, Gerber and Nickerson, 2003; Michelson, 2003; Miller, Bositis and Baer, 1981;
Ramirez, 2005). Several studies have demonstrated similar e↵ects in a variety of
di↵erent countries and settings (Giné and Mansuri, N.d.; Aker, Collier and Vicente,
2013; Bochel and Denver, 1972; Fieldhouse et al., 2013; John and Brannan, 2008;
Guan and Green, 2006). Our hypothesis was therefore that the reminder treatment
would raise turnout.
In the “ink prime” treatment interviewers delivered the mobile phone survey
and then read the following script:
“I would like to speak with you briefly about the upcoming election.
Are you aware that there will be a presidential and legislative election
on February 18th? Are you aware that visible ink will be used to mark
the fingers of those who have voted in this election? [Response noted.]
Are you aware that this ink will be easy to see and will last for several
days? [Response noted.] Ink is not just about preventing fraud. It will
11
Table 2: Treatments and Control
Control
Treatment 1
(Reminder)
Short, non-political survey Short, non-political survey
on mobile phone use
on mobile phone use
Treatment 2
(Ink Prime)
Short, non-political survey
on mobile phone use
Message reminding voters
of election date and asking
them to vote
Message reminding voters
of election date and asking
them to vote
N = 376
Message reminding voters
that ink would be used in
election and that ink would
make their decision to vote visible
N = 377
N = 384
also allow everyone to know if you’ve voted or not. It is important to
show your neighbors that you are responsible and that you are involved
in influencing what happens in your community. All registered voters
will have the opportunity to vote in this election and voting is very
important. Since you are a registered voter, please remember to turn
out and vote.”
We constructed the ink prime treatment in the Kawempe experiment to be a
milder version of the pro-turnout letter treatment in Gerber, Green and Larimer
(2008), which threatened to publicize non-voters’ behavior to apply strong social
pressure to encourage voting. Interviewers were trained to deliver the ink prime
treatment in a neural, informative fashion, and to emphasize that they represented
an American research organization and were not in any way associated with Ugandan political parties. Several studies have implemented treatments similar to Gerber, Green and Larimers in a variety of American settings (Davenport et al., 2010;
Gerber, Green and Larimer, 2008; Mann, 2010; Nickerson, 2008; Panagopolous,
2010; Sinclair, 2011), but, as far as we know, none have done so in emerging democracies. As previous GOTV experiments had consistently found that increasing visibility raised turnout, our expectation was that the ink prime would also have this
e↵ect.
Although initially designed to explore the robustness of a well-established ex-
12
perimental result, the ink treatment also provided a rare opportunity to examine
the e↵ects of ink use in elections. It manipulates awareness about the link between
ink and the visibility of turnout rather than the practice of inking itself. The e↵ect
of this treatment depends on participants having incomplete knowledge of the use
of ink or incomplete awareness of the implications of inking for turnout visibility.
In focus groups conducted prior to the experiment, we found that many people
knew ink would be used in the upcoming election, but not all understood that it
would make turnout decisions visible for an extended period of time. We therefore
had reason to believe that the ink treatment would provide new information to a
significant subset of respondents.
Sampling and Randomization
We visited a total of 1137 households in the week prior to the election. Respondents
had the choice of taking the survey in Luganda (one of Uganda’s many languages,
common in Kampala) or English (the national lingua franca).
We selected ten Kawempe South parishes and designated each parish to a supervisor and team of three enumerators.5 Each enumerator received a packet of
surveys. The surveys came in three versions, corresponding to the three treatment
arms in Table 2. We randomized the order of the surveys for each enumerator.
By working through the packet in the prescribed order, the enumerator achieved
treatment randomization.
Most houses in Kawempe do not have conventional street addresses. We therefore relied on a “random walk” methodology to randomize selection of households.
Within each parish, enumerators selected households and individuals according to
the following pattern: Enumerators started from a landmark designated by their
supervisor, walked a set distance in a predetermined direction (determined by the
date and day of the week), and identified a “block” of four consecutive houses.6
5
Kawempe actually has 13 parishes, but three of them (related to Makerere University) had few
residential holdings so we did not to include them. One of the parishes, Bwaise II, was very large and
three enumerator teams were assigned to it.
6
Supervisors were given large maps of their parishes and instructed to pick landmarks that were
dispersed within the parish, to police the borders of parishes, and to ensure that enumerators did not
intersect one another as they visited houses.
13
They approached the first house in the block and asked to speak with someone
over the age of eighteen. If they found someone of the appropriate age, and that
individual was a registered voter and willing to take the survey, the enumerator
delivered the designated treatment. If the enumerator could not locate a registered
voter willing to take the survey, the enumerator moved on to the second house in
the block and repeated the procedure. If the second household did not contain a
willing registered voter, the enumerator moved on to the third house in the block
and so on until the treatment was delivered to the first available and willing registered voter. Once the enumerator found a registered voter and treated him or
her, the enumerator skipped the rest of the houses in the block, skipped an additional two houses (to reduce the likelihood of contamination), and then began
the procedure again at the next block of four houses.7 In addition to delivering
the survey and treatment, the enumerators described and mapped the location of
each block and the treated house within it.8 This procedure ensured randomization
across treatment arms, but not a representative sample of Kawempe. Our sample
is biased toward people who were home during the period in which we delivered
the survey and were willing to participate.
Table 3 evaluates balance across eighteen pre-treatment covariates. We achieved
balance on almost all variables, with significant di↵erences appearing only for the
age and marital status variables (which are highly correlated). Respondents who
received the reminder treatment were somewhat younger and less likely to be married than those who received the mobile phone treatment. Recipients of the ink
prime treatment were also younger than recipients of the mobile phone survey, but
the skew was milder and less significant. We thoroughly evaluate the sensitivity
of all results to the age imbalance (see section on robustness, below); we do not
believe it challenges the validity of our findings.
During the three days following the election, enumerators re-contacted treated
individuals. At this stage of the experiment, they repeated basic questions from the
7
We assigned treatments to blocks rather than households to improve uptake. We found an individual
willing to participate in virtually all designated blocks.
8
Enumerators were not told they would be returning to the house in the future, only that their
supervisor would be conducting intensive spot checks for quality, a standard practice.
14
Table 3: Balance Verification: p-values from Equality Tests
Cell phone survey
Cell Phone Survey
Turnout Reminder
Cell phone survey
=Turnout Reminder
=Turnout Reminder
=Ink Prime
=Ink Prime
0.2367
0.3445
0.4277
0.0190
0.1268
0.0070
0.3699
0.3688
0.5038
0.4781
0.8629
0.4625
0.7694
0.8511
0.9988
0.9991
0.9989
0.9995
0.0862
0.8303
0.5757
0.1051
0.0942
0.4636
0.4403
0.8690
0.8977
0.6778
0.4188
0.3147
0.2072
0.7525
0.8266
0.8699
0.8480
0.9315
0.5877
0.4656
0.8179
0.4728
0.8756
0.0498
0.9041
0.4638
0.5899
0.7708
0.5221
0.7823
0.3294
0.8971
0.8269
0.8702
0.8483
0.9316
=Ink Prime
Educationa
Interview in Englishb
Female
Married
Has children
Agec
Owns mobile phone
Stand alone fenced housing
Stand alone housing
Small block housing
Apartment
Informal housing
Party Display
Museveni campaignedd
Registered voters, 2006e
Museveni votes, 2006e
Margin of victory, 2006e
Turnout 2006e
0.2140
0.6111
0.7175
0.0560
0.1799
0.0203
0.6243
0.6319
0.7770
0.7757
0.6953
0.5823
0.4175
0.9509
0.9686
0.9823
0.9759
0.9951
a: The education variable had eight categories, where lowest was no education and highest was university or post-university.
Our mean respondent had an education level between 4 (secondary not completed) and 5 (secondary completed).
b: Respondents were given a choice of completing the survey in English or Luganda.
c: The age variable had eight categories, where lowest was age 18-24 and highest was 60+. Our mean respondent was aged
between 2 (25-30) and 3 (31-36).
d: Campaign activity by Museveni in 2011, measured at the parish level.
e: Measured at the parish level.
original survey as well as a new set of political and social questions including vote
choice, political party identification, strength of ethnic attachments, and political
and social pressures to vote. Near the end of the survey, enumerators also asked to
check the respondent’s finger for ink and recorded whether the respondent had a
mark (indicating they voted) or lacked a mark (indicating they did not vote). We
achieved a re-contact rate of over 99 percent; less than 2 percent of successfully
re-contacted participants refused to take the follow-up survey, producing an overall
attrition rate of 2.1 percent.9
9
Using rare events logit to examine patterns in the attrition rate, we found treatment status was
unrelated to the likelihood of attrition. We also checked the correlation between questions from the
pre-election component of the experiment and their counterparts in the post-election component (gender
and age) and a perfect match on gender and a correlation of .99 on age (dropping a small group of
people who refused to answer the age question on the second round). We are therefore confident that
we re-interviewed the correct individuals. For results, see Table A-5 in the Appendix.
15
Results
We first examine whether the ink prime had the intended e↵ect of increasing awareness about the public nature of voting. We asked: “Regardless of whether you
actually voted, do you think that your neighbors know whether you turned out or
stayed home, or do they not know?” If the visibility treatment had the intended
e↵ect, then respondents exposed to it should have been more likely to believe that
their neighbors knew their turnout choice. Indeed, the ink prime increased the
probability of believing neighbors knew turnout decision by about seven percentage points, significant at the .05 level (see Table A-2 in Appendix). The simple
reminder treatment served as a placebo for this outcome and, as expected, had no
e↵ect.
Next we present a basic breakdown of the outcome variable — finger marking
status (Table 4). A last minute decision by the election administrators in Uganda
to use permanent markers instead of indelible ink complicated our measurement
strategy. The permanent marker made a more subtle mark on the skin, which
meant that reading fingers required a certain degree of cooperation on the part
of participants. A small number (4 percent) of respondents refused to allow their
fingers to be examined. More respondents (20 percent) made it difficult for enumerators to get a clear reading. Rather than treating these unclear or refused fingers
as a nuisance category, we view them as a strategic choice that reveals important
information about the mindset of Kawempe citizens. In the analysis that follows,
we group unclear fingers and refusals together as “concealers.” We then use multinomial logit to examine concealers, marked, and unmarked fingers as an unordered
three level categorical variable.10
The large “unclear” category means that we cannot map our marked finger
variable directly to turnout because the true turnout level depends on the fraction
of unclear fingers/refusals that actually voted. We can, however, put bounds on
the turnout level. Around 65 percent of our sample had a clearly marked finger and
10
We have also run models with the original four levels of the dependent variable. Collapsing “finger
unclear” and “finger check refused” has no e↵ect on estimates for finger marked and finger unmarked.
Estimates for “finger unclear” are very similar to estimates for the combined “concealer” group, while
“finger check refused” estimates indicate no substantive e↵ects of interest.
16
Table 4: Finger Status
Finger Status
Count (percentage) Control
Finger
Finger
Finger
Finger
Total
738 (64%)
124 (11%)
233 (20%)
42 (4%)
1,137
Marked (Voted)
Not Marked (Did not Vote)
Unclear
Check Refused
248
36
88
12
384
Turnout
Ink
Reminder Prime
255
235
40
48
66
79
15
15
376
377
only 11 percent had a finger that was clearly not marked. This gives us a turnout
range of 65 to 89 percent in our sample.11
Table 5 shows average treatment e↵ects on finger status for the two treatment
variables (and 95 percent confidence intervals).12 Treatment e↵ects represent the
changes in probability for each category of the finger status variable given a particular treatment. Changes across a row of the table must sum to zero. The
simple reminder treatment produced no discernible shifts between categories of the
dependent variable. The ink prime treatment, on the other hand, increased the
probability of not having an inked finger by close to four percentage points, with
95 percent confidence intervals that just nick zero, and did so in spite of the clear
turnout encouraging tone of the prompt.13 The magnitude of the ink prime average treatment e↵ect on the probability of having an unmarked finger is similar to
that of turnout enhancing e↵ects found in U.S.-based GOTV experiments, most of
which fall in the 1–8 percentage point range. As our sample undoubtedly includes
many people who already knew about inking and had factored it into their decisions
11
Our sample was not representative, so this range is not indicative of official turnout in Kawempe
South, which was 43%. Many residents of Kawempe appeared to leave the city prior to the election and
our baseline survey, and this may be one reason why our lowest estimate was higher than official levels.
Figure reports the distribution of turnout levels in this election. Median official turnout by district in
the election was 61%.
12
Table 5 is generated by running multinomial logit on the three level finger status variable, controlling
for the reminder and ink prime treatments (mobile phone treatment is reference category) as well as parish
fixed e↵ects to account for blocking design, and then generating first di↵erences for each treatment. Raw
output of the model is presented in Table A-3 of the Appendix, “baseline model.”
13
We do not know whether this e↵ect occurred because people actually voted less or because they
were more willing to show us that they had not voted. Put another way, we are fairly certain that
the treatment moved people into the not inked category but we are less certain whether it moved them
out of the inked category, the concealed category, or both. In either case, the treatment either reduced
turnout or reduced reluctance to be identified as a non-voter. Both outcomes are consistent with a
turnout suppression dynamic and contrast sharply with findings that increasing the visibility of voting
has a turnout enhancing as in U.S.-focused GOTV experiments.
17
about whether to vote, our results most likely understate the full impact of inking
on behavior.
Table 5: Average Treatment E↵ects, 95 Percent Confidence Intervals in Parentheses
dP (Finger Inked)
dP (Concealed)
dP (Finger Not Inked)
T1 (Reminder)
.031
(-.037, .099)
-.041
(-.101, .019)
.011
(-.030, .055)
T2 (Ink Prime)
-.023
(-.091, .048)
-.014
(-.072, .045)
.037
(-.007, .082)
Model estimated is multinomial logit on the three level finger status variable, controlling for the reminder and ink prime treatments (mobile phone treatment is reference
category) as well as parish fixed e↵ects to account for blocking design, and then generating first di↵erences for each treatment. Raw output of the model is presented in
Table A-3 of the Appendix, “baseline model.”
The ink prime treatment should have had its largest e↵ects on respondents who
were not already aware of the use and implications of ink in elections. Since we
do not have direct measures of information about inking, we examine di↵erential
treatment e↵ects for respondents at varying levels of education and age. We suspect
that more educated residents had access to better information about the electoral
process, including the use of ink. Younger voters, not having voted in previous
elections, may also have had less knowledge about the use of ink and its implications.
We therefore expect the ink prime to have had stronger e↵ects for these groups
versus older, better-educated participants.
Indeed, controlling for education, age, and interaction terms for these variables
and the ink prime substantially sharpens the e↵ects of the ink treatment (it is now
significant at the .01 level). Moreover, age and education appear to have moderated
the inking treatment. Both interactions are significant at the 95 percent level.14
(See Table A-3 in the appendix for full results).
Table 6 shows the e↵ects of the ink prime treatment for people without a formal
14
The z statistic for the interaction terms (on the inked outcome) are 2.18 for education and 2.12 for
age. These remain significant at the 0.10 level when we use the Bonferroni correction for multiple (3)
tests.
18
education versus those with a university degree or higher, and young people (1824) versus older people (60+).15 The e↵ects of the ink treatment on people with
a university degree were indistinguishable from zero while treatment e↵ects for
people without a formal education were substantial: a 27 percentage point drop
in the probability of having a marked finger and an accompanying 22 percentage
point increase in the probability of having an unmarked one (confidence intervals
do not include zero). Similarly, older people had no identifiable response to the
inking treatment — if anything it may have increased their probability of having a
marked finger (this result is not statistically significant, however). Young people,
in contrast, proved very sensitive to the inking treatment, displaying a thirty two
percentage point increase in the probability of having an unmarked and a twenty
two percentage drop in the probability of having a marked one. Figure 1 displays
these e↵ects.
Given the almost symmetric and compensatory responses in the probability
of being inked or not, and changes in the probability of being concealed centered
around zero, we believe that the treatment not only reduced the probability of being
inked, but actually reduced turnout itself — especially in young and less educated
potential voters. At the very least, it made respondents more likely to reveal that
they had not voted, a result most consistent with turnout suppression.
Robustness Checks
Our randomization produced balance across a large number of covariates, but was
somewhat imbalanced on age (especially for the reminder treatment) and marital
status (see Table 3). Younger people were slightly more likely to get the ink prime
treatment relative to the mobile phone control, and quite a bit more likely to get
the reminder treatment relative to the mobile phone control. A similar pattern
pertains for marital status. Because marital status is highly correlated with age,
and age is causally prior to marital status, we focus our robustness checks on age.
15
These age categories were pre-set in the survey and are standard in many African surveys.
19
Table 6: Heterogeneous Ink Prime Treatment E↵ects for Age and Education
Finger Status:
Finger Inked
Concealed
Finger Not Inked
No Formal
Education
-.22
(-.44, -.01)
-.05
(-.20, .13)
.27
(.08, .51)
University
Degree +
-.07
(-.22, .08)
.05
(-.07, .18)
.02
(-.06, .12)
18-24
-.22
(-.44, .00)
-.10
(-.26, .08)
.32
(.10, .56)
60+
.05
(-.05, .15)
-.05
(-.14, .06)
-.01
(-.03, .01)
Education Level
Age
Note: Simulated from a multinomial logit model that included parish fixed e↵ects, N = 1124.
95 percent confidence intervals in parentheses. See Table A-3 in Appendix for raw coefficients
and full details.
In order for the imbalance in age across treatment groups to generate the observed treatment e↵ects, age would have to be positively correlated with having an
inked finger. We looked for evidence of a correlation between age and finger status
by examining a group of non-treated survey respondents interviewed in the days
immediately following the election.16 Reassuringly, age is uncorrelated with finger
status in this group.
The imbalance in age originates in three of the ten parishes in the study. When
these parishes are removed, we have balance across all of the covariates in Table
3. To further probe robustness, we dropped the three problematic parishes and
replicated our analysis. Results were as strong or stronger in the limited sample17
and education and age continue to mediate the inking treatment, with less education and youth being associated with larger negative responses to the inking. We
therefore believe it unlikely that the modest imbalance in our sample confounds our
16
Respondents were interviewed at the same time and in the same fashion as the three treatment
groups.
17
The change in probability for not being inked in the full sample is .037, with 95 percent confidence
intervals spanning from -.007 to .082. In the reduced sample of balanced parishes it is .049, with
confidence intervals spanning from -.015 to .120.
20
Change In Probability Relative To Control
-.7 -.6 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 .6 .7
Figure 1: Heterogenous Treatment E↵ects by Age and Level of Education
Finger Inked
Concealed
High Age
.4
.3
.2
.1
0
-.5 -.4 -.3 -.2 -.1
Change In Probability Relative To Control
.5
Low Age
Finger Not Inked
Finger Inked
Concealed
Low Education
Finger Not Inked
High Education
Note: Points represent estimated probabilities, and whiskers represent 95% Confidence Intervals.
Black horizontal line represents control group probabilities levels set to 0.
findings.
21
Turnout Discouragement?
We speculated at the outset of this paper that the common electoral practice of
marking voters with ink might alter incentives for citizens to vote, with turnout
manipulation strategies as the causal mechanism connecting ink with behavioral
outcomes. We suggested that inking facilitates turnout manipulation strategies by
allowing political actors to condition rewards or punishments on the observation or
absence of an ink stain. Such strategies aim to decrease or increase turnout, either
to enforce community norms or to further the narrow electoral goals of politicians.
Depending on the mix of strategies in play in a particular area inking may either
suppress or raise turnout levels.
We have shown that an experimental ink prime treatment reduced turnout in
Kawempe during the 2011 election, especially among younger and less educated
potential voters. These results are consistent with the employment of turnout
discouragement strategies in Kawempe during this election, but obviously cannot
provide direct evidence of it. Are there additional indications that such strategies may have in fact been in use? And, if turnout discouragement did occur in
Kawempe, was it of a normative or electoral variety?
On the last point first, we are not aware of any civic campaigns to boycott the
vote during the 2011 Ugandan election and we have no reason to believe Uganda
departs from other areas of Africa in its embrace of norms favoring electoral participation. In a focus group discussion conducted after the election, most participants
voiced the opinion that voting was important.18 Hence, if turnout discouragement
occurred, it was most likely electorally motivated.
Certainly the ruling NRM and its leader Yoweri Museveni had electoral incentives to discourage turnout in Kawempe. Kampala represents one place in which
Museveni could not count on support. Lower turnout there would bolster his electoral dominance and (perhaps more importantly) starve the opposition of one of
its few reliable sources of votes.
We do not have evidence of intentional, centrally directed, turnout punishment
in Kampala. We are not aware of any explicit e↵orts to use ink to scare voters,
18
Focus group conducted in Kawempe on December 21, 2011.
22
as was done in elections in Afghanistan, India, and Zimbabwe. Intimidation can
be difficult to detect, however. Intimidators may have ways to signal their intent
to their desired audience without revealing it to other observers. Moreover, it may
only take a modest threat to a↵ect turnout levels, and that threat need not be
intentionally generated to have an impact. Voting confers low individual benefits;
small shifts in perceptions about the probability of sanctions may be all that is
necessary to shift behavior (Driscoll and Hidalgo, 2014). Hence, the absence of
overt acts of intimidation is not evidence that voters did not feel intimidated.
There is, in fact, ample evidence that many voters felt intimidated during the
2011 election. Tension su↵used Uganda’s pre-election environment, particularly in
Kampala. The previous election in 2006 had seen several instances of violence, as
had the NRM primaries in the summer of 2010 (Conroy-Krutz and Logan, 2012).
Museveni suppressed demonstrations in Kampala in September 2009 by sending
in the police, who shot and killed 20 protestors. Reports of human rights abuses
by the army and police increased in the period between the elections (Barkan,
2011). In 2011, Museveni advertised his military strength throughout the city,
campaigning on a strong man image. He stationed extra police and military units
around Kampala — including Kawempe — in the weeks before the election and on
Election Day (Izama and Wilkerson, 2011). Posters garbed him as Rambo, complete
with military helicopters and tanks in the background, playing on his reputation
as a military man who fought his way to power. Election observers from various
missions noted that the substantial use of military troops on election day had
potentially intimidated voters (Conroy-Krutz and Logan, 2012). As one Ugandan
explained “he came from the bush [a reference to fighting to power through a rural
insurgency movement], and back to the bush he will go [he will resort to bush tactics
if necessary].” In interviews we conducted after the election, Kawempe residents
and local opposition MPs complained of intimidation, bribery, rigging, and vote
stuffing by security forces and NRM representatives during the election.19
In 2008, more than two years prior to the election, 45 percent of Kampala’s
residents polled by the Afrobarometer answered that they personally feared be19
Interviews conducted in Kawempe, October 2011.
23
coming a victim of election related political intimidation or violence.20 Numbers
were similarly high in late 2010 and early 2011, when a majority of Afrobarometer
respondents also reported a high expected likelihood of post-election violence. Fears
were especially prominent amongst opposition supporters (Conroy-Krutz and Logan, 2012). In this context, Museveni may have been able to discourage turnout in
opposition areas without engaging in overt and conspicuous e↵orts to do so. Even
if intimidation did not change hearts, minds, and votes (Conroy-Krutz and Logan,
2012), it may have kept opposition voters at home.21
If Museveni did attempt to discourage opposition voters from going to the polls,
we would expect to see a negative correlation between opposition support and
turnout in the 2011 election. We also would expect opposition support to correlate
with use of violence and intimidation. To test these expectations, we turn to data
collected in a pre-election survey in January and February 2011 (citation redacted).
These data are from 1100 randomly selected polling stations from a sampling
frame that included all polling stations in Uganda and then randomly sampled
five registered voters around each station. The survey included questions on intended vote choice, religious identity, education level, and development variables
like whether the household was electrified, respondent employment, income, and
newspaper readership. It also asked respondents if they thought violence was likely
around their polling station in the upcoming election. By averaging across respondents for each polling station, we get a coarse but unbiased measure of intended
vote choice and expectations of election violence at the polling station level, as
well as numerous socioeconomic control variables. We matched the survey data to
actual polling station turnout in the 2011 election.
Table 7 shows the results of four models. The dependent variable in columns
1-2 is expected violence at the polling station; the dependent variable in columns
3-4 is polling station turnout in 2011. In all models, we control for religious and
20
See Afrobarometer Round 4, Uganda.
We do not mean to imply that Museveni returned to power only or primarily through intimidation.
As others have noted, the primary sources of his victory in 2011 probably include: general satisfaction
with his provision of services (including new districts), good economic performance, improved security
in the North, disorganization and weakness of the opposition, and a massive and expensive election
campaign (Conroy-Krutz and Logan, 2012; Juma, 2011). However, intimidation likely played a role in
some locales, especially opposition leaning areas like Kampala.
21
24
socioeconomic variables; in Models 2 and 4 we control for region fixed e↵ects. We
find that the level of opposition support positively correlates with expectations
of violence, with borderline statistical significance. We also find that areas with
higher numbers of opposition support had lower turnout levels. These results are
consistent with turnout discouragement of opposition voters during the election.
Table 7: OLS Models of Expected Violence and Turnout at the
Polling Station Level for the 2011 Ugandan Elections
Violence Anticipated Turnout
(1)
(2)
(3)
⇤
Opposition
0.06
0.05
-0.10⇤⇤⇤
(1.70)
(1.51)
(-6.79)
Catholic
0.03
-0.01
-0.10⇤⇤⇤
(0.86)
(-0.17)
(-6.10)
Muslim
-0.09
-0.10⇤
-0.09⇤⇤⇤
(-1.59)
(-1.73)
(-2.89)
⇤⇤⇤
⇤⇤⇤
Education
0.08
0.09
0.00
(7.17)
(7.14)
(0.28)
Electrified
-0.04
-0.02
-0.09⇤⇤⇤
(-0.95)
(-0.61)
(-3.47)
Employment
0.03
0.02
0.01
(0.82)
(0.73)
(0.38)
Mean Income
-0.00
-0.01
0.00
(-0.92)
(-1.56)
(1.17)
Newspaper
-0.00
0.00
-0.02⇤
(-0.14)
(0.13)
(-1.91)
⇤⇤⇤
⇤⇤⇤
Constant
0.19
0.21
0.65⇤⇤⇤
(4.25)
(4.63)
(28.33)
Observations
1095
1077
984
R2
0.0639
0.0723
0.1123
Region Fixed E↵ects
No
Yes
No
in 2011
(4)
-0.07⇤⇤⇤
(-5.28)
-0.01
(-0.46)
-0.11⇤⇤⇤
(-3.49)
0.00
(0.26)
-0.08⇤⇤⇤
(-3.71)
-0.01
(-0.68)
0.00
(0.11)
-0.02⇤⇤
(-2.46)
0.62⇤⇤⇤
(28.33)
969
0.2379
Yes
Note: ⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
t statistics based on robust standard errors in parentheses. All variables
except 2011 polling station turnout come from an original pre-election
survey of registered voters conducted in January 2011. For more details
on the survey see (citation redacted).
On a final note, if turnout discouragement strategies occurred in Kawempe, their
most likely targets would have been visible, known opposition partisans: individuals
who could not easily hide or deny their opposition leanings.22 Partisanship is not
easy to discern in Kampala. In this particular election, ethnicity had little correla22
We did not measure opposition support pre-treatment because we feared it would prime respondents
in ways that could interfere with the experiment. We also strongly suspected that misrepresentation and
non-response to the question would be high. Therefore, direct tests of this hypothesis are impossible.
25
tion with voting behavior in Kawempe
23
and, in any case, ethnicity is not easily
perceived in urban Kampala (Habyarimana et al. 2007). One possible exception to
the low identifiability of partisanship in this setting, however, is age. Youth were
less likely than other age groups in Kawempe to support Museveni.24 The salience
of youth as an opposition marker may have been especially high in February 2011,
as this was during the upheavals in Egypt and elsewhere, which were extensively
publicized in Kampala. As we noted earlier, young people responded particularly
strongly to the ink prompt. While this may have been a purely informational e↵ect
(as discussed earlier, young voters were less likely to know about inking), it is also
consistent with younger voters being more fearful of being inked.
In sum, electoral discouragement is a plausible mechanism linking the experimental ink prime to lower turnout in Kawempe. A powerful incumbent had incentives to discourage a known opposition population from turning out to vote. Such
discouragement need not have been conspicuous to be e↵ective.
Conclusion
During the 2011 Ugandan general election in a lively opposition neighborhood in a
coercive single party dominant state, raising awareness about the use of election ink
resulted in fewer inked fingers, especially among young and less educated voters.
This result is consistent with turnout suppression, and supports our guiding intuition that electoral inking may have undesirable behavioral consequences in some
electoral contexts. Kawempe is in many ways unusual in Uganda, and we make
not claims that these results generalzie to other parts of the country. Indeed, our
expectation is that the e↵ects of inking on turnout are highly context specific. In
Museveni strongholds, where nearly everyone supports the ruling party, we would
expect inking to have the opposite e↵ect of encouraging turnout.
We hope our findings elicit greater consideration of the largely unquestioned
practice of inking in emerging democracies. We do not advocate the whole-scale
23
Using data from a survey collected a few days after the election in Kawempe, we find that ethnic
dummy variables do a poor job of predicting support for Museveni. See Table A-4 in Appendix.
24
In the Kawempe survey data, each age level (there were eight total) is associated with about a five
percentage point increase in the probability of supporting Museveni. See Table A-4 in Appendix.
26
abandonment of ink in these elections. Ink represents a low cost and easy to implement solution to important problems of electoral administration. In many elections,
its behavioral e↵ects are likely to be mild and benign. However, during elections
featuring turnout manipulation strategies, electoral administrators may want to
consider whether inking facilitates coercion and, if so, whether the risks of using
ink outweigh the benefits.
Our results also suggest that the outcomes reported in American GOTV studies
should be reconsidered in the light of broader theory. The findings of these studies
likely depend on a particular context, i.e. one in which community norms favor
participation and electoral strategies do not target turnout. In other contexts, the
public revelation of turnout may have far from benign results. A variety of mechanisms and practices may reveal otherwise private political behavior: experimental
letter treatments, use of election ink, geographic segregation of politically similar
individuals, or violations of ballot secrecy. When politicians are able to estimate
the likely behavior of individuals with a high degree of accuracy, they gain access to
a range of powerful strategies like intimidation, vote buying, and turnout manipulation (Kitschelt and Wilkinson, 2007; Sha↵er, 2007; Stokes, 2005; Nichter, 2008). We
hope our work inspires further theoretical and empirical analysis of the interaction
between such revelatory mechanisms and practices and political context.
Finally, our results highlight the importance of broadening the study of African
political behavior to more explicitly consider electoral context. Prior work on
turnout in Africa has focused primarily on individual level factors like education,
income, and partisanship, in many cases borrowing a set of independent variables
from American studies of behavior and examining them in African survey samples
(Bratton, 1999; Bratton, Mattes and Gyimah-Boadi, 2005; Bratton, 2008; Isaksson,
2010; Kuenzi and Lambright, 2005, 2007). To truly understand the African voter,
however, we must look beyond individuals to how electoral practices, community
norms, and political strategies interact to shape behavior.
27
References
Adams, William C. and Dennis J. Smith. 1980. “E↵ects of Telephone Canvassing on
Turnout and Preferences: A Field Experiment.” The Public Opinion Quarterly
44(3):389 – 395.
Aker, Jenny C., Paul Collier and Pedro C. Vicente. 2013. “Is Information Power?
Using Mobile Phones and Free Newspapers during an Election is Mozambique.”
Unpublished .
Ariely, Anat Bracha and Stephan Meier. 2009. “Doing Good or Doing Well? Image
Motivation and Monetary Incentives in Behaving Prosocially.” The American
Economic Review 99(1):544 – 555.
Baldauf, Scott. 2004. “An Afghan ‘Hanging Chad’ Dispute.”.
URL: http://www.csmonitor.com/2004/1012/p06s01-wosc.html/(page)/2
Barkan, Joel D. 2011. Uganda: Assessing Risks to Stability. Washington, D.C.:
Center for Strategic & International Studies.
BBC News. 2008. “Tsvangirai regects ‘sham’ ballot.”.
URL: http://news.bbc.co.uk/2/hi/africa/7478399.stm
Beaulieu, Emily. 2014. Electoral Protest and Democracy in eh Developing World.
New York: Cambridge University Press.
Becker, G. and G. Stigler. 1974. “Law Enforcement, Malfeasance, and the Compensation of Enforcers.” Journal of Legal Studies 3:1 – 19.
Becker, Gary J. 1968. “Crime and Punishment: An Economic Approach.” Journal
of Political Economy 76:169–217.
Bochel, J. M. and D. T. Denver. 1972. “The Impact of the Campaign on the Result
of Local Government Elections.” British Journal of Political Science 2(2):239 –
244.
Bratton, Michael. 1999. “Political Participation in a New Democracy: Institutional
Considerations from Zambia.” Comparative Political Studies 32(5):549 – 588.
28
Bratton, Michael. 2008. Poor People and Democratic Citizenship in Africa. New
York: Cambridge University Press.
Bratton, Michael, Robert Mattes and Emmanuel Gyimah-Boadi. 2005. Public Opinion, Democracy, and Market Reform in Africa. Cambridge, UK: Cambridge
University Press.
Cialdini, Robert B. and Melanie R. Trost. 1998. Social Influence: Social Norms,
Conformity, and Compliance. Boston: McGraw-Hill pp. 151 –192.
Cialdini, Robert B. and Noah J. Goldstien. 2004. “Social Influence: Compliance
and Conformity.” Annual Review of Psychology 55:591 – 621.
Cody, Edward. 1985. “Hondurans Elect President Today; Dispute Over Election
Procedure Not Settled.” The Washington Post p. A21.
Collier, Paul and Pedro C. Vicente. 2008. “Votes and Violence: Evidence from a
Field Experiment in Nigeria.” Working Paper.
Conroy-Krutz, Je↵rey and Carolyn Logan. 2012. “Museveni and the 2011 Ugandan
election: did the money matter?” Journal of Modern African Studies 50(4):625–
655.
Cox, Gary W. and J. Morgan Kousser. 1981. “Turnout and Rural Corruption: New
York as a Test Case.” American Journal of Political Science 25(4):646 – 663.
Davenport, Ti↵any C., Alan S. Gerber, Donald P. Green, Christopher W. Larimer,
Christopher B. Mann and Costas Panagopoulos. 2010. “The Enduring E↵ects
of Social Pressure: Tracking Campaign Experiments Over a Series of Elections.”
Political Behavior 32(3):423 – 430.
Democracy International. 2010.
U.S. Election Observation Mission to the
Afghanistan Presidential and Provincial Council Elections 2009. Technical report.
Democracy International. 2011. Democracy International Election Observation Mission: Afghanistan Parliamentary Elections 2010. Technical report.
29
Driscoll, Jesse and F. Daniel Hidalgo. 2014. “Intended and Unintended Consequences of Democracy Promotion Assistance to Georgia after the Rose Revolution.” Research & Politics 1:1–13.
Eldersveld, S. J. 1956. “Experimental Propaganda Techniques and Voting Behavior.” American Political Science Review 50:154 – 165.
Fieldhouse, Edward, David Cutts, Paul Widdop and Peter John. 2013. “Do Impersonal Mobilisation Methods Work? Evidence form a nationwide Get-Out-theVote experiment in England.” Electoral Studies 1:113 – 123.
Gerber, Alan, Donald Green and Christopher Larimer. 2008. “Social Pressure
and Voter Turnout: Evidence from a Large-Scale Field Experiment.” American
Political Science Review 102(1):33 – 48.
Gerber, Alan S. and Donald P. Green. 2000. “The E↵ects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment.” American
Political Science Review 94(3):653 – 663.
Gerber, Alan S. and Donald P. Green. 2001. “Do Phone Calls Increase Voter
Turnout? A Field Experiment.” Public Opinion Quarterly 65:75 – 85.
Giné, Xavier and Ghazala Mansuri. N.d. “Together We Will: Experimental Evidence on Female Voting Behavior in Pakistan.” World Bank Policy Research
Working Paper 5692.
Green, Donald P., Alan S. Gerber and David W. Nickerson. 2003. “Getting Out
the Vote in Local Elections: Results from Six Door-to-Door Canvassing Experiments.” Journal of Politics 65:1083 – 1096.
Guan, Mei and Donald P. Green. 2006.
“Noncoercive Mobilization in State-
Controlled Election: An Experimental Study in Beijing.” Comparative Political
Studies 39(100):1175 – 1193.
Habyarimana, James, Macartan Humphreys, Daniel N. Posner and Jeremy M. Weinstein. 2007. “Why Does Ethnic Diversity Undermine Public Goods Provision?”
American Political Science Review 101(4):709 – 725.
30
Habyarimana, James, Macartan Humphreys, Daniel N. Posner and Jeremy M. Weinstein. 2009. Coethnicity: Diversity and the Dilemmas of Collective Aciton. New
York: Russell Sage Foundation.
Isaksson, Ann-Sofie. 2010. “Political Participation in Africa: Participatory Inequalitites and the Role of Resources.” Afrobarometer Working Paper No. 121.
Izama, Angelo and John Wilkerson. 2011. “Uganda: Museveni’s Triumph and
Weakness.” Journal of Democracy 22(3):64–78.
Jha, Prem Shankar. 2002. “Kashmir:Voting Amid Violence.” World Press Review
49(12).
John, Peter and Tessa Brannan. 2008. “How Di↵erent Are Telephoning and Canvassing? Results from a ’Get Out the Vote’ Field Experiment in the British 2005
General Election.” British Journal of Political Science 48(3):565 – 574.
Juma, Kakuba Sultan. 2011. “Voting Behaviour in Uganda since 1996: An investigation into the factors likely to determine voters’ choices in the 2011 elections.”
African Journal of Political Science and International Relations 51(1):21–29.
Kitschelt, Herbert and Steven I. Wilkinson, eds. 2007. Patrons, Clients, and Politics: Patterns of Democratic Accountability and Political Competition. Cambridge, UK: Cambridge University Press.
Kousser, Morgan J. 1974. The Shaping of Southern Politics: Su↵rage Restriction
and the Establishment of the One-Party South, 1880-1910. New Haven: Yale
University Press.
Kuenzi, Michelle and Gina Lambright. 2005. “Who Votes in Africa? An Examination of Electoral Turnout in 10 African Countries.” Afrobarometer Working
Paper No. 51.
Kuenzi, Michelle and Gina Lambright. 2007. “Voter Turnout in Africa’s Multiparty
Regimes.” Comparative Political Studies 40(6):665 – 690.
31
Makara, Sabiti, Lise Rakner and Lars Svåsand. 2009.
“Turnaround: the Na-
tional Resistance Movement and the Reintroduction of a Multiparty System in
Uganda.” International Political Science Review 30:185–204.
Mann, C. B. 2010. “Is There Backlash to Social Pressure? A Large-scale Field
Experiment on Voter Mobilization.” Political Behavior 32:387 – 407.
Meislin, Richar J. 1982. “Rural Voters, Despite Fears, Hike For Miles.” The New
York Times p. A1.
Michelson, Melissa R. 2003. “Getting Out the Latino Vote: How Door-to-Door
Canvassing Influences Voter Turnout in Rural Central California.” Political Behavior 25(3):247 – 263.
Miller, Roy E., David A. Bositis and Denise L. Baer. 1981. “Stimulating Voter
Turnout in a Primary: Field Experiment with a Preccinct Committeeman.” International Political Science Review 2(4):445 – 460.
Mwenda, Andrew M. 2007. “Personalizing Power in Uganda.” Journal of Democracy 18(3):23–37.
Nichter, Simeon. 2008. “Vote Buying or Turnout Buying? Machine Politics and
the Secret Ballot.” American Political Science Review 102(1):19 – 31.
Nickerson, David W. 2008. “Is Voting Contagious? Evidence from Two Field
Experiments.” American Political Science Review 102:49 – 57.
Organization for Security and Cooperation in Europe (OSCE). 2010. Election Observation Handbook, Sixth Edition. Warsaw, Poland: Office for Democratic Institutions and Human Rights.
Ostwald, Kai. 2013.
“How to Win a Lost Election: Malapportionment and
Malaysia’s 2013 General Election.” The Round Table: The Commonwealth Journal of International A↵airs 102(6):521 – 532.
Panagopolous, Costas. 2010. “A↵ect, Social Pressure, and Prosocial Motivation:
Field Experiment Evidence of the Mobilizing E↵ects of Pride, Shame, and Publicizing Voting Behavior.” Political Behavior 32(3):369 – 386.
32
Posner, Richard A and Eric B. Rasmusen. 1999. “Creating and Enforcing Norms,
with Special Reference to Sanctions.” International Review of Law and Economics 19:369 – 382.
Raman, Sunil. 2004. “India Link to Afghan Ink Stain.” BBC News .
URL: http://news.bbc.co.uk/2/hi/southa sia/3736590.stm
Ramirez, Ricardo. 2005. “Giving Voice to Latino Voters: A Field Experiment on
the E↵ectiveness of a National Nonpartisan Mobilization E↵ort.” The Annals of
the American Academy of Political and Social Science 601:66 – 84.
Rind, Bruce and Daniel Benjamin. 1994. “E↵ects of Public Image Concerns and
Self-Image on Compliance.” Journal of Social Psychology 134:19 – 25.
Schedler, Andreas. 2002. “Might Cleaning Up Elections Keep People Away From
the Polls? Historical and Comparative Perspectives.” International Political Science Review 23(1):69 – 84.
Schultz, P. Wesley. 1999. “Changing Behavior with Normative Feedback Interventions: A Field Experiment on Curbside Recycling.” Basic and Applied Social
Psychology 21:25 – 36.
Schultz, P. Wesley, J. Nolan, R. Cialdini, N. Goldstein and V. Griskevicius. 2007.
“The Constructive, Destructive, and Recontructive Power of Social Norms.” Psychological Science 18:429 – 434.
Sha↵er, Frederic C., ed. 2007. Elections for Sale: The Causes and Consequences of
Vote Buying. Boulder, CO: Lynne Rienner Publishers.
Sinclair, Betsey. 2011. The Social Citizen. University of Chicago Press.
Sinclair, Betsy. 2012. The Social Citizen: Peer Networks and Political Behavior.
Chicago: University of Chicago Press.
Stokes, Susan. 2005. “Perverse Accountability: A Formal Model of Machine Politics
with Evidence from Argentina.” American Political Science Review 99(3):215–
326.
33
Tangri, Roger and Andrew M. Mwenda. 2010. “President Museveni and the Politics
of Presidential Tenure in Uganda.” Journal of Contemporary African Studies
28(1):31–49.
Taubman, Philip. 1982. “C.I.A. Chief Tells of Attempt to Aid Salvador Vote.” The
New York Times p. A6.
Tripp, Aili M. 2010. Museveni’s Uganda: Paradoxes of Power in a Hybrid Regime.
Colorado: Lynne Rienner Publishers.
Wattenberg, Ben J. 1986. “The Whole World is Watching.” Newsweek p. 25.
Weiss, M. 2009. “Edging Toward a New Politics in Malaysia: Civil Society at the
Gate?” Asian Survey 19(5):741 – 758.
Welsh, B. 2013. “Malaysia’s Elections: A Step Backward.” Journal of Democracy
24(4):136 – 150.
Whatley, M. A., J.M. Webster, R.H. Smith and A. Rhodes. 1999. “The E↵ect of
a Favor on Public and Private Compliance: How Internalized is the Norm of
Reciprocity?” Basic and Applied Social Psychology 21:251 – 259.
34
Appendices
Online appendix of additional Tables and Figures
Table A-1: Summary statistics
Variable
Mean
Std. Dev.
Min.
Max.
N
Finger Status
1.460
0.683
1
3
1137
T1 (Turnout Remainder)
0.330
0.470
0
1
1137
T2 (Ink Prime)
0.330
0.470
0
1
1137
Bwaise II (Parish 1)
0.252
0.435
0
1
1129
Bwaise III (Parish 2)
0.084
0.277
0
1
1129
Makerere I (Parish 3)
0.083
0.275
0
1
1129
Makerere II (Parish 4)
0.085
0.278
0
1
1129
Makerere III (Parish 5)
0.074
0.262
0
1
1129
Mulago I (Parish 6)
0.084
0.277
0
1
1129
Mulago II (Parish 7)
0.085
0.278
0
1
1129
Mulago III (Parish 8)
0.084
0.277
0
1
1129
Muluka II - Makerere University (Parish 9)
0.086
0.28
0
1
1129
Education
4.419
1.868
1
8
1099
Age
2.310
1.752
1
8
1098
Female = 1
0.630
0.484
0
1
1137
Gifts important
0.705
0.456
0
1
1143
Visible display of NRM affiliation
0.231
0.422
0
1
1143
Pentacostal
0.105
0.307
0
1
1043
Mutoro
0.027
0.163
0
1
1100
Muganda
0.595
0.491
0
1
1100
Interview in English
0.405
0.491
0
1
1137
Married = 1
0.447
0.497
0
1
1137
Has children = 1
0.532
0.499
0
1
1137
Owns mobile phone
0.820
0.385
0
1
1137
Continued on next page...
35
... table A-1 continued
Variable
Mean
Std. Dev.
Min.
Max.
N
Stand alone fenced housing
0.142
0.350
0
1
1143
Stand alone housing
0.072
0.258
0
1
1143
Small block housing
0.168
0.374
0
1
1143
Apartment
0.237
0.425
0
1
1143
Informal housing
0.376
0.485
0
1
1143
Visible display of party affiliation
0.377
0.485
0
1
1137
Museveni campaigned in area in 2011
0.180
0.3823
0
1
1129
Catholic
0.360
0.480
0
1
1043
CUG
0.286
0.452
0
1
1043
Muslim
0.203
0.403
0
1
1043
36
Table A-2: Believes Neighbors Know Turnout Choice
Dependent variable:
Believes Neighbors Know Turnout Choice
(1)
(2)
Including All Parishes
Excluding Some Parishes
0.03
0.01
(0.74)
(0.28)
0.07⇤
0.09⇤
(2.00)
(2.20)
0.6⇤⇤⇤
0.6⇤⇤⇤
(11.48)
(10.93)
Parish Fixed E↵ects
Yes
Yes
Observations
1129
654
Turnout reminder (T1)
Ink prime (T2)
Constant
⇤p
Notes: t statistics in parentheses.
37
< 0.05, ⇤⇤p < 0.01,
⇤⇤⇤ p
< 0.001
Table A-3: Full Model with Education, Age, and Parish Fixed
E↵ects
Turnout reminder (T1)
Ink prime (T2)
(1)
(2)
Baseline
Age & Education
Finger Status:
Finger Status:
Inked
Concealed
Inked
Concealed
-0.06
-0.31
0.05
-0.25
(-0.23)
(-1.08)
(0.19)
(-0.85)
-0.38
-0.40
-2.45⇤⇤
-2.42⇤
(-1.55)
(-1.47)
(-2.94)
(-2.56)
(0.15)
(3.68)
(1.39)
(3.82)
-0.05
-0.10
(-0.61)
(-1.11)
0.28⇤
0.32⇤
(2.18)
(2.16)
0.41⇤⇤⇤
0.21
(3.52)
(1.63)
0.53⇤
0.42
(2.12)
(1.54)
Educationa
Educationa * Ink prime
Ageb
Ageb * Ink prime
Constant
1.15⇤⇤⇤
0.44
0.33
0.42
(3.70)
(1.23)
(0.60)
(0.67)
Parish Fixed E↵ects
Yes
Yes
Observations
1124
1085
⇤
Notes: t statistics in parentheses.
p < 0.05,
⇤⇤
p < 0.01,
⇤⇤⇤
p < 0.001
a: The education variable had eight categories, where lowest was no education and
highest was university or post-university. Our mean respondent had an education level
between 4 (secondary not completed) and 5 (secondary completed).
b: Respondents were given a choice of completing the survey in English or Luganda.
38
0
5
Percent
10
15
20
25
Figure A-1: 2011 Turnout by District
.4
.5
.6
2011 Turnout
.7
.8
Data come from the Electoral Commission of Uganda. District level turnout calculated, the y axis displays the percent of districts with that turnout rate. The
vertical line indicates the confirmed turnout rate in the non-representative experimental sample (the “Finger Inked” category).
39
Table A-4: Museveni Voters in Kampala, Linear Probability Model
Probability of Voting for Museveni
Muganda
-0.091
(0.061)
Mutoro
-0.097
(0.195)
Munyankole
0.065
(0.109)
Agea
0.046⇤⇤
(0.019)
Educationb
-0.009
(0.017)
Female Respondant
0.035
(0.058)
Catholic
0.031
(0.114)
CUG
0.023
(0.115)
Pentacostal
0.218⇤
(0.131)
Muslim
-0.010
(0.122)
Parish Fixed
E↵ects
Yes
0.461⇤⇤
(0.189)
Constant
Observations
Residual Std. Error
Note:
340
0.596
⇤
p<0.1;
⇤⇤
p<0.05;
⇤⇤⇤
p<0.01
a: The education variable had eight categories, where lowest was no education and
highest was university or post-university. Our mean respondent had an education level
between 4 (secondary not completed) and 5 (secondary completed).
b: Respondents were given a choice of completing the survey in English or Luganda.
40
Table A-5: Attrition
(1)
(2)
Attrition
Attrition
Reminder
-0.913
-0.854
(T1)
(0.59)
(0.69)
Ink prime
-0.402
-0.239
(T2)
(0.49)
(0.55)
Restricted Sample
No
Yes
Parish FE
No
Yes
Notes: Recontact refusal rate: 1.44%, Recontact
failure rate: 0.7%, Total attrition 2.1%.
Rare events logit, omitted category is cell phone
treatment t statistics in parentheses.
⇤p
< 0.05, ⇤⇤p < 0.01,
⇤⇤⇤ p
41
< 0.001