Who Registers? Village Networks,Household Dynamics,and Voter Registration in Rural Uganda

Abstract

Who registers to vote? Although extensive research has examined the question of who votes, our understanding of the determinants of political participation will be limited until we know who is missing from the voter register. Studying voter registration in lower-income settings is particularly challenging due to data constraints. We link the official voter register with a complete social network census of 16 villages to analyze the correlates of voter registration in rural Uganda, examining the role of individual-level attributes and social ties. We find evidence that social ties are important for explaining registration status within and across households. Village leaders—and through them, household heads—play an important role in explaining the registration status of others in the village, suggesting a diffuse process of social influence. Socioeconomic factors such as income and education do not explain registration in this setting. Together these findings suggest an alternate theory of participation is required.

Keywords

African politics elections public opinion voting behavior representation electoral systems

Introduction

Who votes? While this is one of the most central questions in political science, the voluminous literature on voter turnout often glosses over an important prior question: who registers to vote? Research on voter registration has tended to take place in the context of wealthy democracies where voter registration is not automatic, particularly the United States. While some have suggested that the costs of registration are especially likely to affect those with less education and lower income, evidence that these individual attributes matter for voter registration is mixed (Hershey, 2009; Leighley & Nagler, 2013). A related strand of work examines the role of social ties—relational rather than individual attributes—in explaining variation in political participation. This work has demonstrated that social ties matter for voter turnout, a relationship observed in both high-income (Rolfe, 2012) and low-income settings (Eubank et al., 2021). However, we know little about whether or how these ties also matter for the prior step of voter registration.

While knowledge gaps exist with respect to voter registration in rich democracies, even less is known in lower-income countries. A recent study suggests that individual attributes such as poverty might make constituents more sensitive to registration costs (Harris et al., 2021), but basic descriptive facts about who registers to vote, and importantly, who remains unregistered, are elusive. A vexing problem is the lack of reliable data about the population of adults eligible to vote—a challenge that is particularly severe in lower-income countries where census data are collected infrequently and are of variable quality (Lee & Zhang, 2017).

In this paper, we combine census and official voter register data as well as extensive original survey data from 16 villages in rural Uganda to examine how and whether individual and relational attributes are associated with registration status.¹ Through an in-person survey with all eligible voters, we collect both individual-level demographic information (e.g., gender, education, and age), as well as social network data, which allows us to measure how village residents are related to one another across various domains of interest.

We report two sets of findings. First, there is a limited role for individual attributes in explaining variation in registration status in this setting. Specifically, factors such as gender, education, and income are not significantly correlated with registration status in rural Uganda. Among individual-level factors, only age, being a head of household, and holding a leadership position in the village are clearly associated with registration status, and the latter two are arguably relational attributes. We find that there is a curvilinear relationship between age and registration status. Villagers who are household heads and those who hold leadership positions (for example, elected local leaders, or traditional and religious leaders), are also more likely to be registered to vote.

Second, we find that social ties are important, both within and across households, suggesting a diffuse process of social influence. Within households, irrespective of gender, household heads strongly influence the registration status of other adult household members. Across households, individuals who are closer to the village head and those who occupy a central position within the village network are more likely to be registered. In our study area, village heads are typically members of political parties and thus can mobilize potential voters on their behalf. We find suggestive evidence that registration status is influenced not only through direct but also indirect ties to village heads. Finally, those with ties to non-family lenders are more likely to be registered themselves. These findings show that relational ties are important in explaining registration status, but that ties to multiple types of individuals matter. Our findings further suggest that those individuals who are likely to be party brokers—village heads—affect the voter registration process, but this process likely involves both indirect and direct influence.

Our study contributes sorely needed descriptive facts to a nascent literature exploring voter registration status, rather than turnout, in lower-income countries (Harris et al., 2021; Ichino & Schündeln, 2012; Mvukiyehe & Samii, 2017). It also complements a growing body of work examining the role of relational attributes in explaining political behavior. Existing work in this latter domain has examined the role of social networks in voter turnout (Eubank et al., 2021; Nickerson, 2008; Rolfe, 2012), but not for voter registration, a behavior for which there may be a distinct process of social influence.

Existing studies of registration in high-income settings on the one hand, and of voter turnout in lower-income settings on the other, may not necessarily be informative about who is missing from voter register in the lower-income, rural settings that characterize much of the electorate around the world. Indeed, due to the challenges of accessing reliable population data and social desirability bias in registration status self-reports, we are most likely vastly overestimating voter registration in lower-income countries and inadvertently turning a blind eye to those citizens who are absent from electoral politics altogether. Knowing who is not able to participate on election day—because they were never registered—is key to understanding who is represented through electoral politics.

Determinants of Registration: Theory and Evidence

Our study builds on two seminal bodies of work that explain political behavior: one focusing on individual attributes and the other on relational attributes and social networks.

Individual Attributes and Political Behavior

Following Downs (1957), an extensive literature examines how individual attributes influence both the motivation and ability to vote, including factors affecting the cost of political participation. A resource model of political participation suggests that those with fewer resources in terms of time, money, and civic skills will be less likely to participate, and that these resources are correlated with socioeconomic status, especially education and income (Brady et al., 1995). Other work examines additional individual attributes that may correlate with resources, particularly membership within marginalized groups, such as women or religious or ethnic minorities (Grossman et al., 2014). On the basis of this voluminous body of work, there are reasons to believe these individual attributes matter in explaining voter registration status.

However, the role of these factors in explaining political participation, and particularly voter turnout, varies within and across countries, and therefore may operate differently in lower-income (Kasara & Suryanarayan, 2015) and less democratic contexts (Croke et al., 2016). Further, relatively little research has examined whether these factors matter for registration specifically, a process which typically takes place prior to election day and a behavior for which there are no immediate benefits.

Historically, research on the relationship between individual attributes and political behavior has focused disproportionally on the United States, and in this context has repeatedly revealed a strong positive association between political participation and both education and income. However, recent comparative work has demonstrated that the role and relative importance of sociodemographic factors vary across countries, and thus that the relationship between individual attributes and political participation is conditioned by contextual and institutional factors. For example, Gallego (2014) found that while income and education are highly correlated with turnout in the United States, this is not the case in other wealthy democracies like Spain, Denmark, and South Korea. Similarly, Nevitte et al. (2009) found that age is correlated with voting in three-quarters of the 33 elections examined, education in about half, and income in only about one third. The sample was comprised of mostly middle- and high-income countries for which survey data were available.

Research in low- and lower-middle income countries further suggests that factors such as education and income either matter less for political participation or operate differently in poorer or less democratic settings. Isaksson et al. (2014) found that across 20 African countries, individuals who are poorer are, if anything, more likely to vote, cutting against the resource model of political participation. Similar patterns have been found in India and Latin America (Krishna, 2008).

Attributes such as education may also operate differently in less democratic contexts as compared to consolidated democracies. While education is often found to be positively associated with voting in democratic contexts, more educated individuals may deliberately avoid turning out to vote in autocracies as a form of dissent (Croke et al., 2016). In authoritarian settings, the less educated may also be more easily influenced by vote buying (Blaydes, 2010).

The extent to which findings from the voluminous literature on voter turnout apply to the prior step of voter registration is poorly understood and also contested. Here again, much of the research focuses on the case of the United States, where there exists great variation in laws governing the registration process across states. Where it is not automatic, registration, like turning out to vote, involves costs, which are weighed against the benefits of political participation. It is conceivable, therefore, that those with fewer resources will be less likely to register, and more likely to respond to policy changes that reduce the cost of registration. However, it is challenging to gather reliable individual-level data on unregistered voters, and there is only mixed evidence that reducing the cost of registration disproportionately benefits the poorer or less educated.

Grumbach and Hill (2021) demonstrate that the introduction of election day registration (EDR), which is designed to simplify the registration process, leads to higher registration rates and turnout, especially among young Americans, but Leighley and Nagler (2013) found that EDR expansion did not increase the relative share of low-education individuals among voters. Likewise, Nagler (1991) found that restrictive registration laws were associated with lower turnout, but this effect was similar for high- and low-education constituents. While such studies of registration are rare, their findings are consistent with the much larger literature on voting behavior in American politics, which generally finds mixed evidence as to whether innovations that reduce voting costs affect voter turnout. For example, universal vote-by-mail increased overall turnout (Thompson et al., 2020), and this increase was found to be especially pronounced among low-propensity voters like less educated, minority, and young voters (Bonica et al., 2021). However, some earlier studies found that convenience voting is, in general, less helpful for low-propensity voters than for those who were already likely to vote (Berinsky, 2005).

Turning to the context of lower-income countries, Harris et al. (2021) found that an intervention reducing the cost of registration in Kenya—conducting mobile registration rather than requiring citizens to register in-person—increased registration, particularly in poorer communities. However, this analysis was conducted at the aggregate rather than individual level. It could be that potential voters are more sensitive to registration costs in poorer areas. However, it could also be that those in lower-income settings are more sensitive to clientelistic strategies, such as turnout buying (Nichter, 2008). Indeed, how and whether brokers mobilize voters for voter registration, as compared to better-documented voter turnout, is unclear, a point to which we return below.

Together this body of work provides a set of suppositions about individual sociodemographic attributes that might matter for voter registration. There are reasons for skepticism, however, both because of the distinct way these attributes might operate in poorer and less democratic settings, and because voting theories might not apply to the act of registration. Moreover, some of the predictions about how individual attributes such as education and income matter for registration are cross-cutting.

Social Ties and Political Behavior

The second strand of work we build on examines the role of social ties and social influence in explaining political behavior (Eubank et al., 2021). Social networks could matter in three ways that are relevant for voter registration: information, influence and mobilization. As for information, citizens must learn to follow a set of rules about how and when to register. Connections to politically knowledgeable individuals can reduce information search costs. As for influence, according to social context theory, citizens are more likely to participate politically when they observe sufficient levels of participation among their peers (Rolfe, 2012; Siegel, 2009). Finally, participation might increase when one is connected to mobilizers with a strong interest in making sure that many others vote (Nichter, 2008). This literature focuses on within-household (Nickerson, 2008) or within-family dynamics (Cruz et al., 2017), as well as on ties to various types of brokers (Stokes et al., 2013).

Brokers and other party operatives are thought to play an especially important role in mobilizing potential voters. Party brokers can facilitate political participation in numerous ways, typically by reducing the costs and/or by increasing the benefits of participation for individuals (Stokes, 2005) or communities (Rueda, 2015). For example, they can provide transportation to polling or registration places, they can distribute information about candidates and parties, and they can distribute resources in exchange for turning out to vote, with the goal of affecting voting behavior (Brierley & Kramon, 2020; Larreguy et al., 2016). Thus, the nature of ties to village brokers may be an important factor in explaining variation in registration status.

Often, brokers are thought to directly influence voters, and are assumed to have a high degree of information about them (Stokes, 2005). This implies a model of broker behavior that involves brokers directly influencing many of the voters within their sphere of influence. In the context of Uganda, a recent study found that brokers are well-known figures in a given community (Blattman et al., 2019). Brierley and Nathan (2021) found, however, that in the context of Ghana, party brokers are not necessarily well-connected to voters, nor do they have more information than the average voter. This work suggests that brokers may need to work indirectly to maximize their influence on voting outcomes.

Indeed, a strategy that involves directly reaching out to all intended targets may be inefficient. If brokers can influence key individuals who then influence their friends, family, and other social ties, brokers will be able to reach far more voters. In other words, party brokers may themselves need intermediaries, and the relational process of political mobilization may be more diffuse than previously conceptualized. By conducting a complete census of the villages in the study area, we are able to examine whether and which social ties matter for voter registration, and in so doing shed additional light on the role brokers may play in this setting.

Empirical Challenges in the Study of Registration

Much of the aforementioned research on both individual attributes and relational factors in shaping political participation focuses on turnout rather than the prior step of voter registration. One major reason for the dearth of research on registration is that accessing accurate data on registration rates, much less individual-level correlates of registration, is difficult, and particularly so in lower-income country settings.

The calculation of voter registration rates requires two inputs: (a) the number of registered voters, and (b) the number of people who are of voting age in a given year. Reliable estimates for both of these input measures are often not readily available.

First, the denominator, voting-age population, is best estimated using census data. However, census data are typically collected every 10 years at best and can be of questionable quality. Estimation of adult population figures between census years requires interpolation, which itself entails a number of assumptions. For these reasons, the reliability of voting-age population estimates in a given election year is questionable even in the best of conditions (Radcliff, 1996). Subnational measures of registration rates are even more difficult to calculate, as they require additional assumptions about rates of population growth and migration across subnational units. Estimating the voting-age population in a given year at the subnational level is especially problematic for a country like Uganda, with a high population growth rate, variable growth rates across regions, and substantial levels of internal migration and urbanization.

Estimating the number of registered voters, the numerator, is even more problematic than estimating the voting-age population. First, most countries do not make voter registers available to the public. Second, even when they are available, registers are likely to be outdated: they may include individuals who have died or moved away from a given locality, or they may fail to capture those who have recently moved into a locality. Even in high-income countries like the United States, where one can obtain well-curated individual-level voter files and a wealth of individual-level data collected about registered voters from marketing firms, individual-level census data are unavailable for reasons of privacy. As a result, we know very little about the unregistered individuals who do not appear in the voter file or accompanying commercial data sets (Jackman & Spahn, 2015).

In the absence of accurate measures of the either the numerator or the denominator, it is difficult or impossible to calculate an accurate voter registration rate even at a national level, much less identify or study those who are unregistered at the micro level. This problem may be particularly acute in lower-income countries. For example, Kuenzi and Lambright (2007) show that for one-third of African countries in their sample, the number of official registered voters is larger than the estimated voting-age population, suggesting either wide-spread inflation of voter registers, underestimates of population counts, or both. The result is that the “true” registration rate is unknown. Worse, without knowing which figure is more or less accurate, even the direction of potential bias in this figure is unknown.

Consider, for example, the case of Uganda. In an exploratory exercise, using 2014 census data and 2016 election results published at the parish level, we found that a quarter of the 7500 parishes in Uganda exhibited a registration rate of 100% or greater.² This was the case even after increasing the voting-age population to account for voters who came of age between the census and voter registration period. This exploratory exercise underscores the difficulty of using official records to estimate voter registration and the necessity of alternative strategies if we are to empirically address the question of who registers.

How can these challenges be overcome? In this paper, we collect original survey data from a complete census we conducted in 16 rural villages in Uganda, and merge it with the official voter register in these villages.³ Specifically, we match individuals in our village census to the voter register, which provides a major advantage, allowing us to advance the empirical examination of determinants of voter registration.⁴ Most importantly, our measure of voter registration is behavioral rather than self-reported. An individual in our sample is coded as registered if and only if her name exists in the final certified register itself. Most existing measures of both turnout and registration in similar contexts rely on self-reports, which typically vastly overestimate these behaviors.⁵

By combining our own village census with the official voter register we are able to calculate a more accurate registration rate than is possible using official data or survey data in isolation.⁶ To the best of our knowledge, ours is the first study to merge a complete bespoke census with a large voter registration database, allowing for the examination of a rich set of covariates not only for registered voters, but crucially, for unregistered voters as well. With these data in hand, we are able to contribute to a better understanding of the variation in individuals’ voter registration status in a low-income country setting, examining two types of explanations that have been posited in the existing literature: individual factors such as SES, and relational factors such as social ties.

Context

We conduct our study in Uganda, a predominantly rural, low-income country governed by a dominant-party regime. Dominant-party regimes comprise an increasing share of regimes globally and are thus substantively important to investigate. Uganda’s ruling party, the National Resistance Movement (NRM), has been in power since 1986, and has maintained wide pockets of popular support over the past three decades, especially in rural areas.

Elections under the NRM have not been fully free and fair—the NRM has on multiple occasions resorted to the manipulation of state resources, intimidation, and politicized prosecutions of opposition leaders (Izama & Wilkerson, 2011), and in the most recent election, the alleged abduction of hundreds of suspected opposition supporters (Human Rights Watch, 2021). However, irregularities in vote tallies are typically concentrated in a relatively small number of polling stations, usually NRM strongholds or areas that are difficult for other election observers to access (Callen et al., 2016; Klimek et al., 2012). Overall, turnout rates in general elections range between 60 and 70% of registered voters.⁷

Existing research has documented several turnout strategies in Ugandan elections. For example, candidates use both targeted benefits and promises of group benefits to increase electoral support (Grossman & Lewis, 2014). Political parties in Uganda also make use of paid local brokers, who are often well-known figures in the area. According to Blattman et al. (2019), virtually all Members of Parliament in Uganda employed brokers in the 2016 parliamentary elections. However, unlike some other contexts (Brierley & Nathan, 2021), brokers are not selected by a formalized process. As such, much remains unknown about who these brokers are and how they operate in practice. Further, relatively less is known about the mobilization strategies employed for voter registration, which takes place months before a given election, rather than turnout, which takes place over a shorter period of time and where there are typically many more domestic and international observers present.

In Uganda, like many countries, the burden of registering to vote falls on individuals, as there is no automatic voter registration. At the time of our study, to register to vote, Ugandans were required to travel to a registration center and present proof of citizenship and age (18 years and older).⁸ Apart from travel costs and waiting time, registration to vote is free of charge.⁹ Voter files in Uganda are carried from previous elections, and citizens are encouraged to check their registration status prior to upcoming elections.¹⁰ Voter files are updated prior to elections to remove those who passed away, update the information of those who have migrated and would like to change the location of their assigned polling station, and to add new voters. The register for the 2016 election was compiled and cleaned between 2014 and May 2015, with a final display of the register at each polling station in July and August 2015.¹¹ We conducted our village censuses in March and April 2016, shortly after the February 2016 general election.

Figure 1 shows the registration rate across the 16 study area villages, compiled by combining our village survey data with the official voter register for Arua district, where all the villages are located. Arua is a relatively rural district, located in the northwest of the country, bordering the Democratic Republic of Congo to the west and just over 400 km by road from the country’s capital. Average official district-level turnout in Uganda during the 2016 elections was 67%, and for Arua district was 59%. Average district-level vote share for the incumbent president was 64% nationally and 58% in Arua district. Our estimates of village-level voting patterns indicate that the majority of sampled villages are NRM leaning, but are not ruling party strongholds.¹² The median estimated vote share for the incumbent president in the 2016 elections was 60%. This suggests that party brokers may have reasons to try to be selective in their mobilization strategies, rather than targeting all village members.

The mean voter registration rate across villages in our sample is 69%, indicating that there is still a substantial portion of the adult population unregistered. Village-level census data beyond Arua were not available to us, but we compare parish-level census data in the parishes where the sampled villages are located to parishes in the rest of rural Uganda, excluding cities and municipalities. The parishes in which the study was conducted are similar to other rural parishes in most respects, including population size, educational attainment, access to electricity, reliance on subsistence farming, and prevalence of malnourishment, as shown in Supplementary Material Figure SI-1. Sampled parishes are somewhat more likely to have thatched roofs as compared to iron sheets, which could be an indicator of relatively lower economic status, and have a slightly higher share of women.

Importantly, we find that within-village variation in registration rates is much larger than across-village variation: the intraclass correlation is 0.007% (full sample) and 0.011% (for households (HHs) of size 2+). We thus focus our inquiry on within-village variation, using village fixed effects in all regression models to absorb village characteristics that could be correlated with voter registration status.

Data

Next we describe the data and estimation strategy we use to explain within-village variation in voter registration status in rural Uganda. Our key dependent variable is an individual’s registration status, which is a binary variable that equals one if the respondent is registered to vote in the 2016 elections, and zero otherwise.

While studies of registration are rare, past research suggests that some of the individual- and household-level characteristics associated with turnout in middle- and high-income countries are also important in lower-income countries. These include income (Kasara & Suryanarayan, 2015), age (Kuenzi & Lambright, 2011), gender (Robinson & Gottlieb, 2021), education (Croke et al., 2016), and pro-sociality (Blattman, 2009). By focusing on the registration phase, we explore the possibility that these factors operate in part through their association with voter registration. While we are able to accurately measure registration status, we do not have information on whether an individual voted on election day, as this information is not publicly available in Uganda.

We measure the following individual-level covariates: age is a continuous measure of the respondent’s age in years; female is an indicator of the respondent’s sex; secondary education is a binary variable that equals 1 if the respondent attained at least secondary education and 0 otherwise; Catholic is a binary variable that equals 1 if the respondent identifies as Catholic, which is the dominant religion in our study area, and 0 if she identifies with another religion (usually Protestant); Lugbara is a binary variable that equals 1 if the respondent identifies as Lugbara, which is the dominant ethnic group in our study area,¹³ and 0 if she identifies with another ethnic group; and income is a subjective 5-points categorical measure of wealth ranging from 1 (low) to 5 (high).^14,¹⁵

We also measure individual covariates that capture involvement in communal affairs. These include: leader, a binary variable that equals 1 if the respondent occupies a leadership position within the village and 0 otherwise;¹⁶ participation, a continuous summary index aggregating across various recent non-electoral forms of communal engagement;¹⁷ and pro-sociality, a behavioral proxy-measure of care for the community, measured as the share of a 10 monetary units endowment contributed in a standard dictator game that respondents played with an anonymous same-district recipient.

In addition, we measure two location-based covariates. Distance (km) is a continuous measure of the Euclidean distance in kilometers between respondent’s geocoded dwelling and the nearest voter registration center. Geography captures indirect social influence using a spatial lag that counts the number of registered voters within the village besides individual i, and assigns less weight to those who reside farther away from that individual.¹⁸

Finally, we include three household-level factors: household head is a self-reported binary measure that is equal to 1 for the head of the household and 0 for all other household members; degree HH, which is a count variable of the number of adults within a single household, and % registered HH, which is our proxy measure of peer effects at the household level. Households are as defined by local leaders with whom we worked to conduct the household listing and enumeration. Figure 2 shows the distribution of household sizes in the study villages and their registration rates, where the size of the household is determined by the number of adults living in the household.

To explore the possibility that registration is driven by network ties beyond the household, we rely on a network module we administered to all village residents as part of the census. This module—administered as part of a longer survey that also included modules on respondent demographics and household characteristics—included a set of questions about four different types of social ties respondents may have with other co-villagers. We use a simple name generator technique (Knoke & Yang, 2008) to elicit information on four kinds of relationships: (1) family ties, (2) friendship ties, (3) lenders: to whom they would go to borrow money, and (4) problem solvers: to whom they would go to solve a problem regarding public services in the village. This technique involves asking respondents for up to five names of those individuals whom they consider social ties in each of these categories.¹⁹ Through this exercise, it could be that i named j as a peer, while j did not. In other words, this exercise gives us four “directed” networks, that is, networks that allow for non-reciprocal ties. We then collapse those directed networks into undirected networks; that is, networks in which i and j are either tied or not. To do so, we say that there is a tie between i and j if at least one named the other as a peer.

Because network concepts have simpler definitions with undirected networks, these networks will be the main focus of our analysis. However, we also report results in which we consider directed networks. Comparing household roster data with network surveys, we believe we reached over 80% of village residents. Following standard practice (Larson & Lewis, 2017), individuals who did not complete a network survey were dropped from the analysis (see Supplementary Material Section A for details). Table 1 provides descriptive information for the variables used in the empirical analysis.

Table 1.

Descriptive Statistics. Column 2 (“Sample”) reports descriptive statistics for all members of households of size 2 and above. We use this sample in our regression analysis below. The column Δ₀ reports the difference between the columns “HH Size 1” (those excluded from the empirical analysis) and “Sample.” The column Δ₁ reports the difference between the columns “HH Head” and “Non-Head,” among households of size 2 and above. Standard errors are clustered at the village level. *p < 0.1; **p < 0.05; ***p < 0.01

Variable	HH Size 1	Sample	Δ₀	HH Head	Non-Head	Δ₁
Dependent variable
% Registered	0.701	0.696	0.005	0.78	0.639	0.141***
Individual
% Female	0.72	0.526	0.194***	0.163	0.772	−0.609***
Age	39.322	37.02	2.302**	43.247	32.809	10.439***
Income	2.394	2.586	−0.191***	2.48	2.657	−0.178**
% Secondary education	0.216	0.235	−0.019	0.309	0.184	0.125***
% Catholic	0.629	0.581	0.048	0.561	0.594	−0.033
% Lugbara	0.948	0.946	0.002	0.942	0.949	−0.007
Pro-sociality	0.202	0.201	0.001	0.213	0.193	0.02**
Leader	0.125	0.151	−0.026	0.24	0.091	0.148***
Participation index	−0.076	0.037	−0.113***	0.194	−0.069	0.263***
Distance to registration center (km)	0.847	1.017	−0.17**	0.952	1.062	−0.11**
Network
Degree centrality	12.34	17.75	−5.409***	23.843	13.629	10.214***
Closeness centrality	0.002	0.002	0	0.003	0.002	0***
Clustering coefficient	0.421	0.371	0.05***	0.333	0.396	−0.063***
Eigenvector centrality	0.128	0.169	−0.041***	0.221	0.134	0.088***
Sample size
N	773	2184		881	1303

Estimation

To explore the relationship between registration status, which is our outcome of interest, and the above individual attributes and relational ties, we fit the following linear probability model

y_{i j} = α_{j} + x_{i}' β + f_{j} (i)' γ + ϵ_{i j}

with y_ij a binary variable that equals 1 if respondent i in village j is registered, and equals 0 otherwise. The vector x_i is a vector of individual attributes. With N_j the set of respondents in village j, the function

f_{j} : N_{j} \to ℝ^{k}

maps to each respondent i ∈ N_j a vector of network statistics derived from the network of village j. As such, while parameters β capture the association between individual attributes and registration, parameters γ capture the association between registration and network attributes. Among those attributes, two are particularly important: degree, the number of peers of respondent i, and the percentage of registered peers. Because the dynamics of registration may depend on unobserved village-level characteristics, we perform within-village comparisons and include a village-level fixed effect α_j. We cluster standard errors at the village level and use a wild bootstrapping procedure due to the small number of clusters.

Note that in this specification we use the percentage of registered peers rather than the number of registered peers, which amounts to estimating a relative threshold model instead of an absolute threshold model (Centola & Macy, 2007). Indeed, because they only count registered peers, absolute threshold models implicitly assume that only such peers influence outcomes. Conversely, because they consider the share of registered peers, relative threshold models implicitly assume that both registered and non-registered peers may influence the outcome. Theoretically, relative threshold models better capture the dynamics of influence that may underlie registration, since non-registered peers may also influence an individual. As a robustness check (Supplementary Material Table SI-6), we re-estimate our main results using an absolute threshold model instead and find that both measures produce comparable results and fit the data to a similar extent.

Note that this specification cannot be given a causal interpretation. This is especially the case for network parameters γ, which may suffer from three sources of bias: homophily, contextual confounding, and the reflection problem (Aral et al., 2009). Nonetheless, the large number of individual controls included in x_i, as well as the village-level fixed effect α_j, help to alleviate at least some concerns about omitted variable bias. While this exploratory exercise is admittedly correlational, it aims to answer a basic and important question—who registers?—with a goal of generating hypotheses that can inform future research.

Results

We begin by presenting our findings with respect to a set of individual attributes that have been found to be associated with political participation in a variety of contexts (Table 2, columns 1–4).

Table 2.

Determinants of Individual Voter Registration Across 16 Ugandan Villages. Model 1 considers the full sample. Models 2 to 7 focus on households of size 2 and more, for which within-household peer effects are well-defined. Within these households, models 3 and 6 focus on household heads, while models 4 and 7 focus on non-head members. All models include village-level fixed effects.

	Dependent Variable: Registered
	All	All (2+)	Head	Non-Head	All (2+)	Head	Non-Head
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Female	0.017	0.008	0.016	0.000 00	0.014	0.027	0.005
Female	(0.027)	(0.028)	(0.035)	(0.033)	(0.029)	(0.037)	(0.035)
Age	0.015***	0.018***	0.002	0.028***	0.018***	−0.000 1	0.028***
Age	(0.003)	(0.005)	(0.007)	(0.006)	(0.005)	(0.007)	(0.006)
Age²	−0.0002***	−0.0002***	−0.000 04	−0.0003***	−0.0002***	−0.000 02	−0.0003***
Age²	(0.000 04)	(0.000 1)	(0.000 1)	(0.000 1)	(0.000 1)	(0.000 1)	(0.000 1)
Income	−0.006	0.002	0.026**	−0.010	0.001	0.027**	−0.011
Income	(0.007)	(0.007)	(0.011)	(0.009)	(0.006)	(0.011)	(0.009)
Secondary education	0.021	0.025	0.011	0.054	0.024	0.017	0.049
Secondary education	(0.023)	(0.024)	(0.034)	(0.041)	(0.023)	(0.032)	(0.041)
Catholic	0.032	0.037	0.041	0.022	0.037	0.038	0.021
Catholic	(0.040)	(0.044)	(0.060)	(0.044)	(0.042)	(0.061)	(0.041)
Lugbara	0.076	−0.018	0.029	−0.022	−0.016	0.040	−0.025
Lugbara	(0.047)	(0.050)	(0.066)	(0.080)	(0.050)	(0.068)	(0.086)
Leader	0.052**	0.035	0.011	0.092**	0.036	0.010	0.097***
Leader	(0.026)	(0.024)	(0.034)	(0.037)	(0.024)	(0.034)	(0.035)
Pro-sociality	−0.012	−0.016	0.009	−0.042	−0.013	0.015	−0.037
Pro-sociality	(0.056)	(0.048)	(0.057)	(0.083)	(0.045)	(0.056)	(0.080)
Participation index	0.020	0.025	0.003	0.037	0.028	0.003	0.040
Participation index	(0.014)	(0.018)	(0.028)	(0.025)	(0.017)	(0.029)	(0.025)
Distance (km)	0.011	0.018	−0.003	0.035*	0.014	−0.009	0.029
Distance (km)	(0.012)	(0.016)	(0.014)	(0.020)	(0.016)	(0.014)	(0.020)
Geography	−0.005	−0.002	−0.007	0.003	−0.006	−0.013	−0.001
Geography	(0.007)	(0.007)	(0.009)	(0.008)	(0.008)	(0.010)	(0.008)
Household head	0.074***	0.082***			0.095***
Household head	(0.023)	(0.026)			(0.028)
degree HH					0.008	0.022	0.008
degree HH					(0.005)	(0.014)	(0.006)
% Registered HH peers					0.061*	0.089***	0.089**
% Registered HH peers					(0.035)	(0.032)	(0.039)
Observations	2753	1940	785	1155	1940	785	1155
R²	0.049	0.065	0.033	0.094	0.069	0.043	0.099

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

We find little evidence that factors such as education, income, gender, or membership in a minority ethnic or religious group are associated with registration status, casting doubt that resource-based theories of political participation apply to the behavior of registration in this context as they do in rich democracies.

Note, however, that some individual attributes do matter for registration status, including age, being a household head, and holding a leadership position within the village. Regarding age, we find a curvilinear relationship with registration (Figure 3), though the small sample of older respondents means that we should be cautious in interpreting the relationship between age and registration status among this population. It does seem, however, that those above 25 are significantly more likely to be registered than those between 18 (the legal age for registration) and 25. This finding is consistent with a robust curvilinear relationship between age and turnout observed in high-income consolidated democracies (Bhatti et al., 2012).

We estimate that being the head of a household increases the likelihood of being registered to vote by 8 percentage points. Those who hold leadership positions (as defined above) are also more likely to be registered to vote, though this relationship seems driven primarily by those individuals who hold a leadership position but are not household heads, a relatively small group. Still, among this group the effect is quite large, 9.2 percentage points, as shown in Table 2, column 4.

As mentioned, we do not find evidence that traditional proxies of marginalization at the individual-level—such as being female, relatively poor, a member of local religious minority (i.e., not being Catholic) or local minority ethnic group (i.e., not being Lugbara), or living at the village periphery²⁰—matter for registration status among in our rural sample (Table 2, columns 1–4). Similarly, having attained secondary-level education does not correlate with registration status. Importantly our null finding for education is likely not a function of how education attainment is coded.²¹ An examination of the bivariate correlations between these attributes and registration status suggests that multicollinearity does not explain the absence of significant relationships observed in the multivariate model (Supplementary Material Table SI-5). Our index of political participation is not associated with registration in the regression analysis, but it does exhibit a high degree of correlation in the bivariate analysis. The lack of a clear relationship between marginalization and registration are important when considering that past work has demonstrated that these marginalization factors do matter for voting and other forms of participation in Uganda (Grossman et al., 2014).

We urge caution in the interpretation of this latter set of findings, however, for a few reasons. First, it could be that the effects of an attribute like income or education are cross-cutting. For example, low levels of education, particularly when proxying illiteracy, could make it more difficult to register to vote in practical terms, but on the other hand, it could be that those with more education are less likely to engage politically in a non-democratic context because they view such efforts as futile (Croke et al., 2016). Thus, it could be that education does indeed have causal effects on registration, but that these effects cancel each other out. The same logic applies to income (registration costs could reduce, but vote buying increase the propensity of registration for poor constituents).

Second, it could be that there is insufficient variation in either education or income for us to pick up an association between these attributes and registration, but that one would emerge if we included places with greater variation, particularly urban areas. Third, individual attributes may have an effect on registration for some sub-samples of potential voters. For example, when focusing on the sample of household heads, we find some evidence of an income effect: a one unit increase on the 5-point categorical income scale is associated with 2.5 percentage point increase in the probability of registration (Table 2, column 3).

Together, our findings on individual attributes suggest that existing models of political participation—especially resource-based models—do not find strong support in our examination of voter registration in this rural setting. This does not mean that these attributes are entirely irrelevant, but they may matter less for the outcome of registration than for turnout, less in rural Uganda than in more urban or wealthy settings, or perhaps operate differently in lower-income or less democratic settings.

Social Ties and Voter Registration: A Diffuse Process of Influence

Next, we turn to our findings with respect to a set of relational factors. Our results provide evidence that social ties matter and suggest a pattern of diffuse social influence in shaping registration behavior. Specifically, we find evidence of peer effects within and across households, as well as evidence that social ties to some notable individuals, who we believe are most likely to be political brokers, shape registration status.

Within-household peer effects

We focus on households with at least two adult members, since the notion of within-household peer effects is not defined for households of size 1. Above, we found that household heads are more likely than non-heads to be registered. Here, we find that whether the head of household is registered is positively associated with the registration status of other household members, suggesting a role of within-household peer effects, and indicating that there are clusters of registered and unregistered households.

First, we note that voter registration increases dramatically with the share of other household adult members being registered (Table 2, column 5). For both household head (column 6) and non-household head members (column 7), moving from no other household members registered to all registered increases registration by about 8.9 percentage points when accounting for household size (degree HH), as is customary. In sum, within-household peer effects on political engagement are consequential.

Second, we investigate how much influence is exerted by different household members (Table 3). To do so, we focus on non-head members in households with at least 2 adult members (that is to say, the full sample of multi-person households), and consider first the role of any member (column 1), then the role of the household head only (column 2). Considering column 2, we find that if the household head is registered to vote, this increases the likelihood that other non-head adults are registered from 0.57 to 0.67 (9.6 pp, or 17%), a magnitude that is comparable to the joint effect of any member (column 1).

Table 3.

Effect of Household Head. This table only considers non-head members. Column 1 reproduces column 7, Table 2 to facilitate comparisons. All models include village-level fixed effects and the controls discussed in Data.

	Dependent Variable: Registered
	Size 2+	Size 2+	Size 3+	Size 3+	Size 3+
	(1)	(2)	(3)	(4)	(5)
Degree HH	0.008	0.011**	0.006***	0.008***	0.008***
Degree HH	(0.006)	(0.005)	(0.002)	(0.002)	(0.002)
% Registered	0.089**		−0.031
% Registered	(0.039)		(0.108)
Head registered		0.096***		0.058	0.061
Head registered		(0.033)		(0.047)	(0.051)
% Registered non-head					−0.042
% Registered non-head					(0.091)
Controls	✓	✓	✓	✓	✓
Observations	1155	1155	586	586	586
R²	0.099	0.102	0.129	0.131	0.132

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

Focusing on large households (3 adult members or more; columns 3–5) allows us to evaluate more directly the finding that household heads convey disproportionate influence, by comparing the influence of the household head against that of non-head household members. Note that the sample size is rather small for households with three or more adult members. Although results are not statistically significant, we replicate our findings from the full sample of multi-person households (Table 3, columns 3 and 4). In Table 3, column 5, we estimate both effects jointly and find that there is a large and positive coefficient, though not statistically significant, for the variable indicating whether the household head is registered, but negative for the percent of non-heads who are registered. This suggests that within-household registration dynamics are driven by household heads, and not by non-head members.

Network centrality and registration status

In addition to examining the role of within-household peer effects, the unique nature of our network data allows moving beyond within-household peer effects, and exploring the relationship between individuals’ position in their village network and registration status.

First, we find that individuals with a greater number of connections in the community are more likely to be registered: in Table 4, the coefficient for “degree”—that is, the number of peers that a respondent has—is always positive and significant. The first column in Table 4 is our within-household model, and “degree” simply refers to the size of the household. Columns 2 through 6 consider outside-household peer effects, first considering any kind of tie to a non-household member (column 2), and then, respectively, the respondent’s reported ties to family, friends, lenders/borrowers, and people that respondent would turn to, or that would turn to that respondent, to solve a problem.²² The coefficients for “degree” in columns 2 through 6 indicate that no matter which types of ties we consider, those who are more connected to others in the community are more likely to be registered. We demonstrate in the appendix that this finding is robust to using alternative ways of capturing an individual’s network centrality, such as closeness, clustering and eigenvector centrality (see Supplementary Material Table SI-8, and Supplementary Appendix A.1 for a glossary of those measures).

Table 4.

Comparison of Within-Household Dynamics with Other Possible Peer Effects. Sample is restricted to household with size 2 and above. Definitions of ties can be found in the Supplementary Material Section A.3. All models include village-level fixed effects and the controls discussed in Data.

	Dependent Variable: Registered
	Household	Non-HH	Family	Friends	Lender	Solver
	(1)	(2)	(3)	(4)	(5)	(6)
Degree	0.008	0.004***	0.016***	0.033***	0.014***	0.001**
Degree	(0.005)	(0.001)	(0.002)	(0.003)	(0.002)	(0.001)
% Registered peers	0.061*	−0.028	0.041	0.038	0.079***	0.032
% Registered peers	(0.035)	(0.068)	(0.032)	(0.036)	(0.026)	(0.048)
Controls	✓	✓	✓	✓	✓	✓
Observations	1940	1940	1940	1940	1940	1940
R²	0.069	0.078	0.086	0.113	0.082	0.067

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

It is not surprising that better-known individuals, who occupy relatively central positions in the village social network, are more likely to be registered to vote, and perhaps more politically active in general. Yet full network data of this kind are rare, and we know of no other study documenting the relationship between network position and voter registration. Moreover, this finding also serves as an important reality check on both our outcome measures, as well as our measures of network ties.

Second, we find that some outside-household social ties are more consequential than others. The models presented in Table 4 also include the “percent of registered peers” for the respondent in each type of network. The coefficients are indistinguishable from zero for the family, friends, and problem-solver networks. While individuals who are central in these networks are more likely to be registered, we do not see evidence of influence such that individuals with larger shares of registered peers are more likely to be registered.

However, we do find strong evidence of such influence for one specific type of network—that of lender-borrower relationships. In fact, a comparison of columns 1 and 5 in Table 4 indicates that this type of peer effect is similar in magnitude to the rather large within-household peer effects described above.²³

If politicians are relying on local intermediaries to mobilize registration among likely allies, we would expect to see influence flowing not necessarily through friends or family members, but through a class of influential and politically connected community members. Below, we consider more formally the possibility that those identified as local lenders are playing such a mobilization role.

Investigating the role of political brokers

Existing research suggests that political brokers play a central role in mobilizing potential voters. If this is the case, we would expect to find that those who have social ties with brokers are more likely to be registered than those who are not. While we do not have a measure of whether a given individual is a broker, past work in Uganda suggests that elected village and subcounty heads are likely to be brokers (Blattman et al., 2019). Additionally, our finding that, out of non-household ties, only the lending network carries significant peer effects (shown in Table 4, column 5) hints to the possibility that lenders and borrowers may have a patron-client relationship, in which lenders act as more proximate political brokers. As such, we examine two kinds of relationships: lending relationships, as well as relationships to the village head, also called a Local Council 1 chairperson (LC1).

We first confirm that lenders likely act as patrons to their borrowers. As discussed above, only among the network of lenders, and not among other types of networks, do we find evidence that social ties are associated with registration status. Lenders are also significantly more likely to be registered to vote (79%) than the average person (69%), lending further support to the possibility that at least some lenders act as something akin to political brokers.²⁴

However, the analysis shown in Table 4, column 5, made two simplifying assumptions. First, it combined individuals who have no lending ties with individuals who have lending ties but whose alters—the other individual with whom someone reports a social tie—are not registered to vote. Second, the analysis in Table 4 considered undirected lending ties, thereby omitting information about who was the lender and who was the borrower for a given social tie. To relax these simplifying assumptions and further investigate the idea that village lenders likely act as political brokers, we conduct additional analyses. First, we explicitly model those individuals that have at least one lending tie, shown in Table 5, column 1. Second, in column 2, we also take into account the direction of the tie, considering specifically outward ties, that is, ties to a lender. Addressing both issues confirms our previous finding (Table 4, column 5) that lenders exert influence on their borrowers (column 2).

Table 5.

Effect of Lender. Model 3 inspects the effect of the most popular lender of agent i is registered. The reference category for this variable is “i has no lender.” All models include village-level fixed effects and the controls discussed in Data.

	Dependent Variable: Registered
	(1)	(2)	(3)
Degree >0	0.041
Degree >0	(0.059)
Degree	0.014***
Degree	(0.002)
% Registered lenders	0.058*
% Registered lenders	(0.033)
Out degree > 0		−0.033
Out degree > 0		(0.041)
Out degree		0.014***
Out degree		(0.005)
% Registered out lenders		0.082***
% Registered out lenders		(0.024)
Most popular lender not registered			0.028
Most popular lender not registered			(0.040)
Most popular lender registered			0.072**
Most popular lender registered			(0.031)
Controls	✓	✓	✓
Observations	1940	1940	1940
R²	0.082	0.071	0.069

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

Furthermore, individuals may borrow from several lenders, whose “popularity” may vary, in the sense that such lenders may have a different number of borrowers. Should lenders act as brokers, then more popular lenders should wield more influence. To test this implication, we zoom in on each individual’s most popular lender; that is, their lender with the most borrowers (column 3). Comparing the impact of all lenders to that of one’s most popular lender (i.e., columns 2 vs. 3), we find that most of the effect of lenders is simply a function of whether or not one’s most popular lender is registered to vote. Finally, we show that the only type of lenders that matter for voter registration are those who are not also family members, further suggesting that it is only the lenders who are most likely to be brokers (rather than simply family members) that exert political influence and mobilize registration (see Supplementary Material Table SI-10).

In Table 6 we examine the association between ties to the LC1 and voter registration. Examining relationships to the LC1 is challenging since LC1s are well-connected, leaving few people without ties to the LC1. As such, we restrict our attention to two analyses; one that considers direct lending relationships to the LC1 and one that focuses on indirect relationships, since influence likely trickles down from the LC1 to an individual through a series of patron-client relationships.

Table 6.

Effect of Distance to LC1 on Registration Status. In model 1, the reference category for the variable “Distance” is “i has no tie to an LC1” (i.e., i is more than 1 step away from an LC1). In model 2, the reference category for that variable is “i is more than 3 steps away from an LC1.” All models include village-level fixed effects and the controls discussed in Data.

	Dependent Variable: Registered
	(1)	(2)
Distance = 1	0.012	0.087***
Distance = 1	(0.023)	(0.030)
Distance = 2		0.072**
Distance = 2		(0.034)
Distance = 3		0.075**
Distance = 3		(0.034)
Controls	✓	✓
Observations	1940	1940
R²	0.065	0.070

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

First, we examine directed lending ties to LC1s. Specifically, we compare individuals who report that they will reach out to the LC1 if in need of a loan to those who do not; that is, we compare individuals who are one step away from the LC1 to individuals who are further away. This comparison suggests that those individuals are not significantly more likely to register to vote than villagers who are further away, as shown in Table 6, column 1. While at first sight, this may suggest that LC1s hold little influence on voter registration, it is also possible that their political influence extends further than their immediate connections.

As such, we then investigate whether the LC1’s influence extends further than his immediate connections (i.e., borrowers), by considering separately individuals that are 2 and 3 steps away from the LC1 (column 2). We find that the LC1’s influence extends up to 3 steps away, with little decay from steps 1 to 3, implying a sharp decay in influence for those individuals that are farther than 3 steps away from the LC1; that is, the reference category (Table 6, column 2).²⁵

Together, our findings on lenders, LC1s, and within-household ties are consistent with a diffuse process of influence, whereby LC1s act as patrons to a first circle of clients, who then act as patrons to a second circle of clients, and so on. This process extends up to three degrees away from the LC1. Note that those circles of patron-client relationships that radiate from the LC1 are likely to involve household heads. While we do not find evidence that household heads are more likely to be influenced by LC1s than non-heads,²⁶ we do find that household heads are more likely to be connected to LC1s in the first place, both in the lending and strong ties networks (Supplementary Material Table SI-11).

For this reason, it appears that social influence travels from the LC1 to household heads through a patron-client network, ultimately reaching other family members through the disproportionate influence that household heads exert upon other household members. While existing work suggests brokers work through direct contact with potential voters, we find evidence of a much more indirect process of influence—and one that is potentially more efficient.

Discussion

In this study, we examine the relationship between voter registration status and both individual attributes and relational ties in a sample of Ugandan villages. Our analysis is premised on the idea that learning about who registers to vote can shed light on the logic of electoral mobilization in lower-income settings, as well as identifying whose voices are not being heard and represented in these contexts. We make three key contributions.

First, we highlight the challenges of measuring voter registration, particularly in lower-income settings, and provide a novel strategy to accurately measure voter registration in a sample of villages that does not depend on self-reports nor on adult population estimates. By combining the official voter register with survey data conducted with all adults in the villages in our sample, we are able to estimate a more accurate voter registration rate for these villages as well as examine individual predictors of voter registration. We find that actual registration rates are significantly lower than registration estimates based on both self-reports (using publicly available surveys such as the Afrobarometer) and on rates based on official registration statistics. We thus urge caution in relying on official voter registration rates given limitations in the accuracy of information about the voting-age population and about the number of people who have been officially registered to vote, and particularly registration rates reported at subnational levels.

Second, we do not find evidence that individual attributes commonly thought to affect political participation, including education and income, are associated with registration status in our study area. Neither do we find evidence that groups thought of as being politically marginalized, including women or members of ethnic or religious minorities, are less likely to be registered. These findings underscore the limitations of generalizing core findings from high-income countries, and particularly the United States, to lower-income countries and to less democratic settings. An interesting question for future analysis is whether proxies of marginalization such as gender, income, or educational disparities emerge as consequential at the stage of voting in rural Uganda, even though such disparities do not seem to be highly consequential at the registration stage.²⁷

Nevertheless, we urge caution in interpreting our results as suggesting that factors such as education and income do not matter for voter registration in lower-income settings. These factors could have positive and negative effects on voter registration, canceling each other out, and may matter more in urban settings, which constitute a growing share of the population in rapidly urbanizing lower-income countries. Future research could go further in examining the conditions under which, if at all, sociodemographic factors such as education and income matter for voter registration in lower-income settings.

Third, voter registration status in rural Uganda is strongly related not only to one’s social ties to other members of one’s household (Table 2), but also to other members of the community (Table 4). Importantly, we find that intra-household dynamics are crucial for understanding voter registration, and that this relationship is driven by the household head, irrespective of the head’s gender. In addition, registration is increasing in how connected an individual is to others, and how central an individual is in her village’s social network (Supplementary Material Table SI-8), which is consistent with social rather than individualized mobilization (Eubank et al., 2021).

Further, our analysis reveals evidence of a diffuse process of social influence in the registration process, with individuals who lend money and elected village leaders (LC1s) playing key but often indirect roles in mobilizing residents to register to vote. Previous work attributed much influence to a small number of core brokers. We find that while the distance to the village leader matters in explaining registration status, there is little decay until one is quite removed, suggesting a large role for indirect influence. Household heads are more likely than the average person to have a close tie to the village head, and household heads themselves exert influence on the registration of other household members. This indirect and diffuse process of influence is intuitive in the sense that it would be inefficient for a broker to target each and every potential voter. Rather, a diffuse and indirect process allows brokers themselves to rely on intermediaries to influence others to register. Future work could employ a similar strategy to examine whether the results reported here travel to other contexts, as well as the implications of the distinct mobilization strategies underlying voter registration and voter turnout.

Supplemental Material

sj-pdf-1-cps-10.1177_00104140211036048 – Supplemental Material for Who Registers? Village Networks, Household Dynamics, and Voter Registration in Rural Uganda

Supplemental Material, sj-pdf-1-cps-10.1177_00104140211036048 for Who Registers? Village Networks, Household Dynamics, and Voter Registration in Rural Uganda by Romain Ferrali, Guy Grossman, Melina R. Platas, and Jonathan Rodden in Comparative Political Studies

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the United States Agency for International Development (USAID) and Stanford Institute for Innovation in Developing Economies (SEED).

ORCID iDs

Guy Grossman

Melina R. Platas

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Romain Ferrali is an Assistant Professor of Economics at the Aix-Marseille School of Economics.

Guy Grossman is a Professor of Political Science at the University of Pennsylvania.

Melina R. Platas is an Assistant Professor of Political Science at New York University Abu Dhabi.

Jonathan Rodden is a Professor of Political Science at Stanford University.

References

Anderson

M. L.

(2008). Multiple inference and gender differences in the effects of early intervention: A reevaluation of the abecedarian, perry preschool, and early training projects. Journal of the American Statistical Association, 103(484), 1481–1495. https://doi.org/10.1198/016214508000000841

Aral

Muchnik

Sundararajan

(2009). Distinguishing influence-based contagion from homophily-driven diffusion in dynamic networks. Proceedings of the National Academy of Sciences, 106(51), 21544–21549. https://doi.org/10.1073/pnas.0908800106

Berinsky

A. J.

(2005). The perverse consequences of electoral reform in the United States. American Politics Research, 33(4), 471–491. https://doi.org/10.1177/1532673x04269419

Bhatti

Hansen

K. M.

Wass

(2012). The relationship between age and turnout: A roller-coaster ride. Electoral Studies, 31(3), 588–593. https://doi.org/10.1016/j.electstud.2012.05.007

Blattman

(2009). From violence to voting: War and political participation in Uganda. American Political Science Review, 103(2), 231–247. https://doi.org/10.1017/s0003055409090212

Blattman

Larreguy

Marx

Reid

O. R.

(2019). Eat widely, vote wisely? Lessons from a campaign against vote buying in Uganda. NBER Working Paper No. 26293.

Blaydes

(2010). Elections and distributive politics in Mubarak’s Egypt. Cambridge University Press.

Bonica

Grumbach

J. M.

Hill

Jefferson

(2021). All-mail voting in colorado increases turnout and reduces turnout inequality. Electoral Studies, 72(August), 102363. https://doi.org/10.1016/j.electstud.2021.102363

Brady

H. E.

Verba

Schlozman

K. L.

(1995). Beyond SES: A resource model of political participation. American Political Science Review, 89(2), 271–294. https://doi.org/10.2307/2082425

10.

Brierley

Kramon

(2020). Party campaign strategies: Rallies, canvassing and handouts in Ghana. African Affairs, 119(477), 587–603. https://doi.org/10.1093/afraf/adaa024

11.

Brierley

Nathan

(2021). The connections of party brokers. Journal of Politics, 83(3), 884–901.

12.

Callen

Gibson

C. C.

Jung

D. F.

Long

J. D.

(2016). Improving electoral integrity with information and communications technology. Journal of Experimental Political Science, 3(1), 4–17. https://doi.org/10.1017/xps.2015.14

13.

Centola

Macy

(2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113(3), 702–734. https://doi.org/10.1086/521848

14.

Croke

Grossman

Larreguy

H. A.

Marshall

(2016). Deliberate disengagement: How education can decrease political participation in electoral authoritarian regimes. American Political Science Review, 110(3), 579–600. https://doi.org/10.1017/s0003055416000253

15.

Cruz

Labonne

QuerubÍn

(2017). Politician family networks and electoral outcomes: Evidence from the Philippines. American Economic Review, 107(10), 3006–3037. https://doi.org/10.1257/aer.20150343

16.

Downs

(1957). An economic theory of democracy. Harper & Row

17.

Eubank

Grossman

Platas

Rodden

(2021). Viral voting: Social networks and political participation. Quarterly Journal of Political Science, 16(3), 265–284. https://doi.org/10.1561/100.00019092

18.

Ferrali

Grossman

Platas

M. R.

Rodden

(2021). Replication data for: Who registers? Village networks, household dynamics, and voter registration in rural Uganda.

19.

Gallego

(2014). Unequal participation Worldwide. Cambridge University Press

20.

Grossman

Humphreys

Sacramone-Lutz

(2014). “I wld like u wmp to extend electricity 2 our village”: On information technology and interest articulation. American Political Science Review, 108(3), 688–705. https://doi.org/10.1017/s0003055414000331.

21.

Grossman

Lewis

J. I.

(2014). Administrative unit proliferation. American Political Science Review, 108(1), 196-217. https://doi.org/10.1017/s0003055413000567.

22.

Grumbach

Hill

(2021). Rock the registration: Same day registration increases turnout of young voters. . https://doi.org/10.1086/714776.Journal of Politics

23.

Harris

J. A.

Kamindo

van der Windt

(2021). Electoral administration in fledgling democracies: Experimental evidence from Kenya. Journal of Politics. . https://doi.org/10.1086/710785

24.

Hershey

M. R.

(2009). What we know about voter-ID laws, registration, and turnout. PS: Political Science and Politics, 42(1), 87–91. https://doi.org/10.1017/s1049096509090234

25.

Ichino

Schündeln

(2012). Deterring or displacing electoral irregularities? Spillover effects of observers in a randomized field experiment in Ghana. The Journal of Politics, 74(1), 292–307. https://doi.org/10.1017/s0022381611001368

26.

Isaksson

A.-S.

Kotsadam

Nerman

(2014). The gender gap in African political participation: Testing theories of individual and contextual determinants. Journal of Development Studies, 50(2), 302–318. https://doi.org/10.1080/00220388.2013.833321

27.

Izama

Wilkerson

(2011). Uganda: Museveni’s Triumph and weakness. Journal of Democracy, 22(3), 64–78. https://doi.org/10.1353/jod.2011.0044

28.

Jackman

Spahn

(2015). Unlisted in America. Unpublished manuscript

29.

Kasara

Suryanarayan

(2015). When do the rich vote less than the poor and why? Explaining turnout inequality across the world. American Journal of Political Science, 59(3), 613–627. https://doi.org/10.1111/ajps.12134

30.

Kim

J.-O.

Petrocik

J. R.

Enokson

S. N.

(1975). Voter turnout among the American States: Systemic and individual components. American Political Science Review, 69(1), 107–123. https://doi.org/10.2307/1957888

31.

Klimek

Yegorov

Hanel

Thurner

(2012). Statistical detection of systematic election irregularities. Proceedings of the National Academy of Sciences, 109(41), 16469–16473. https://doi.org/10.1073/pnas.1210722109

32.

Knoke

Yang

(2008). Social network analysis. Sage Publications.

33.

Krishna

(2008). Poverty, participation, and democracy: A global perspective. Cambridge University Press.

34.

Kuenzi

Lambright

G. M. S.

(2007). Voter turnout in Africa’s multiparty regimes. Comparative Political Studies, 40(6), 665–690. https://doi.org/10.1177/0010414006288969

35.

Kuenzi

Lambright

G. M. S.

(2011). Who votes in Africa? An examination of electoral participation in 10 African countries. Party Politics, 17(6), 767–799. https://doi.org/10.1177/1354068810376779

36.

Larreguy

Marshall

Querubín

(2016). Parties, brokers, and voter mobilization: how turnout buying depends upon the party’s capacity to monitor brokers. American Political Science Review, 110(1), 160–179. https://doi.org/10.1017/s0003055415000593

37.

Larson

J. M.

Lewis

J. I.

(2017). Ethnic networks. American Journal of Political Science, 61(2), 350–364. https://doi.org/10.1111/ajps.12282

38.

Lee

M. M.

Zhang

(2017). Legibility and the informational foundations of state capacity. Journal of Politics, 79(1), 118–132. https://doi.org/10.1086/688053

39.

Leighley

J. E.

Nagler

(2013). Who votes now?: Demographics, issues, inequality, and turnout in the United States. Princeton University Press.

40.

Mvukiyehe

Samii

(2017). Promoting democracy in fragile states: Field experimental evidence from Liberia. World Development, 95(C), 254–267. https://doi.org/10.1016/j.worlddev.2017.02.014

41.

Nagler

(1991). The effect of registration laws and education on us voter turnout (Vol. 85, pp. 1393–1405). The American Political Science Review. https://doi.org/10.2307/1963952

42.

Nevitte

Blais

Gidengil

Nadeau

(2009). Socioeconomic status and nonvoting: A cross-national comparative analysis (pp. 85–108). The comparative study of electoral systems

43.

Nichter

(2008). Vote buying or turnout buying? Machine politics and the secret ballot. American Political Science Review, 102(1), 19–31. https://doi.org/10.1017/s0003055408080106

44.

Nickerson

D. W.

(2008). Is voting contagious? Evidence from two field experiments. American Political Science Review, 102(1), 49–57. https://doi.org/10.1017/s0003055408080039

45.

Radcliff

(1996). Turnout and the economy: A comment on jackman and miller’s ”voter turnout in the industrial democracies in the 1980s. Comparative Political Studies, 29(6), 718–723. https://doi.org/10.1177/0010414096029006005

46.

Robinson

A. L.

Gottlieb

(2021). How to close the gender gap in political participation: Lessons from matrilineal societies in Africa. British Journal of Political Science, 51(1), 68–92. https://doi.org/10.1017/s0007123418000650

47.

Rolfe

(2012). Voter turnout: A social theory of political participation. Cambridge University Press.

48.

Rueda

M. R.

(2015). Buying votes with imperfect local knowledge and a secret ballot. Journal of Theoretical Politics, 27(3), 428–456. https://doi.org/10.1177/0951629814533841

49.

Siegel

D. A.

(2009). Social networks and collective action. American Journal of Political Science, 53(1), 122–138. https://doi.org/10.1111/j.1540-5907.2008.00361.x

50.

Stokes

S. C.

(2005). Perverse accountability: A formal model of machine politics with evidence from Argentina. American Political Science Review, 99(3), 315–325. https://doi.org/10.1017/s0003055405051683

51.

Stokes

S. C.

Dunning

Nazareno

Brusco

(2013). Brokers, voters, and clientelism: The puzzle of distributive politics. Cambridge University Press.

52.

Thompson

D. M.

J. A.

Yoder

Hall

A. B.

(2020). Universal vote-by-mail has no impact on partisan turnout or vote share. Proceedings of the National Academy of Sciences, 117(25), 14052–14056. https://doi.org/10.1073/pnas.2007249117

53.

Human Rights Watch (2021). Uganda: End enforced disappearances of opponents. https://www.hrw.org/news/2021/03/11/uganda-end-enforced-disappearances-opponents#

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.46 MB