When Does Information Influence Voters? The Joint Importance of Salience and Coordination

Abstract

Scholars argue that access to information about a politician’s programmatic performance helps voters reward good performers and punish poor ones. But in places where resources are made conditional on collective electoral behavior, voters may not want to defect to vote for a strong legislative performer if they do not believe that others will. We argue that two conditions must hold for information about politician performance to affect voter behavior: Voters must care about the information and believe that others in their constituency care as well. In a field experiment around legislative elections in Benin, voters rewarded good programmatic performance only when information was both made relevant to voters and widely disseminated within the electoral district. Otherwise, access to positive legislative performance information actually lowered vote share for the incumbent’s party. These results demonstrate the joint importance of Salience and voter coordination in shaping information’s impact in clientelistic democracies.

Keywords

African politics elections public opinion voting behavior experimental research political economy

Introduction

Electoral accountability is a central idea in democratic politics, but, in practice, poorly performing politicians are often voted into office again and again. A prevailing explanation for this pattern is that voters lack basic information that would allow them to distinguish among candidates based on performance.¹ Increase access to information about how incumbent politicians perform in office, and they will reward the better performing candidates, or so the logic goes.

Yet, recent empirical investigations of the effect of information on electoral behavior in low- and middle-income democracies have uncovered mixed results. In some cases, information provision has led voters to punish poorly performing politicians.² But, in other cases, it has had no effect,³ and sometimes voters punished challengers as well as incumbents.⁴ These findings raise a puzzle: Under what conditions will voters reward (punish) incumbents on the basis of information about their performance while in office?

We address this question in the context of clientelistic democracies, settings where politicians and parties generally seek votes through the conditional provision of selective and particularistic benefits, rather than policies, programs, and national public goods.⁵ Following Shefter (1977), we assume politicians can employ one of two basic strategies to attract voters: distribute divisible targeted benefits to supporters,⁶ or distribute universalistic benefits. Given the inefficiencies of targeted distribution, a growing literature studies the conditions under which polities transition from more clientelistic political appeals to what we refer to as programmatic, or universalistic, appeals (Asunka, 2016; Lizzeri & Persico, 2004; Weitz-Shapiro, 2012). Our focus in this article is on voter responsiveness to information about incumbent performance on a programmatic, rather than a targeted, dimension.

Adding to this literature, we ask, Under what conditions will voters in a clientelistic democracy condition their votes on information about programmatic performance? Clientelism is not necessarily normatively problematic, but, as other authors have argued, the shift between clientelistic and programmatic politics is puzzling and worthy of scrutiny.⁷ Asking this question contributes to the broader literature seeking to identify the conditions under which clientelistic politics shift to more programmatic politics, a central area of research in comparative political economy.

Two features of clientelist democracies motivate our argument. First, when politicians seek votes through the conditional provision of particularistic goods instead of through programmatic policies, voters may care little about politician legislative performance. In addition, the conditional targeting of goods may heighten the stakes of coordination at the local and constituency levels: Voters have strong incentives to coordinate around the expected winner—especially if they are located around the same polling station where vote tallies are easily observable—so that they can attract bloc rewards and avoid bloc punishment (Rueda, 2016; Smith & Bueno De Mesquita, 2012). As a result, voters will act on new performance information about a candidate only if they believe that sufficient others across the constituency will too, making that politician a likely winner. Absent wider coordination across the constituency around this new performance information and its importance to voters’ well-being, an individual voter’s choice to support the strong legislative performer risks lowering the local community’s vote share for the winning clientelist candidate, inviting punishment as a result. In other words, if voters do not expect others to value legislative performance and to act on the new performance information, then going at it alone may be perceived as a risky defection from the status quo.

We thus argue that two conditions must hold for voters in clientelistic democracies to condition their votes on programmatic performance information.⁸ First, voters must care about the programmatic performance dimension about which they receive information, believing that it is relevant for their welfare. We call this condition Salience.⁹ Second, if most voters do not already value the programmatic performance dimension, then voters must additionally overcome the coordination problem posed by the disincentive to switch to voting along a new programmatic performance dimension unless others do so as well. That is, voters must believe other voters across the constituency also care about the programmatic performance dimension. We call this condition coordination and emphasize that coordination can occur as long as voters hold certain beliefs about the behavior of others.¹⁰ Although physical communication and/or mobilization can also facilitate coordination, shifting expectations about others’ behavior can be sufficient.

We conducted our study around the 2015 National Assembly (legislative) elections in Benin, a clientelistic democracy. As in other former French and Portuguese colonies, members of Benin’s legislature are charged with legislative and executive oversight duties, rather than with constituency development which is more subject to targeting. We conducted a large-scale field experiment that delivered information about programmatic legislative performance of incumbent politicians running in the election.¹¹ Villages were randomly assigned to receive this legislative performance information (or not). As our baseline data suggest, voters do not initially perceive investment in legislative activities as a signal of good performance, whereas they do perceive constituency development as a positive attribute.¹²

To rigorously test the impact of our two proposed moderators on information uptake, both the content and method of information delivery were randomly assigned. To assess the importance of information Salience, some villages additionally received a message illustrating the relevance of the legislative performance dimension to voter welfare. To assess the importance of constituency-level voter coordination, some villages were the only ones in their areas to receive performance information (and were told as much), whereas others were one of the many villages in their areas to receive the information (and were told as much). Our argument, that both Salience and coordination must hold to observe a move toward programmatic voting, is explicitly tested where the information was widely disseminated with a Salience message.

Using official village-level data on incumbent party vote share, we find that voters rewarded (punished) good (poor) legislative performance only when the information was provided as part of a widely disseminated Salience campaign: Voters who received the information with a Salience message and were informed that people in other villages in their constituency had received the same information rewarded (punished) good (poor) legislative performers. We also find that when positive legislative performance information was not widely disseminated across the constituency, this information unexpectedly lowered support for the incumbent.

This latter result was unexpected and absent from our preanalysis plan, and so we leverage additional qualitative and quantitative data collected through focus groups, elite interviews, and our baseline survey to offer an explanation inductively. These data produce two key insights that, taken together, offer a plausible explanation for the unexpected finding. First, in the absence of a Salience message, voters generally value targeted transfers, not legislative performance. On its own, this unsurprising feature of Beninese voters cannot explain the negative effect of information. As argued in our preanalysis plan, weak preferences for legislative performance should lead to null, not negative, results where strong performance information was not widely disseminated. However, second, and important, some voters also perceive legislative performance and the ability to provide transfers as substitutes rather than complements. That is, incumbents that perform well legislatively are expected to be relatively bad at providing transfers. Under these conditions, “good” legislative performance information may be interpreted as bad news about the incumbent, absent interventions that increase the Salience of legislative performance among a wider distribution of voters. A more general point is that voters may value some dimensions over others, and information about one dimension may have unexpected effects on beliefs about alternative dimensions. This ex post insight could be the subject of future research.

Our findings highlight the joint importance of information relevance and voter coordination in conditioning the impact of programmatic performance information on voter behavior in clientelistic contexts where there are two dimensions of politician performance and voters value one over the other.¹³ Where voters expect bloc rewards for local levels of incumbent support, information about a particular performance dimension will affect electoral behavior only if voters expect a sufficient number of others in the constituency to similarly condition their votes on that performance dimension. In the face of null or surprising results,¹⁴ the literature has called for more attention to the moderators of information’s impact. We advance this literature not by considering any single moderator in isolation but rather by considering the implications of their interaction.

Our experiment also brings together the literature on clientelism with the literature on information and accountability. Scholarship on clientelism often focuses on structural conditions¹⁵ or policies¹⁶ that lead to programmatic politics. Meanwhile, the literature on information and accountability studies individual-level behavior, but less often focuses on switches from personalistic to programmatic voting.¹⁷ Our study investigates micro-level influences on personalistic and programmatic voting behavior and suggests that information Salience and voter coordination challenges jointly constrain programmatic voting in clientelistic contexts.

Information Salience, Strategic Coordination, and Vote Choice

Our understanding of the relationship between voter information and politician behavior is informed by standard moral hazard models of electoral accountability.¹⁸ These models assume that an elected politician is motivated both by being in office and by any personal benefits he can obtain through the misappropriation of public funds for private purposes. Voter preferences are unaligned with politicians inasmuch as they want to minimize misappropriation. As in standard principal–agent setups, the better the information the principal (voter) has about the agent (politician), the better the principal is able to control the behavior of the agent.

Voters with imperfect information follow a decision rule that allows them to translate perceived welfare into vote choice.¹⁹ Increased access to information about actual government policy or incumbent behavior then reduces the error with which voters translate their own welfare into beliefs about actions of the incumbent government. This logic yields the following standard expectation, one of the main hypotheses specified in our preanalysis plan:

Hypothesis 1: Increasing access to positive (negative) performance information about an incumbent will have a positive (negative) effect on electoral support for that incumbent.

This expectation, however, is based upon two implicit assumptions that we argue may not always hold, particularly in clientelistic democracies. First, the hypothesis assumes that voters care about the signal contained in the information, or that they care enough to change their behavior. Second, it assumes that voters are nonstrategic and do not condition their votes on expectations about the behavior of others. In cases where voters are thought to evaluate multiple dimensions of candidate performance, it is particularly important to scrutinize these assumptions, which we do below.

Information Salience and Dimensions of Politician Performance

The first assumption implicit in Hypothesis 1 is that voters believe the new performance information is relevant for their welfare—that the information is salient, in our terminology. However, there are generally myriad dimensions of politician performance, and these dimensions may vary in their Salience in the minds of voters. In clientelistic democracies, for example, politicians often spend—and are expected to spend—significant time and effort performing informal tasks that sustain clientelistic relationships.²⁰ Yet, this activity may not be what legislators are tasked with de jure. As such, information about a weakly relevant dimension may not change voter behavior, especially if voters have strong priors about a different performance dimension that they perceive to be highly relevant to their own welfare.

For the purposes of illustration and application to our empirical context, we distinguish between two performance dimensions that voters potentially consider differentially relevant to their own welfare: legislation and transfers. By legislation, we mean performance in activities related to making laws and holding other branches of government accountable, for example, participating in policy debates, overseeing the executive, and representing constituent interests during the policymaking process. By transfers, we instead mean activities related to handling the needs of individuals or particular groups or areas, that is, resources, inducements, or favors that are targeted by the politician directly to individuals or a locality, with or without the expectation of quid pro quo at election time.²¹

Given these two dimensions, a voter’s decision about her politician depends on her evaluation of the politician along each dimension and the weight she assigns each dimension. We use a simple decision-theoretic framework to illustrate our intuition. The voter’s decision calculus is given by the following equation where $Y$ is the total value the voter places on the candidate, $p$ is the weight she places on the legislative dimension, and ${LE}^{'}$ and ${TR}^{'}$ are her prior evaluations of the legislative and transfers dimensions, respectively:

Y = p ({LE}^{'}) + (1 - p) ({TR}^{'}) .

(1)

New information about legislative performance could lead voters to update their prior beliefs ${LE}^{'}$ to $LE$ . However, if $p$ is small, for example, and voters believe that legislative performance is not sufficiently relevant to their welfare (which is true in many places and which we will show is true in our study context), then there may be an unnoticeable shift in $Y$ . In other words, information about legislative performance alone may have little impact on voter behavior. More generally, performance information will only influence voters when they believe it to be relevant to their own welfare. We thus test the following hypothesis, specified in our preanalysis plan:

Hypothesis 2: Increasing the Salience of the legislative performance dimension will amplify the effect of information about legislative performance on voter behavior.

Our experiment tests this hypothesis with a treatment arm that provides legislative performance information in conjunction with a message explicitly designed to increase $p$ , or the extent to which a voter perceives the performance dimension to be relevant to her welfare.

The Problem of Voter Coordination

Also implicit in Hypothesis 1 is that voters are nonstrategic, or at least that information will affect a voter’s calculus independently of how it affects others’ evaluations of the candidate. If, however, voters are strategic, then information may not have an effect if it changes only a voter’s personal valuation of the candidate but does not change their beliefs about other voters’ behavior (Mvukiyehe & Samii, 2017).

Voters should be motivated to coordinate with one another on a particular candidate when there are strategic complementarities—or perceived benefits to voting at high rates for a particular candidate. In a primarily programmatic electoral system, voters are considered to be strategic when they choose a candidate they think is more likely to win at the expense of a more ideologically preferred candidate.²² In primarily clientelistic systems where there are fewer ideological attachments and where the provision of transfers is often contingent on collective voting behavior, voters may instead be motivated to coordinate their votes on a perceived winner because of the expectation of targeted collective goods if they do.²³ In either case, strategic voters seek to coordinate on a particular candidate. Complicating the voter’s calculus, in a case of multiple salient performance dimensions, strategic voters will also have to coordinate on which dimension of performance drives their electoral decision.²⁴

The method and extent of information dissemination might affect voters’ ability to coordinate. Morris and Shin (2002) identify that, in the presence of strategic complementarities, public dissemination of information will have an independent effect from the content of the information itself. Because of the coordination motive, the common knowledge produced by public information—the belief that others are also taking up the knowledge and that everyone knows it—increases its potential impact. Arias, Balan, Larreguy, Marshall, and Querubin (2017) show that such a coordination effect can be facilitated by information dissemination through interactions within tight-knit social networks. Yet, as long as voters know enough about fellow constituents to update their beliefs about how others will act in response to new information, actual interaction is not required. How widespread public information has to be for these effects to be felt depends on the nature of the coordination problem. We address two levels of coordination problems that voters in a clientelistic democracy might face.

Within-constituency coordination

Consider a common setup where voters receive political information within their locality, for example, their village or town. Voters in a clientelistic democracy face a coordination problem within their greater electoral constituency when responding to this new information. Even if voters in one locality are motivated to shift the dimension along which they vote from the status quo dimension (transfers) to a new dimension along which information is provided (legislative), then $Y$ has sufficiently increased, and they will likely suffer in a clientelistic setting if they are the only community to make this electoral choice.

For example, if candidates can monitor villages via polling station returns, then winning candidates can withhold collective transfers from the village that bloc voted on the stronger legislative performer. Our focus group respondents in Benin reported that electoral rewards and punishments are distributed to the village as a bloc: those that vote in large numbers for the incumbent are typically rewarded with “gifts” and money; those that do not are punished with neglect. This kind of bloc punishment is particularly feasible in contexts like ours where election results are made available at the village level and party agents are permitted at each polling station, ensuring close monitoring of local-level support.²⁵ This scenario requires coordination across the electoral district. Thus, voters may condition their votes on performance information only if they believe sufficient others in their electoral district will do the same such that their preferred candidate has a reasonable chance of winning.

In sum, we expect strategic voters to not only have priors about other people’s candidate preferences but also about other people’s value of $p$ , or their likelihood of voting on the performance dimension. Under the clientelistic status quo we have described, voters will expect other voters to care a lot about the transfers dimension and very little about the programmatic dimension. Thus, if a voter were to receive information that caused her to increase her own value of $p$ , she will not update her belief about others’ value of $p$ unless she thinks they, too, received the information. This logic implies that increasing the proportion of voters within an electoral district with access to information—and generating common knowledge about that proportion—will amplify the impact of that information on voting behavior. We test the following hypothesis, prespecified in our preanalysis plan:

Hypothesis 3: Increasing the proportion of communities in an electoral district with access to information will increase the impact of that information.

We test this within-constituency Coordination hypothesis by randomly varying the proportion of communities in an electoral district that receive legislative performance information, and critically for an argument that relies on voter beliefs, informing voters accordingly.

Within-locality coordination

For voters to solve the coordination problem posed by switching to the programmatic performance dimension in a clientelistic democracy, we have argued that producing common knowledge at the level of the constituency is required. Of course, voters could also fail to coordinate more locally if they do not perceive that others in their bloc are receiving information or switching performance dimensions. The way in which information is disseminated at the local level could moderate these perceptions. We thus prespecified:

Hypothesis 4: Public provision of information within the village will have a greater impact on voter behavior than private provision of information to individual voters.

In our experiment, we test this within-locality Coordination hypothesis with treatment arms that provide information to voters either in private in their homes or in public at a community event. We also suspected, ex ante, that villages in Benin might be tight-knit enough to begin with that coordination at this level might not need facilitation. To test for this possibility, we assigned some residents of “private” villages to a control condition such that they were surveyed at baseline and endline but received no information treatment. We found that these individuals adjusted their beliefs about politician performance in ways similar to those treated with private information, confirming our prior expectations.

Setting

We test our hypotheses in Benin, a West African democracy. Beninese politics have been widely described as clientelistic, where voters often pay more attention to transfers or constituency service than to legislative performance.²⁶ Scholars have found that this is also the case in Ghana, where MPs are held accountable for constituency service and clientelistic transfers more than legislative activity (Lindberg, 2003, 2010) And in Mali, voters focus on the transfers dimension, even though activities in this area are entirely informal (Soumano, 2011).

Benin’s political landscape is also characterized by ethnic voting (Adida, 2015; Dowd & Driessen, 2008; Koter, 2013; Wantchekon, 2003), a feature it shares with the political landscape of many, though certainly not all, other developing democracies.²⁷ Koter (2013) explains ethnic voting in Benin as a legacy of local leaders being either removed (if they resisted) or subdued by French colonizers, a legacy that weakened existing hierarchies and political organization, and left the country without traditional local elites to cultivate future electoral mobilization strategies. As a result, “ethnic affiliations have been an important means of political mobilization since independence” (Koter, 2013, p. 206). Yet, although Benin exhibits significant ethnic voting (Dowd & Driessen, 2008), and although coethnicity is an important consideration for voters as they evaluate political candidates (Adida, Gottlieb, Kramon, and McClendon, 2017), coethnic ties are not the only influence on voter support. According to Koter, who relies on Dowd and Driessen’s (2008) index of ethnic politics, around 40% of vote choice in Benin, is predicted by ethnicity (p. 214). But this rate leaves room for other considerations (including politician performance information) to affect electoral behavior.

Benin is a useful empirical case in which to test our argument because members of the legislature do not have autonomous control over budgets and thus are less able to target resources to particular constituencies, generating a sharp distinction between efforts that would be perceived as targeted rather than programmatic. As in other former French and Portuguese colonies,²⁸ Benin’s legislature is charged only with legislative and executive oversight duties.²⁹

A disadvantage of our study setting, which is also true of other former French and Portuguese colonies, is that its electoral rules—closed-list proportional representation (PR)—complicate the relationship between information about the incumbent politician and the voter’s actual electoral choice: voting for a party. To mitigate this constraint, we take advantage of the fact that in Benin, although they are elected by proportional representation to formally represent an entire constituency, national legislators informally “take care of” a discrete locality. Such a pattern has also been identified in Colombia, which Crisp and Desposato (2004) suggest is sustained by collusion among parties. In Benin, voters elect an average of 3.5 deputies per constituency. With 77 total communes distributed among the 24 legislative constituencies, there are 3.2 communes per constituency, on average. This makes feasible, as a rule of thumb for voters and legislators, a one-to-one mapping of communes to legislators.³⁰ This mapping in practice is consistent with expert evaluations of the party system in Benin as fragmented and weak, which reinforces the persistence of personalistic rather than party voting (Banégas, 1998; Gazibo, 2012). To wit, new parties and coalitions form at almost every electoral cycle; in addition, Benin’s last two presidents have both run as independents, unaffiliated with existing political parties. Before conducting the experiment, we took additional steps to verify this one-to-one mapping. Preexperiment focus groups confirmed to us that villagers typically can name and agree on a single legislator as their incumbent representative. In addition, our baseline survey confirmed that 58% of our sample named the incumbent we used in our design as the deputy “most responsible for their area” with most others saying that they did not know who their deputy was (rather than naming a different politician).³¹ We restricted our sample of communes to those 30 communes where incumbents were running again and where our implementation team relied on local knowledge to confirm a deputy-commune one-to-one mapping. We also conducted robustness tests (Supplemental Appendix Table G.1) in which we analyzed treatment effects in the survey data only among respondents who at baseline named the incumbent we used in our intervention. The treatment effects of providing positive information about incumbent performance within this restricted sample are consistent with the administrative results presented below.³²

Administratively, Benin is comprised of villages or quartiers (the urban equivalent) nested within communes, which are themselves nested within electoral constituencies. As described below, our interventions occur at the village/quartier(s) level (Performance Information and within-group Coordination treatments) and at the commune level (within-constituency Coordination treatment). Based on the one-to-one mapping of voter attention to a single incumbent in our sample of communes, we assume voters will care most about electoral behavior within their commune and vary dosage at that level. In the conclusion, we discuss the implications of this sampling decision for the generalizability of our results.

Research Design

This section details our experimental design. In collaboration with a local nongovernmental organization (NGO), the experiment involved the dissemination of information about incumbent legislative performance in advance of Benin’s April 26, 2015, National Assembly elections, which were not held concurrently with Presidential elections (held in 2016).³³ We begin this section by describing the details of each treatment condition. The scripts used in each of the treatment videos are provided in Supplemental Appendix B. Second, we describe our randomization procedure. Finally, we discuss the implementation.

Performance Information Treatment

Treated participants in the study were given information about their incumbent legislator’s relative performance in the National Assembly. The information was provided in the form of a video to hold constant the exact wording and tone of delivery across treatment conditions while making the information accessible to people of all education levels, literate and illiterate. In the video, a male actor read a script in a neutral tone, as a news caster or radio host might. The video included graphics to illustrate key points. It was recorded in French and then dubbed in local languages as necessary.

The information provided was drawn from official reports of the Office of the President of the National Assembly.³⁴ The video provided performance information about an incumbent legislator’s: (a) rate of attendance at legislative sessions, (b) rate of posing questions during legislative sessions, (c) rate of attendance in committees, and (d) productivity of committee work (the number of laws considered by the committee). The video provided raw data for each of these four performance indicators and presented two summary indicators. The first, an index of plenary performance on a scale of 1 to 10, took the average of normalized scores on the first two indicators: attendance and participation during full legislative plenary sessions. The second, an index of committee performance also on a scale of 1 to 10, took an average of the normalized scores on the second two indicators: attendance at committee meetings and productivity. To further synthesize the performance information, a global performance index which averaged scores from the first two indices was provided. In Supplemental Appendix D, we discuss our efforts to validate these indices using a separate data set on politician wealth and interviews with legislators.

Figure 1 shows two examples of how the information was presented.³⁵ Bar graphs highlight the performance of the legislator responsible for that commune relative to other legislators in the department (a local average)³⁶ and the country (national average). Red (Green) bars were used when the incumbent’s performance fell below (above) the local average.³⁷

Figure 1.

Example intervention bar graph.

Salience Condition

To test the importance of information Salience, the experiment also varied whether voters heard information about the relevance of the legislative performance dimension to voter welfare.³⁸ Treated participants were shown a video with either only the information about relative legislator performance (Info Only), or that same information plus this Salience treatment (Salience). The Salience treatment described the main responsibilities of legislative deputies. It then provided three concrete examples of how legislative performance (or lack thereof) can impact voter welfare. A positive example of good legislation was the passage of an anti-graft law requiring public servants to disclose assets. A negative example of a missed opportunity was the failure of the legislature to vote on and pass a health insurance scheme that was proposed in 2008. Finally, a positive example of executive oversight detailed how the legislature opposed changes to the Constitution proposed by the president that would expand his power. The Salience treatment was provided to treated participants immediately before they received the legislative performance information.

Coordination Conditions

To test our predictions about voter coordination—Hypotheses 3 and 4, respectively—we also varied the method by which the information was disseminated. First, participants were told during the intervention how many other villages in their commune were receiving legislative performance information. This high-dosage treatment was designed to test the hypothesis about beliefs about within-constituency coordination, and is described in greater detail below. Second, treated participants received the intervention either privately by watching a video on a smartphone in the respondent’s household (Private) or publicly through the screening of the same video via a projector in a public location in the village (Public). The public treatment was designed to test the hypothesis about beliefs about within-village coordination.

Randomization

We implemented a two-stage randomization procedure. First, we randomly assigned each of the 30 communes in our sample to either the low- or the high-dosage condition, blocking on incumbent legislative performance, which is observed at the commune level, and on north/south, because being in the culturally distinct north or south of the country is an important moderator of political behavior in Benin.³⁹ Within four blocks (high and low performance in the north and south) of communes, we assigned half to high-dosage and half to low-dosage treatment.

Second, we randomly assigned treatment conditions within communes. With the inclusion of commune fixed effects in our subsequent analyses, this randomization procedure allows us to derive our estimates of treatment effects primarily from within-incumbent variation in voting behavior and thus to rule out the possibility that average treatment effects are driven by observational differences across legislators. The unit of randomization within communes was the rural village or its equivalent urban quartier, the lowest level of social and territorial organization. In high-dosage communes, we randomly assigned each village/quartier to one of the five conditions: (a) Information Only/Private, (b) Information Only/Public, (c) Information + Salience/Private, (d) Information + Salience/Public, or (e) Control. Three villages/quartiers in each of the 15 high-dosage communes were randomly assigned to one of the four treatment conditions, and the remainder villages/quartiers in the commune were assigned to the control group. Thus, in the high-dosage communes, we have a $2 \times 2$ factorial design with a pure control group. To increase statistical efficiency, we assigned villages to experimental conditions while stratifying on urban/rural status and electoral competitiveness of the village in the previous legislative election. In the low-dosage communes, where only one village/quartier(s) received treatment, we randomly assigned units to one of the two conditions: (a) Information + Salience/Public, or (b) Control. One village/quartier(s) was assigned to the treatment condition, whereas the remainder were assigned to control. We use the Information + Salience/Public condition in the low-dosage communes because we believed a priori that it would have the strongest effect, thus making it harder for us to detect dosage effects.

Table 1 summarizes the experimental design. In low-dosage communes, there are 15 treated units (villages/quartiers) and 643 controls. In high-dosage communes, there are 45 treated units in each of the four treatment conditions (180 treated in total) and 486 controls. As prespecified, we use all nontreated villages/quartiers in our sample communes as controls, which substantially increases our statistical power. A baseline survey was conducted in all treated villages/quartiers (180 in high dosage and 15 in low dosage), in three control villages in each high-dosage commune (45 total), and in one control village in each low-dosage commune (15 total), for a total of 225 units in the survey sample.

Table 1.

Experimental Design.

	Low dosage (15 communes)			High dosage (15 communes)
	Control	Private (T3)	Public (T4)	Control	Private (T3)	Public (T4)
Control	643			486
Info Only (T1)					45	45
Salience (T2)			15		45	45

Cells represent the number of units (rural villages or urban quartiers) in each experimental condition. In private condition units, 40 randomly selected individuals were shown the treatment videos. In public condition units, 60 randomly selected individuals were invited to the public screening of the treatment videos, and on average about 50 individuals attended.

In each private condition village/quartier(s), 40 participants were shown one of the treatment videos. In each public condition village/quartier(s), 60 people were recruited and invited to the public screening, and on average about 50 participants attended (standard deviation is 13).⁴⁰ The average village/quartier(s) in our sample contains about 320 households, meaning that an individual from 12% to 15% of households was treated.⁴¹ We expected substantial within-household information transmission and we also found experimental and qualitative evidence of significant information transmission within villages (results discussed below), which means that many more people in each community were likely exposed to the information in the treatment.

More details on our randomization procedure can be found in Supplemental Appendix A. In particular, Supplemental Appendix Figure A.1 provides a CONSORT diagram, outlining the sampling and randomization procedure and shows each of the experimental conditions with the sample size of villages and survey respondents in each condition. Then, Supplemental Appendix Figure A.2 geographically plots the sampled villages and quartiers in each of the 30 sample communes.

Implementation

We designed and conducted the experiment in collaboration with the Centre de Promotion de la Démocratie et du Développement (CEPRODE), an independent, nongovernmental, nonpartisan Benin-based organization, whose enumerators introduced themselves as such and followed explicit instructions to deliver the treatments in a nonpartisan manner. Indeed, most of our enumerators were college students with significant experience fielding surveys in a professional manner. To avoid overlap with the 2-week period of campaigning prior to the election, the experiment and baseline survey were conducted from March 9 to April 9, 2015.

Two representatives from CEPRODE implemented the experiment in each treated village/quartier(s). Upon arrival, they sought permission from the local chief or leader to conduct the study, informing them it was a collaboration between CEPRODE and researchers from American universities.⁴² Individuals from the community were then randomly sampled, given information about the project, informed that it was a collaboration between a local nonpartisan NGO and American researchers, and given an opportunity to consent to participate.

A subset of consenting participants took the baseline survey. Following the survey (or immediately following consent), participants in the private condition were shown the video (either Info Only or Info Only + Salience) on a smartphone. Participants in the public condition were invited to attend a community-level screening of the videos later in the day (where either Info Only or Info Only + Salience were shown).

Data and Estimation Strategy

To measure the effect of the treatments on aggregate outcomes at the level of treatment assignment, we collected administrative data on party vote shares at the polling station level and then aggregated to the village level. We were able to match 2015 polling station data to all villages in our experimental sample except for one treated village and two surveyed control villages, which we drop from the analysis.⁴³ Including control villages that were not surveyed, among all villages and quartiers in our original sample of 30 communes, we were able to match 88% to the 2015 outcome data. On most pretreatment characteristics, unmatched and matched villages are statistically indistinguishable (see Table AC.1 in Supplemental Appendix C). In Supplemental Appendix C, we also show evidence of balance across high- and low-dosage communes as well as between treatment and control groups within the low- and high-dosage communes.

We conducted a panel survey of villagers in all treatment villages and a subset of control villages (see discussion above and Supplemental Appendix A). The endline survey contained a measure of self-reported voting behavior. We do not rely on the endline survey data here for two reasons: First, we uncovered significant patterns of differential attrition across treatment conditions making certain inferences subject to bias; and second, we expect our self-reported measure of vote choice in the endline survey to be subject to social desirability bias after having provided positive and negative performance information about legislators (evidence of both are illustrated in Supplemental Appendix I and in Adida, Gottlieb, Kramon, and McClendon (2019)). We prespecified that we would privilege behavioral data over survey data in the case in which our self-reported outcomes suffer from such problems. Our behavioral measure helps avoid these problems and is conservative in the sense that we uncover far fewer significant treatment effects when we analyze the official results (the full survey-data results are presented in Supplemental Appendix J).

Defining Positive and Negative Information

Our theoretical predictions, which we prespecified, are conditional on the nature of the information provided. That is, we expect voters’ behavior to be pushed in a different direction by information about legislators that is positive rather than by information that is negative.⁴⁴ As the information provided explicitly compares the incumbent legislator’s performance to the performance of legislators in the surrounding area (those in the same department: the same constituency plus a neighboring constituency), we code positive and negative information relative to this local benchmark. More specifically, we define the information as positive if the incumbent’s overall score is better than that of other deputies in the department. Poor legislative performers are those whose overall legislative score is worse than that of other legislators in this local area. This coding rule was prespecified in our preanalysis plan.⁴⁵

Model Specification

As prespecified, we divide the sample into communes where the incumbent is a strong legislative performer and communes where the incumbent is a poor legislative performer and run analyses separately. To analyze treatment effects of receiving good (bad) news about one’s incumbent, we estimate the following model using ordinary least squares (OLS):

E (Y_{i j} | {Positive}_{j} [{Negative}_{j}]) = β_{0} + β_{1} T_{i j} + μ_{k} .

(2)

where $Y_{i j}$ represents the vote share of the incumbent deputy’s party in village $i$ of commune $j$ , and µ _k represents a complete set of block fixed effects.⁴⁶ Our use of block fixed effects ensures that our experimental estimates are being driven by comparisons of similar villages/quartiers within the same commune and with the same incumbent.

$T$ is a generic treatment variable indicating the treatment status of village $i$ in commune $j$ . In some models, we pool all treatment conditions and estimate the impact of receiving any information. In others, we use indicators for different treatment arms, with Control as the omitted category. We cluster standard errors by commune-treatment condition. We prespecified that we would run all analyses with and without additional pretreatment covariates. Because the results are substantively unchanged with covariates and because we lose the majority of our sample when we include covariates from the survey, we present results without controls in the body of the article and results with controls in Supplemental Appendix G.4.

Results

We begin by presenting the mean incumbent party vote share (with standard deviations) in each experimental condition in Table 2 for the sample of voters who received positive information about their legislator’s performance. Several patterns stand out. First, in low-dosage communes, vote share for the incumbent party is substantially lower, by about 15 percentage points, in treatment than control villages.⁴⁷ In high-dosage communes, vote share for the incumbent party is also lower in the Info Only (T1) conditions than in Control. These patterns suggest that positive legislative performance information on its own generates a negative effect on incumbent support. However, second, vote share for the incumbent party is higher in the Salience (T2) condition relative to Control. Thus, positive information seems to increase incumbent party vote share only when information is disseminated widely across the commune (high dosage) and when the information is coupled with the Salience message.

Table 2.

Mean Incumbent Party Vote Share by Treatment Arm for Positive Information.

	ControlLD	T2T4LD	Control	T1T3	T1T4	T2T3	T2T4
Vote share for the incumbent	0.298	0.149	0.340	0.297	0.294	0.391	0.327
Vote share for the incumbent	(0.188)	(0.148)	(0.232)	(0.188)	(0.219)	(0.239)	(0.217)
Observations	259	6	259	27	26	27	27

Standard deviations in parentheses. LD in the column title signifies low-dosage treatment. T1 is the Info Only condition. T2 is the Salience condition. T3 is the Private condition. T4 is the Public condition.

Table 3 presents the mean incumbent party vote shares in the group of communes that received negative legislative performance information. A few patterns stand out. First, it is notable that, observationally, levels of incumbent support are higher in control areas where voters have a weak legislative performer as an incumbent, than in control areas where voters have a strong legislative performer as an incumbent (comparing Table 2 with Table 3). In the absence of treatment, those who do well in legislative performance get lower vote shares for their parties. This pattern suggests that, without treatment, those who likely spend time on other tasks (e.g., transfers) rather than on their formal, legislative duties, do better with voters. Second, the descriptive differences across experimental conditions in negative information communes are smaller than across experimental conditions in positive information communes. Although in low-dosage communes, incumbent party vote share is somewhat lower in treatment than in control, the difference is much smaller than in positive information communes. And, in high-dosage communes, there are no major differences between treatment and control villages.

Table 3.

Mean Incumbent Party Vote Share by Treatment Arm for Negative Information.

	ControlLD	T2T4LD	Control	T1T3	T1T4	T2T3	T2T4
Vote share for the incumbent	0.332	0.257	0.402	0.413	0.429	0.389	0.425
Vote share for the incumbent	(0.229)	(0.261)	(0.212)	(0.260)	(0.154)	(0.221)	(0.233)
Observations	381	9	225	18	18	18	18

Comparing Table 2 with Table 3 also underscores that support for incumbents is moderately low across the board. Mean vote share for incumbent parties across conditions is 42.9% at its highest. Like other developing democracies, Benin exhibits notable turnover at the legislative level (Ferree, 2010). This pattern has two implications for our study. First, it means that there was room for treatments to change support for incumbents. Support for incumbents was not so unshakeable, nor incumbency advantage so strong, that treatments had no chance of having an effect on voter behavior. However, second, this pattern also means that there was a possibility that voters in some areas would ignore information about incumbent quality, whether good or bad, if they were looking simply to elect a new representative, regardless of quality. This possibility makes the fact that we found treatment effects from some interventions more noteworthy. We also note that we blocked treatment assignment on competitiveness and find similar results across blocks.

Table 4 presents regression results estimating the impact of providing positive performance information on incumbent party vote share.⁴⁸ We see that the only instance in which legislative performance information changes voting behavior in the expected direction is when it is made salient and disseminated to a significant proportion of the constituency. Column 1 tests the impact of receiving any information treatment in the full sample of communes. The coefficient is very close to zero, suggesting no average impact of the provision of positive information. Column 2 shows that access to positive legislative information in low-dosage communes has a negative, and statistically significant, effect on incumbent party vote share. The magnitude is about 14 percentage points (95% confidence interval [CI] = [–24, –4]). Column 3 shows that positive information has a very small and statistically insignificant effect on incumbent party vote share in high-dosage communes. In sum, the evidence does not corroborate Hypothesis 1. Rather, positive information decreases vote share for the incumbent party in low-dosage communes, a result we return to in the next section.

Table 4.

The Impact of Positive Information on Incumbent Party Vote Share.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
Variables	All	Low dose	High dose	High dose	High dose	All
Received any treatment	−0.001 (0.014)	−0.143* (0.046)	0.012 (0.014)
Performance information only				−0.020 (0.014)		−0.020 (0.013)
Civic + performance information				0.042* (0.019)		−0.143* (0.045)
Public treatment					−0.004 (0.018)
Private treatment					0.027 (0.019)
High Dosage × Civics						0.186* (0.049)
Constant	0.318* (0.003)	0.298* (0.001)	0.332* (0.004)	0.332* (0.004)	0.332* (0.004)	0.318* (0.002)
Observations	630	265	365	365	364	630
R ²	.457	.149	.611	.616	.611	.465

In parentheses, robust standard errors clustered by commune in Model 1 and Commune × Treatment, otherwise. Models include block fixed effects.

p < .05.

In Column 4, we test the Salience hypothesis, Hypothesis 2. In the Info Only condition, positive information has a negative but not statistically significant impact on incumbent party vote share. By contrast, positive performance information has a positive and significant effect in the Salience condition. Increasing the perceived relevance of legislative performance to voter welfare increases the vote share of incumbent parties by 4.2 percentage points in high-dosage communes where legislative performance was strong (95% CI = [0.5, 8]). We can also reject the null hypothesis that the effect of Salience is equivalent to the effect of Info Only $(p = 0.004)$ . This evidence corroborates Hypothesis 2 for the subsample of high-dosage communes. When legislative performance information is made more salient and is distributed to a large proportion of the electoral district, its impact increases. Otherwise, the impact of positive information has a null or even a negative effect.

Turning to the Coordination hypotheses, the interaction model in Column 6 tests Hypothesis 3 on within-constituency coordination. The results show that the effects of the Salience treatment are statistically different in low- and high-dosage communes (as indicated by the coefficient on Salience × High Dosage).⁴⁹ The Salience condition thus improves the party vote share of good legislative performers, but only when information is disseminated widely across the commune. However, it fails to benefit good legislative performers in low-dosage communes, a result which is consistent with our theory that voters are acting strategically.⁵⁰ Even though treated voters in low dosage may now privately prefer a good legislative performer, that an insufficient number of other voters have similarly updated will make it too risky to defect from the status quo. Column 5 presents tests of Hypothesis 4 on the importance of within-locality coordination. There is little difference between the private and public effects, and we cannot reject the null that the effects are the same $(p = 0.18)$ .

As prespecified, we also conducted the main analyses on the sample of communes in which participants received negative performance information. Overall, we do not find a strong effect of access to negative information about legislative performance (see complete results in Supplemental Appendix Table AF.1). However, when accompanied by a widely disseminated Salience message, access to negative performance information lowers support for the worst performing politicians. This result is shown in Figure 2 where we separate the sample into four subgroups corresponding to the information provided to participants in the experiment: information that the incumbent was “much worse,” “worse,” “better,” or “much better” than the local average. Categories were defined using quartiles of the performance score in each department.⁵¹ We find the high-dosage Salience treatment had a negative effect on the party vote share of those who performed “much worse,” with a coefficient statistically significant at the $p < 0.1$ level $(p = 0.09)$ . Thus, voters do punish the worst performing incumbents, but only when the information is disseminated in combination with the Salience treatment and widely across the commune.

Figure 2.

Effect of joint Coordination and Salience treatments on incumbent party vote share, by incumbent performance (with 95% confidence intervals).

To further illustrate the substantive importance of the main results, Figure 3 presents the predicted incumbent party vote shares that are derived from the main analyses in high-dosage communes. The predicted vote shares correspond to the models in Column 4 of Table 4 and Supplemental Appendix Table F.1. The patterns confirm the descriptive trends presented in Tables 2 and 3: In the control group, poor legislative performers do substantially better electorally than do good legislative performers. The Information Only treatment has no impact on this gap in electoral support. By contrast, the Salience treatment closes the gap significantly.

Figure 3.

Predicted values by treatment condition and incumbent performance in high-dosage communes (with 95% confidence intervals).

In a final set of analyses, presented in Supplemental Appendix K, we relax the binary definition of good and bad news and allow the type of information to vary over the 10-point index. Two important patterns emerge strengthening our results. First, the impact of treatment in low-dosage communes becomes more negative as the incumbent’s performance score increases (the interaction term is not significant). Second, the impact of Salience is increasing and eventually becomes positive as the incumbent’s performance score increases (this interaction term is statistically significant).

Robustness

The negative effect of positive information in low-dosage communes was not anticipated, and so we conducted a number of different analyses to assess its robustness. First, to ensure that the result is not being driven by one or a few anomalous but high influence communes, we re-ran the low-dosage analysis removing each commune from the sample iteratively. In Supplemental Appendix G, we show that the results are robust to these tests.

Second, we were concerned that, despite the random assignment of dosage, there may have been idiosyncratic differences between good performing incumbents in the high- and low-dosage communes. We thus investigated each of the incumbents individually, and found no differences in terms of their political party affiliations or connections to the president’s party.

Third, we considered that the “good news” provided in the low-dosage communes may have been less positive than the “good news” provided in the high-dosage communes. We find, however, that on average the good performers in low-dosage communes are better performers than the good performers in high-dosage communes (7.26 average score in low dosage vs. 6.07 average score in high dosage). Supplemental Appendix Figure G.1 displays the distributions of the performance scores of good performers in high- and low-dosage communes, confirming that the good performers in low-dosage communes actually scored better than the good performers in high-dosage communes. Finally, Supplemental Appendix G also shows that the results are not sensitive to the use of block fixed effects and are robust to the inclusion of weights that account for differences in the sizes of each block.

Discussion

This study generated three key sets of results. First, we showed that access to information leads voters to reward good legislative performance and punish poor legislative performance only when the legislative performance dimension is salient to voters and the information is widely disseminated, facilitating voter beliefs about cross-village coordination. Second, we found that access to positive legislative performance information alone actually lowered vote share for the incumbent party. Third, we found no difference between the public and private methods of information delivery.

Our interpretation that the first finding is being driven, in part, by a change in the Salience of the legislative performance dimension is substantiated by additional endline survey evidence showing that participants’ valuation of this dimension increased only in the Salience treatment condition. In Supplemental Appendix N.2, we present the results of a survey experiment in which respondents at endline were presented with one of the three randomly assigned candidate descriptions. We manipulated whether the candidate was high-performing on our two dimensions of interest—transfers and legislative performance—or neither. The Salience treatment increased respondent valuation of the high legislative performer relative to both Control and Info Only, but did not generate a similar increase in respondent valuation of the candidate who performed well on the transfers dimension. Consistent with our expectations, the transfers dimension is preferred over any other in the Control condition.

Our interpretation that increased coordination within the constituency is also driving the first finding is substantiated by both a manipulation check and evidence of voter beliefs measured during the endline survey. (Note that, although we do not use the endline survey to estimate treatment effects on self-reported votes for the incumbent due to our concerns about response bias in that question, questions soliciting beliefs about others’ behavior are typically thought to be less subject to response bias, Lusk & Norwood, 2009.)⁵² First, voters in the low- and high-dosage treatment were very aware of the differential breadth of the two treatments. Eighty-four percent of people in high-dosage (correctly) answered that about a quartier(s) of villages in their commune received the performance information, and 75% of people in low-dosage answered that only one to two villages in the commune received treatment. Second, when asked to guess how many of the other 19 surveyed respondents in the village voted for the incumbent, treatment caused participants to expect others to increase their support for the incumbent, but only when good news was provided in conjunction with a Salience message that was widely disseminated within the constituency (see the marginal effect of Salience plus good news in high-dosage communes in Figure 4). This is consistent with the explanation that the positive and significant impact of treatment on voting behavior is facilitated by voters across the constituency being better able to coordinate on a new performance dimension in the high-dosage condition because it was common knowledge that a substantial proportion of the constituency was receiving that same intervention.

Figure 4.

Effect of Salience treatment on expected incumbent party vote share in village, by dosage level and incumbent performance (with 95% confidence intervals).

In the remainder of this section, we leverage additional qualitative and quantitative evidence to interpret our two unanticipated findings. The claims we make below are inductive and useful for theory generation. However, they require additional testing and therefore remain tentative here. Our evidence is from the following sources. In January 2015 (2 months before experimental treatments were administered), we conducted focus groups with a random sample of 160 voters across eight representative villages.⁵³ Then, in March 2015, we conducted interviews with nine journalists from public, private, and community radio stations with differing political orientations (national, opposition, and centrist) in both Northern (Parakou) and Southern (Cotonou, Ouidah) regions.⁵⁴ We also conducted two phases of legislator interviews.⁵⁵ Last, in March 2015, we conducted a baseline survey of more than 6,000 eligible voters participating in our experiment.⁵⁶

The Negative Effect of Positive Information

Two key insights into voter beliefs and preferences about politician performance come from our nonexperimental data. First, voters place little value on the legislative performance dimension, especially relative to the transfers dimension. And second, some voters and politicians perceive performance along the legislative and transfers dimensions to be substitutes.

In focus group discussions, most participants claimed that they consider, first and foremost, whether or not the candidate will help develop the local community (e.g., build schools and install potable water infrastructure) when making voting decisions. A nontrivial number also said they pay attention to who gives the most presents or money during the campaign. Interviews with politicians and radio hosts further confirmed voter preferences for constituency transfers. One accomplished legislator reported that voters challenge him because he has not built roads or schools directly in their villages. Radio hosts said listeners do not care about legislative performance and, as a result, though journalists have access to legislative sessions, they seldom report on the performance of legislators during those sessions.⁵⁷

In Supplemental Appendix N.2, a survey experiment in our baseline survey that described a hypothetical candidate as either providing transfers, performing legislative duties, or neither shows that respondents prefer candidates who provide transfers, but performing legislative duties is not rewarded by respondents.⁵⁸ In sum, our evidence is consistent with past research on Benin which shows that voters often value transfers over legislative performance.⁵⁹

Our baseline data further suggests that legislative performance may be taken as a negative signal about transfer capabilities. In evaluations of real candidates in the baseline survey (see Table 5), we find that positive legislative performance is negatively associated with evaluations of the incumbent, while support for the incumbent is positively associated with the receipt of personal transfers.⁶⁰ In other words, patterns in the baseline survey data are consistent with voters treating legislative performance and transfer capabilities as substitutes.

Table 5.

Pretreatment Correlates of Voting for the Incumbent.

Variables	(1)
Overall performance score	−0.026 (0.014)
Time incumbent spends in village	0.017 (0.023)
Received personal assistance from incumbent	0.607* (0.168)
Vote Margin	0.107 (0.068)
Urban	0.030 (0.034)
Constant	0.403* (0.082)
Observations	248

Robust standard errors clustered at the commune level in parentheses.

p < .05.

Our interviews with legislators also suggest that politicians do indeed have limited time and resources, and that, for at least some, time spent on legislative duties is time not spent on transfers. A handful of legislators lamented this trade-off, indicating that their responsibilities in Porto Novo prevent them from helping their community. One went so far as to recognize that by prioritizing legislative duties, he is jeopardizing his own chances of reelection. Note, however, that a handful of legislators diverge from this view: They divide their time between constituency transfers and legislative tasks, and do not perceive any issue with this division of labor. Even so, all legislators characterize this as an allocation decision in the face of fixed time and resources.

A plausible explanation: substitutes versus complements

It is thus plausible in the Benin context that the negative effect of access to positive legislative performance information on vote share in the absence of a widely disseminated Salience campaign is due to voters hearing the “good news” about strong legislative performance as bad news about their preferred type of politician activity (transfers). Using our unexpected finding as an opportunity for theory generation, we thus propose a new hypothesis that fits the Benin case but must be subjected to future investigation: When two dimensions of candidate performance are perceived as substitutes rather than complements, increasing access to positive performance information about the less salient one will have a negative effect on electoral support for that incumbent. In effect, this highlights a third assumption in the standard model that does not appear to be met in our context: that information about one dimension of performance has no impact on voter perceptions of performance in other, potentially more valued, dimensions.

We derive this hypothesis more formally in Supplemental Appendix E, but provide some of the intuition here. A voter in any electoral system in which constituency transfers are salient may perceive information about the legislative dimension to also be a signal of the politician’s performance along the transfers dimension. For instance, a voter may perceive performance on legislative activities and constituent transfers to be positively correlated, or complements, if she believes that good performance on one dimension translates to good performance on all dimensions. In this case, positive information about the legislative dimension should convey a positive signal about transfers, and the expectations of the standard model would hold. Alternatively, a voter may perceive the dimensions to be negatively correlated, or substitutes. As in our case, she may see the politician as having a budget constraint on his time, such that spending effort along one dimension will necessarily detract from the other. Or, she might believe that the types of politicians that perform well on one dimension will lack the ability or disposition to perform well on another. In this case, positive information about the legislative dimension could convey a negative signal about the transfers dimension, which could actually lead voters to vote against the incumbent.

Public Versus Private Dissemination

We also found that public dissemination of information within villages had little impact relative to private delivery, even though the number of villages treated across constituencies mattered. One explanation for this finding is the tight-knit characteristic of villages in Benin: that, facilitated by close relationships and frequent conversations among villagers, information easily spreads within villages. We randomly assigned some individuals to receive a survey and no intervention (control) and some individuals to receive a survey with the intervention (treatment) in private condition villages. When comparing reported support for and views of the incumbent across these treatment and control individuals within private villages, we observe no detectable treatment effects. These results, which are presented in more detail in Supplemental Appendix L, provide evidence of strong preexisting information transmission mechanisms within the village. Indeed, the responses of people who were not directly treated were equivalent to the responses of people directly given treatment within the same village. Barriers to coordination within the village appear to be fairly minimal in Benin. By contrast, our results on the impact of dosage show that coordination within the constituency (through changes in expectations about other voters’ behavior) poses a significant constraint.

Conclusion

Social science research investigating the relationship between access to information about politician performance and electoral outcomes makes straightforward theoretical predictions but yields inconclusive empirical results, particularly among clientelistic democracies. Our article advances knowledge about information and electoral behavior in these contexts by examining the joint moderating roles of information Salience and voter coordination. With a field experiment in Benin, we found that only when information was provided with (a) a Salience message highlighting the relevance of legislative performance to voter welfare and (b) a coordination message highlighting the fact that other villages in the constituency were receiving the same information, did it translate as expected into electoral behavior.

Although we provide evidence from Benin, where the role of ethnicity in politics and the proportional representation electoral system are important features, the central logic of the argument can apply more broadly to democracies where clientelism is a primary strategy of political elites. This condition is likely to arise in younger democracies where programmatic promises are less credible (Keefer, 2007) and where local brokers have influence in coordinating blocs of votes—or where communities are able to coordinate on their own. Thus, we would expect the logic of the argument—centering on information Salience and voter coordination—to extend to contexts where clientelistic politics is prevalent but ethnicity is not salient, such as Senegal (Koter, 2013), and to other clientelistic democracies with different electoral systems and institutions, such as Mexico (Arias et al., 2017).

When our treatment increased the Salience of the legislative dimension but did not facilitate cross-village coordination (low dosage), positive performance information had a statistically significant negative impact on electoral support for the incumbent. We investigate reasons for this unexpected finding using evidence from elite interviews, focus groups, and a baseline survey of more than $6, 000$ eligible voters. Two relevant but preliminary findings emerge: Beninese voters generally value transfers over national legislation and view legislative performance as a substitute for efforts to provide such transfers. We thus propose as a plausible new implication of our argument that substantiates the surprising finding: Information about “good” legislative performance will be received as bad news about the incumbent when two dimensions of candidate performance are perceived as substitutes rather than complements and information is provided about the less salient one. Such a proposition underscores a third weakness of the standard model—its assumption that information about one dimension of performance does not affect voter perceptions of other performance dimensions—which must be tested in other contexts before being confirmed.

In highlighting the joint importance of Salience and strategic voting as moderators of information provision, this study advances a number of important literatures. First, our findings have implications for the literature on information and accountability, which has found mixed results on the impact of information access. Although Salience appears in the literature as a constraint to information’s effect on voter behavior, we emphasize the idea that strategic considerations and voter coordination⁶¹ are additionally important. Furthermore, our study is the first to randomly vary these conditions in a real election to study their impact on voter behavior. Our finding that both conditions must hold for information to influence voters can be generalized to specific contexts: clientelistic democracies where legislators engage in both formal, programmatic activities and informal, personalistic ones.

Second, our results have implications for the conditions under which citizens move from clientelistic to more programmatic voting behavior. Although modernization and related theories imply that shifts in voter preferences play an important role in shaping incentives for clientelism, our results emphasize that voter coordination problems constrain voters’ ability to act on these preferences. Related, our findings speak to the potential role of mass media expansion in shaping clientelistic politics. An observational study of Benin showed that access to communal radio messages highlighting the advantages of public goods diminished support for transfers relative to public goods.⁶² Our study offers a theoretical explanation for this claim: Mass media can provide not only persuasive messages raising the Salience of programmatic performance but also common knowledge to facilitate coordination around new political preferences. Our study additionally provides experimental evidence that both Salience and coordination condition the impact of performance information on voter behavior.

The findings also open up fruitful avenues for future research. For instance, we do not make claims here about general equilibrium effects on electoral accountability. We use information about legislative performance in the previous term and do not attempt to assess whether politicians will change their legislative behavior in the future now that this information has been provided (cf. Humphreys & Weinstein, 2012). The long-term effects of our intervention on accountability could be assessed in future research. Furthermore, our specific results may travel best to other countries where legislators do not have constituency development funds or are not formally tasked with constituency development, which appears to be the case for most non-Anglophone African countries, but the generalizability of the results could be tested in future studies.⁶³ Our findings may also travel best to contexts in which there is a one-to-one mapping of incumbents to geographic areas (either formally as in the case of single-member districts, or informally as in our and other contexts⁶⁴), but the generalizability of our results to other electoral contexts could be tested. Finally, our proposed theory that, under certain conditions, positive performance information can have a negative effect on incumbent party vote share could be tested in new settings.

Supplemental Material

Adida_et_al_Replication_Files – Supplemental material for When Does Information Influence Voters? The Joint Importance of Salience and Coordination

Supplemental material, Adida_et_al_Replication_Files for When Does Information Influence Voters? The Joint Importance of Salience and Coordination by Claire Adida, Jessica Gottlieb, Eric Kramon and Gwyneth McClendon in Comparative Political Studies

Supplemental Material

AGKM_CPSappendix – Supplemental material for When Does Information Influence Voters? The Joint Importance of Salience and Coordination

Supplemental material, AGKM_CPSappendix for When Does Information Influence Voters? The Joint Importance of Salience and Coordination by Claire Adida, Jessica Gottlieb, Eric Kramon and Gwyneth McClendon in Comparative Political Studies

Footnotes

Acknowledgements

The authors thank Metaketa leadership for support on the research design, Amanda Pinkston for sharing 2011 legislative election data, and Ana Quiroz for excellent research assistance. The authors further thank Adam Chabi Bouko for his leadership in executing the field experiment, along with the entire CEPRODE team. The authors are also grateful to participants at ASA 2015, the Princeton University Research in Experimental Social Science Workshop, the UToronto Comparative Politics Speaker Series, Columbia University’s Seminar for the Study of Development Strategies, Washington University in St. Louis’ Comparative Politics Annual Conference, the DIME seminar at the World Bank, and MIT’s Political Behavior of Development conference for helpful comments.

Authors’ Note

This research was conducted in collaboration with the Centre de Promotion de la Démocratie et du Développement (CEPRODE) and its Director, Adam Chabi Bouko. Our project received ethics approval from the authors’ home institutions. The authors also obtained permission to conduct the study from the President of the National Assembly of Benin. In each study village, permission to conduct research was obtained from the chief and consent was obtained from each surveyed participant in the study. This study is part of the larger Metaketa initiative to accumulate knowledge about the relationship between information and accountability across country contexts. The registered preanalysis plan for this study can be found at: .

Declaration of Conflicting Interests

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

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by EGAP.

ORCID iDs

Claire Adida

Jessica Gottlieb

Supplemental Material

Supplemental material for this article is available online at the CPS website

Notes

Author Biographies

Claire Adida is an associate professor of Political Science at UC San Diego. She investigates the determinants of migrant integration and identity politics in a comparative context.

Jessica Gottlieb is an assistant professor in the Bush School of Government & Public Service at Texas A&M University. She studies the determinants of democratic accountability in low-income countries with a focus on West Africa.

Eric Kramon is an assistant professor of Political Science and International affairs at George Washington University. His research focuses on electoral politics, clientelism, and ethnic politics, with a regional focus on Africa.

Gwyneth McClendon is an assistant professor in the Wilf Family Department of Politics at New York University. Her research focuses on comparative political behavior, religious and ethnic politics, and political participation.

References

Adida

(2015). Do African voters favor coethnics? Evidence from a survey experiment in Benin. Journal of Experimental Political Science, 2(1), 1-11.

Adida

Gottlieb

Kramon

McClendon

(2017). Reducing or reinforcing coethnic preferences? An experiment on information and coethnic voting. Quarterly Journal of Political Science, 12, 437-477.

Adida

Gottlieb

Kramon

McClendon

(2019). Response bias in survey measures of voter behavior: Implications for measurement and inference. Journal of Experimental Political Science, 6, 192-198. Retrieved from https://doi.org/10.1017/XPS.2019.9

Arias

Balan

Larreguy

Marshall

Querubin

(2017, September). How social networks help voters coordinate around information provision to improve electoral accountability: Evidence from Mexico (Working paper). Retrieved from https://pdfs.semanticscholar.org/b236/44769bd41022a783a525aff550113bae07aa.pdf

Asunka

(2016). Partisanship and political accountability in new democracies: Explaining compliance with formal rules and procedures in Ghana. Research & Politics, 3(1), 2053168016633907.

Banégas

(1998). Marchandisation du vote, citoyenneté et consolidation démocratique au Bénin. Politique Africaine, 69(1), 75-88.

Banerjee

Kumar

Pande

(2011). Do informed voters make better choices? Experimental evidence from urban India. Unpublished manuscript.

Barkan

Mattes

(2014). Why constituency development funds in Africa? Representation vs. constituency service? In Baskin

Mezey

M. L.

(Eds.), Distributive Politics in Developing Countries: Almost Pork (pp. 27-48). Lanham, MD: Boulder, CO; New York; London: Lexington Books.

Barro

R. J.

(1973). The control of politicians: An economic model. Public Choice, 14(1), 19-42.

10.

Chandra

(2004). Why ethnic parties succeed: Patronage and ethnic head counts in India. New York, NY: Cambridge University Press.

11.

Chong

De La O

A. L.

Karlan

Wantchekon

(2015). Does corruption information inspire the fight or quash the hope? A field experiment in Mexico on voter turnout, choice, and party identification. The Journal of Politics, 77(1), 55-71.

12.

Cox

G. W.

(1994). Strategic voting equilibria under the single nontransferable vote. American Political Science Review, 88, 608-621.

13.

Crisp

B. F.

Desposato

S. W.

(2004). Constituency building in multimember districts: Collusion or conflict? Journal of Politics, 66, 136-156.

14.

De La O

A. L

. (2015). Crafting policies to end poverty in Latin America. New York, NY: Cambridge University Press.

15.

Dixit

Londregan

(1996). The determinants of success of special interests in redistributive politics. Journal of Politics, 58, 1132-1155.

16.

Dowd

R. A.

Driessen

(2008, January). Ethnically dominated party systems and the quality of democracy: Evidence from sub-Saharan Africa (Afrobarometer Working Paper 92). Retrieved from https://www.files.ethz.ch/isn/91882/AfropaperNo92.pdf

17.

Dunning

Grossman

Humphreys

Hyde

S. D.

McIntosh

Nellis

(2019). Information, accountability, and cumulative learning lessons from Metaketa I. New York, NY: Cambridge University Press.

18.

Fearon

(1999). Electoral accountability and the control of politicians: Selecting good types versus sanctioning poor performance. In Przeworski

Stokes

S. C.

Manin

(Eds.), Democracy, accountability, and representation (pp. 55-97). New York, NY: Cambridge: Cambridge University Press.

19.

Ferejohn

(1986). Incumbent performance and electoral control. Public Choice, 50(1), 5-25.

20.

Ferraz

Finan

(2008). Exposing corrupt politicians: The effects of Brazil’s publicly released audits on electoral outcomes. The Quarterly Journal of Economics, 123, 703-745.

21.

Ferree

K. E.

(2010). The social origins of electoral volatility in Africa. British Journal of Political Science, 40, 759-779.

22.

Fujiwara

Wantchekon

(2013). Can informed public deliberation overcome clientelism? Experimental evidence from Benin. American Economic Journal: Applied Economics, 5, 241-255.

23.

Gazibo

(2012). Beyond electoral democracy: Foreign aid and the challenge of deepening democracy in Benin (Wider Working Paper 2012/33) Helsinki: UNU-WIDER.

24.

Gottlieb

(2016). Common knowledge and voter coordination: Experimental evidence from Mali. In Blais

Laslier

J.-F.

Van der Straeten

(Eds.), Voting experiments (pp. 89-116). New York: Springer.

25.

Gottlieb

Larreguy

(2020). An Informational Theory of Electoral Targeting in Young Clientelistic Democracies: Evidence from Senegal. Quarterly Journal of Political Science, 15, 73-104.

26.

Hicken

(2011). Clientelism. Annual Review of Political Science, 14, 289-310.

27.

Hounkpe

Warren

(2012). Alp Benin interim country report (Technical report, African Legislatures Project).

28.

Humphreys

Weinstein

(2012). Policing politicians: Citizen empowerment and political accountability in Uganda preliminary analysis (International Growth Centre Working Paper S-5021-UGA-1).

29.

Keefer

(2007, October). Clientelism, credibility, and the policy choices of young democracies. American Journal of Political Science, 51, 804-821.

30.

Keefer

Khemani

(2014). Radio’s impact on preferences for patronage benefits (Policy Research Working Paper 6932). The World Bank. Retrieved from http://documents.worldbank.org/curated/en/840781468178733531/pdf/WPS6932.pdf

31.

Kitschelt

Kselman

(2013). Economic development, democratic experience, and political parties’ linkage strategies. Comparative Political Studies, 46, 1453-1484.

32.

Kitschelt

Wilkinson

S. I.

(2007). Patrons, clients, and policies: Patterns of democratic accountability and political competition. New York, NY: Cambridge University Press.

33.

Koter

(2013). King makers: Local leaders and ethnic politics in Africa. World Politics, 65, 187-232.

34.

Koter

(2017). Costly electoral campaigns and the changing composition and quality of parliament: Evidence from Benin. African Affairs, 116(465), 573-596.

35.

Larreguy

Marshall

Trucco

(2015, September). Breaking clientelism or rewarding incumbents? Evidence from an urban titling program in Mexico (Technical report, Harvard University working paper). Retrieved from https://scholar.harvard.edu/files/jmarshall/files/corett_v18.pdf

36.

Lieberman

E. S.

Posner

D. N.

Tsai

L. L.

(2014). Does information lead to more active citizenship? Evidence from an education intervention in rural Kenya. World Development, 60, 69-38.

37.

Lindberg

S. I.

(2003). “It’s our time to ‘chop’”: Do elections in Africa feed neo-patrimonialism rather than counter-act it? Democratization, 10, 121-140.

38.

Lindberg

S. I.

(2010). What accountability pressures do mps in Africa face and how do they respond? Evidence from Ghana. The Journal of Modern African Studies, 48, 117-142.

39.

Lizzeri

Persico

(2004). Why did the elites extend the suffrage? Democracy and the scope of government, with an application to Britain’s “age of reform.” The Quarterly Journal of Economics, 119, 707-765.

40.

Lusk

Norwood

(2009). An inferred valuation method. Land Economics, 85, 500-514.

41.

Magaloni

(2006). Voting for autocracy: Hegemonic party survival and its demise in Mexico. New York, NY: Cambridge University Press.

42.

McKelvey

Ordeshook

(1972). A general theory of the calculus of voting. Mathematical Applications in Political Science, 6, 32-78.

43.

Morris

Shin

(2002). Social value of public information. The American Economic Review, 92, 1521-1521.

44.

Mvukiyehe

Samii

(2017). Promoting democracy in fragile states: Field experimental evidence from Liberia. World Development, 95, 254-267.

45.

Pande

(2011). Can informed voters enforce better governance? Experiments in low-income democracies. Annual Review of Economics, 3, 215-237.

46.

Rueda

M. R.

(2016). Small aggregates, big manipulation: Vote buying enforcement and collective monitoring. American Journal of Political Science, 61, 163-177.

47.

Shefter

(1977). Party and patronage: Germany, England, and Italy. Politics & Society, 7, 403-451.

48.

Smith

Bueno De Mesquita

(2012). Contingent prize allocation and pivotal voting. British Journal of Political Science, 42, 371-392.

49.

Soumano

(2011). Mali interim country report (Technical report, Africa Legislatures Project).

50.

Stokes

S. C.

Dunning

Nazareno

Brusco

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

51.

Wantchekon

(2003). Clientelism and voting behavior: Evidence from a field experiment in Benin. World Politics, 55, 399-422.

52.

Weitz-Shapiro

(2012). What wins votes: Why some politicians opt out of clientelism. American Journal of Political Science, 56, 568-583.

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.

20.13 MB

0.00 MB

0.80 MB