Do Informed Voters Make Better Choices? Experimental Evidence from Urban India

Transcription

1 Do Informed Voters Make Better Choices? Experimental Evidence from Urban India Abhijit V. Banerjee, Selvan Kumar, Rohini Pande and Felix Su October 20, 2010 Abstract Disclosure laws for politicians exist in over a hundred countries. But can disclosures about politician performance and qualifications influence electoral accountability in settings characterized by weak institutions and less educated populations? In the run-up to elections in Delhi we implemented a field experiment where we provided slum dwellers with newspapers containing report cards with information on candidate qualifications and legislator performance obtained under India s disclosure laws. We observe striking changes along three dimensions. Access to report cards increased voter turnout, reduced the incidence of cash-based vote buying and caused electoral gains for better performing incumbents. The turnout effects are more pronounced in more competitive jurisdictions and when the incumbent is a worse performer. We also observe significant voter sophistication voters make comparisons across candidates to overcome political agency problems and reward better performing incumbents. The authors are from MIT (Banerjee), Yale (Kumar) and Harvard University (Pande and Su). We thank our partners Satark Nagrik Sangathan, Delhi NGO Network and Hindustan Times and especially Anjali Bharadwaj, Amrita Johri and Mrinal Pande for enabling this study and Shobhini Mukherji for providing field oversight. We thank Hewlett Foundation and the Empowerment Lab at CID, Harvard for financial support and Tim Besley, Pascaline Dupas, Esther Duflo and seminar participants for helpful comments. 1

2 1 Introduction The poor numerically dominate the electorate in many low-income democracies, yet have largely failed to translate their political weight into effective service delivery and other economic gains (see, for instance, Mauro (1995); Hall and Jones (1999); UNDP (2002)). Explanations abound. Ethnicity-based clientelism may cause poor voters to value the politicians group identity (Horowitz, 1985; Chandra, 2004) and have a weak preference for honesty per se (Banerjee and Pande, 2009). Weak electoral institutions ballot stuffing, vote buying, voter intimidation may allow the political elite to subvert democracy (Acemoglu et al., 2010; Simpser, 2008). Yet another possibility is that voters are unable to identify politicians who would serve them well, either because they lack the information or because they are unable to interpret the available information. The empirical challenge of distinguishing between these views comes from the fact that weak institutions, clientelistic policies and poorly informed voter populations often coexist. In this paper, we make use of a large field experiment in urban India to provide rigorous evidence on one channel of influence information about politician performance and qualifications. Building on insights from political agency models, we test whether providing such information via the media influences voter behavior. Our field experiment occurred in the run-up to the 2008 state legislature elections in Delhi, India s capital. Legislators in Delhi are elected from single-member jurisdictions. Recent disclosure laws allowed our partner NGO to obtain detailed information on legislator performance and characteristics. 1 We published this information in the form of jurisdiction-specific report cards in a leading vernacular newspaper. Each report card contained information about incumbent performance along three dimensions legislative activity, committee attendance and spending of discretionary funds (divided into up to ten spending categories). It also provided information on the wealth, education and criminal record of the incumbent and two main challengers in that jurisdiction. In a random sample of two hundred slums households received a pamphlet on legislator responsibilities and a free copy of the newspaper that featured the report card for their jurisdiction. Households in the 575 control slums did not receive any informational material. Political agency models identify several reasons for why electoral accountability improves politician performance. In the incentive view, the threat of being voted out of office constrains politicians to act in the social interest (Barro, 1973; Ferejohn, 1986). In the selection view voters use elections to select politicians who they expect to serve their interests better in the future (Fearon, 1999; Besley, 2005). There is also the possibility of a natural interaction between these 1 These include the 2005 Indian Right to Information Act and a 2003 Indian Supreme Court ruling. 2

3 views if both the politician s type and his actions influence what they voters observe about them (see, for instance, Austen-Smith and Banks (1999); Banks and jan Sundaram (1998)). The publication of report cards was unanticipated by politicians and occurred after the last date for candidate entry. Hence, the primary channel of influence we identify is selection by voters. We observe change along three dimensions. The campaign increased voter turnout by 3.5 percentage points (from 57.5% to 59.5%). While it did not influence average incumbent vote share, worse performing incumbents and those facing better qualified challengers received significantly fewer votes. We do not observe significant changes in party campaigning. However, cash-based vote-buying was 19 percentage points less likely to occur in treatment polling stations. Survey data shows that the treatment increased citizens knowledge of legislator responsibilities and performance. As one would expect with a newspaper based campaign, the increases are concentrated among the literate population. We are also able to exploit features of our information campaign to rule out the possibility that the campaign worked by motivating rathe than informing voters. Across treatment slums, information about incumbent performance led to non-linear turnout effects, with voters more energized to turnout by worse performers. This finding is consistent with results reported in the literature (Washington, 2006; Bloom and Price, 1975; Hastings et al., 2007). We also observe stronger turnout impacts in treatment slums located in jurisdictions where the elections are more competitive. Alongside the incumbent vote share responded to information in a predictable manner. On performance, voters reward incumbents who spend more money in slums and have better attendance in oversight committees (in contrast, the extent of legislative activism left incumbent vote share unaffected). Both the electoral and survey data show that voters had a good sense of how different spending categories (roads, drains, sewage) translated into spending in slums in their jurisdiction. They used these data in deciding whether to reward incumbents for their spending decisions. In contrast, for committee attendance they use the performance of the incumbent in the neighboring jurisdiction who featured in the same newspaper as a yardstick. Similarly, in judging the relevance of qualifications voters compare incumbent and challenger qualifications within the jurisdiction. Incumbents who were richer or less well educated than their challengers received fewer votes. Our paper contributes to a growing empirical literature on electoral accountability. Several of these papers identify the net impact (selection and incentive) of information on policy-making in democratic setting (Besley and Burgess, 2002; Stromberg, 2004; Snyder and Stromberg, 2010). Broadly, increased media penetration makes governments more responsive to voter needs, as measured by policy outcomes. More closely related to our paper is Ferraz and Finan (2008) who show that electoral outcomes respond to information about incumbent corruption record. We extend this literature in several ways. 3

4 First, unlike the existing literature our intervention holds fixed party response and isolates the pure selection effect of information. Second, by considering measures of politician performance and qualifications we can directly evaluate the relevance of disclosure laws which make broad information about politician qualifications and performance available to voters. Our results demonstrate that voters are able to process performance information in a relatively sophisticated manner. Our results, thus, support the optimistic view of the power of information disclosures suggested by Djankov et al. (2010), based on the negative cross-country correlation between disclosure laws and corruption. By using newspapers as the channel of dissemination we also provide direct evidence on how the print media can increase accountability in a low income setting. In this sense, our paper is also related to several recent studies that relate information from media sources to turnout and governance outcomes (Stromberg, 2004; Gentzkow, 2006; della Vigna and Kaplan, 2007). More broadly, our findings further emphasize the importance of an independent and credible media source in enhancing the quality of government (Besley and Prat, 2006; Djankov et al., 2003). Finally, our results contribute to the growing experimental literature on voter behavior in low income countries (for an overview, see Pande (2011)). This literature builds on the insights of the US based Get Out the Vote literature (Gerber and Green, 2000). Papers in this literature shows significant turnout effects of non-partisan motivational campaigns (Gine and Mansuri, 2010; Banerjee et al., 2010) and to campaigns that exhort voters to use their electoral influence to protest against malpractices (Collier and Vicente, 2008). To the best of our knowledge, our field experiment is the first to evaluate the impact of performance and qualification information on voter behavior. The rest of this paper is structured as follows: Section 2 provides a conceptual framework to help interpret what we find. Section 3 describes the context, the experimental intervention and empirical design. Section 4 provides the results and Section 5 concludes. 2 Conceptual Framework We present a simple framework to identify how the distribution of report cards will influence citizens voting behavior. The release of report cards was unanticipated and occurred after candidate entry. Hence, we don t expect the campaign to influence politicians performance incentives. Thus, we anticipate that the availability of better information should influence electoral outcomes by improving voter inference. We focus on two aspects of voter decisionmaking: whether to vote and who to vote for. 4

5 2.1 Basic Model The economy consists of a single jurisdiction populated by three groups of voters. A fraction µ are committed voters. A fraction ξ of the committed voters are pro-incumbent and always vote for the incumbent. Since most incumbents in our data are from the Congress this can also be seen as the fractions of party loyalists. A fraction 1 ξ always vote for the challenger. The remainder (1 µ) are swing voters who are non-strategic and face a cost of voting. The cost of voting for voter k is c k, which is drawn independently from the distribution N(0, σ 2 c). Swing voters have preferences over politician performance. Politician i s performance depends on his/her type θ i and a cost of performance shock. Politician type is drawn from the distribution N(0, σ 2 θ) and the cost shock ε i is drawn from a distribution N(0, σ 2 ε). Citizens do not observe politician type or the cost shock. However, they know that a politician s expected performance (yi e ) is a linear function of her expected characteristic θ e i. y e i = θ e i + ε i By using this formulation we explicitly suppress the typical incentive effects in a political agency model and focus on the selection effects. The timing of the model is as follows. We begin with an incumbent in office. Nature determines his type and the cost shock. An election then occurs with the winning candidate choosing policy in period two. We assume the incumbent stands for re-election and that the challenger is randomly drawn with characteristic θ j. Citizens update their beliefs about candidates on the basis of signals received before the election. Citizen k votes for the candidate with a higher θ e i if she votes. She votes if θ e i -θ e j >c k and abstains otherwise. We consider the following structure for signals. For the incumbent each voter observes a noisy signal of yi 0, his true past period performance, given by y s ik = y 0 i + η ik where η ik is distributed as N(0, σ 2 η) and voters signals are iids For both the incumbent and the challenger each voter also observes a noisy signal of their true θ i. θ S ik = θ i + γ ik θ S jk = θ j + γ jk 5

6 where both γ ik and γ jk are drawn independently from N(0, σ 2 γ) for each k. We model the delivery of report cards as improving the precision of the signals. We assume throughout that σ 2 η < σ 2 γ. We are interested in how turnout and incumbent vote shares are influenced by the release of report cards. Incumbent Vote sharethe slope with respect to θ i and yi 0 of the fraction of votes going to the incumbent should go up for any fixed θ j. The slope with respect to θ j of the fraction of votes going to the incumbent should go up for any fixed θ i and yi 0. Turnout: if better information reinforces prior regarding relative performance of incumbent then turnout will go up. If better information changes relative ranking then effect ambiguous and turnout may fall Corollary: If better performance information causes incumbent vote share to increase and turnout to falls, then the model suggests that citizens, on average, had a more negative view of the incumbent than the challenger. 3 Experimental Design and Data 3.1 Setting A. Elections in Delhi Delhi is India s national capital and second-largest metropolis. It is also designated as an Indian state, and has an independent legislature composed of seventy elected legislators. Elections occur every five years with each legislator directly elected (via plurality rule) from a single member jurisdiction. It is easy to see why the hypothesis of limited information has prima facie plausibility in the context of the Delhi election and elections in developing countries, more generally: First, each legislator represents over a hundred thousand citizens: Few have talked to him or even met him. Second, a large majority of the poorer voters even in a place like Delhi, which has high literacy rates by Indian standards, do not regularly read newspaper, which are the main source of relatively unbiased information about politics and politicians. 2 Third, there has been a steady accretion in the responsibilities assigned to the legislators over the last two decades, as a part of an overall push towards decentralization and devolution of powers away from the 2 In a household survey among slum dwellers in our sample 40% of the men and 66% of the women stated that they do not read newspapers. 6

7 bureaucracy, with the consequence that voters do not know exactly what they should expect from their legislator. Our field experiment occurred in the run-up to the November 2008 State election. The three major parties contesting were the incumbent party Congress and two opposition parties Bhartiya Janata Party (BJP) and Bahujan Samaj Party (BSP). All three parties saw issues relating to the urban poor as central to their campaign. Congress campaigned on a platform of local development, and emphasized the regularization of slums undertaken since In contrast, BJP campaigned on the platform of controlling price rise and combatting terrorism. 4 The third party, BSP, did not have a particularly clear campaign platform other than the explicit claim of working for the benefit of poorer segments of society. Our campaign was timed to coincide with the two week official campaign period (ending 48 hours before polling starts). During this period party campaigning was widespread, especially in slums. According to newspapers and local observers, a couple of days prior to the election, political parties engaged in more direct vote-buying in slums and plied voters with bribes, most often in the form of liquor and cash. This was, for instance, reflected in the fact that the police reported a 400 percent rise in liquor smuggling cases two weeks prior to the election, with the Delhi excise department registering over 1,500 bootlegging cases in the month prior to the election. B. Public Disclosure Laws in India Our experiment makes use of two Indian disclosure laws. In October 2005 the Indian Right to Information (RTI) Act was implemented. The Act gives Indian citizens access to all non-classified government records. Under the provisions of the Act, any citizen may request information from a public authority and be legally entitled to an expeditious reply (typically within thirty days). It is estimated that roughly a million RTI petitions have been filed annually since 2005 (PricewaterhouseCoopers, 2009). Our partner NGO filed over 70 RTIs in 2008, through which it obtained information about the responsibilities of legislators in Delhi and incumbent performance in all jurisdictions. These disclosures allowed us to identify legislator responsibilities and collate three types of information about incumbent performance. We also use information available under the 2003 Supreme Court ruling that made it com- 3 In 2007, the Congress government initiated slum regularization, whereby slum inhabitants could purchase property rights to the government land they illegally inhabited at a heavily discounted rate. The regularization process also included a government drive to provide basic amenities to illegal settlements located on both public and private land, such as water supply, sanitation, drainage, and roads. 4 Their main campaign slogan was Mehengi Padi Congress ( Congress is Expensive ); during 2008, Delhi saw a sharp price rise in food items largely driven by the oil price surge). Coincidentally, Delhi elections occurred three days after the 26/11 attacks, when terrorists killed 173 people and wounded over 308 in Mumbai. Many people predicted that the Mumbai attacks would bolster the BJP in elections. 7

8 pulsory for candidates contesting national and state elections to submit affidavits at the time of filing their nomination paper, containing information on criminal charges, assets and liabilities and educational qualifications. 5 Our measures of candidate qualifications are based on affidavit information for candidates of the three major parties. 3.2 Report Cards The central plank of our information campaign was door to door distribution of newspapers containing report cards on legislator characteristics and performance. Here, we describe the performance and qualification indicators which were featured in the report cards. Table 1 summarizes these data. A. Performance measures The RTI responses told us that Delhi legislators have three main responsibilities. First, to attend the legislature and act as a voice for their constituency during the legislative process and to draw attention to the problems of their constituents when necessary. 6 Table 1 shows that mean attendance in the legislature in 2007 was 16.9 out of 18 sessions and nearly 70% of legislators had perfect attendance.there is significantly more variation in participation - during the 2007 legislative session, approximately half of the legislators asked no questions, while the remaining half asked between 2 and 77 questions each. Second, legislators participate in committees that provide oversight of local government institutions. The three main committees are the Ration Vigilance Committee, the Police Vigilance Committee, and the District Development Committee. The Ration Vigilance Committee is tasked with ensuring that the local ration shops, which provide subsidized food to belowpoverty-line residents, function effectively. The Police Vigilance Committee is supposed to provide oversight over the local police station, and ensure that the police do not harass locals, take bribes, or engage in other types of corruption, a common problem in India. 7 Across Delhi we observe that 70% of legislators attended their jurisdictions most recent ration committee meeting, 46% attended the police vigilance committee meeting and only 29% attended the district 5 This judgment was implemented by the Indian Election Commission, which stated that if a candidate does not submit the affidavit along with their nomination form, then this shall be considered to be violation of the Order of the Honorable Supreme Court and the nomination of the candidate concerned shall be liable to rejection by the Returning Officer at the time of scrutiny of nominations for such non-furnishing of the affidavit. Filing an affidavit disclosing this information is, therefore, a precondition for contesting. 6 Delhi legislature rules state that any question raised must relate to a matter of administration for which the Government is responsible. 7 The District Development Committee is a district-level committee that provides oversight of development projects. This committee is not convened by the legislator, though legislators do play a role as members. 8

9 development committee meetings. However, in our experimental sample no legislator attended the district development committee and we, therefore, disregard this category in the analysis. Finally, each legislator receives 20 million Rupees a year (roughly $ 45,000) to spend on development in their jurisdiction, known as the MLA LADS (Local Area Development Scheme) Fund, along with five million Rupees annually to be spent exclusively on water development (known as the Delhi Water Board Priority Fund). The legislator is responsible for fund allocation, with implementation undertaken by the involved municipal corporation. 8 A first indicator of spending performance is total spending under the scheme, since unallocated funds are lost to the jurisdiction (unspent money can be rolled over into the next year, and at the end of the term is returned to the exchequer). Our report card provided information on how legislator spent money between 2004 and The average legislator spent half the funds available to him. We also observe significant variation in MLALADS allocation across sectors. The most popular spending category was road and sidewalk development. Across Delhi, spending on roads and sidewalks accounted for an average of 60% of a legislator s total allocation, (on average, lakh out of lakh total). In contrast to MLALADS spending, we observe no variation in spending under Delhi Water Board Scheme and we, therefore, drop this from analysis. Our analysis exploits variation across report cards for our sample of ten jurisdictions along three dimensions: First, attendance and number of questions asked in the legislature in Second, legislator attendance at the most recent meeting of two committees: Ration Vigilance Committee and Police Vigilance Committee. Third, legislator spending from MLA LADS from 2004 to 2007 under the following categories; roads (including sidewalks), water (referring to water supply infrastructure such as borewells, pumps, and tanks), parks and statues, sewage (sewage pipes and public toilets), drains, lights, community halls, and boundary walls and others. B. Qualification Indicators Affidavits filed by candidates provided information on value of assets owned by the candidate and his/her spouse, criminal charges and educational qualifications. Ninety one candidates in the 2008 Delhi election had pending criminal charges. These candidates featured prominently on the rolls of the major parties (a quarter of the major party candidates faced criminal charges). Since some criminal charges may be extremely petty or 8 The procedure is as follows: after deciding on a particular development project the legislator drafts a proposal and gets an informal cost estimate from the relevant municipal corporation which will be responsible for actually implementing the project. Delhi has several different municipal bodies that are responsible for implementing development projects in different sectors. He then allocates money for that project and submits a formal proposal to the relevant municipal corporation. After performing a feasibility and cost analysis, the municipal corporation will either initiate work or ask for revisions to the proposal. 9

10 arguably related to being an active politician (e.g. rioting charges), we focus on heinous charges these are criminal charges which carry a minimum conviction of at least five years. Panel B of Table 1 shows that roughly 20% of incumbents in our sample faced henious charges. A common characterization of wealth in India is being a crorepati, i.e. have assets in excess of Rs. 10 million. In the 2008 election close to 20% of the candidates (153 candidates) were crorepatis. Delhi Election Watch, a consortium of NGOs that independently monitors elections, analyzed the change in personal assets of the 45 incumbents who were recontesting. The average increase in assets per MLA over a single five-year term was 211%, amounting to an average of almost 1.8 crore. Finally, candidates in Delhi are relatively well-educated. Overall, only 3% (18 candidates) were illiterate. 18% had up to 10 years of schooling, and 19% had up to twelve years of schooling. 19% held a college degree and 15% a post graduate or professional degree. The report card for a jurisdiction provided information on total assets, criminal charges and education qualifications for the three major party candidates (which always included the incumbent). 3.3 Experimental Design A. Sample Our sample comes from ten jurisdictions with a high density of slums and where the incumbent was standing for re-election. Table 1, column (5) shows that incumbent performance and characteristics in treatment jurisdictions was very similar to other jurisdictions. Within a jurisdiction our sample frame consisted of polling stations located in slums. Households are assigned to polling stations on the basis of a door to door survey conducted by the Indian Election commission. A polling station serves roughly 400 households (1000 adult voters) who live in the same or adjacent neighborhoods. In our jurisdictions we identified slum polling stations in conjunction with our partner NGO. Our sampling frame consisted of the 775 polling stations in the ten jurisdiction which served voters living in slums. In each jurisdiction, we randomly selected twenty polling stations for treatment. This yielded a sample of 200 treatment slum polling stations and 575 control polling stations. B. Intervention Our intervention targeted households with at least one adult voter who featured on the voter list of a treatment polling station. These households were exposed to a three-phase voter information campaign. 10

11 Door to Door Campaign The first phase was a door to door campaign, in which treatment households received a pamphlet containing three types of information. First, information about the voting process including the actual mechanics of voting, such as how to determine if one is eligible to vote, accepted forms of identity proof, and what to do if one felt they were being unfairly denied the right to vote. Voters were reminded that vote-buying is illegal and they should not let party workers accompany them to the polling station. and roles and responsibilities of an MLA (see Figure 2). Second, legislator responsibilities were listed. Third, voters were informed about politician disclosure laws and encouraged to read our partner newspaper to learn about candidates backgrounds. During this phase, no candidate-specific information was provided. On average, a two-member NGO team covered the households associated with a polling station in one and half days. Monitoring reports show that, on average, two-thirds of the households in a polling station were reached with the NGO spending 15 minutes at a household. Newspaper Campaign Between 20 and 25 November 2008 (roughly ten days before the election) our partner newspaper published report cards on incumbent performance and candidate qualifications. Each day the newspaper featured two report cards (an example is provided in Figure 3). The choice of the two jurisdictions featured in a single newspaper was made by the research team on the basis of geographic proximity. On the morning of the newspaper publication NGO workers placed a free copy of the newspaper on the doorstep of each household included in the treatment slums. Four hundred newspapers were disseminated in each slum, yielding a total delivery of 80,000 newspapers. After the distribution of newspapers, independently hired monitors visited 20 households in 172 of the 200 treatment polling stations to check for the presence of the Hindustan newspaper and door to door pamphlets. The free Hindustan newspapers were observed in 80% of households, and 76% of households either had a copy available or reported seeing the pamphlet that was distributed in the door to door campaign. Focus Group Discussions In the final campaign phase, NGO workers organized informal streetside focus groups, where they facilitated a detailed public discussion of the information provided in the newspapers and door-to-door campaign. These discussions occurred within 48 hours of newspaper distribution, and were motivated mainly by a concern that slum dwellers might not be able to contextually interpret the considerable depth of information that was being presented to them in the newspapers. At least one focus group discussion was conducted in neighborhood(s) associated with each treatment polling station. These discussions were held in a public locality in the neighborhood. Monitors were assigned to observe 155 focus group discussions. In 130 instances, monitors successfully located and observed the focus group discussions. These 11

12 meetings lasted an average of one and a half hours. On average, 20.4 women and 13.9 men were present at a focus group discussion. The monitoring reports indicate that the observed discussions mainly focused on the newspaper content, and complied with the campaign s directive to remain non-partisan. 3.4 Data Our empirical analysis utilizes several datasets. The first is official electoral returns which are available at the polling station-level. Here, the two outcomes of interest are voter turnout and incumbent vote share. Average voter turnout in the control polling stations was 57%, and the average incumbent vote share was 46%. Nine of the ten incumbents were from the ruling party (Congress). Nine of the ten incumbents in our sample won the election, and the margin of victory (in our sample) varied from 0.53% to 30%. The second is from a separate observational survey: in 29 treatment and 32 control polling stations, a surveyor spent approximately four hours on the eve of the election observing and noting whether there was any visible evidence of political party campaiging and also whether any individual or party worker is observed distributing cash, liquor, food, clothes or milk/refreshments as enticement for voting for a specific party. We use these data to construct several outcome variables. We measure whether the observer saw any door to door campaigning or any public rallies. Both types of campaigning was widespread. 95% of the polling stations witnessed door to door campaigning and public rallies were conducted in over 70%. Next, we examine whether the observer noted any instances of either cash or non-cash gift giving by outsiders in the slum. Third, we use data from a household survey that was conducted in the six day interval between election day and when results were announced. The survey was conducted in the 200 treatment polling stations and a randomly selected 200 control polling station localities. In each polling station ten individuals were randomly selected and administered a brief pop-quiz to test their level of civic knowledge. Our analysis also uses the report card data to identify whether voter responsiveness varied with incumbent performance and qualifications. On spending, the report card provided categorywise information. It is likely that categories differ in their relevance for slum dwellers for example, road spending may be less useful for slum dwellers who have unpaved roads. Therefore, after the elections we recruited the NGOs that had previously carried out the door-to-door campaign to identify whether each of the spending items that incumbent legislator had allocated money towards benefited slum residents. Each NGO was provided a list of all projects in their jurisdiction that had been allocated funding by the incumbent. The NGO then dispatched fieldworkers to visit the location of each individual spending item and assess whether or not it 12

13 mainly benefitted residents living in areas we had defined as slums. This allows us to identify ex post the fraction of spending that was in slums (overall and category-wise). In Table 2 we report a randomization check. Panel A uses electoral roll data and Panel B survey data from the household survey conducted just after the election. The average polling station had a thousand electors, and panel B shows that these electors are relatively poor the average per capita household income is a dollar a day. We observe balance on all covariates across treatment and control polling stations (we report the randomization check for three samples: the electoral data sample (775 polling stations), the observational data sample (61 polling stations) and the household survey sample (3896 respondents across 388 jurisdictions). 4 Results Since we are evaluating a randomized intervention, our basic empirical analysis is straightforward. A. Information Campaign and Voter Knowledge We start by using our survey data to examine whether the campaign influenced voter knowledge. We have two sets of outcomes. First, we asked each respondent eleven questions about the responsibilities of legislator, his performance and the qualifications of candidates in the election. We code the response for each question as correct or incorrect. Second, for a subset of spending categories we asked respondents to rank the work done by the incumbent. Here we consider the two largest categories of spending - roads and drains and create an indicator variable for whether the respondent believes the incumbent has done a lot of work. The impact of a newspaper-based campaign is likely to be more pronounced for literate respondents. Therefore, in addition to the average campaign effect we also examine whether the effect varies by respondent s literacy status (seventy six percent of the respondents in our sample are literate). For respondent i we estimate Y isj = α j + βt isj + ɛ isj (1) and Y isj = α j + β 1 T sj + β 2 L isj + β 3 T sj L isj + ɛ isj (2) where s denotes polling station and j jurisdiction. T sj is a dummy indicating whether the polling station was assigned to treatment. L isj is a literacy dummy. Since we stratified our treatment by jurisdiction we always include a jurisdiction fixed effect α j and cluster standard errors by 13

14 polling station. Table 3 reports the results. We start with the average score across the eleven questions. Respondent knowledge levels are low and the average respondent has a score of 2.3 out of 11. Overall, the campaign had a positive but insignificant effect on aggregate quiz score (column 1). However, we observe significant heterogeneity when we allow the effects to vary by respondent s literacy status (column 2). Literate respondents are able to answer half a question more and this effect is significantly accentuated by the treatment. In columns (3)-(5) we break down the aggregate score into three component parts knowledge about legislator responsibilities, candidate qualifications and spending behavior. In all cases, literate respondents demonstrate higher knowledge with the treatment accentuating this difference. The difference, however, is noisily estimated for spending and qualifications. To examine respondent knowledge about spending on public goods further we turn to category-specific responses. 32% of the respondents state that the incumbent had done a lot of work on roads. In column (6) we see that the likelihood that treatment causes the respondent to believe the incumbent did a lot of work is increasing in the fraction of total road spending that occurred in slums. In column (8) we see the same pattern for drains. These results are striking because the report cards provided category-specific spending but not broken down by whether it occurred in slums. In columns (7) and (9) we examine whether these effects vary by iteracy status and find such evidence for drains but not roads. B. Information Campaign and Electoral Outcomes: Average Effects The survey data in Table 3 shows that the campaign improved slum dwellers knowledge of incumbent performance. Given this, we turn to the campaign s impact on voter turnout and incumbent vote share. We also check whether the campaign influenced party campaigning and incidence of vote-buying. We estimate regressions of the form given by equation (3), where our outcome variables are measured at the polling station level. The results are in Table 4. In columns (1) and (2) we use data from official electoral returns at the polling station level as outcomes. First, we examine the impact on turnout. In column (1) the dependent variable is log votes (and log registered votes is included as a control variable). The campaign increased turnout by 3.6%. In column (2) we consider incumbent vote share as the outcome variable and do not observe any effect on average incumbent vote share. In columns (3) to (6) we use the observational data on party campaigning and vote-buying which is available for a much smaller sample of 61 slums (randomly selected from the treatment and control slum samples). Columns (3) and (4) show that campaigning was widespread on election eve but unaffected by treatment. In contrast in column (5) we observe a 19 percentage point 14

15 decline in the incidence of cash bribes in treatment polling stations. We do not observe a decline in non-cash vote-buying, prominent among which is distribution of liquor. These findings are consistent with several theories of voter choice. For instance, the information campaign may have simply increased awareness of democratic practices. This caused more citizens to vote and reject vote-buying practices but did not influence voter preferences for the incumbent. This explanation would be consistent with the fact that parties did not change campaigning practices. Alternatively, the absence of an average campaign effect may reflect voter sophistication in processing information. If citizens voted to reward better performing incumbents and punish those who are doing badly, as the political agency model would predict, we would not expect any obvious effect on the average incumbent vote share across jurisdictions. Below, we examine this hypothesis. C. Do Campaign Effects Vary with Incumbent Performance? Our report cards provided information along three dimensions of incumbent performance - legislative behavior, committee attendance and spending (both overall and by category). To summarize incumbent performance we conduct a principal component analysis using data on legislative performance, committee attendance, total spending and fraction slum spending. We use the highest eigenvalue as a summary statistic. For this summary indicator and for each separate performance indicator we estimate specification: Y sj = α j + β 1 T sj + β 2 P j T sj + ɛ sj (3) where Y sj is the incumbent vote-share (or turnout) and P j is the performance indicator for the incumbent. Columns (1) and (2) show that turnout was higher when the incumbent was a worse performer and the incumbent s vote share is increasing in his performance. In columns (3)-(10) we separately evaluate the different components. Columns (3)-(4) show that information about an incumbent s attendance in the legislature and his record of asking questions in the legislative assembly did not influence voting outcomes. This is consistent with the view that poor slum dwellers see the main responsibilities of their legislator as relating to local development and grievance redressal not the enacting of bills. Columns (5)-(6) consider committee attendance. The report cards provided information on whether last meeting of the committee was held according to schedule and whether the incumbent attended the meeting. We observe no impact of committee attendance on turnout. However, in column (6) we see that the average attendance of the incumbent is a significant predictor of incumbent vote share. All else equal, committee attendance increases the incumbent s 15

16 vote share by 5 percentage points. In Columns (7)-(8) we examine whether turnout and an incumbent vote share is sensitive to the extent of discretionary fund spending. Over their five year term, the incumbent could have spent upto Rs. 100 million. However, the average incumbent in our sample only spent Rs. 50 million arguably, lack of spending measures an unwillingness on the part of the incumbent to exert effort. 9 We do not observe any responsiveness of voters to the total amount spent by the incumbent. Next, we investigate the thesis that voters care about the nature of spending rather than the total amount spent. The report card listed incumbent spending by category - these included roads, drainage, light, parks etc (the full list is in the report cards shown in Figure 3). As a first cut, we parse these data through the lens of whether the spending was more likely to occur in a slum. Columns (9) and (10) present the results. Turnout is decreasing in amount spent in slum while incumbent vote share is increasing in slum spending. This suggests that the big suprises in terms of incumbent performance tended to be negative. In Table 6 we examine whether these results are present when we consider category-wise spending. Since legislators varied in how they allocated funds we restrict attention to the three largest categories of spending - roads, parks and drains. All legislators spent funds in these three categories. Looking across the table, we see that for all spending items, voters care about slum-specific spending in the category. Turnout is higher when incumbent spends less in slums while incumbent vote share is increasing in slum spending in each category. D. Do Campaign Effects Vary with Challenger Qualifications? In addition to performance information the report cards also provided data on candidate qualifications. We summarize the data by creating an average rank for the incumbent. This averages the incumbent s rank on each of three qualifications college education, net assets and criminal charge when we compare him to the the two main challengers in the jurisdiction. For this indicator, we estimate a regression of the form given by equation (3). In addition, for each of the three qualification categories we estimate Y sj = α j + β 1 T sj + β 2 Q j T sj + β 3 CQ j T sj + β 4 CQ i T sj + ɛ sj (4) where Q j is the relevant qualification and CQ j is the fraction of challengers who have this qualification. As a robustness check we also include the fraction of challengers in the neighboring jurisdiction i with the qualification CQ i This information should be irrelevant for the voter. As 9 Legislators also had access to a separate fund for spending on water related issues. Interestingly, all incumbents spent the entire amount of this funding. 16

17 before, the outcome in odd columns of the Table is turnout and in even columns is vote share. Table 7 reports the effects of candidate qualifications. Columns (1) and (2) consider the summary rank measure. We observe no significant impact on turnout. However, incumbents with a worse rank receive fewer votes. Columns (3)-(8) consider different qualifications. In columns (3) and (4), the incumbent s characteristic is having a heinous criminal charge against him (heinous criminal charges are those that come with at least 5 years of prison time if convicted). The challenger variable is the fraction who also have such a charge. We observe no significant effect, perhaps because the incidence is relatively low. 10 In columns (5)-(6) the incumbent qualification is his wealth and the challenger variable is their average wealth. Here, we choose to use a wealth indicator that is salient in the particular cultural context. An incumbent s qualification is having more than a crore (10 million) rupees in declared wealth, while the challenger variable is the fraction of challengers who do not have that much. Column (5) tells us that turnout is responsive to challenger qualifications similar to our results for performance indicators, turnout is higher when challengers are better qualified (i.e. when the incumbent looks worse). In column (6) we see that the incumbent also receives significantly more votes when his opponents are crorepati s. If it is disclosed that both his opponents are crorepatis his vote share is 7.2% higher than when it is discolosed that both of them are crorepatis and he is not. Finally, in columns (7)-(8) we consider education. Interestingly, unlike the performance indicators here turnout is increasing in incumbent education. In column (8) we see that not being as educated as the opponents hurts the incumbent: the incumbent variable is whether he does not have a college degree and the challenger variable is the fraction of them who don t have a college degree. If neither opponent has a college degree then the incumbent s vote share is 6.8% higher. These results, with the possible exception of the weak criminality results, accord with intuition. The poor are suspicious of rich candidates, either because they feel that the rich are less likely to care about what they care about or because they see wealth as a signal of corruption, but like candidates who are educated, probably because education signals competence. They also suggest that voters are quite sophisticated in how they interpret evidence: For one we only get significant results when we control separately for the incumbent s and challenger s characteritics. This suggests that people are not reacting naively to the incumbent s wealth or education, but comparing him with the challengers and favoring the one that looks better to them. 10 The results are similar when we use any criminal charge but since any criminal charges includes charges that politicians often end up with while doing their job (being a demonstration, for example) this variable is intrinsically less interesting and is not reported. 17

18 Another important check on voter rationality involves including the qualifications of candidates in the neighboring jurisdiction that was also featured in the same issue of the newspaper in the qualification regressions. Given that those candidates are not in the choice set, their presence should not matter, and this is indeed what we find. E. How Do Voters Compare Across Dimensions? The qualification regressions suggest that voters make comparisons across candidates in deciding how to vote. The newspapers we distributed contained two report cards next to each other. We can, therefore, examine whether voters also engage in yardstick competition. TBC: see Table8 5 Conclusion The idea that voters in an otherwise well-functioning democracy might be severely constrained by information about the candidates qualifications and past record is both striking and important. We see that voters when given the information move quite substantially and if this information had reached the entire jurisdiction, outcomes may have been quite different. We also see evidence that voters are somewhat sophisticated in how they use the information, allaying fears that information would simply confuse them. References Acemoglu, D., J. Robinson, and R. Santos-Villagran (2010). The monopoly of violence: Evidence from colombia. Austen-Smith, D. and J. Banks (1999). Electoral Accountability and Incumbency. Models of Strategic Choice in Politics. Ann Arbour: University of Michigan Press. Banerjee, A., D. Green, J. Green, and R. Pande (2010). Can voters be primed to choose better legislators? Experimental evidence from rural India. Banerjee, A. V. and R. Pande (2009). Corruption. mimeo, Harvard. Parochial Politics Ethnic Preferences and Politician Banks, J. and R. jan Sundaram (1998). Optimal retention in agency problems. Journal of Economic Theory 82, Barro, R. (1973). The Control of Politicians: An Economic Model. Public Choice (). 18

19 Besley, T. (2005). Political selection. Journal of Economic Perspectives 19, Besley, T. and R. Burgess (2002, November). The political economy of government responsiveness: Theory and evidence from india*. Quarterly Journal of Economics 117 (4), Besley, T. and A. Case (1995). Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition. American Economic Review 85 (1), Besley, T. and A. Prat (2006). Handcuffs for the grabbing hand? media capture and government accountability. American Economic Review 96 (3), Bloom, H. and D. Price (1975). Voter response to short run economic conditions: The asymmetric effect of prosperity and recession. American Political Science Review 69, Chandra, K. (2004). Why Ethnic Parties Succeed: Patronage and Ethnic Headcounts in India. Cambridge: Cambridge University Press. Collier, P. and P. Vicente (2008). Votes and violence: Evidence from a field experiment in Nigeria. Center for The Study of African Economies Working Paper Series. della Vigna, S. and E. Kaplan (2007). The fox news effect: Media bias and voting. Quarterly Journal of Economics 122, Djankov, S., R. LaPorta, F. L. de Silanes, and A. Shleifer (2010). Disclosure by politicians. American Economic Journal: Applied Economics, forthcoming. Djankov, S., C. McLiesh, T. Nenova, and A. Shlefier (2003). Who owns the media? Journal of Law and Economics 46 (2), Fearon, J. (1999). Electoral Accountability and the Control of Politicians: Selecting Good Types Versus Sanctioning Poor Performance. Democracy, Accountability and Representation. Cambridge: Cambridge University Press. Ferejohn, J. A. (1986). Incumbent Performance and Electoral Control. Public Choice 30 (). Ferraz, C. and F. Finan (2008). Exposing corrupt politicians: The effects of brazil s publicly released audits on electoral outcomes. Quarterly Journal of Economics 123 (2), Gentzkow, M. (2006). Television and voter turnout. The Quarterly Journal of Economics 121 (3), Gerber, A. and D. Green (2000). The effects of canvassing, telephone calls, and direct mail on voter turnout: A field experiment. The American Political Science Review 94 (3),

20 Gine, X. and G. Mansuri (2010). Together we will:evidence froma field experiment on female voter turnout in pakistan. Hall, R. and C. Jones (1999). Why do some countries produce so much more output per worker than others? Quarterly Journal of Economics 114 (1), Hastings, J., T. Kane, D. Staiger, and J. Weinstein (2007). The effect of randomized school admissions on voter participation. Journal of Public Economics. Holmstrom, B. (1982). Moral Hazard in Teams. The Bell Journal of Economics 13 (2), Horowitz, D. L. (1985). Ethnic Groups in Conflict. Berkeley: University of California Press. Mauro, P. (1995). Corruption and growth. Quarterly Journal of Economics 110 (3), Pande, R. (2011). Can informed voters enforce better governance? experiments in low income democracies. Annual Review of Economics 3, forthcoming. PricewaterhouseCoopers (2009). Final report: Understanding the key issues and constraints in implementing the RTI act. Simpser, A. (2008). Cheating big: On the logic of electoral corruption in developing countries. Snyder, J. and D. Stromberg (2010). Press coverage and electoral accountability. Journal of Political Economy 118 (1), Stromberg, D. (2004). Radio s impact on public spending. The Quarterly Journal of Economics 119 (1), UNDP (2002). Human development report 2002: Deepening democracy in a fragmented world. Washington, E. (2006). How Black Candidates Affect Voter Turnout. Quarterly Journal of Economics 121 (3),

21 :"-%$&;(&!1%58*5"#&36"/"*.$/578*7&&"#4&<1.$/&2$6"=51/(&>+,,"/?&>."878*7 K(%8#TB6#L8:5&](843 $%&'"(9#%89#8B8Z K%a=# b6%:ab8#c# 3%&'"(9#"(123"%4B63`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

22 <?&&69)*=&( RIS6+,-6)#A6+.-T6 RIS6+0-6)#A6+1-T6 DE9(%F)G"#69)*=&( RIS6+3-6)#A6+4-T6!"#$%"& '%()$*(#$ =UF)&?(6!"#$%"& '%()$*(#$ =UF)&?(6!"#$%"& '%()$*(#$ =UF)&?(6!"#$%&'(&)%$*+,-"%&.,%%/ +, / '"$)&67&(8$"%9,:,/;533,:::;1,: :;034,:,3;.,,,::.;102 :;2:3 5//;:50,:00;:/0 :;/21 +/.,;:,:- +/:1;254- +/.4;0/3- +/:1;4/3- +/35;10.- +/30;403- '"$)&6<(*)&(67&(8$"%9 0.0; ;051 :;34/ 0,4;/51 0.4;2:/ :;043 /4.;1/, 00:;,:/ :;../ +,0.;/5.- +,/4;,.0- +,/5;5::- +,/3;51,- +,01;/:3- +,3/;2..- 0;0:0 0;10. :;044 0;/3/ 0;112 :;/10 0;.5. 0;54: :;/:3 +.;:25- +.;,/1- +.; ;,1/- +.;415- +,;253-7&(8$"%6BC( /1;1:, /1;/// :;5.2 /1;.,, /1;/:3 :;205 /0;242 /1;025 :;.01 +.;/0:- +.;0:4- +.; ;0,.- +,; ;54:- DE9(%F)G"#9 131.::,50,50 /..5!"#$%&0(&12-3$4&5"+" <(*)&( :;052 :;055 :;3/5 +:;1::- +:;1::- BC( /2;052 /1;5,4 :;.5, +,/;/,5- +,.;5,.- >"?9(6HIJ( 1;51/ 2;:53 :;,01 +.;3:.- +.;5.0- 2/42; ;,41 :;.23 +1/.0;/ //;435- O)G"#6!)%A6>"&A(% :;4,3 :;4,5 :;523 +:;/43- +:;/41- PI$(%)$( :;32, :;332 :;01: +:;0.2- +:;0,3- K?9&I* :;,1/ :;,4. :;.1, +:;/2:- +:;/42- P"Q6!)9$( :;14: :;14/ :;5:3 +:;050- +:;05/- $Q"68"&?*#9;6W)#(&6B68)&8?&)G"#96I#8&?A(6Z?%I9AI8G"#6[\(A6(S(8$96)#A6%"E?9$69$)#A)%A6(%%"%9;6W)#(&6Y668)&8?&)G"#96I#8&?A(6Z?%I9AI8G"#6[\(A6(S(8$96)#A69$)#A)%A6(%%"%968&?9$(%(A6EL6 ="&&I#C69$)G"#;

23 +,-)'./-*%&# 0$"&1'&2 73J))7+,"#*%&#7K")%L: : 3;<7=7;>: 34<=74>: QC;7:=;8: E5:;C7:; QC;7:=;8:1/1E5:;C7:; QC;7:=;8:1/1Q>:7I1DEF1EFAY1 YH;8958G1!_O`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aL/9;<39= 6,4,"&,4,%$ J0F9/E!,4,%%#!,4,$(# -./012/315Q<22FR//==,4,$%,4,))55 AR/3E03K/!,4,'%#!,4,%+# 6,4,$+,4,,* 6,4,"& 6,4,,( CD/3EF3G=!HIJ=K.<./#!,4,$)#!,4,"&#!,4,%,#!,4,"'# 6,4,"'55,4,"&55 CD/3EF3G=F3=C>L29!,4,,(#!,4,,(# 789/.:0;<39 ))' ))' ))' ))' ))' ))' ))' ))' ))' ))' 5")/67 N.<2=0=D.F3KFD0>=K<2D<3/31=030>[9F9=<N=\]9K<./!1<10>=9D/3EF3G#Y=\]9K<./!1<10>=9>L2=9D/3EF3G#Y=.0;<3=K<22FR//=0R/3E03K/Y=D<>FK/=K<22FR//=0R/3E03K/Y=03[=>/GF9>01L./=^L/9;<39=09_/EY=03E= \]9K<./!>/GF9>01L./=0R/3E03K/#4=`=9K<./9=L9/=2/039=03E=9103E0.E=E/:F0;<39=N.<2=1O/=",6TL.F9EFK;<3=902D>/4 %4=-O/=E/D/3E/31=:0.F08>/=N<.=<EE63L28/./E=K<>L239=F9=><G=3L28/.=<N=:<1/.9=01=D<>>F3G=910;<3P=1O/9/=./G./99F<39=F3K>LE/=0=K<31.<>=N<.=><G=3L28/.=<N=./GF91/./E=:<1/.94=-O/=E/D/3E/31=:0.F08>/= N<.=/:/363L28/./E=K<>L239=F9=D/.K/31=<N==:<1/9=N<.=F3KL28/31=01=D<>>F3G=910;<34

26 A&";, B"'C, )'"*+, '"()*+,-"+#. /-0+12$-#34"#$3567,$ '"()*+,-"+#. /-0+12$-#34"#$3567,$ '"()*+,-"+#. /-0+12$-#34"#$3567,$!"#!$#!%#!&#!'#!(#!)#!*#!+#!",#!""#!"$# -./012/31,4,&%,4,),,4,$),4,,',4,&&55,4,$(,4,,",4,"',4,$',4,$* 6,4,"( 6,4,"*!,4,'%#!,4,')#!,4,%,#!,4,%$#!,4,$"#!,4,$'#!,4,"(#!,4,"*#!,4,$+#!,4,$+#!,4,$$#!,4,$"# 6,4,,% 6,4,,& 6,4,,) 6,4,,( 6,4,"%,4,&&,4,,& 6,4,&,5,4,$,,4,(,,4,&,,4,"( MG/3BE3F?E3?=01/F<.J?!TUV?D.<./#!,4,"&#!,4,"&#!,4,,+#!,4,,+#!,4,"+#!,4,&&#!,4,"&#!,4,$'#!,4,("#!,4,()#!,4,%'#!,4,%(# 6,4,$+55,4,$'55 6,4,)"5,4,'(55 6,4"''555,4,+$55 MG/3BE3F?E3?M>C29?E3?=01/F<.J!,4,""#!,4,"$#!,4,%+#!,4,$%#!,4,'&#!,4,&$# 789/.:0;<39 ))' ))' ))' ))' ))' ))' ))' ))' ))' ))' ))' ))' "4?K/?./G<.1?7LM?./F./99E<39?D<310E3E3F?NC.E9BED;<3?OP/B?/Q/D19?03B?.<8C91?9103B0.B?/..<.94?R>>?D<>C239?C9/?/>/D1<.0>?.<>>?03B?./G<.1?D0.B?B0104?R>>?:0.E08>/9?0./?09?B/O3/B?E3?3<1/9?1<?-08>/?'4 $4?=<31.<>?2/039?A<.?<BB63C28/./B?D<>C239?E3D>CB/?D<31.<>6F.<CG?G<>>E3F?910;<39?HI<9/?E3DC28/31?9G/31?0?>/99?1I03?0:/.0F/?02<C31?<3?1I01?D01/F<.J4?=<31.<>?2/039?A<.?/:/363C28/./B?D<>C239? E3D>CB/?D<31.<>6F.<CG?G<>>E3F?910;<39?HI<9/?E3DC28/31?9G/31?0?>/99?1I03?0:/.0F/?02<C31?<3?1I01?D01/F<.J?8<1I?<:/.0>>?03B?E3?9>C294

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`?,-L,-99D<19?F<1/.D1D1L?\C,D9HDF;<1?Ea-H?-b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

29 Figure 1: MLALADS Booking of Slum-Area Projects across 10 Sample Assembly Constituencies (*Road figures divided by 10 for scale)

30 Figure 2: Pamphlet Distributed by the Delhi NGO Network

31 Figure 3: Report Cards in The Hindustan Times on November 24, 2008