Diminishing the Effect of Vote-buying on Electoral Outcomes in India: A Pilot RCT to Test the Effectiveness of Radio Messages

Vote-buying is a widespread practice in developing countries that may undermine the accountability and responsiveness of elected officials, especially to the needs of the poor. Vote-buying may derive its effectiveness in influencing voter behavior by appealing to internalized norms of reciprocity, conveying information about candidates or because of lack of faith in the secret ballot. One way to reduce the incidence of vote-buying is by diminishing its effectiveness in influencing electoral outcomes. We designed and deployed a radio campaign during the April-May 2014 general election period in India.

Our radio campaign consisted of a series of three 60-second spots. Each radio spot consisted of a satirical vignette about the hidden social costs of vote buying. The radio spots highlighted the incentive of vote-buying candidates, which is to earn returns on their “investment” on getting elected. Further, it raised questions on the credibility of their election promises. Finally, it reminded voters of the secret ballot and urged them to vote for honest candidates and punish vote-buying candidates. Through the radio messages our goal was to diminish positive voter reciprocity (indeed we appeal to their vindictiveness by urging them to punish vote-buying candidates), undermine the signaling value of the bribes and remind voters of the secrecy of the ballot. We produced the 3 spots in 5 major languages and broadcast the spots in the local language of the radio station. Each spot aired 16 times (8 primetime and 8 non-primetime) over a period of 3 consecutive days (starting 2 days prior to polling and including the day of polling). Thus each treatment radio station broadcasted 48 spots over 3 days. The Appendix contains scripts of the radio spots aired.

2014-04-08

2014-05-07

The outcome of primary interest in this study is the effect of the treatment on the vote-share of candidates, particular vote-buying candidates. The other outcome of interest is the effect of the treatment on voter turnout, particularly in constituencies with high levels of vote-buying. In order to improve the precision with which the average treatment effects are estimated, we will present both the unadjusted difference-in-means estimate of each average treatment effect as well as a regression-adjusted estimate using lagged outcome variables (i.e., vote-share or voter turnout in the previous (2009) general election, respectively) as covariates.

Since vote-buying cannot be observed directly, we use different proxies for vote-buying by candidates and for aggregate levels of vote-buying in constituencies. This is described is detail in the pre-analysis plan.

We randomly assigned 30 radio stations from a list of 60 to receive an anti-vote buying radio campaign. We interviewed local journalists on aspects of campaigning and vote-buying in constituencies they cover. We attempted to conduct all the interviews before the radio campaign started and did not, during the course of the interviews, reveal any information about the intervention. We aimed to get responses from two journalists for each constituency. Since vote-buying cannot be observed directly, we use different proxies for vote-buying by candidates and for aggregate levels of vote-buying in constituencies.

The radio stations range in power between 1 KW and 20 KW, with 40 radio stations of 6 KW power. Of the 60 stations, 50 are FM and 10 are AM stations. The expected radii of the coverage areas lie between 50 km and 125 km. The coverage areas of the radio stations contain parts of different constituencies (“segments”). This constituency segment is the level of analysis in this study. We divided the 60 radio stations into 5 blocks based on the expected day of election in that region. Half of the radio stations in each block were randomly assigned to receive the radio campaign (“treatment” group), and the other half were to not receive it (“control” group). Thus, the probability of assignment to treatment is equal to half and is the same in all 5 blocks.
There are 543 constituencies in the country, out of which 338 are in states in our sample. Constituencies that are not completely covered by any of the 60 stations will be excluded from the sample. We do not currently have a detailed mapping of the radio propagation zones, but we expect our final sample to consist of approximately 275 constituency segments. There are on average around 1.5 million registered voters, 15 candidates and 1500 polling stations per constituency. Our focus will be on turnout and on voting for main candidates. Electoral returns are released at the polling station level. We have the locations for all polling stations from the Election Commission of India (ECI). We will aggregate the electoral returns from the polling stations that lie within each constituency segment. We identified the main candidates based on the journalist interviews. The experimental design is discussed in detail in the pre-analysis plan.

The randomization was done using Stata code.

The unit of randomization was the radio station. Radio stations were divided into 5 blocks. Half of the radio stations in each block were assigned to treatment. The probability of assignment to treatment is half and is the same across blocks.

Yes

60 radio sations

Approximately 275 constituencies

Approximately half in treatment and half in control.

Pre-analysis+plan+AEA+registration.pdf