Establishing Responsive Linkages between Politicians and Voters

In the first stage, we randomize integrated voice response robo-calls from the politician to voters in his constituency. In the second stage, we present politicians the aggregated responses of citizens and then randomize the type of response from the politician back to the voters. We repeat this over several months to create a new channel of communication between a politician and his constituents.

Politicians individually script their own calls, with our assistance. The calls consist of questions about upcoming decisions that the politician has to make, his recent activities, and requests for feedback on both parliamentary and constituent affairs. Feedback can be given using the touch pad or through recorded messages to the politician.

2017-08-01

2018-12-31

Measures of voter efficacy, politician effort, accountability, and voting behavior.

We randomize our intervention at three levels: the politician level, the polling station level, and the household level.

At the politician level, we randomize whether or not a politician has access to the new technology. From a list of provincial parliamentarians with whom we are able to collaborate, we select 20 politicians to be in the treated group who we will work with over the course of this intervention and sort 20 politicians into a control condition.

For each treated politician, we randomly select 5 polling station areas in his constituency where we carry out the intervention. In these polling stations, we collect phone numbers at baseline and then place robo-calls and collect feedback from constituents. In control polling station areas, we carry out no activities at the level of voters. We split our 5 treated polling station areas into two groups: 3 high saturation and 2 low saturation polling stations. In high saturation polling stations, we survey 50% of households in our baseline and treat a subset of those. In low saturation polling station areas, we survey 25% of the households in our baseline. This saturation design allows us to test for spillover effects of our intervention.

Within treated polling station areas, we treat roughly 60% of the sampled households. For example, this means in the high saturation polling stations (where we sample 50% of the households), we end up treating a total of 30% of the households in that polling station area. In the low saturation polling stations (where we sample 25% of the households), we end up treating a total of 15% of the households in that polling station area. At the individual level, we randomize: (1) the number of messages they receive; (2) the types of questions they are asked; (3) whether they give no feedback, feedback just using their touch pad, or recorded feedback; and (3) how responsive the MPA is to their feedback in the form of a follow up robo-call or personalized message.

Because our respondents are all male heads of households, but women vote in separate polling stations, we can also measure intra-household spillover by examining outcomes at female-only polling stations in treated and control areas.

Randomizations using computer.

Some treatments will be administered at the politician level, some at the polling station level, and some at the household level. We randomize at all three levels.

Yes

We put 20 politicians into treatment and 20 into control; treating 5 polling stations per politician; treating roughly 75 or 150 households per polling station area.

20 treated politicians and a variable number of control politicians depending on the outcome, 100 treated and a variable number of control polling stations depending on the outcome, and 12,000 male heads of household.

20 treated politicians and a variable number of control politicians depending on the outcome, 100 treated and a variable number of control polling stations depending on the outcome, and 12,000 male heads of household divided among several treatment arms.