At least not in Benin. This week, I take a look at interesting paper  by Leonard Wantchekon documenting an experiment he did in Benin with this year’s presidential election. In this paper, Leonard compares the results from a deliberative sharing of a candidate’s platform in a local town hall against a one-way communication of the candidate (or his broker) with a big rally.
How does he do it? Leonard has an interesting history, and this gives him the unique ability to try things with national party politics. In this instance, he is trying to get some insight into how to reduce clientelism. So, working with the campaign managers of three top contenders for president, and taking a pretty much nationwide sample, villages are randomized into treatment and control. Treatment in this case is the town hall model: a research assistant and a campaign representative organize two town hall meetings -- one on education and health and a second on rural infrastructure and employment. Villagers debate policy proposals and these results are then transmitted on up the campaign hierarchy. In the control areas, there are 2-3 rallies, organized by a local political broker (e.g. an MP, or a local mayor) where either the candidate or the broker makes a speech covering the policies of the candidate. “There was no debate, but instead a festive atmosphere of celebration with drinks, music, and sometimes cash and gadget distribution.” Not surprisingly, a lot more folks come to these.
What was surprising to me is the relative cost. Rallies cost about $15 per participant while town halls cost about $2. And about 40% of the rally cost is a direct transfer to the broker (it’s also interesting to see that, over the course of the election, cash and gifts to voters appear to be evenly distributed in the end across both treatment and control – so direct payments to voters isn’t going to be driving things). One final wrinkle on the treatment – each of the treatment villages are also paired up with the incumbent or one of his two challengers to allow us to look at effects by candidate.
So what does he find? Turnout is higher in treatment areas, by around 5%. So this is the big headline: it’s just not efficient (from a getting votes out point of view) to do these big, expensive rallies. Stick to the town halls, and have a good discussion.
Now, Leonard and his team collected two types of data: village electoral returns and post-election surveys. These two data sources give somewhat different pictures of the effects. For voter turnout, while they both show an impact on overall levels, the electoral results suggest that turnout was only significantly higher for the challengers, not the incumbent. For the individual survey data, the results hold for both treatment and incumbent. There is also a difference in whether or not the treatment leads to higher electoral results for the treated candidate. In the village level data, there is no effect. But for the individual survey data, the effect of the town hall approach was to boost votes by 16 percent for the treated candidates – and this appears to be driven by a boost for the opposition, not the incumbent. So from a methodological point of view, it would be interesting to understand more about why these measures give us different results – but the paper doesn’t discuss this (in its current version).
Oh, and the election results? The incumbent won by a significant margin. And the former IMF official came in third. `
P.S. Many thanks for the interesting comments on last week’s data access post . Please keep those coming, and I hope to do a follow up post in awhile.