Does Political Corruption Cumulate in Its Effects?

We will conduct an individually randomized control trial in Oaxaca City, Mexico. The experiment involves visiting close to 1750 voting age Mexicans in Oaxaca who already consented to participate in our previous study (https://doi.org/10.1257/rct.7770-1.0.) After asking baseline questions, we will randomly assign half of the urban sample to view a video of news footage of a second brother of the President taking a bribe. As detailed in the previous study, the urban sample includes 200 who had already seen a previous video of the first brother, as well as 200 who had received a video of opposition politicians, as well as 400 who had received a `nation-building' video, along with a financial exposure intervention. 150 had received a video about Mexico's economic performance, and 400 had received only a stock intervention. 400 were control. We cross-randomize the new treatment, assigning half in all these cells.

The aim of this new treatment is to examine the effects of information about corruption in a post-election setting, and also to examine whether effects of corruption cumulate with randomly assigned additional evidence. A cumulating effect is consistent with gradual Bayesian updating with strong initial priors. However, there are two reasons why there might be no cumulating effect. The first is that individuals may have already converged in their beliefs after the first dose of information, while the second is that they are unswayed by such information. we can test all of these, comparing individuals responses to the two waves of videos. It may also be a dormant effect: even if individuals do not immediately recall the first treatment, it may be the case that the `second brother’ video will have greater effects for those who had already been treated relative to those in the control who are exposed to this video.

We also anticipate differential treatment effects for ex ante stronger incumbent supporters. For those exposed to the incumbent treatment, these effects may be strengthened (due to greater updating) or weakened (due to motivated reasoning). However, since only 100 in the sample will have received both incumbent corruption videos, this intervention is underpowered, and therefore we consider it exploratory.

2021-09-13

2021-09-20

We will ask the following survey questions after the video.
1. Preference for democracy using questions adapted from Latinobarometer, including:
“In general, would you say you are very satisfied, somewhat satisfied, a little satisfied or not satisfied with the democracy in Mexico?”,
“On a scale from 1-10 we ask you to evaluate how democratic is Mexico”, etc.

2. Perceptions about the prevalence of corruption, including:
“How much progress do you think has been made in reducing corruption within State institutions in these past 2 years?”,
“Which party do you think is the most corrupt”, etc.

3. Survey measures of voting, including:
“If presidential elections were to be held this Sunday, which party would you vote for?”

We will ask them for which party they voted in the past election, before the video, to evaluate the effect of the first videos on real voting. We will also ask a question about trust in specific institutions: “Please state, for each one of the groups, institutions or people mentioned in the following list, how much trust do you have in them”

We will randomly assign 1/2 of a sample of 1750 individuals news footage of corruption.

Randomization done in an office by a computer.

Randomization at the individual level.

No

1750 individually randomized adults.

1750 subjects: adults with a valid voting ID

We will have 875 per treatment arm (though fewer in specific sub-treatment cells, see below).

Some approximate power calculations (that don’t include control variables) suggest that the power to detect changes in one of the main democracy variables may be fine overall but limited for examining differential effects across subtreatments. The democracy outcome variable we tested is a dummy variable with a mean of 0.27. In the first round of the experiment we found that the corruption video shown decreases preference for democracy as measured by this variable by 0.07. Based on that, we calculate that, to detect an effect of such size with 80% power and 95% confidence, we would need 600 observations per arm (that is 600 assigned to the 2nd video, and 600 assigned to no-2nd video). If instead we had say 200 observations per arm, the power we would have will be close to 33%. And if we had 400 observations per arm, power would be close to 61%. Given this, we seem to be well powered to compare the “overall” effect of the second video, across all previous treatment conditions. But our power to test for instance whether the effect of this new corruption video for those that got a first corruption video in the first round, is limited, and around 33%.