Are Voters Updating when Audit Reports are Informative?

Our focus here is on individual-level priors regarding malfeasance in municipal governance and whether individuals update when they are informed about malfeasance in their local municipality. The information in question here is the 2020 municipal audits that will be conducted in 168 of the Chilean municipalities. The content of these audits will provide the basis for the information treatments that we implement in this experiment.

The malfeasance information treatment essentially summarizes the level of malfeasance the Contraloria reports for each municipality. All of the treated subjects see two pieces of information: One element is a simple count of the number of irregularities identified by the Contraloria audit. And a second provides a summary score that is generated by the Contraloria and reflects both the irregularities committed but also their seriousness and their importance relative to the size of the municipal budget. These are presented to subjects on a horizontal colour-coded scale anchored on the left by 0 (Green/Very Good) and on the right by 100 (Red/Very Bad).
%-- does a message about malfeasance in government matter? Scaling up is another matter. But ideally to address this first issue we want to compare a large N of individuals who are randomly assigned to treatment (and who we know are treated) to a large N of individuals randomly assigned to control (and who we know have not been treated).

Subjects are also informed about the relative malfeasance performance of their municipality. There are two versions of the information treatment: ``spatial'' and ``temporal''. A temporal benchmark compares the municipality's performance (irregularities plus score) to its previous audits. We do this by placing the benchmark information on an identical horizontal axis just below the actual audit results. A spatial benchmark version compares the municipality's performance to an average of municipalities from the last audit.

2020-07-20

2020-10-24

-Beliefs about levels of malfeasance
-Attitudes towards the municipal government
-Vote intention

The experimental design builds on an annual audit that is conducted by the CGO of municipal governments. For the year 2020, the CGO has selected 167 municipalities for an audit. Our audit information treatments are designed in conjunction with the CGO's roll-out of these municipal audits.

The audit outcomes will provide the content for the information treatments that we will implement via WhatsApp. We are leveraging the fact that the audit outcomes will vary -- the number of reported irregularities varies quite significantly across municipalities. Figure 1 in our pre-analysis summarizes the distribution of municipality irregularities reported by the Contraloria audits in 2016. On average, the audits report about 30 irregularities. The sample frame aims to have a balance of subjects in municipalities with above average, ``negative'', audit reports and with below-average, ``positive'', audit reports.

The treatment condition in the experiment is receiving a short video summarizing the results of the CGO audit report. Our subjects are individuals residing in any of the 167 municipalities audited by the CGO in 2020. The video treatments are customized for each municipality. As a result, we expect the treatments to range from very positive -- virtually no irregularities -- to very negative -- in the neighborhood of 90-100 irregularities. We can think of the treatment as a positive or negative "dosage'' rather than a dichotomous treatment arm. As we pointed before, the outcomes of interest are the subjects' beliefs about the prevalence of corruption in municipal government; their attitudes regarding municipal governance; and their vote intentions. The outcome variables are measured at three different periods in the experiment: pre-treatment; post-treatment; and pre-election.

The experimental design builds on an annual audit that is conducted by the CGO of municipal governments. For the year 2020, the CGO has selected 167 municipalities for an audit. Our audit information treatments are designed in conjunction with the CGO's roll-out of these municipal audits.

The audit outcomes will provide the content for the information treatments that we will implement via WhatsApp. We are leveraging the fact that the audit outcomes will vary -- the number of reported irregularities varies quite significantly across municipalities. Figure 1 in our pre-analysis summarizes the distribution of municipality irregularities reported by the Contraloria audits in 2016. On average, the audits report about 30 irregularities. The sample frame aims to have a balance of subjects in municipalities with above average, ``negative'', audit reports and with below-average, ``positive'', audit reports.

The treatment condition in the experiment is receiving a short video summarizing the results of the CGO audit report. Our subjects are individuals residing in any of the 167 municipalities audited by the CGO in 2020. The video treatments are customized for each municipality. As a result, we expect the treatments to range from very positive -- virtually no irregularities -- to very negative -- in the neighbourhood of 90-100 irregularities. We can think of the treatment as a positive or negative "dosage'' rather than a dichotomous treatment arm. An important consideration in recruiting the subject pool is ensuring balance over the range of irregularities illustrated in Figure \ref{fig:irregular}. As we pointed before, the outcomes of interest are the subjects' beliefs about the prevalence of corruption in municipal government; their attitudes regarding municipal governance; and their vote intentions. The outcome variables are measured at three different periods in the experiment: pre-treatment; post-treatment; and pre-election.

Randomisation will be done using R.

Individuals eligible to vote

No

10,000 eligible voters

10,000 eligible voters

1,500 voters Spatial/Low
1,500 voters Temporal/Low
1,500 voters = 1,500

750 voters Spatial/High
750 voters Temporal/High
500 voters Placebo

2,000 placebo pure control(non-audited municipalities)

For more details, please find our power calculations in our pre-analysis plan.