Carolina Torreblanca
University of Pennsylvania
Global Development: Intermediate Topics in Politics, Policy, and Data
PSCI 3200 - Spring 2026
For electoral accountability to work:
Today we test both sides
Today’s question: Can electoral accountability reduce corruption in the Global South?
For electoral accountability to discipline corruption, we need:
“A corrupt mayor who faces the possibility of reelection can exploit this information asymmetry to increase reelection chances by refraining from rent-seeking and behaving as a noncorrupt mayor”
How do you measure corruption?
Main measure: Share of audited resources found to involve corruption
Three types of corruption identified in audits:
Key feature: the CGU audit lottery randomly selects which municipalities get audited . . .
Comparison: first-term mayors (can be reelected) vs second-term mayors (cannot)
\[r_i = \beta I_i + \mathbf{X}_i \varphi + \mathbf{Z}_i \gamma + \varepsilon_i\]
Is this an experiment?
“Retrospective voting models assume that offering more information to voters about their incumbents’ performance strengthens electoral accountability”
Two competing predictions:
“Inspire the fight”
“Quash the hope”
Why might information backfire?
The intervention: door-to-door distribution of flyers, one week before 2009 municipal elections
Outcomes (from official electoral data, precinct-level):
\[Y = \beta_0 + \beta_1 \, CorruptionInfo + \beta_2 \, NoInfo + M_j + \epsilon\]
Does the effect depend on how much corruption there is?
\[\begin{aligned} Y = \beta_0 &+ \beta_1 \, Corr \times Low + \beta_2 \, Corr \times Med \\ &+ \beta_3 \, Corr \times High + \beta_4 \, NoInfo + M_j + \epsilon \end{aligned}\]
Information here “quashed the hope”.
| F&F (2011) | Chong et al. (2015) | |
|---|---|---|
| Country | Brazil | Mexico |
| Tests | Politician incentives | Voter information |
| Term limits | 2 terms | 1 term |
| Finding | Reelection incentives reduce corruption 27% | Corruption info decreases turnout and challenger votes |
Electoral accountability requires both incentives and information — but information alone may not be enough
Chong et al. find that for medium levels of corruption (33–66%), the effect of corruption information on turnout is:
\[\hat{\beta} = -0.30, \quad SE = 0.44\]
No stars! Recall: *, **, *** indicate \(p < 0.10\), \(p < 0.05\), \(p < 0.01\)
This coefficient is not statistically different from zero
So… is the effect zero?
Can we conclude the effect is zero?
Recall the interpretation of p-values:
Power is the probability of correctly accepting the alternative hypothesis
The probability of a true positive
Equals (1 - probability of type II error)
The common threshold in the discipline is 80% power
You can check out the EGAP power calculator to understand better
Which is “worse” in research?