REPORT BoliviaElectionsLatin America and the CaribbeanWorld

Nickels Before Dimes: Misleading Design in Escobari and Hoover’s “Natural Experiment”

Executive Summary

In October 2019, the Plurinational State of Bolivia held legislative and presidential elections. To win in the first-round vote, the leading presidential candidate would need to win a majority of the valid votes cast, or 40 percent with a 10 percentage point lead. Because of division within the opposition, surveys suggested that support for the incumbent, Evo Morales, lay close to the latter threshold.

With 84 percent of a rapid, preliminary count of the election results completed, the Tribunal Supremo Electoral (TSE) announced partial results. At that point, Morales led the next-closest candidate, Carlos Mesa, by 7.9 percentage points. However, the outstanding polling stations were expected to strongly favor the incumbent. A Morales victory in the first round was predictable based on the data available at the time of the announcement. Indeed, the final, official result showed Morales with a sufficient 10.56 percentage point lead.

However, the Electoral Observation Mission (EOM) of the Organization of American States (OAS) cried foul, declaring the updated results represented “an inexplicable change in trend that drastically modifies the fate of the election.”1 This lent international institutional legitimacy to opposition protests disputing the results. These protests turned violent, threatening the lives of high-level government officials and their families alike. And following a mutiny of police forces and a public call for Morales’s resignation by the head of the military, Morales and his vice president resigned and fled to Mexico, where the Mexican foreign ministry had granted Morales asylum given “the emergency situation that he faces in Bolivia, where his life and safety are at risk.”2 Shortly afterward, an extra-constitutional government headed by Jeanine Áñez was installed and Bolivian security forces acting under the de facto government’s direction violently repressed Indigenous protesters and perpetrated massacres at Senkata and Sacaba.

Since the OAS statement, several researchers have attempted to offer statistical evidence for the OAS claims. In October 2020, just days before new elections, Diego Escobari and Gary Hoover released an extensive paper arguing that they had identified fraud sufficient to change the outcome of the 2019 election.

None of Escobari and Hoover’s claims hold up to scrutiny. In one sense, this could be an issue of semantics. If, in choosing to work from overly simplistic models, Escobari and Hoover inferred merely that their models do not fully explain the trend of increased support for Morales in the later-reported election results, then their conclusions would be less controversial. In labeling this modeling failure “fraud,” they implicitly assert that they have exhausted all benign explanations. However, they make no actual claim of exhaustion.

Only at first do they employ a “natural experiment” approach to analyzing fraud. There, they assume that the partial results presented by the TSE the evening of the election were free of fraud, while the remaining, uncounted polling stations are potentially contaminated. However, the extent of fraud they claim to measure in these late polling stations is implausible. The official results in precincts counted entirely after the announcement would need to overstate Morales’s support by 84 percentage points for their conclusions to hold.

Escobari and Hoover later abandon the claim to a natural experiment and extend their search for fraud to every single polling station in the election by contrasting results with a nationwide referendum from 2016. However, the models they apply are not well suited for the task of measuring election fraud.

Rather, when applied to simple fraud-free synthetic election data, their methods fail, reliably detecting fraud where none exists. The natural conclusions are that their approaches to fraud detection are faulty and that their claims to have measured fraud in the 2019 Bolivian elections are not supported.

Read the full report here.

    Support Cepr

    If you value CEPR's work, support us by making a financial contribution.