As Bolivia gears up for a do-over election on May 3, the country remains in unrest following the Nov. 10 military-backed coup against incumbent President Evo Morales.
The military-installed government charged Morales with sedition and terrorism. A European Union monitoring report noted that some 40 former electoral officials have been arrested and face criminal charges of sedition and subversion, and 35 people have died in the post-electoral conflict. The highest-polling presidential candidate, a member of Morales’s Movimiento al Socialismo (MAS-IPSP) party, has received a summons from prosecutors for undisclosed crimes, a move some analysts suspect was aimed to keep him off the ballot.
The media has largely reported the allegations of fraud as fact. And many commentators have justified the coup as a response to electoral fraud by MAS-IPSP. However, as specialists in election integrity, we find that the statistical evidence does not support the claim of fraud in Bolivia’s October election.
The OAS claimed that election fraud had happened
The primary support for claims of fraud was the OAS report. The organization’s auditors claimed to have found evidence of fraud following a halt in the preliminary count — the nonbinding election-night results meant to track progress before the official count.
The Bolivian constitution requires that a candidate either earn an outright electoral majority or 40 percent of the votes, with at least a 10-percentage-point lead. Otherwise, a runoff election will take place. The preliminary count halted with 84 percent of the vote counted, when Morales had a 7.87 percentage-point lead. Though the halt was consistent with election officials’ earlier promise to count at least 80 percent of the preliminary vote on election night and continue through the official count, the OAS quickly expressed concern over the stop. When the preliminary count resumed, Morales’s margin was above the 10-percentage-point threshold.
The OAS claimed that halting the preliminary count resulted in a “highly unlikely” trend in the margin in favor of MAS-IPSP when the count resumed. The OAS reported “deep concern and surprise at the drastic and hard-to-explain change in the trend of the preliminary results.” Adopting a novel approach to fraud analysis, the OAS claimed that high deviations in data reported before and after the cutoff would indicate potential evidence of fraud.
But the statistical analysis behind this claim is problematic
The OAS report is in part based on forensic evidence that OAS analysts say points to irregularities, which includes allegations of forged signatures and alteration of tally sheets, a deficient chain of custody, and a halt in the preliminary vote count. Crucially, the OAS claimed in reference to the halt in the preliminary vote count that “an irregularity on that scale is a determining factor in the outcome” in favor of Morales, which acted as the primary quantitative evidence to their allegations of “clear manipulation of the TREP system … which affected the results of both that system and the final count.”
We do not evaluate whether these irregularities point to deliberate interference — or reflect the problems of an underfunded system with poorly trained election officials. Instead, we comment on the statistical evidence.
Since Morales had surpassed the 40-percent threshold, the key question was whether his vote tally was 10 percentage points higher than that of his closest competitor. If not, then Morales would be forced into a runoff election against his closest competitor — former president Carlos Mesa.
Our results were straightforward. There does not seem to be a statistically significant difference in the margin before and after the halt of the preliminary vote. Instead, it is highly likely that Morales surpassed the 10-percentage-point margin in the first round.
How did we get there? The OAS approach relies on dual assumptions: that the unofficial count accurately reflects the vote continuously measured, and that reported voter preferences do not vary by the time of day. If these assumptions are true, then a change in the trend to favor one party over time could potentially indicate fraud had occurred.
The OAS cites no previous research demonstrating that these assumptions hold. There are reasons to believe that voter preferences and reporting can vary over time: with people who work voting later in the day, for instance. Areas where impoverished voters are clustered may have longer lines and less ability to count and report vote totals quickly. These factors may well apply in Bolivia, where there are severe gaps in infrastructure and income between urban and rural areas.
Was there a discontinuity between the votes counted before and after the unofficial count? For sure, discontinuities might be evidence of tampering. In Russia, for instance, one allegation is that local election officials stuff ballot boxes to meet preset targets.
If the OAS finding was correct, we would expect to see Morales’s vote margin spike shortly after the preliminary vote count halted — and the resulting election margin over his closest competitor would be too large to be explained by his performance before preliminary count stopped. We might expect to see other anomalies, such as sudden shifts in votes for Morales from precincts that were previously less inclined to vote for him.
We didn’t find any evidence of any of these anomalies, as this figure shows. We find a 0.946 correlation between Morales’s margin between results before and after the cutoff in precincts counted before and after the cutoff. There is little observable difference between precincts in the results before and after the count halt, suggesting that there weren’t any significant irregularities. We and other scholars within the field reached out to the OAS for comment; the OAS did not respond.
We also ran 1,000 simulations to see if the difference between Morales’s vote and the tally for the second-place candidate could be predicted, using only the votes verified before the preliminary count halted. In our simulations, we found that Morales could expect at least a 10.49 point lead over his closest competitor, above the necessary 10-percentage-point threshold necessary to win outright. Again, this suggests that any increase in Morales’s margin after the stop can be explained entirely by the votes already counted.
There isn’t statistical support for the claims of vote fraud
There is not any statistical evidence of fraud that we can find — the trends in the preliminary count, the lack of any big jump in support for Morales after the halt, and the size of Morales’s margin all appear legitimate. All in all, the OAS’s statistical analysis and conclusions would appear deeply flawed.
Previous research published here in the Monkey Cage finds that economic and racial differences make it difficult to verify voter registration in the United States, resulting in higher use of provisional ballots among Democrats — and greater support for Democratic candidates among votes counted after Election Day. Under the OAS criteria for fraud, it’s possible that U.S. elections in which votes that are counted later tend to lean Democratic might also be classified as fraudulent. Of course, electoral fraud is a serious problem, but relying on unverified tests as proof of fraud is a serious threat to any democracy.
Update: The authors of this article were contracted by the Center for Economic and Policy Research (CEPR) to verify the numerical and statistical results of CEPR’s November 2019 study of the Bolivian election.
John Curiel is a research scientist with MIT’s Election Data and Science Lab. He earned his PhD in political science from the University of North Carolina at Chapel Hill.
Jack R. Williams is a researcher with MIT’s Election Data and Science Lab.