Irfan Nooruddin reports for The Washington Post:
The last 5 percent of the vote count, which favored Morales substantially, is very different from the trendline for the other 95 percent of the count.
On Oct. 20, 2019, Bolivia held a presidential election. To win on the first round, the winner had to secure an outright electoral majority or at least 40 percent of the vote and a 10 percent margin over the second-place candidate. While incumbent Evo Morales had a comfortable lead over his principal opponent, Carlos Mesa, the gap was below this threshold until a sharp late break in the vote trends went in favor of Morales.
As part of its audit of the results, the Organization of American States (OAS) commissioned me to conduct an independent analysis of the vote returns. I concluded the vote shift that pushed Morales over the 10 percent margin needed to avoid a second-round election was unusual and suspicious. This finding was consistent with a separate analysis of the Bolivian election that concluded even more forcefully that the election was fraudulent.
On Feb. 27, four months after the election and just shy of three months after the OAS published its final report, scholars affiliated with the MIT Election Data Science Lab published an article questioning the OAS report and claiming that their analysis on the 10 percent question found no evidence of fraud.
In contrast to what these authors say, the statistical evidence suggests irregularities — or worse.
The late break in the data strains credulity
On election day last October, the Bolivian Supreme Election Tribunal (TSE) published a preliminary vote count. With some 84 percent of polling stations counted, Morales led by 7.29 percent, well below the critical 10 percent margin. At this point, the preliminary vote count was suspended, which is not in itself problematic. The MIT analysis makes two main arguments: first, that an extrapolation of the vote trends is consistent with Morales eventually cresting the 10 percent margin (see Figures 1 and 2 of their report); and second, that there is no discontinuity in the data at the point where the preliminary count was halted.
Whether the extrapolation of the data series is plausible depends on whether the steep slope required in Morales’s vote share is plausible. Moreover, the worrisome discontinuity comes later. Below I plot the average polling station level vote share for Morales’s MAS party and Mesa’s Civic Community party, the two major parties, over the course of the preliminary vote count, using the time stamps at which the TSE recorded the actas — the tally sheets for each polling station.
The divergence in the two parties’ fortunes begins early in the count, with Morales developing a comfortable lead. The first vertical line indicates the point at which the preliminary vote count was halted. At this point Morales had a 7-point lead, and the slopes of both trendlines changes. The second vertical line marks the 95 percent cumulative vote count, at which point Morales was still a point shy of the 10 percent margin required to avoid a second round of voting.
This final stretch, where Morales gains enough votes to crest the threshold, deserves greater scrutiny. So, below, I plot every one of the 35,000 polling stations in Bolivia. I focus here on the 95 percent mark.
One explanation is that late-reporting polling stations were more likely to be from Morales strongholds. This is possible, but the real question is whether the size of the vote margins is plausible, and provides evidence to support this discontinuity in the trend line for the MAS at the 95 percent threshold. Even if the late-reporting polling stations were more likely rural areas that favored Morales, a sharp discontinuity around an arbitrary point such as the 95 percent threshold demands explanation. This last portion of the vote count, which favored Morales substantially, is not just different to earlier in the evening but also sharply different than the trend just on the other side of the 95 percent threshold.
This becomes clearer if we separate the data by Bolivia’s nine departments (departamentos) — the administrative divisions for the country’s 112 provinces. All polling stations from three departments had reported their results early in the count. In the other six departments, shown below, it is possible to consider the average polling-station-level vote share in favor of Morales before and after the 95 percent cumulative-votes-counted threshold.
In every department where there are substantial numbers of polling stations reporting late, the MAS does much better in the final 5 percent of the vote count than in the previous 95 percent. In Beni, where the two candidates are roughly even throughout the count for the first 95 percent of the cumulative vote, the MAS leaps to an average 15 percent edge in the last 5 percent. In Chuquisaca, where Mesa enjoyed an average 12 percent edge at the polling station level for the first 95 percent of the vote, this flips to a 44 percent MAS advantage on average in the last 5 percent. That’s a 50 percent average vote share reversal for the two parties! And so it goes, department after department.
It is the final 5 percent of the vote count that is critical, for here Morales’s advantage nationwide rises from just under 9 percent to 10.57 percent — this would require his advantage over Mesa to increase by around 115,000 votes. How does this happen? Well, in that final 5 percent of the vote count, Morales’s vote count grows by 167,000 votes, but Mesa’s vote count grows only by 50,000. It is this great divergence, unpredicted and unanticipated by any previous part of the election trends, that pushed Morales over the 10 percent margin to outright victory.
This suggests irregularities — at the very least
None of us can know what exactly happened in Bolivia on Oct. 20, 2019. But the comprehensive OAS audit report is consistent with irregularities — or worse — that undermine the legitimacy of the process of that day’s election. Bolivians will have the chance to return to the polls in May 2020. Perhaps then they will have a free and fair election.
Irfan Nooruddin (@irfannooruddin) is a professor in the School of Foreign Service at Georgetown University, and author of Elections in Hard Times: Building Stronger Democracies in the 21st Century (Cambridge University Press, 2016).