Skip to content

Should we be suspicious of the vote counting in Bolivia?

I recently was sent two documents regarding the recent Bolivian presidential election.


Andrés Castro pointed me to this report from the Organization of American States reporting audit results from the election.

The conclusions of this report are pretty harsh. For example:

Given all the irregularities observed, it is impossible to guarantee the integrity of the data and certify the accuracy of the results.


It should be borne in mind that the irregularities we have pointed out are those we observed in a short period of time. It is also important to point out that it was not possible to analyze the original tally sheets for the departments of Potosí, Chuquisaca, and Santa Cruz as part of the documentation had been burned. In all likelihood, given more time to process documentation, even more irregularities would surface.

On the other hand, I’ve heard that election security is pretty bad in the U.S. too, so it’s hard for me to know what to make of all this.

They also have some graphs showing vote counts changing over time. They don’t do themselves any favors with model fits such as this plot of the vote share received by incumbent president Evo Morales over time:

The discontinuity analysis is a joke, maybe suitable for publication in the Proceedings of the National Academy of Sciences but I wouldn’t expect to see it any serious report. But in any case it’s clear that the votes were different, on average, at different points in the vote-collecting process.

The report states: “Morales average vote share increased by over 15% compared to the previous 95%, while Mesa’s average vote share plunged by about the same percentage. This pattern is highly unusual.” I’m not sure what is the baseline of what is considered usual here. I guess it depends on how the votes were counted.


David Rosnick sends along this report, coauthored with Guillaume Long, Cavan Kharrazian, and Kevin Cashman, “What Happened in Bolivia’s 2019 Vote Count? The Role of the OAS Electoral Observation Mission.”

They give some background:

The TSE has two vote-counting systems. The first is a quick count known as the Transmisión de Resultados Electorales Preliminares (TREP, hereafter referred to as the quick count). This is a system that Bolivia and several other Latin American countries have implemented following OAS recommendations. . . . and is designed to deliver a swift — but incomplete and not definitive — result on the night of the elections to give the media an indication of the voting tendency and to inform the public. The TSE is unlikely to process 100 percent of the results in the quick count in nationwide votes due to logistical limitations and the amount processed can vary widely by geography and the type of ballot. . . .

The second vote-counting system is the official count (or cómputo), which is legally binding under Bolivian law. The official count is more thorough and precise and takes longer. It is the only valid vote tallying system, and the TSE uses it to determine and announce the final election results. . . .

In these elections, the results of the official count generally coincided with those of the quick count, which ended once 95.63 percent of tally sheets were counted, with Morales having a lead of 46.86 percent to Mesa’s 36.72. The final official count, with 100 percent of votes counted, resulted in Morales winning the election in the first round with 47.08 percent, to Mesa’s 36.51 percent.

And then they argue against the conclusions of the OAS report:

In line with the quick count process in previous elections, the TSE had ended the quick count at 83.85 percent of tally sheets verified. This tally showed MAS-IPSP receiving 45.71 percent of the presidential votes, and CC receiving 37.84 percent, a difference of 7.87 percentage points. Two days later, the OAS mission issued its preliminary report on the elections, which briefly repeated the criticism that “the changes in the TREP [quick count] trend were hard to explain and did not match the other measurements available.” . . .

The results from the quick count for the first 83.85 percent of the vote count are consistent with a final projected result of Morales winning the election outright with a more than 10 percentage point victory . . .

Neither the quick count nor the official count exhibit significant changes in voting trends in the final results; rather, the same well-known trend, explainable by differences in voter preferences in different geographical areas, is evident in both counts . . .

The official count was never interrupted and was regularly updated online without any significant interruption. Any potential irregularity would have had to affect the official count and not only the quick count in order to affect the final result.

And they offer this graph:

I am no Bolivia expert so I won’t try to weigh the evidence. I agree with the authors of the second report that the vote trend does not seem like enough evidence to claim irregularities. But I don’t know enough to have any opinion on the particular claims regarding problems with vote counting.


  1. Terry says:


    Very interesting story. Politically motivated parties possibly playing fast and loose with the evidence. Whodathunkit.

    Lots of curious details. Morales had to win by 10% to avoid a runoff (which is why the red line in the second graph is at 10%). Court ruling based on “human rights”.

    Good article in the NYT:

    • Ben says:

      I dunno about that NYTimes article. They’re doing a sort of “nothing to see here” argument. But there is clearly something to see there! According to the OAS there was something to see!

      But in some sense the NYTimes article and OAS article are sending similar messages to the outside. I don’t know, but I doubt the OAS article had much swing in the rank and file of the Bolivian military. Presumably the army could have carried out a coup without it. The OAS article seems to be saying to non-Bolivians, “hey, don’t pay attention to this thing that’s happening, the elections were a con anyway”.

  2. John says:

    I would not find an analysis credible unless it falsified the theory that MAS vote share and late reporting have the common cause of being a rural precinct.

  3. apoorva says:

    As an aside, the proliferation of bad RD plots recently is a good demonstration of why software defaults matter.
    Rdplot (from the rdrobust library) uses a 4th order polynomial by default . The jump at 85 in that plot seems big enough that even a linear spline would catch it, but the gratuitous wobbly graphs definitely send alarm bells ringing.

  4. Adrian says:

    It would be helpful to have similar plots for the previous Bolivian elections since 2009 to make a clearer call. I’d be surprised if there had been such a large change in the pattern of vote counting and the underlying political preferences since the last one.

  5. jim says:

    “Should we be suspicious of the vote counting in Bolivia?”

    Nope. It’s all on the up ‘n up. Elections being monitored by the ODERS (Organization of Dedicated Election Rigged States, currently the rotating chair is held by DPRK), verification assured.

  6. A third study:

    “Evo Morales and Electoral Fraud in Bolivia: A Natural Experiment Estimate” –

  7. DMac says:

    Maybe I’m confused, but I really don’t understand how one could justifiably use a regression discontinuity in the case of the plots in the original OAS report. I thought you needed some kind of discontinuity induced by a plausibly exogenous forcing variable. Reaching an arbitrary 81% or 95% of votes counted doesn’t seem like such a discontinuity to me.

    • Adrian says:

      That depends on how your interpretation of a regression discontinuity – the simplest and most basic one is the literal way to interpret it, namely, if there is a discontinuity at the given point of the running variable. In that case, the interpretation is just that – checking if there is a discontinuity, with no judgment on causality.

      The causal inference one requires more assumptions, and I would not be sure it is relevant in any way in this case.

      Now, coming back to the topic at hand: Just because there may be a discontinuity at the 80-something percent of progress in the counting of the votes, that doesn’t necessarily mean there was fraud. There is no reason to assume that the underlying quantity at hand should, or shouldn’t, be smooth at all points. That’s why equivalent plots for the last presidential elections would be useful to get a better sense of what’s going on. If the pattern isn’t all that different, I would hesitate to use that as evidence of election fraud, if they were – well, I’d be a lot more skeptical about the election results.

      TL;DR: It’s more reasonable to assume that the vote counting function should be similar for past elections, rather than assuming the underlying function should fulfill some smoothness conditions, and if there was a large change in the fitted nonparametric curve and the plot of the raw data itself, it would warrant a more thorough investigation.

  8. alex says:

    Ignored by the international media, but much covered domestically, was a lengthy report produced by a Bolivian IT engineer in the early days after the election: The actual report was published somewhere

    This was forwarded to the OAS audit mission, and introduced a number of other arguments.

    Leaving aside the fact that more than 250 reports of electoral fraud were submitted by different parties and observers–none of which have been analyzed and reported on yet–much hinges on the assertion that the final 5-10% of votes came from geographical areas that tended to vote heavily in favour of the MAS (it would be more appropriate to talk about demographics than geography, but in Bolivia at least this is implicitly understood).

    This assertion was almost immediately made by Morales when the TSE proclaimed him the winner, but has not yet been proven or disproven; I haven’t yet read a commentator or expert who has either accepted or rejected this proposition based on actual evidence.

    Yet it should be possible to analyze the return polls chronologically, and to see (a) whether the geographical distribution of the last returns corroborate the geography/demographics argument advanced by Morales and the MAS and (b) whether the vote distribution by party in those last polls is plausible.

    • Adrian says:

      …On the other hand, this is more credible. It includes a decrease in the number (level, not share) of votes for the opposition in the presidential elections and, similarly, an increase in the number of votes for Morales through various means. They also claim that at some point in time there was a decrease in the percent of voted counted.

      Allegations like these, which are based on checking the measurements directly, are more credible than those based in statistical models (even nonparametric ones).

  9. Elio Campitelli says:

    The OAS did themselves a disfavour by publishing such a a flimsy and silly analysis. Now the spotlight is on this shaky evidence instead on the many other irregularities that are described in the report, such as some voting stations reporting more than 100% turnout, or the fact that they used a previously-undisclosed external secret (and unaudited) server to handle part of the counting.

Leave a Reply