I can't bear this sanctimonious bullshit they put at the end of all these op-eds. Escobari had one of his own at the bottom of his Project Syndicate piece, too, where he says that outsiders shouldn't make trouble about Bolivia because it's in a delicate political situation. You asshole. That whole stupid fucking paper was, in effect, an attempt to consolidate elite consensus around an unelected, coup government. It was the OAS's job to get to the bottom of what happened and inform the public about it. Everything they have done has been to obscure the facts about what happened in order to advance political agendas. If Irfan had any decency at all, then he would release his data, but his career is more important than any other consideration, so he puts out this cynical bullshit. There are people who are in jail right now in part because of these lies.

And fuck the Post for running this trash. They should've forced him to at least share his data as a precondition of the publication of the piece, or obligated him to respond directly to the criticisms, instead of evading debate by cutting and pasting what was already in the OAS report.

Fuck all these people. Que se pudran todos.

This is asinine. Why would the final 5% look like a broad average of the previous 95%? At least compare it to the previous five percent or something if you are absolutely intent on advancing these sorts of stupid claims.

It never gets remarked on much, but I don't really understand his decision to use this 'polling station-level vote share' metric. These polling stations, or 'mesas', are not equally sized. They tend to have around 230 voters each, with one last table in the precinct holding the remainder. However, precincts in small rural towns may well only have one mesa, so you don't get even one full mesa with ~230 voters. Naturally, using this kind of unweighted average overstates the effect of these small mesas.

Also, and this is extra information: He is doing his calculations in an unusual way. He is calculating 'vote share' as: (votes_received_by_party / (valid_votes + blank_votes + null_votes)). I don't even have a huge problem with using, in effect, 'votes cast' as the denominator in such a metric, but you should not call it 'vote share' and you should specify what you are doing. Another detail, maybe useful, is that he is discarding mesas in which the party in question received 0 votes. He did this because there are 4 mesas which were annulled and so appear in the data as having no votes. When he went to do his 'polling station-level vote share' numbers, it blew up in his face because the annulled mesas created a divide-by-zero error. He made this error go away not by excluding mesas which had zero votes, and thus a zero in their denominator, but rather by excluding mesas in which the given party received zero votes, which creates a zero in the numerator. He is incompetent and I had to study his tables in extremely close detail to even pry these tiny secrets from them, and that was only based on the lone column that could be checked.

This is quite a bizarre statement. It was extremely 'problematic'. The count was stopped when some subordinate of the Supreme Electoral Tribunal literally cut the power to the floor of the Civil Registry Service where the vote totals were being verified. And this was after the head of the Civil Registry Service refused to halt the count. This action of interrupting the preliminary count generated social convulsions that fed a protest movement that later toppled the government. You can't stop the damn count. In recent news articles, it came out that shortly after the stoppage, a MAS official was making death threats to the administrators of the electoral tribunals so they'd hurry up and complete the preliminary count. It was an absolute catastrophe.

This is incorrect. The margin at the time of the last pre-interruption spreadsheet, which was at 19:40:57 PM, was 7.78%. I do not have any idea where his numbers come from, as they are at odds with the public data. You may think I am splitting hairs, but statistics is all hairs. Also, by refusing to share his data and methods, he asking others to trust him. But if he is wrong in the few details that can be checked, then trusting him seems foolish.

According to the MIT researchers, the reason their analysis was delayed for so long was because they got stonewalled by the OAS and the current Bolivian government in their requests about data and methods. Irfan does this too. You can try to email him and ask for his data and methods, but he absolutely will not respond. None of them will. They know that they don't need others to replicate their results to be believed because the media accepts their claims on faith. Also, CEPR effectively pre-refuted Irfan's analysis back in, hell, I don't know, October? He simply tries to talk past people and copy-pastes his contributions to the OAS report.

This whole thing is a cut-and-paste from the OAS report, and it's really got me confused because that part, aside from being wildly incorrect, is also specific to the analysis of the Cómputo, not the TREP. It literally appears right below a table based entirely on Cómputo data and time stamps. All the previous parts in this opinion piece seem to be from the TREP section. The first graph and the other one with the discontinuity? They're from the TREP section. The bar graph with the departments? It seems similar to that explanation from the TREP section. At what point did things turn to the Cómputo? Whatever the case, this part is, like, excruciatingly bad. It makes you wonder if he has any shame or even cares about being correct. Oh well.

This really isn't so surprising, given that the political unit being described is one level below a country. Large areas contain all different kinds of people and Bolivia in particular has stark demographic splits sure to be reflected in voting trends. For example, if you look at the capital of Potosí versus the rest of the department, you'll find something like 90 percentage points of difference between CC and the MAS. That is only one example of many. I expect you could find a similarly wide split in, say, Cochabamba and elsewhere. These people do not look or think like one another, even though they are from the same department.

He says they were even 'throughout the count', which indicates that the difference between the two didn't move over the entire prior progression of the count for Beni, but this sounds unlikely to me? Like, not at all? It just seems like he is inaccurately summarizing his graph.

First, for reference here is his graph. What he is saying appears to be inconsistent with what's in the OAS report. In that report, he claimed that at the 95% TREP cut-point, there were mesas remaining from seven different departments, not six, and that the other two had either reported all or nearly all of their votes. What's more, Tarija is listed as having 115 mesas remaining in the report, but here it is excluded from the graph as one of the departments that had 'reported' early. (Again, not sure what is meant by this.) Also, I checked the TREP data and no department had all of their mesas verified in the TREP. It's not clear what's going on here.

You can't check this claim for the TREP, since that timing data is unavailable, and you honestly can't really check it for the Cómputo because Irfan's time stamps don't actually match the public data, but Irfan made a similar claim about the last 5% of the Cómputo, and it was so clearly wrong that you could refute it merely by appealing to his own table. That is, he gave readers just about the scantest possible look at his data and it was enough to utterly condemn his analysis. I'm sure his claims about the 95/5 TREP cutpoint are legit, though.

And I've gotta say here, too, that I have my own theory about possible trend shifts at the end of the TREP. The TREP, which is also called 'the quick count', is all about speed. You want to get a result out to the public as soon as possible. This means that when they come across a difficult tally sheet, they set it aside and move on to easier ones. According to the system administrator, these set aside tally sheets are resolved 'when possible', which probably means near the end when there are no more easy tally sheets to verify. This would explain a shift in trend at the end of the count that also happens to align well with previously reported voting from the same precincts.

One kind of overlooked detail about the mix of mesas left at the end of the count is that while rural areas were overrepresented, they were definitely a minority of the vote. However, the urban areas left to vote also tended to support Morales more than the early-reporting urban vote. Mind you, not the same precincts or neighborhoods - just from mass population centers, which, uh, contain multitudes.

This is stupid. The more useful comparison is not with whatever immediately preceded these mesas in the progression of the vote count, but rather on what earlier reported voting from these same precincts looked like. Williams and Curiel already showed that there is an extremely close relationship between these two, as one would expect. There are a few precincts that only reported after the interruption, but I analyzed these and they look, on average, extremely similar to 2016, which was a proxy for another Morales term. It's not even clear how Irfan is creating these graphs. He gives no explanation for them at all. Out of curiosity, I extracted the data points from his three TREP-time graphs and put them in the same graph. The lines at the end are very different, both before the break point and after the break point. As usual, he provides zero explanation of what he has done to generate this discontinuity that didn't exist in his previous graph, and of course nobody has the data to check these findings independently, which is how Irfan prefers it.

If only presidential TREP columns are represented on his graph, then it turns out, somewhat amusingly, that his 95/5 breakpoint is likely the true moment at which the TREP was interrupted. This is based on SERECI's claim that 89.29% of all acta columns had been verified by 8:00PM, as well as several other boring details and bits of arithmetic nobody cares about.

This graph is not available on Outline, but you can find it here. What is so confusing to me is that he says he is using TREP time stamps, but only 95.63% of presidential acta columns were verified in the TREP. So 84% of the TREP would be 0.84 * 0.9563 = 80.33% of all presidential actas. Is he supplementing the TREP times with ones from the Cómputo? What on Earth is happening here?

This is kind of a fussy point, but 'recorded' means 'verified' here, correct? There are two times for each of these actas: a time when the servers received the acta and a time when the vote totals were verified. I assume he is talking about the latter, but you can never be sure with this dude. His part of the OAS report is full of contradictory details and mislabeled data.

I've gotta say: I have really come to loathe this word 'irregularities'. For something to be irregular, there must of course be a baseline of regularity. A careful researcher would, say, look to previous elections for patterns, but instead nearly everyone operates from some platonic ideal in which all the trend lines are perfectly straight and reporting order is effectively random. It's the most insane thing in the world and is just about the farthest thing from serious study that you could imagine, but is readily approved by a complacent news media.

The Escobari and Hoover paper was refuted way back in December: https://cepr.net/wp-content/uploads/2019/12/escobari-hoover-bolivia-2019-12.pdf

Escobari had to rely on public data for his analysis, which meant everyone was working from a shared set of facts and there could be a debate on the merits. This explains why his paper was so quickly refuted. On the other hand, Irfan, who's much less capable than Escobari, works from non-public data which he refuses to share and which doesn't align with the public data, so is work is unverifiable and direct criticism of it is nearly impossible.