28/02/20

Regressando à questão sobre se houve fraude na Bolívia

Segundo um estudo estatístico feito por dois peritos do MIT, não parece haver nada de estranho nos resultados divulgados após o apagão:

Bolivia dismissed its October elections as fraudulent. Our research found no reason to suspect fraud, por John Curiel and Jack R. Williams , no Monkey Cage (o blogue de ciência política do jornal Washigton Post):
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. (...)

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. (...)

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.
O texto original pode ser difícil de ler poque está por detrás de um firewall, mas se não conseguirem em modo normal, possivelmente conseguem se abrirem como "janela privada".