Casa, Belo Horizonte Nunca compre uma rifa de um Aiatolá

O governo iraniano anunciou que o Ahmadinejad ganhou a eleição ontem com quase o dobro de votos de seu oponente mais próximo, o reformista Mousavi. A oposição, e inúmeros blogueiros e analistas acusaram fraude, e enquanto escrevo o pau come solto nas ruas de Teerã. São três as principais evidências citadas de pilantragem:

1 - A comissão eleitoral supostamente teria dito à campanha do Mousavi que ele ganhara a eleição, mas que ele não deveria anunciar a vitória até o comunicado oficial após todos os trâmites oficiais

2 - Horas depois, e atropelando o processo normal de validação da eleição, o supremo lider, Aialtolá Khamenei, anunciou Ahmadinejad como vencedor, com base em 80% dos votos contados, por uma margem muito maior que a esperada

3 - Em cada um dos comunicados com os totais parciais dos votos, a razão entre o número de votos dos dois candidatos permaneceu quase exatamente constante, ao contrário do que aconteceu em eleições passadas. A preferencia por um e outro (e, de forma mais geral, pela reforma ou pelo status-quo) é aparentemente bastante divergente entre as diferentes províncias, e entre iranianos urbanos e rurais. A comparação direta entre os resultados regionais parece mostrar a mesma razão constante de votos.



É a respeito de (3) (veja o gráfico acima) que o pau come solto na internet atualmente. Entre supostas provas e refutações (da prova, não da fraude), eu escrevi o seguinte (em inglês mesmo porquê não vou traduzir isto tudo as 2 das manhã):

Neither the proof nor the refutation of fraud are proper statistical analyses. The near-constant ratio between the candidate's vote totals would make sense only if the votes in each partial result were approximately an representative sample of the total vote. Which would mean that either:


a) Iranian precincts or districts count and report votes at approximately the same rate.

or

b) For each partial result the tallied vote ratios just happen to coincide with those of the complete result, because reporting places which are more pro-opposition than average just happen to balance out pro-government places

or

c) There is a reporting bias, i.e., some kinds of precints/voting sites/districts count and report votes at different rates*, but voting patterns are approximately the same across the country and between cities and countryside.

Now, a) is conceivable in theory, but from what I read reporting was not uniform. b) requires a series of unlikely coincidences, and c) would mean voting patterns have changed significantly (and congruently) between past elections and the last one.

None of this proves fraud, but certainly seem to suggest it. A proper statistical analysis would take as the null hypothesis the non-existence of fraud, and try to quantify how unlikely that would be. It would be virtually impossible to disguise massive fraud from a full statistical analysis of voting patterns, if detailed results were available. For some reason, I suspect they won't be. But even with coarser data, one could

- Prove or disprove a) by simple inspection of the places included in each partial tally

- Prove or disprove b) by inspecting subtotal in each tally

- Test c) by comparing reported voting patterns (as a function of region, urbanization and whatnot) in this election with those of past elections.

The catch is that, while we can in principle exclude a) and b), we can at most say about c) that 'Either there was fraud, or voting patterns changes in such-and-such way'. One is then left with the somewhat subjective task of estimating the likelihood of such a change. This is what Iranian experts are for, and this is where the debate should occur. Of course, were we to be told that, for instance, the ratio between Obama and McCain votes wa the same in both Illinois and Arizona, we would be pretty sure there was something wrong with the results. From the fragmentary reports I've read, which seem to be based as of yet on very incomplete data, the reported voting patterns in Iran seem equally implausible. We should not rush to judgment, but my preliminary impression is that this election does indeed look fishy.


__________________
* E.g. smaller rural precincts count quicker than big urban ones, or eastern regions count earlier than western because most people vote before sunset