In the previous section, we used a regression analysis to try to find any suspect patterns in how the supplementary lists impacted the vote share of Victor Ponta and Klaus Iohannis. We also found a few precincts where our model’s prediction falls far off the mark — where Ponta or Iohannis got significantly more votes than our regression would expect. Do these outliers represent places where something suspicious was happening? Maybe, maybe not.
- In part 1 we discovered that, in the first round of the elections, Ponta seemed to do no better or worse in precincts with a greater proportion of supplementary list voters. In the second round, he did worse in such precincts. Iohannis, in contrast, tended to perform better in the polling stations with a greater share of temporary voters — in both rounds.
- The regression we did in part 2 gave us a better picture of what happened in the second round. On the whole, Iohannis gained far more support from new voters showing up for the November 16 runoff than Ponta — irrespectively of whether they popped up as temporary list voters or not. For every 100 extra voters, Iohannis got about 70 extra votes, our model estimated.
- Our regression showed almost no difference between the permanent list vote and the temporary list vote in terms of how the two split between the candidates. No evidence of a widespread and potentially suspect pattern, then.
The yellow names in the image below are some of the most obvious outliers from the regression about the second-round Iohannis vote:
And here are the major outliers from the Ponta model:
The estimates for this regression model are the best possible under (1) the obvious constrain that no potential source of the second round votes can contribute more than 100 and less than 0 percent of its votes, plus (2) the simplifying assumption the same model holds true across all of the 18,553 precincts in the country. But it is natural that there should be some variation in, for instance, how many of the first-round Tariceanu or Kelemen votes are transferred to Ponta and how many to Iohannis in the second round. The precincts we highlighted on the graphs are places where the actual election results deviated substantially from our model-based prediction.
To explore why this might have happened, we inspected closely the election results in dozens of the most deviant precincts and found just two where temporary votes could conceivably provide the explanation. Now we will discuss these two cases.
Colonesti – OLT 154
The first one is a rather notorious polling station — precinct number 154 in Olt County, located in the village of Colonesti. This place has acquired a dubious reputation as a result of many observers and journalistic reports noting irregularities. During the second round of the election, there were reports of observer intimidation and „electoral tourism” in Colonesti. Plus, after the voting ended at 9 pm, several individuals took the urn containing the ballots to the town hall — although it is illegal to do so, as the votes must be counted in the precinct, in the presence of all committee members. After the police apprehended them, the urn was taken to the office of the County Electoral Bureau, who counted the votes. The polling results in Colonesti are considered valid by the authorities and have been included in the final national tally, though there is a criminal investigation under way.
Setting aside this information for a moment, though, what does our model pick up about Colonesti? Basically, in this precinct, Iohannis obtained far fewer votes in the second round than our model would have predicted. Between the first and second election rounds, the number of voters on the supplementary list increased from 103 to 283 persons. Knowing this, our model would have estimated that Iohannis would get 343 votes in the second round. Instead, he got 135. What does this mean?
From the best estimate our regression could come up with, the expectation would have been that 76% of the new supplementary list voters would go to Iohannis (i.e. 139). Given the data from the precinct, this is only mathematically possible if an unusually large proportion of the new permanent list voters (of whom there were 100) as well as many voters of the eliminated candidates, opted for Ponta. Is this conceivable? Yes, if Iohannis is particularly unpopular in this place. Given that in this precinct Ponta got almost four times as many votes as Iohannis did in the first round, it is quite possible, and in fact very likely, that he could secure many more second round votes among all sorts of voters than our predictive model assumes, following the nation-wide trend. It is also mathematically possible that Ponta was exceptionally popular with the new temporary list voters, which would not necessarily signal fraudulent behavior. Fraud remains one of several possible explanations though.
Caminul de Batrani Lacu Sarat – BRAILA 34
Another outlier case is the polling station located at the Lacu Sarat Retirement Home in Braila. Here, Iohannis actually did better than our model would have predicted, although the number of temporary list voters dropped.
In the second round, the number of people who voted on the supplementary list dropped by 179. Our nation-wide estimate would have expected this to translate into a loss of votes for Iohannis (because according to the model, 76% of the change in temporary votes would affect Iohannis). While the model would have predicted a total of 49 votes for Iohannis, he in reality got 142. Why?
It is again possible — and quite likely — that the deviation from our prediction occurs because much fewer than 76% of the temporary voters in the first round voted for Iohannis. If that is so, then the drop in their numbers did not negatively impact Iohannis as much as the model would assume. In fact, since Ponta got almost 60% of all first round votes in this precinct, it is quite plausible that he could have 60% or more of the first round temporary votes. Then it would have been mostly he, and not Iohannis, that would suffer as a result of the drop in the turnout of the temporary list voters. Were the first round temporary list voters fraudulent electoral tourists, then? There is nothing to suggest that in this story, but the statistics alone cannot exclude that possibility.
There are thus several plausible explanations for both of these large deviations. With the available information, it would be rash to claim that, at least in these unusual cases that deviate from the prediction of a model that fits so well the data from the whole country, we identified sure signs of fraud. But these places may warrant further investigation — in particular Colonesti, where there have already been reports of irregularities.