HOW much credence should Brazilians be giving to opinion polls? Before the first round of the elections on October 3rd they predicted Dilma Rousseff of the left-wing Workers’ Party would win with an absolute majority. In the event she fell short by three percentage points, more than the margin of error. Pollsters mis-called quite a few other races too, most egregiously in São Paulo, where Aloysio Nunes topped the poll for the state’s two Senate seats with the largest personal vote of any Brazilian senatorial candidate ever. He was predicted to come third.
The accuracy of forecasts matters not only for those who want to know what is likely to happen in the presidential run-off on October 31st. It may also affect the results themselves. The polls make the political news, points out Marco Antonio Villa, a historian at the Federal University of São Carlos. That means they influence both the candidates’ strategies and the choices of the electorate.
The last few polls have put Ms Rousseff on between 53% and 57%, a big difference in a two-horse race (she is facing José Serra of the centrist Party of Brazilian Social Democracy). Those figures can’t both be right. Are either of them?
On the side of scepticism is the sheer difficulty of predicting the behaviour of an electorate of roughly 100m spread over half a continent, including some of the world’s richest and poorest people, by asking just a few thousand of them. Brazilian pollsters may be making it even harder than necessary by neglecting to ask respondents whether they actually intend to vote. This is standard practice in most places, and the responses of those who say they are not sure are usually weighed more lightly or discarded altogether. In Brazil, pollsters don’t ask because voting is compulsory—but in practice, around a fifth of the electorate stays home on the day.
Against that is an interesting argument that at least in the first round of the presidential election, the opinion polls represent public opinion more accurately than the results themselves. Many Brazilians have had very little education and a tenth are illiterate, points out Alberto Almeida, a political scientist in São Paulo. He argues that by the time some of them had chosen one candidate for state governor and two each for senator and federal deputy—all identified by number, and with in some cases hordes of candidates to choose between—quite a few had unknowingly spoilt their vote or lost track of what they are doing entirely. And since poorer, less educated folk are generally more likely to plump for the person they think is already winning, the consequence is lower-than-predicted results for front-runners. In 2002 opinion polls gave Lula three percentage points more votes than the first-round result; in 2006, the difference was four points.
In favour of Mr Almeida’s theory is that he used it to predict publicly that Ms Rousseff would fail to win on the first round. If he is right, there is no particular reason to doubt the second-round polls, since the choices are much simpler (between two presidential candidates, and two candidates for governor in the states where no one won outright). That would give Ms Rousseff the victory, most likely with somewhere between 53% and 57%.