Whizzy statistical models were ten-a-penny during the presidential election in the United States in 2012, when Nate Silver and fellow “forecaster-pundits” produced uncannily accurate predictions of a solid win for Barack Obama. In Brazil, where voters will cast first-round ballots on October 5th, they remain scarce. Election-watchers prefer instead to divine trends, poll by poll. The latest show Dilma Rousseff, the left-wing incumbent, opening up a lead of six to eight percentage points over Marina Silva, her nearest challenger, in a likely second-round run-off later this month. Models come in three broad types. The first, made famous by Mr Silver, takes polls released each week, then aggregates and weights them to come up with a prediction. Another looks at how “fundamentals”—which can mean anything from unemployment to government-approval ratings—have shaped past elections. A hybrid approach mixes the two, increasing the weight of the polling data as election day draws nearer.
All three methods hit snags in Brazil, says Clifford Young of Ipsos, a research company. Because direct elections for president were introduced in Brazil only in 1989, fundamentals models have just six data points to play with, against 17 in the United States (where accurate opinion surveys started in 1948). The power of poll aggregation, too, is curbed by the dearth of data. Fewer than 40 usable national polls have been published in Brazil so far, says Mr Young; Mr Silver had at least 4,000 to hand in 2012 (including robust state-level ones). By their nature, hybrid models have both problems.
A handful of number-crunchers have nevertheless given it a whirl. João Pinho de Mello of Insper business school uses consumer-price levels and government-approval ratings (which are on the rise as Ms Rousseff has used ample media exposure to play up her successes). Mr Young considers only the latter, but factors in incumbency and augments the small historical sample by drawing on elections elsewhere in the world. Neale El-Dash of Polling Data, a website, plumped for poll-aggregation.
Their models confirm that momentum is with Ms Rousseff. Mr El-Dash’s model flipped from 62% in favour of Ms Silva late last month to 74% for Ms Rousseff as The Economist went to press. That brings it into line with Mr Young’s forecast, which has always put the chances of a Rousseff victory at 60-78%, and Mr Pinho de Mello, who estimates the president’s probability of winning at 55%.