Data wizards put pundits’ predictions to shame
Going against almost all political pundits, statistician Nate Silver has been branded a genius and a witch. But his secret is simple: mathematics.
Barack Obama was not the only victor in this week’s US Presidential election. While the candidates battled for votes, political pundits were engaged in their own struggle: to accurately predict the final result.
Some forecast a landslide for Republican Mitt Romney, citing ‘determination among Evangelical Christians’ and young voters’ disaffection. Some listened to the confident noises coming from one campaign or the other, and gave them a narrow victory. Others, faced by tight polling numbers, simply admitted that the race was ‘too close to call’.
Statistician Nate Silver was not so coy. In the months leading up to the election he had been feeding every major opinion poll into a computer and crunching the data using his polished model. As other pundits swung this way and that, he had given Obama a narrow but consistent lead.
On the morning of election day, Silver released his final prediction. Obama would be decisively re-elected, he said, holding all but two of the states he carried in 2008 . The only state he was not confident about was Florida, where he gave the President a 50.3% chance of success.
Republicans rounded on Silver. But as state after state fell exactly as he had called it, his critics were slowly humbled. In not one case was Silver wrong: others had put Florida in the Romney camp, but the state was so tight the votes were still being counted two days later.
Is Nate Silver a witch? ‘Probably’, according to satirical website isnatesilverawitch.com. But his powers come from maths, not magic: while his rivals pore over speeches and comment pieces, he picks through the smallprint of polls.
Silver is not the first statistician to upstage so-called ‘experts’. In the mid-1990s, for instance, the managers of baseball team Oakland Athletics introduced a radical transfer policy: instead of judging players by sending experts to watch them, they would simply analyse their statistics. Traditionalists poured scorn over their player picks – yet the Athletics outperformed expectations over and over again.
These data heads might have some clever tricks, say more traditional pundits; but they lack the deep insights of a dedicated expert. Politics, sports and life in general are full of unpredictable and unquantifiable things, they say: plot twists, character flaws, human emotions. Simple number crunching can never account for all this.
But it doesn’t need to, say statisticians. Where experts are blinkered by prejudice and ‘gut feeling’, numbers tell no lies. They cannot tell us everything, true, but what they can tell is at least objective and pure. For drama and opinions, they say, stick to traditional punditry; but if it is facts you want, choose maths.
- Who would you trust more to predict the result of a football match: a top football manager, or a statistician?
- Does focusing on data distract attention from the issues that really matter?
- Conduct your own opinion poll: create a survey asking members of your school how they would vote in an election, and make a graph to present the information.
- List five factors to beware of when looking at an opinion poll, which might make the statistics unreliable.
Some People Say...
“You can prove anything with statistics.”
What do you think?
Q & A
- I don’t care about baseball or American politics.
- Alright then, how about the future of our planet? The statistical models used by people like Nate Silver are very similar to the ones used by scientists predicting the effects of global warming, or meteorologists predicting weather patterns. None of their forecasts are infallible. But they are the most objective measure we have, and we rely on them to understand and respond to the huge and potentially catastrophic challenges that await.
- That still seems very abstract. Are statistics at all relevant to my everyday life?
- Statistics are everywhere: weather forecasts, insurance rates, life expectancy. Supermarkets choose what goods to stock based on statistics, and statistics determine the effectiveness of every medicine you take.
- Polished model
- Called ‘538’, after the number of points a Presidential candidate must obtain to win an election. Rather than finding a simple average of available polls, the 538 model gives each poll a different weight based on how recent it is, how much it agrees with other polls and how accurate the pollster has been in the past. Silver first became famous by using similar methods to predict baseball results.
- 50.3% chance
- Models like 538 work by running many simulations of an event based on the data available, with variations based on how reliable the data is. In 50.3% of Silver’s simulations, Obama won Florida; he won the election in 90.9% of them.
- Smallprint of polls
- Pollsters do not simply pick a random selection of people and ask them who they will vote for. Instead, they take a ‘stratified sample’ that they believe represents the people who are likely to vote. How well pollsters make these judgements determine how accurate the poll is likely to be.
- Radical transfer policy
- Many of the players whom Oakland Athletics picked had been turned down by other teams because their style looked awkward or their build wrong. What scouts could not see is that these players were consistently doing unspectacular but important things extremely well.