Event Summary: ‘The Signal and the Noise: The Art and Science of Prediction’

By

This is a summary of an event with Nate Silver, Statistician and Political Forecaster, on 1 May 2013, it reflects the views expressed by the speaker and not those of The Henry Jackson Society or its staff.

To view the full transcript of the event, click here

 

With more data than ever at our disposal, how can it be used effectively to make accurate predictions? Addressing an audience in Parliament on 1st May, political forecaster and statistician Nate Silver spoke about the different ways data can be used to make valuable predictions, and the various pitfalls to be avoided. Mr Silver, famous for his accurate American presidential election forecasts, also gave an insight into predicting the results of US elections.

According to Mr Silver:

Although there is more data available now than ever before, the potential for human error remains. For example:

  • There has been an exponential growth in web page numbers since the invention of the internet, and this is similar to the information revolution that followed the invention of the printing press;
  • This has led some to overpromise on what this allows analysts and researchers to predict, and fails to take human error into account.
  • In 2011, an earthquake measuring 9.0 on the Richter scale hit Japan despite theories that the country would never suffer an earthquake of that magnitude. These incorrect predictions had damaging implications as the Fukushima plant had only been designed to withstand an earthquake with a magnitude of up to 8.6 on the scale.
  • Academia also faces this problem, with mathematical analysis of findings in medical journals and other peer reviews suggesting that many predictions do not stand up to scrutiny.  

Errors caused by the volume of data can also be compounded by increases in the speed of data transmission:

  • This was demonstrated when the Twitter feed of AP was hacked, claiming that there had been a terrorist attack on the White House;
  • Despite a lack of corroboration from traditional media sources, the US stick market dropped by over $1 billion in a minute, and then recovered equally quickly.

Whilst there is some value in the increased amount of data, we should also be conscious of our inability to spot it, and the possibility of error in the computer programmes we use. For example:

  • Silver recently noticed his taxi driver taking him on an unusual route to his destination. He eventually realised that this was because the driver was relying on a GPS system that was providing poor directions;
  • In 1997 the supercomputer Deep Blue made a seemingly pointless move during a game with Gary Kasparov and went on to lose;
  •  After the game, Kasparov studied the move and concluded that Deep Blue had made a sophisticated calculation beyond his own understanding. In fact, according to one of Deep Blue’s creators, the move was merely the result of a bug in the programme.

Mr Silver also discussed his forecasting of the 2012 US Election, and highlighted several issues, including:

When predicting presidential elections, it is important to avoid partisan interpretations of opinion polls:

  • In the 2012 US presidential election, there were often between 20 and 30 opinion polls conducted in every state;
  • Rather than looking for general patters emerging from the results of these polls, Democrat and Republican-leaning news organisations tended to report only the results of polls that suggested that their preferred candidate was in the lead.

Money tends to have little impact on presidential elections, but can influence House of Representatives elections:

  • Presidential candidates receive so much financial backing that one candidate having slightly more campaign money than another is unlikely to make a significant difference;
  • If a voter sees one candidate’s advert 28 times and another candidate’s advert 33 times, he or she is unlikely to be swayed merely by seeing one candidate’s advert an additional five times;
  • By contrast, less money is involved in House of Representatives elections. Therefore, one candidate can appear considerably more often in the media than another candidate if he or she receives greater funding for election campaigns.
HJS



Lost your password?

Not a member? Please click here