Share this post
FaceBook  Twitter  
Recently, Nate Silver challenged Joe Scarborough to a $2,000 bet on the presidential election outcome after the latter called the New York Times blogger a partisan for predicting the Obama will win reelection. With Polly the election forecasting parrot on his shoulder, "Cap'n" Scott Armstrong has offered to join "Long" Nate Silver in his bet against Joe "Talking" Scarborough. Armstrong says that the MSNBC talking head apparently does not know about the SeerSucker Theory. In fact, all five methods used in the PollyVote predict that Obama will win a popular vote victory, so Nate Silver is on safe ground.

It is true that a number of pundits, such as Michael Barone, Karl Rove, and George Will, among others, see it differently. But unaided forecasts by individual experts (unaided, that is, by evidence-based procedures) are clearly inferior to those by quantitative methods. This is the conclusion from analyzing enormous amounts of data across seven decades. (See Armstrong's Moneyball review and Tetlock's large-scale study.) This is not to deny that on a single event like this, an expert might be lucky and beat an evidence-based method. Properly structured, however, experts' forecasts can be usefully incorporated into a larger family of forecasts. That's what Polly does. For this election, a panel of experts was surveyed on a monthly basis and the trimmed mean averaged into the PollyVote along with polls, econometric models, and the Iowa Electronic Market vote share market quotes. Actually, the PollyPanel of experts acquitted itself quite well in the 2004 and 2008 elections. Polly's latest expert survey will be out later today.

The PollyVote is designed for forecasts that involve much uncertainty. This means that it is most useful months or years prior to the election, not for an Election Eve forecast. Polly has always predicted the correct winner—in 2004, starting eight months in advance, and in 2008 fourteen months ahead. This time around Polly has continuously predicted Obama to win the two-party vote since its first forecast was released on January 1, 2011, almost two years before the election.

Election Eve Forecasts

One or more of Polly's components will be more accurate than Polly—most likely the 7-day average of the IEM and the poll-of polls. Note that Polly's error was only 0.1% lower than the average error of it components on Election Eve in 2004 and 2008. Polly's 7-day average of the IEM was the most accurate in 2004 and 2008.

Forecasts

AE

MAE

Election Eve

2004

2008

2004

2008

PollyVote

51.7

47.1

0.5

0.8

0.7

IEM

51.2

46.4

0

0.1

0.1

Polls

51.4

47.7

0.2

1.4

0.8

Experts

50.1

47.3

1.1

1

1.1

Models

53.8

46.9

2.6

0.6

1.6

AE: average error; MAE: mean absolute error