Share this post
FaceBook  Twitter  

PollyVote uses the powerful method of combining forecasts. In the case of the polls, combining can be used within a method, which typically reduces forecast error by one-eighth. But it is most powerful when combining across methods. When there are many methods, such as six methods used for U.S. presidential elections, forecast errors errors have been cut by about one-half that of the typical error.

The benefits of combining have been known for over a century, and the evidence has mounted over time. However, it is resisted by governments and corporations.

So why is the combining method resisted and why do people choose less accurate methods?

One reason is that combining seems counterintuitive. Intelligent people think that if you combine forecasts you get only average accuracy, so they attempt to pick what they believe to be the best method. However, the result of combining is never worse than the typical error, which means there is little risk in using it. Thus, New York City would have saved money in January 2015 by combining two forecasts for a snowstorm—they used the least accurate forecast. Because forecasts from different methods often bracket the true value, combining is more accurate than the typical forecast. Most surprising, combined forecasts are sometimes more accurate than the best component forecast.

Another reason is that many people who commission or use forecasts are not interested in accuracy. They want forecasts that support a decision that they want to make or an outcome that they prefer. In the case of political election forecasts, voters are looking for hope and entertainment. They get this from individual polls. On this basis, the PollyVote is boring. Polly's forecast is stable and has very little error, as you can see from the PollyGraph. Polly has been correct about who would win for each of the last 100 days of the last six U.S. Presidential elections.