Following on from the first prediction in our series yesterday, here’s a fairly common animal in the zoo of failed predictions: the surprise election. Thanks to this iconic photo, there’s probably no example more famous than when Democratic nominee Harry Truman beat Republican nominee Thomas Dewey by a comfortable margin in the 1948 US presidential election.

Virtually every expert prediction got this one wrong, including those from individuals and groups sympathetic to the then-incumbent. The unfortunate Chicago Tribune had relied on the forecast of its political correspondent (who had gotten four out of five election predictions right in his career) and the polls, nearly all of which indicated a Dewey victory. This was relatively early days in terms of the science of public polling, and the election error prompted some major advances in methodology.

Misled by evidence from elections prior, pollsters had falsely determined that US presidential election voters always make up their minds by September. As a result, for the 1948 election they simply stopped polling by the end of August, thus missing a late swing for Truman. Pollsters also began to switch away from quota-based sampling to random sampling, and would place a newfound importance on predicting not just candidate preference but also likelihood of turnout.

The science of forecasting has come on a long way since then of course: witness Nate Silver’s impressive 50/50 prediction for the most recent US presidential elections (from a model based on principles which aren’t a far cry from those of almanis). But we still get it wrong: nearly every expert and poll – including Mr. Silver – predicted another Hung Parliament in the UK’s general election this year. The chance of either party forming a majority was seen as slim at best. What went wrong?

Why don’t you test your forecasting skills by signing up to almanis today at


Comments are closed.