Ever since the Chicago Daily Tribune’s iconic blunder of projecting Thomas Dewey as the victor in the 1948 election, polling has evolved into a more precise science, one so accurate that scoffers and critics began to argue that the media was swaying public opinion through polls.
Exit polling has always been crucial, especially during the days of newspapers with deadlines to meet, for the media to be able to project winners in a timely manner.
But something has changed, and voters and viewers have perhaps noticed that the major media outlets have become slower at projecting winners in primary races this election year than ever before in the modern age.
You’d expect faster predictions, with better computer modelling, faster communications, easier travel to the polling places, etc., but the fundamental difference is in how Americans are now voting.
According to the Associated Press, last election saw over one-third of all votes cast by mail or advanced balloting, and this year’s primary season is indicating that this number is growing by leaps and bounds.
Florida showed excellent skill in having the mail-in balloting counted before the primary polling sites were even closed on March 15, having the information ready for the media within minutes of the first closures. They are definitely leading the nation in handling advanced ballots after the debacle of the 2000 election coming down — at least in part — to uncounted mail-in ballots, but other states have not been as responsive to the changes.
With expanded mail-in voting across the United States, a paradigm is forming: election day exit polls can't be trusted.
Right now, the Associated Press is testing several different ideas on how to measure this new paradigm, in particular testing the states of Kentucky, Mississippi, Illinois, and Georgia with a new model of likely-voter polling to predict the outcomes.
While they are not using the model in this year’s exit polling or analysis, the outcomes thus far on the Kentucky-Mississippi model have been outstanding, a 3 percent absolute error (that is error discovered after the fact), beating all previous models’ ability to predict election outcomes. (The previous model had an absolute error of 6.1-percent)
But there’s a twist to this — not a single person was polled at the actual election sites, but by telephone during the 8 days before the election was held.
If this model holds true (which it will be tested and retested many times — the AP isn’t projecting this model will be used until 2018 at the very earliest) they could have almost ‘iron-clad’ information about the winners before election day.
If such reliable information is available prior to election day, the media is going to have to address the ethics of how they use it and when. But the bottom line still remains, there’s simply no way to ‘guess’ right in a tight election — sometimes we have to wait until the last ballots are counted.