On the wall of the Truman Lounge in the National Press Club here in Washington hangs the picture of an election headline that’s familiar to every student of presidential politics. The picture is of former President Harry S. Truman holding the front page of the Chicago Tribune with the headline, “Dewey defeats Truman” in huge bold letters.
The photo is there to remind press club members of the danger of using information they don’t know to be true. In this case, the Tribune went with the Dewey election win based in part on faulty polls and in part on assumptions made as a result of early returns. But regardless of the reason the headline was wrong, the newly re-elected Truman gleefully waved the erroneous newspaper headline in triumph.
Now, 68 years later, the public and the press once again are relying on voter sentiment polls as a way to divine who is going to win the presidential election of 2016. Unfortunately, the polls don’t seem to be much help.
A look at the roundup of polls on the non-partisan news site RealClear Politics shows the results are all over the place, with a difference of as much as 14 points regarding the winner of next week’s election and each major candidate being ahead by as much as 7 points, depending on the poll.
The differences stem in part from the people chosen to be part of the polling sample, the manner in which the polls are conducted and, to some extent, differences in sample size.
But these differences are important. For example, many online polls are self-selecting, which means that anyone who goes to a polling website can enter their preference. The results can be skewed fairly easily when one group rounds up enough people to sign on to the site and answer questions in support of one candidate or another.
Other polls are selective because they depend on characteristics that can skew the results. For example, telephone polls that use only landlines can miss a relatively large percentage of younger voters who use only wireless phones. Such polls also can miss people who either don’t have phones at all or who don’t respond to polling requests.
Finally, there is the fact that people sometimes lie when they talk to pollsters. They may be embarrassed to admit that they’re in favor of a particular candidate or they are just too cantankerous to state honestly how they will actually vote.
So the obvious question is whether there’s really a more reliable way to find out how the election is going. The answer may be tracking the activity of people on social media. The reason that may work is based on the assumption that voters don’t lie to their friends.
At eWEEK we found that social media analysis produced excellent results during the party primaries starting in the summer of 2015. There we worked with the folks at ICG Solutions and their LUX2016 analysis engine to determine who won the primary debates. We found that the analysis results were as accurate as the best polls and it produced results much faster.
Now the question is whether social media can tell us who is winning in the 2016 presidential race. I’ve spoken with two different sets of social media analysis companies that use different methodology in determining voter preferences. The results, taken over time, are surprising.