eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
Hanging on the wall of the National Press Club’s Truman Lounge is a testament to the fallibility of polling. It’s the famous photo of a triumphant President Harry S. Truman waving a copy of the Chicago Tribune with the banner headline, “Dewey Defeats Truman” two days after the 1948 election.
It was perhaps the first big flop for predictive polling in American politics. It wouldn’t be the last.
This is not to suggest that scientific polling properly conducted is a waste of time, because it’s not. Polling can give an indication of what might happen in a future election, and in many cases it’s reasonably accurate. But sometimes, such as in the March 8 Michigan primary elections for Democratic and Republican presidential candidates, they can also be very wrong.
In the case of Michigan, the polls had Bernie Sanders losing to Hillary Clinton by at least 11 percent. Some polls predicted a larger margin. Yet Sanders pulled out a narrow win.
Likewise, while the polls correctly predicted that Donald Trump would win the primary election for the Republicans, they did not predict that former Ohio governor John Kasich would virtually tie Senator Ted Cruz for second in the delegate count. Nor did the polls suggest that Senator Marco Rubio would finish a distant fourth.
Part of the problem with polling, as it’s performed by the news media and by campaign organizations, is that it’s not happening in real time. This means that poll results that are released on a Monday will probably reflect contact with potential voters sometime in the previous week. At best, they reflect voter sentiment as it was perhaps days in the past.
Another part of the problem is that the polling doesn’t necessarily reflect reality, at least in terms of what a specific voter might do when it’s actually time to cast a ballot. There are several reasons for this lack of accuracy.
First, poll participants lie. Really, they don’t necessarily tell the pollster what they really believe or what they might actually do when it comes to voting or purchasing a product.
Sometimes the lack of truthfulness is because the person doesn’t actually know the answer, but doesn’t want to admit that. Or the voter in an election poll might change their mind not even having a clear idea of who the candidates are. Or, their choice isn’t available, so they make up an answer since they can’t choose Jeb Bush or Lindsay Graham, for example, when the poll is performed.
And sometimes, for whatever reason, the polling subject doesn’t want to truthfully say who their favorite really is. They might say they are undecided, for example, when they’re not. The reasons range from feeling that it’s not the pollster’s business to wanting all the attention that goes with being an undecided voter in a swing state.
There are other reasons. Polling organizations make an adjustment to the results to reflect the levels of participation they think will take place. This is what happened in polls preceding the Democratic primary in Michigan in which some pollsters discounted the under-35 vote as being unlikely to show up.
In fact, those younger voters did show up in large numbers, according to the exit poll data shared by the broadcast organizations covering the election, and the fact that, Sanders, who consistently polls well among younger voters, won him the Michigan primary much to the surprise of all the political prognosticators who expected a big win by rival Hillary Clinton.
Why the Polls Got the Michigan Primaries Wrong
In the case of the Republican primary, the reasons for the polls missing are more complex. The GOP debate that happened shortly before the election found many contemplating the low level of discourse among three of the four candidates, and the more they thought about it, the more sentiment swung to the only candidate that remained above the fray, John Kasich.
Another factor was the loss of other candidates shortly before the primary. As a result, voters who were undecided in many cases voted for Kasich, and others who had decided for a candidate changed their minds when their favorite candidate dropped out.
These changes were clear for anyone who checked voter sentiment in real time, but few news organizations did that. Had they checked the publicly available real-time polls of voter sentiment, the surge in voter sentiment favoring Sanders and Kasich would have been clear. Even the day after the March 8 primary, favorable voter sentiment for both Kasich and Sanders remained high, which is additional evidence that the voter sentiment was both real and observable.
You can see the voter sentiment levels on the LUX2016 site, which tracks real-time sentiment using social media. LUX2016, from ICG Solutions, is a real-time big data analysis engine that is capable of processing and analyzing vast quantities of real-time data.
In this case the engine was aimed at the entire output of several social media sites simultaneously. For the analysis of sentiment during the March 8 primary, the data came from Twitter and Facebook, where it was analyzed for positive and negative sentiment regarding each candidate, along with sentiment regarding important issues being considered in the campaigns.
The data from the entire output was analyzed for probable gender, party affiliation, location, and age group. That information was also stored in aggregate and is available through a simple click to query dashboard. If you click on the link above, the LUX2016 engine is reporting voter and issue sentiment in real time at the bottom of the page.
The exit polls conducted by several news organizations also showed that the contest in Michigan wasn’t turning out as predicted, but those from the news media aren’t normally available until after the polls close, at which point they’re useful for analysis, but not for prediction.
The weaknesses of polling are important to know about far beyond the political arena. When your organization conducts a survey, the very best result it can hope for is a measure of sentiment at the moment the questions are answered. That sentiment can’t be counted on to hold true for long and it may not be right in the first place, since after all, people sometimes lie.
But whether it’s a Web survey, a scientific survey conducted by a polling company, or a focus group, you can’t count on the result as being absolutely reliable. In fact, there are times when listening to answers in a focus group or in a survey doesn’t tell you as much as watching what happens while you’re getting the answers. Especially in focus groups, participants say and do a lot of things that aren’t on the answer questionnaires, but can provide valuable information.
But it’s also important to know that studies of sentiment, whether it’s from voters or customers, can provide valuable insight. Sometimes that insight can be critical to the success of your company. But it’s also important to know that no matter how good a poll is, it can’t predict the future.