Understanding Polls

Welcome to an election year. As bad as the television ads are, I feel even more inundated with political messaging on Facebook. Although I rarely watch television these days, political messaging is seemingly everywhere during an election year and it is not going to let up between now and November. One of the staples of the political messaging tool-kit is the poll.

Polls are, among other things, an attempt to record people’s opinions. They can be done a number of different ways, and some of them are more accurate than others. One of the biggest differences between a poll and a survey is in the way the sample, or the group of people who are asked the questions, are found. Surveys are often used for longer term projects, and therefore require a very careful statistical analysis of the group of people you ask your questions. Polls, on the other hand, are designed to ask groups of people the same question or two and crunch the numbers quickly so that you can measure change over even a short period of time. That makes them ideal to measure who is favored to win an election. Usually, election polls are expected to be accurate within a three percent margin of error.

So what went wrong in Michigan’s presidential primary last March? There, Clinton was favored to win by a large margin of between 11 and 37 points depending on the poll. Instead, Sanders won.  Not by much, but by enough to make people question the veracity of polls in general.

The problem with polls is who is getting asked the questions. Let’s go back to the Michigan example. Reputable polling companies (Gallup being the best known) follows federal law, which stipulates that only landline phones can be called with the poll questions. That law is rooted in the technology and cellular plans of the nineties when people only got a few minutes of cell phone calls a month and did not want to spend them on political polls! But what worked in the 1990s and early 2000s does not work in 2016 when many people gave up landlines for unlimited cellphone talk time. Restricting polling to landlines means that pollsters are going to be literally unable to reach a significant chunk of the population.

Fortunately for those conducting research, this is changing.  In fact, more than a third of households no longer have landlines. But part of what makes this so skewed is which households those are. In a 2013 study by the Centers for Disease Control (CDC), 61% of rental homes do not have landline phones and 45% of homes with children do not have landlines. By not polling people who only have cellphones, you are automatically skewing your data by not reaching a proportional number of younger people, families with children, or those who rent. All of these are demographic groups who were more likely to vote for Sanders in the primary.

The second problem with the Michigan Poll was also related to who was asked. The poll in question was targeted to registered Democrats. While that sounds fine in theory, the fact of the matter is that a lot of people are registered with a political party and yet haven’t voted in the last election for a variety of reasons. Changing party registration can happen right at the poll, so getting a list of registered voters or registered members of a party does not mean that they are likely to vote in any given election, even a presidential primary. A better solution is to instead look at recent voters, people who voted in the last election, or in two of the last four elections. This way you hopefully manage to avoid the people who are registered, but don’t bother to go to the polls.

So, there you have it. Some of the concerns about the numbers you see in polls and why they might not reflect what actually happens when people vote. Keep these in mind as you hear the latest polling data leading up to November 8th. And remember, a poll or a survey’s “generalizability”  is only as good as the sample. Or as is so often the case, you get out what you put in!

Happy Polling!

Laura Fuller is a County Extension Educator (Noble County & Buckeye Hills EERA).