The Ins and Outs of Polling Data

How many times a day do you open Instagram Stories to find someone you follow has posted a poll? More than once I’d guess. We are constantly being inundated with polls, political and otherwise. Especially leading up to the election, it seems like all the headlines are touting numbers. However, unlike social media polls, election polls are nuanced, and quite frankly confusing most of the time.

After polls failed to predict Donald Trump as president in 2016, many of us were left wondering if we should trust the data. So really, how reliable are polls? Better yet, how much weight should we put on the headlines we’re seeing about the outcome of next week’s election? This is what we’re going to dive into today.

How do polls work? Who conducts the polls? How is it decided who gets asked? Is this a good representation of sample voters? 

Many of us think of polls in a cut and dry manner, maybe the way we learned about them in a high school or college statistics course. But when it comes to elections, factors like ‘the bigger the sample size the better’ don’t always ring true. According to the Pew Research Center, “the real environment in which polls are conducted bears little resemblance to the idealized settings presented in textbooks.” 

The first part to understand is how not all polls are conducted the same way. For example, some polls are conducted by telephone using live interviewers, others require participants to opt-in online, and some conduct polls online using a panel of respondents recruited offline. Furthermore, we are seeing more and more fast and cheap polls done via robocalls. 

The most common method for collecting survey responses is through random digit dialing (RDD). A database of telephone numbers - both landline and cell phones - is analyzed by computers to figure out all active blocks of numbers, which are area codes and exchanges (the second three digits) actively in use. The computers then randomly dial every possible number combination in each active block. To achieve a completely random sample polling orgazations also need to select random respondents within the households they reach. 

Once the person is on the phone, the questions begin. “Which candidate will you vote for in this election? “Do you approve of President Trump?” “Did you vote in the last election?” Studies show that the order of the questions affects the accuracy of the poll. Thus, most political polls first start with asking the respondents who they would vote for as not to unknowingly influence the person’s answer. 

The goal of political polling is to show the opinion of the entire country without speaking to every registered voter. This is where representation comes in. Polling needs to accurately represent the larger population, by only polling a small group. Mathematical models which take a random sample of data are adjusted to match the characteristics of the population it should represent. This is where weighting comes in. The model must adjust - or weight - the sample to match the most recent census data about the sex, age, race, education, and geographical breakdown of the U.S. population. But this is where things can get tricky. Racial and ethnic minorities are generally underrepresented in polls and those who do participate from communities of color tend to have a higher education. The thing is, highly educated people are already more likely to vote so that skews the data a bit.

What happened in 2016?

The biggest problem in 2016 was many polls didn’t correct the overrepresentation of college-educated voters, meaning education as a variable of representation was weighted incorrectly. At the national level everything looked accurate, but at the state level some polls incorrectly showed Hillary Clinton with a lead. Historically, the less well-educated voted Democratic, but in 2016 that group of voters supported President Trump. They were not weighted correctly.

Don’t forget about the Electoral College. 

Polls represent populations, but the Electoral College is what decides presidential elections. So while President Trump lost the popular vote in 2016, he won more seats from the Electoral College, which actually determined the outcome. Stanford political scientist, David Brady, suggests paying attention to the polls from battleground states, “I think when it comes to the Electoral College, the best thing to look at is individual state polls and the states that went for Trump in 2016 – Michigan, Wisconsin and Pennsylvania – which essentially gave him the Electoral College win.”

So, should we trust the polls? 

Yes and no. Political polls are generally a good representation of the opinions of the American people but not necessarily an accurate prediction of the winning candidate. This might make you ask why polling numbers get so many headlines if they don’t actually tell us who will win the presidency, and the reason is if polls only polled the Electoral College, then the views of the majority of Americans (about 80%) who don’t live in battleground states would not be considered. 

In heated political times like we’re living through at this very moment, everyone wants to know what will happen on Tuesday and in the days that follow. The anxiety of waiting is excruciating at times but to a certain extent, we must wait and control what we can - and that is voting! There is some new research that shows when the polls show one candidate is likely to win, eligible voters may be less likely to vote. We cannot let this happen, we need you to make a plan to vote and encourage those around you to do the same. 

In case you’re looking for more resources on polls, a well-respected resource is Real Clear Politics, a nonpartisan website that aggregates data from national and state polls, which averages between six and nine different well-trusted polls. There are also a couple helpful articles here:


Don't like what the polls say? Get out and vote! #IGNITEtheVote