Recently the College of Coastal Georgia introduced a new Bachelor of Science in Data Science degree. Students in the program will learn through statistical analysis, computer programming, and mathematical modeling how to add value to data through analysis. In other words, they will learn how to produce consumable information. This is a highly valuable skill and will make these students very competitive in both the public and private sectors.
Most of us, however, find ourselves in the data consumption camp. Data and interpreted data come at us in large doses and very quickly. How we understand this information can, in turn, shape how we view the world. So, understanding how to be good data consumers is of great consequence.
In this week’s column, I want to explore one particular area of data consumption: Presidential polling numbers. Where Presidential candidates land in the polls has long been fodder for the media and political pundits. As voters, we often rely on this information to tell us how campaigns are going. The 2020 Presidential election cycle is no exception, particularly in light of virtual conventions, diminished rallies, and reliance on media coverage almost exclusively.
We know from the 2016 election, however, that polls are not necessarily great predictors of Presidential outcomes. In August of 2016, Hillary Clinton had an 8-point lead over Donald Trump (50% to 42%) among registered voters in some of the largest Presidential polls. Now, in August of 2020, major polls indicate a 9-point lead in favor of Joe Biden (51% to 42%). Prior to the 2016 election, the Biden campaign probably would have found this relatively strong lead comforting. But, we know that despite these almost identical numbers, Hillary Clinton did not win the election. Why? In short, because polling registered voters on a nationally-elected position like President has some caveats that must be considered.
Most importantly, the President is not elected by popular vote, but rather by an Electoral College. Indeed, Hillary Clinton did win the popular vote in 2016 with 2.9 million more votes than Trump, but that has never been the number that matters in a Presidential race. Only the state-by-state electoral counts will get you a win, and Clinton simply did not have those numbers in the “rust belt” states.
So, while polls can show us how voters on the whole are feeling about the candidates, they do not predict the outcome of the election in and of themselves. This was a major polling downfall in 2016; there was significant national polling, but anemic battle-ground state polling. If you want a more predictive picture of the outcome in 2020, you would do better to follow local polling in states where Trump won by 5 points or less: Florida, Georgia, Michigan, North Carolina, Pennsylvania, and Wisconsin.
Next, unless you have the time to dig down into who actually responded to a poll, it can be hard to know whether the poll is really representative of the population. In polling in general, for instance, there often is an overrepresentation of certain populations that are more likely to respond such as college-educated respondents, older respondents (who have a land-line phone), English-speaking voters, and white voters to name a few. The results of these polls, therefore, may not be as representative as we would like.
Finally, there are a few bias issues that are not readily apparent in a polling news headline. “Social desirability bias” is the idea that people may simply lie to a pollster because they think doing so will reflect well on them. In other words, a respondent may not want to tell the stranger on the phone who they are voting for due to a perceived fear of judgment. There can also be leading questions or biases within the way a poll is delivered. For example, questions can be framed in such a way that leads a respondent toward a certain answer. Another example of a bias issue in polling is the fact that many more people will say they intend to vote than actually vote. There are a number of reasons why this happens, but the result is that a poll suggests that likely voters are leaning in a certain direction, but the outcome turns out quite different.
As we enter party convention month, these specifics of polling will be something to keep in mind. There will likely be a significant uptick in national polling, but unless you are interested in digging into the details of those polls, I recommend looking elsewhere for your data. My hope is that pollsters were stunned enough by the misinterpretation of the 2016 data that they will focus more on state-level, representative (as much as that’s possible) polling.
Dr. Heather Farley is a professor of Public Management in the School of Business and Public Management at College of Coastal Georgia and Chair of the Department of Criminal Justice, Public Policy & Management. She is an associate of the College’s Reg Murphy Center for Economic and Policy Studies.
Reg Murphy Center