Teaching the 2016 Election: Parsing the Polls

by Shawn Healy, PhD, Civic Learning Scholar

In this era of Big Data, we are inundated with polling information, particularly during presidential election cycles. Each new national poll seemingly garners front page headlines and sends supporters of Team Hillary or Trump into ecstasy or a tailspin. How do we make sense of this noise for our own sanity, and most importantly, help our students comprehend this avalanche of quantitative information as Election Day nears?

First, it’s important not to place excessive emphasis on a single poll. True, some are conducted with better methodologies than others, but even the most carefully calibrated poll represents a point in time and a cross section of the electorate. The reality is that the latter is in a constant state of flux, and the only poll that matters takes place in real time on November 8.

Therefore, I prefer an aggregation of polls that minimize the effects of any single sample and demonstrate the trajectory of public opinion across time. RealClearPolitics is my go-to site for this purpose, although Pollster performs a similar function.

Graph from www.realclearpolitics.com

A view of the trend lines in head-to-head polling between Hillary Clinton and Donald Trump demonstrates the volatility of this race. Trump temporarily pulled ahead last week, but Clinton appears to have benefited from a post-convention bump and leads her Republican opponent by an average of 4.4% as of this morning. Note the range of polls that contribute to this average, one with Clinton up 9% and another with Trump ahead by 2%. Clinton leads in the four polls conducted since the Democratic National Convention concluded.

Each of the polls presented sample registered or likely voters, of course not one in the same, although the two are definitely correlated. More than anything, pollsters are attempting to build models that accurately predict turnout by partisanship, race/ethnicity, gender, educational attainment, religiosity, etc. Beyond standard errors endemic to polling, diversity in the predicted construction of the electorate also accounts for variation among polls.

Finally, it’s important to note margins of error. The polling data presented represents a point estimate of the state of the race. However, given its current dead heat dimensions, each of the polls aggregated on RealClearPolitics is statistically tied. Take for example the CBS News poll that has Clinton up 6%, 47%-41%, with a 3% margin of error. Conceivably, the race could be tied by subtracting 3 from Clinton and adding 3 to Trump. Conversely, Clinton could be on her way to a landslide victory if the error was cast in the opposite direction (50%-38%).