Trump rising, Trump falling. Clinton up, Clinton down. The mass of conflicting polls can be maddening and provokes the question: Why can’t the pollsters agree?
The simple answer is that polling is not an exact science. Even though the media outlets who sponsor many of the polls like to present them as the definitive state of the race, the fact is that there is a lot of guesswork involved and plenty of room for error.
The fundamental problem for the pollsters is figuring out just who is going to come out to vote. While voting is a habit and patterns are durable over time, small changes can mean a lot in a close race and each presidential election has its own dynamic that will interest some voters and turn off others. Keep in mind that, from 2004 to 2012 no major demographic category (ethnicity, age, education, income) saw a variation of more than 6-points. In the end every pollster knows that objective statistical analysis will only get them so far. Their turnout models include a throw of the dice.
In tight elections, small differences mean a lot. In the 2012 election white non-Hispanic vote turnout dropped from 66% to 64% while African-American turnout increased from 65% to 66% — those small changes were enough to tilt the race to President Obama. This election is similarly close and that means small variations in turnout models can easily flip a Clinton lead into a Trump lead. For example, the most Trump-favorable poll, USC/LA Times has Trump hovering around 46% with a 3-4 point lead. Monmouth and Quinnipiac show Trump behind, but gaining the same vote share – they seem to agree with USC’s turnout model for Trump, but think more of Clinton’s voters will show up.
Figuring out 2016 turnout presents more difficulties than normal. Two elections with Barack Obama at the head of the ticket skewed traditional voting patterns with millennials turning out in higher numbers in 2008 and African-American voters turning out in higher than normal numbers in both 2008 and 2012, while turnout among white non-Hispanics dipping in 2012. As a result, you have a couple of elections that could either represent the start of a permanent shift or might be anomalies. We just don’t know.
Two of Clinton’s top demographic targets, African-Americans and Hispanics, show both the peril and promise for her. Hispanic turnout is among the lowest of all demographic categories, 47% in 2004, 50% in 2008 and 48% in 2012. One would think that Trump’s rhetoric would markedly increase Hispanic turnout and help Clinton and having half of eligible voters is a lot of opportunity. While this seems likely, it is impossible to determine how much that increase would be. Voting habits are very tough to change. When you consider that the first African-American candidate for President only increase turnout in that demographic by 5-points, simply getting a majority of Hispanics would be an herculean feat.
In 2012 Obama won African-Americans 93% to 7% and achieved a record turnout of 66%. Such a turnout seems highly unlikely for Clinton (even though African-Americans delivered her the Democratic nomination). A return to the 2004 turnout level of 60% would cost her over 1 million votes and possibly the election. Perhaps the efforts of Obama will help her split the difference and get 63% turnout. What seems likely is that African-American turnout will be somewhere above 60%, but below 66% — and that is a pretty big spread.
Trump’s turnout opportunities are more limited, but he has them. Trump has done well with older voters, voters with high school or less education and non-Hispanic whites. Turnout increases in a straight line as both income and education rise. Older voters have averaged over 70% turnout, which leaves Trump with some room for improvement, but not much. Less educated voters present more of an opportunity. Voters with just a high school diploma have averaged 55% turnout – and only 53% in 2012. So, Trump has some room to grow there. Non-Hispanic whites have generally high turnout at 67% in 2004 and 66% in 2008, but that number dipped to 64% in 2012. If Trump can get back to the 2004 turnout level, it would wipe out Clinton’s advantage with minority voters.
Each pollster has to consider all of these demographic groups and make an educated guess as to what that final turnout percentage will be. And the fact is there is no right answer today. The most favorable polls for Trump have been USC/LA Times and Rasmussen. Clinton does well with NBC and Marist/McClatchy. On Election Day some pollsters are going to look like heroes and some like goats, but it won’t be genius, as much as a couple of fortunate hunches.