Group 5: Kelsey Bumgardner, Kyle Morrison, Andreas Tonckens, and Molly Foley
Tonight is the first presidential debate between Hillary Clinton and Donald Trump and it is expected to be the most watched debate in history. With the election just over two months away, it seems that more American people have taken interest in the political drama between these two candidates than most other contests. In addition to the debate, polls have gained a growing amount of attention from the public. Yet, why do polls have so many different results?
There are a couple of different reasons why polls may differ in results. Response bias can impact whether a sample is truly representative of a population. Selection bias can determine if a sample is random or not. But what if you can control all of these confounding factors and variables? Or what if you gave the same data to different pollsters and asked them to interpret it. The New York Times did exactly this by giving raw data from a poll of Florida voters to four expert pollsters. All four pollsters had different results for who was ahead in the presidential election. Their answers illustrate the significance of identifying likely voters.
Most polls in today’s society use participants who are likely voters rather than registered voters. The reality is that not all of the registered voters will actually go out to vote. Getting data from likely voters will show a more accurate depiction of the results on election day. So how do statisticians determine who is a likely voter? The most obvious way is to ask the participant how likely they are to vote? However, many Americans will lie and say they will vote, while come November 8th they don’t show up to the polls. Americans want to be perceived as good citizens, that it is their freedom, right, and duty to vote. Social desirability bias would also inflate the number of likely voters reported from this question because participants often give the answer they believe they’re supposed to give. While self-reported vote intention is used frequently when conducting surveys, it overestimates voter turnout. Two out of the four pollsters in the New York Times experiment defined likely voters as those who answered they were “almost certain” to vote or self-reported. Interestingly, the results of these two polls had Clinton ahead by a +3 and +4 margin.
The other two pollsters looked at voter history to determine the likelihood of the voter. Past turnout is a strong predictor, so pollsters often look at voter file data from the past to determine who is likely to vote. However, this method may also skew results. For example, a large portion of Trump’s supporters are older, while Clinton has a great share of younger voters. In using voter history, one cannot determine the likelihood of a recently eligible voter that supports Clinton turning out on election day. Clinton’s supporters are also composed of a great number of African Americans and Hispanics, who have historically been less likely to vote. Using voter history to identify likely voters may tilt the election slightly more towards Trump. The two pollsters from the New York Times experiment who defined likely voters using their voting history had Clinton ahead by +1 and Trump ahead by +1, a much closer race.
It is clear that even when using the same sampling methods, identifying who exactly are likely voters can have a huge impact on the results of a poll. RealClearPolitics recognizes the variability in polls and takes an average of ten different polls to calculate who is ahead in the general election. The McClatchy/Marist poll methodology uses a probability turnout model to determine likely voters, which is based upon participants’ interest in the election, past voting participation, and chance of vote. The ABC News/Washington Post poll methodology incorporates intention to vote, attention given to the election, age, past voting history, whether the participant says he/she is registered, knowledge about polling locations, and political party identification. This illustrates the diversified methods in which pollsters determine who is a likely voter. Both of these polls are included in RealClearPolitics’ average and support the notion that not a single poll is perfect. As voters, we must look at a poll with a grain of salt and remember how differently data can be interpreted. RealClearPolitics smartly combines these polls into an average, so we can look at them collectively and hopefully achieve better prediction.