A Look Back: 2012 Presidential Polls in Florida

• Scott Richards, M.A.

During every presidential election, we’re inundated with polls and polling information and the 2012 election was no exception. Now that the election is behind us, it is fair to ask how the pollsters performed: how good were these surveys? Did they accurately predict the election’s outcome?

There are challenges that every pollster faces in their effort to produce accurate poll results. In any survey, you want to be sure you’re interviewing the right person. In a presidential preference poll, this means you need to question the people who are going to vote. There are a lot of techniques and models used to identify these individuals, who are known as “likely voters.” Pollsters want their results to be representative of the electorate at large which means they’re concerned about factors such as gender, race, age, education level, and political affiliation (the ratio of Democrats to Republicans to Independents) among others.

It is obvious that adjusting polling results, while necessary, can be a bit of a balancing act.It is obvious that adjusting polling results, while necessary, can be a bit of a balancing act.

…if a pollster believes that in a certain state, say, 40% of the voters are Republicans and the actual survey just happens to turn up 35% Republicans, each Republican interviewed will be given a weight of 40/35 to correct for the under sampling of Republicans. All pollsters do this to correct for under- or oversampling by party, gender, age, race, income, and other factors. This is not only legitimate, but necessary with the small samples all the pollsters use. (Tanenbaum, 2012)

Survey methodology is not perfect, and pollsters can legitimately disagree about the makeup of the electorate in an upcoming election. This emerged as an issue in the 2012 election, when charges were made that many of the public polling organizations were oversampling groups that tended to support Obama, in particular Democrats, young voters and non-white voters. An analysis of national polls conducted by Fordham University after the election found:

For all the derision directed toward pre-election polling, the final poll estimates were not far off from the actual nationwide voteshares for the two candidates.For all the derision directed toward pre-election polling, the final poll estimates were not far off from the actual nationwide voteshares for the two candidates. On average, pre-election polls from 28 public polling organizations projected a Democratic advantage of 1.1 percentage points on Election Day, which is 2.7 percentage points away from the final estimate of a 3.8-point Obama margin in the national popular vote (Obama 51.0% versus Romney 47.2%) (Papagopoulas, 2013)

Polling challenges can be magnified at the state level. In general, state-level polls are not conducted as frequently as national polls, and they tend to have fewer interviews, which mean the results have a wider margin of error. As a hotly contested battleground state in the 2012 election, Florida received a lot of attention from candidates and pollsters alike. The final outcome was very close: the margin of victory was less than 1% of all votes cast.

In mid-October 2012, the University of Florida Research Center (UFSRC) reported results from a Presidential preference survey. This study was conducted between September 1^st and October 14^th. It differed from most polls, which are usually conducted in a brief two- to five-day period. It used a simple screen to identify Likely Voters: anyone reporting they were “not likely to vote” was excluded. The raw data was then adjusted for population and age. The UFSRC’s survey results for both the raw data (prior to any weighting) and adjusted data (after weighting) are:Let’s take a closer look at some state-level polls that were conducted in Florida for the 2012 Presidential election.

The unadjusted responses in Table 1 show 274 respondents preferred Mitt Romney, while 253 respondents indicated they would vote for Barack Obama: a slight 21-vote (3.7%) edge for Mitt Romney. Adjusting the data for population and age—changes which primarily adjusted for the under sampling of 18-44 year old respondents, and the oversampling of respondents ages 65+ brought the respondent pool more in line with the Florida electorate, shifted the results. After adjustment, there is an 8-vote (1.4%) advantage for Obama.

There are two factors that should improve the accuracy of a poll: one is the timing of the poll, and the other is the number of interviews that are conducted. The closer a poll is conducted to Election Day, the more its results should reflect the overall opinion of the electorate. Completing more interviews should also give a better picture of respondent preferences. Here is a comparison of some of the state-level surveys that were conducted by national polling organizations:

There are two main goals for a poll: to identify the leader (or winner) and to get an idea of how close the race is between the candidates. Most of the polls listed in Table 3 (seven out of nine) show a very close race, with a 1% or 2% margin for the leader. Five of the nine polls correctly predicted an Obama victory; 3 others predicted a Romney win, and one poll had the race as a dead heat. How did the larger polls fare? Four polls completed 900 or more interviews: three of those predicted an Obama win. Of the three polls that were conducted closest to the election in November, two showed Obama winning, and the third showed a tied race.

While the UFSRC’s poll was conducted earlier than the other surveys in this comparison, and it had the second-lowest number of interviews, it showed Obama leading a very tight race. The UFSRC did have a relatively large share of “Other” responses. (This category includes people who are undecided, along with those who say they’re voting for third-party candidates—and it typically declines as you near Election Day.) The UFSRC’s poll is consistent with surveys conducted by other national polling firms.

Overall, the state-level polls conducted in Florida did a good job of showing how close the presidential race was; they weren’t as good at showing the leader. This highlights the need for surveyors to continue to improve their methods of data collection and find better ways to reflect the opinions of a changing electorate.