If the election for president were held today, and Donald Trump were the Republican candidate and Hillary Clinton were the Democratic candidate, for whom would you vote?
When people think of polls, they commonly think of the vote intention question, which asks respondents for whom they would vote if the election were held today.
Strictly speaking, such intention polls do not provide forecasts but rather snapshots of current opinion. Nonetheless, polls are a common means of forecasting election outcomes. Scholars, the news media, and the public commonly interpret polls as forecasts, and project the results to Election Day.
Forecast accuracy of polls
Results from single polls often produce large forecast errors, which are greater, the longer the time until the election. Also, research has shown that different polls conducted by reputable survey organizations at about the same time often reveal considerable levels of variation in their results. Errors caused by sampling problems, non-responses, inaccurate measurements, and faulty processing diminish the accuracy of polls (e.g. Erikson & Wlezien, 2012).
Across the last 100 days prior to the six elections from 1992 to 2012, polls provided on average the least accurate forecasts of all PollyVote component methods. Compared to a single poll published on the same day, the combined PollyVote forecast reduced forecast error by 59% (Graefe et al., 2014).
A simple approach to increasing poll accuracy is to combine polls that are conducted by different organizations at nearly the same time. Using the median of all state-level polls taken within a month of the presidential election, Gott and Colley (2008) correctly predicted Bush’s victory over Kerry in 2004, with an error of only four electoral votes. They also forecasted Obama to win over McCain in 2008 with an error of only two electoral votes. In both elections, the median statistical approach missed the winner in only one state. Simply aggregating polls has also become popular in the news media. Well-known poll aggregators such as RealClearPolitics and FiveThirtyEight update combined polls on an almost daily basis.
Use of polls in the PollyVote
The PollyVote relies on poll aggregators to calculate the combined polls component forecast through simple averaging. For the 2020 election, PollyVote’s combined poll forecast is an average of three poll aggregators, namely those from RealClearPolitics, FiveThirtyEight, and the Economist.
References
- Graefe, A., Armstrong, J. S., Jones, R. J. J., & Cuzán, A. G. (2014). Combining forecasts: An application to elections. International Journal of Forecasting, 30(1), 43-54.
- Erikson, R. S., & Wlezien, C. (2012). The timeline of presidentialelections: How campaigns do (and do not) matter. University of Chicago Press.