The Rise of Poll Aggregation and Election Forecasting

Abstract

This chapter discusses (1) the development of election poll aggregation and its use in popular election forecasts, (2) the technical and statistical demands of using polls this way, and (3) the controversies surrounding aggregation and forecasting. The first section covers how increases in publicly released polls resulted in poll averaging and aggregation websites becoming popular in the early 2000s, then how election forecasting using polls as the biggest predictors became popular in the media. The second section discusses how polls are aggregated and how aggregations vary. The focus then turns to how polls are used in election forecasts. Finally, common questions that arise from poll aggregation and forecasting are addressed: Are averages always better than single polls? Are there too many forecasts; do we need to aggregate the forecasts? Are we expecting too much from polls, which are meant to be snapshots, not predictions, by using them in forecasts?

The Twilight of the Polls? A Review of Trends in Polling Accuracy and the Causes of Polling Misses

“…Further reinforcing the salience and implied certainty of polling has been the rise of poll aggregators and polling-based electoral forecasting (Jackson 2016). Notable early forecasting successes, such as Nate Silver’s 2008 and 2012 election predictions, created an air of infallibility around election forecasting.…”

Forecasting Federal Elections: New Data From 2010–2019 and a Discussion of Alternative and Emerging Methods

“…Forecasters have attempted to improve polling in ways that include poll aggregation (Wang 2015), application of a discount rate to account for early frontrunner bias (Erikson and Wlezien 2012), and adjustment of polls' demographic weighting (Prosser and Mellon 2018). I suggest that the first two improvements have limited benefit because they rely on the same underlying data, which is the root of most polling accuracy issues (Jackson 2018), while weighting adjustments can have mixed effects. Internationally, erroneous weightings reduced forecasting accuracy for the 2017 UK general election (Sturgis and Jennings 2017) and the 2014 US midterm elections (Keeter, Igielnik and Weisel 2016).…”

Forecasting two-horse races in new democracies: Accuracy, precision and error

“…It is likely that the average of the two polls will be a better estimator of the result than any poll chosen at random. Studies have shown that, as a general rule, as more data is considered, accuracy levels improve (Jackson, 2018). And, while experiments and research are still burgeoning, they have already shown that accuracy levels can at least matched those of traditional ones (Graefe et al, 2014).…”

“…And because of their parsimony, they have been on the rise. While their use in media can be traced back to the website FiveThirtyEight, initiatives have since burgeoned (Jackson, 2018). The most basic model is known as the Poll of Polls model.…”