Can Google predict who will win the election?
With E-Day just around the corner, Canada’s pollsters have been out in force, asking Canadians whom they’ll back at the ballot box. But is it possible to predict Monday’s winner by looking at whom people are searching for online?
On Friday, Google put up a blog post offering some insights into what Canadians have searched for during this campaign. It includes a riding-by-riding breakdown of which leaders and which issues were searched for most often. However, the post stopped short of providing a direct comparison of searches for each leader. Here’s what that would look like; the chart goes back to just before Prime Minister Stephen Harper called the election on Aug. 2 and ends on Oct. 14.
(For best results on mobile, turn to landscape.)
While search interest in Liberal Leader Justin Trudeau has picked up speed over the past week, searches for Harper remain robust. Search interest in NDP Leader Tom Mulcair has fallen far behind. (The Google Trend tool does not show absolute search volume, but is a relative comparison of searches over time, with 100 reflecting the peak of intensity. There’s an explainer here.)
For comparison, here is how the three leaders have fared in public opinion polls, where people are explicitly asked whom they support. The chart below shows the poll average compiled by Éric Grenier of ThreeHundredEight.com for CBC’s Poll Tracker tool.
(Again, for best mobile results, flip to landscape.)
The two charts both show Trudeau expanding his lead (in searches and poll support) over Harper, with Mulcair losing momentum.
The question of whether Internet searches can predict election outcomes is one of great debate, for obvious reasons. Big Data, the term given to the vast quantity of information now available, thanks to our hyper-connected lives, has shown to be highly useful in discerning consumer habits and trends. The possibility exists that data generated through the social web could be far more indicative of our voting inclination than what we tell pollsters we intend to do. Given the many high-profile failures by pollsters in recent elections, this would be an important tool for election forecasting.
So far, however, the evidence is mixed. One 2014 study by Oxford Internet Institute—”Can electoral popularity be predicted using socially generated big data?“—looked at the correlation between Google searches, Wikipedia usage and actual election results in elections in three countries, the U.K., Germany and Iran. While searches for individual leaders rather than parties proved more successful (Google and Wikipedia predicted the winners of the 2013 Iranian and 2010 U.K. elections), there were no clear patterns across different elections, and selection bias was pronounced. Not everyone is on the Internet, and they are not randomly distributed across voting districts. “People do not simply search in the same proportions that they vote,” the paper’s authors wrote. The institute found similar results after analyzing social media ahead of the 2015 general election in the U.K.
Some will also argue that Google searches merely reflect what the media are talking about: The more stories there are about a leader, the more likely it is that people will search for that individual. That argument likely overstates the mainstream media’s influence. It also doesn’t account for the relatively even scale of coverage the three leaders have received over the past 90 days. According to the Infomart media database, which captures the majority of Canada’s print, digital, TV and radio outlets, Harper has received the most coverage (35,632 story mentions) with Trudeau (24,844) and Mulcair (23,466) nearly tied for second.
This argument also doesn’t explain the largest discrepancy between the two charts: Mulcair’s early dominance. The NDP’s early lead, reflected in the polls, received extensive news coverage. Yet, in search interest, the NDP leader always lagged Trudeau and Harper.
Come Monday night, we’ll have a better understanding of how accurate both the pollsters and search interest were in forecasting the election. But it will still be some time before researchers are ready to declare Google an election-predicting winner.