Robert Epstein at Quartz: “Because I conduct research on how the Internet affects elections, journalists have lately been asking me about the primaries. Here are the two most common questions I’ve been getting:
- Do Google’s search rankings affect how people vote?
- How well does Google Trends predict the winner of each primary?
My answer to the first question is: Probably, but no one knows for sure. From research I have been conducting in recent years with Ronald E. Robertson, my associate at the American Institute for Behavioral Research and Technology, on the Search Engine Manipulation Effect (SEME, pronounced “seem”), we know that when higher search results make one candidate look better than another, an enormous number of votes will be driven toward the higher-ranked candidate—up to 80% of undecided voters in some demographic groups. This is partly because we have all learned to trust high-ranked search results, but it is mainly because we are lazy; search engine users generally click on just the top one or two items.
Because no one actually tracks search rankings, however—they are ephemeral and personalized, after all, which makes them virtually impossible to track—and because no whistleblowers have yet come forward from any of the search engine companies,
We cannot know for sure whether search rankings are consistently favoring one candidate or another.This means we also cannot know for sure how search rankings are affecting elections. We know the power they have to do so, but that’s it.
As for the question about Google Trends, for a while I was giving a mindless, common-sense answer: Well, I said, Google Trends tells you about search activity, and if lots more people are searching for “Donald Trump” than for “Ted Cruz” just before a primary, then more people will probably vote for Trump.
When you run the numbers, search activity seems to be a pretty good predictor of voting. On primary day in New Hampshire this year, search traffic on Google Trends was highest for Trump, followed by John Kasich, then Cruz—and so went the vote. But careful studies of the predictive power of search activity have actually gotten mixed results. A 2011 study by researchers at Wellesley College in Massachusetts, for example, found that Google Trends was a poor predictor of the outcomes of the 2008 and 2010 elections.
So much for Trends. But then I got to thinking: Why are we struggling so hard to figure out how to use Trends or tweets or shares to predict elections when Google actually knows exactly how we are going to vote. Impossible, you say? Think again….
This leaves us with two questions, one small and practical and the other big and weird.