On February 20 2015 Irish Premier Enda Kenny confirmed that a “yes-no” referendum on same sex marriage would be held on May 22 of the same year. … The electoral outcome turned out to be 62.07% for the yes vote. … Using hourly Google Search data one week prior to the Irish Referendum of May 22 2015 and a simple ratio of “vote yes” to “vote no” searches I demonstrate how the outcome could have been predicted on the nose.
In a bold and risky political move the Greek prime minister Alexis Tsipras called for a referendum on June 27 2015 quitting ongoing negotiations with Greece’s creditors in Brussels. … Due to tense debates and increasing polarisation it became increasingly impossible to rely on traditional polling. Even the first exit polls (performed by phone on Sunday evening) could only see a marginal lead for one or the other vote at different times. … Using Google Trends I could tap into voters’ true and unbiased revealed preferences and nowcast hourly what the ratio of the No vote to the Yes vote is and called an over 60% No vote well ahead of the closing of the voting urns.
Here’s the story from the Greek referendum. Askitas compared the frequency of Google searches for “no and not yes” (in Greek) to the frequencies of “yes and not no.” Google Trends provides the data at a one-hour lag.
Here were the most frequent searches of each category:
Here were the frequencies for each hour during the week before the vote:
Consistently more No than Yes. Askitas then looked at the ratio of No to Yes:
The ratio is consistently above 1, and it is mostly fluctuating around 1.5, which would correspond to a 2:1 vote in favor of No.
Actually I think it would’ve made more sense to plot No/(No + Yes) so you’d get a value between 0 and 1, which is directly comparable to a vote proportion.
There’s no reason to know ahead of time that this Google Trends trick would work — after all, you can search for No or Yes without having that particular political view. And, as Askitas points out, searchers are not representative of the general population, as older and more rural residents are less likely to be searching.
Finally — and to return to to the question in the title of this post — I can see this being used as a tool to predict referendums, but I don’t think it would work so well in, say, the U.S. general election. Why? Because we can already predict the election to within a few percentage points based on historical information. Google Trends seems to be able to distinguish 65 percent support from 50 percent or 80 percent but, based on the graphs above, it wouldn’t seem to have the precision to distinguish between margins of 47-53 and 53-47. Referendums are a particularly good place to use Google Trends because there is often no good baseline.