How Google searches might be able to predict the 2020 Democratic race

Posted at 10:36 AM, Jun 25, 2019
and last updated 2019-06-25 12:36:03-04

We are what we search.

Our reliance on the internet combined with Google’s dominance as our entry point into the web means that our search history reveals lots and lots about not just us but the country and, yes, its politics. (That’s not a new theory. It’s best explained by Seth Stephens-Davidowitz in his amazing 2017 book “Everybody Lies: Big Data, New Data, and What the Internet Can Tell Us About Who We Really Are.”)

Which brings me to the 2020 Democratic presidential race — and my (and our) ongoing attempt to understand not just where the race is today but where it’s been and, most importantly, where it’s going. We typically rely on polling to do that and, generally speaking, it does a solid job. But, the polling industry is in troubled times right now — barraged by cheap competitors with low quality results (that nonetheless draw headlines) an increasing struggle to build representative samples without heavy weighting and the rapid abandonment of land lines by young (and not so young) people.

None of that means we should abandon polling. Far from it. But, what it does suggest is that other measurements are needed. And that’s wherethis amazing chart — via Google’s Simon Rogers— comes in. It tracks Google search interest in 23 of the Democratic candidates for president from December 30, 2018, until June 16.

Here’s the interactive chart:

What’s fascinating — at least to me — is how closely the search patterns for each candidate (and for the field more generally) track with polling trends we’ve seen. Consider:

* Former Vice President Joe Biden, from the minute he formally entered the race in late Aprilis at the top of the pack in terms of search

* Massachusetts Sen. Elizabeth Warren tumbled from No. 1 in search late last year all the way to No. 9 in the spring, a descent paralleled by her struggles to gain a foothold in the race amid ongoing questions about her Native American heritage (or lack thereof). Search also captures Warren’s momentum in recent weeks; as of June 16, she was the second most searched candidate — behind only Biden.

* The rise — and fall — of former Texas Rep. Beto O’Rourke is also captured in the Google search data. In early March, O’Rourke was at the top of the field in terms of search. As of mid-June? He’s in seventh. Not good.

Now, searching for a candidate on Google isn’t the same thing as voting for them. You could, for example, look up “Elizabeth Warren” because you don’t want her to win and need some talking points to push back on your pro-Warren friends. But, the massive number of Google searches that are being measured during this time period would tend to wash away any statistical noise caused by that sort of scenario.

As the Cook Political Report, a non-partisan political handicapping site, notes: “Google search data is obviously different than polling. A voter could be curious about a candidate but have no intention of voting for them. For the purposes of the 2020 election cycle, Google search data is most useful in showing trends over time and interest in specific events.”

Most interesting to me is whether or not Google searches can be predictive of future support. For example, businessman Andrew Yang as been steadily in the top 5 in terms of Google searches since late February. And yet, he remains around 1% territory in most national polling on the race. Does Yang’s elevated search interest simply tell us that he has a larger online following than other candidates? Or is the search interest a leading indicator of a Yang surge?