Google searches are not a great indicator of electoral success


This is the era of having more data at our disposal than we know what to do with. And so it is that I came to wonder: Do Google searches correlate well to election results?

This is not entirely an idle question. Google, through its Trends team, publishes data and analysis on what people — in the United States and internationally — are searching for. And since Google searches have become the default mechanism by which people get answers to their questions, it’s reasonable to think that those searches might reveal something about intent. This is Google’s business model after all: If you search for “new car,” Google presumes you want to buy a new vehicle and Google-sold ads for new cars will pop up on every page you open. It’s not complicated.

Since Google is good about sharing its data, the Trends team provided me with the breakdown of searches in the last week and the last day of the campaign by candidate for a number of contested Senate and gubernatorial races nationally, allowing me to answer the question in the first paragraph above. (Google, as it turns out, is even good at second-order question-answering.)

That answer? No, search interest doesn’t overlap much with electoral results.

Sign up for How To Read This Chart, a weekly data newsletter from Philip Bump

We can start by simply comparing the actual results of each contest (as of Wednesday afternoon) with search interest. On the chart below, each candidate (Democrats in blue and Republicans in red) are shown relative to the percentage of the vote they earned (from bottom to top) and to the percentage of search interest they got on Nov. 7 (from left to right).

If search interest matched election results perfectly, every state would sit on the diagonal line. Instead, the result is more of a cloud. The correlation isn’t that strong.

It struck me, though: Maybe this is somehow missing people who’ve already voted? So I took data from Arizona, where results are broken out by vote type (including Election-Day-only) votes, and compared the data. The correlation was stronger. On Nov. 7, for example, Republican gubernatorial candidate Kari Lake was the focus of 73 percent of the search results between her and her opponent, Katie Hobbs. The following day, Lake got 70 percent of votes cast. But in the Senate race, the correlation broke down.

Google also gave me a slightly wider window to look at. If we compare searches over the last week to the election results, we see that the correlation is slightly better, but not much. (Visually, we see that states below are generally a bit closer to the diagonal line.) This is still not a great predictor.

But we can also be more concrete. The last-week Google search results did correctly call 45 of the 65 races I looked at. (I excluded races like the contest in Alaska, where the result is determined by ranked-choice balloting.) That’s a decent-but-not-great two-thirds accuracy.

It also includes a lot of blowouts where predicting the winner wasn’t useful. In contests where the actual margin was under 10 points, Google’s search data only got 21 of 37 races right, about 57 percent of the total. Slightly better than a coin toss.

Google and Google Trends are very useful tools, obviously. If, however, you’re looking to them to tell you who’s going to win a close election, you’d be about as well served by flipping a quarter in the air and seeing how it lands.

Source link

Tags: No tags

Leave A Comment

Your email address will not be published. Required fields are marked *

Shopping Basket
Social Media Auto Publish Powered By :