Can Google searches predict where coronavirus cases will soon emerge?

On Tuesday, author Dan Sinker came across a correlation with a bit more heft. He compared Google searches for “loss of taste” in the United States with the number of coronavirus cases confirmed each day. The shape of the two curves match.

There is good reason to think that these two things are linked. Loss of taste and smell are, by now, recognized as common symptoms of infection with the novel coronavirus. Should someone lose one of those senses, it’s understandable why they’d head to Google to figure out what was happening. The correlation between the two seems as though it probably has a causal link.

Google’s explored the idea that its search data could be used to trace illness before. At one point, the company explored using searches for influenza-related terms as a way to track the spread of the illness. It abandoned the experiment after finding that its predicted number of cases were substantially higher than reality.

Sinker’s tweet nonetheless made me curious about whether there was a consistent relationship between searches for those terms and case totals nationally or in states. Using Google’s online Trends tool and The Post’s coronavirus data set, I compared the two.

Sometimes data analysis yields a truly stunning result. This was such a time.

Google provides search data as an index of peak activity in a period. So if, say, there are 3,000 searches for dinner Tuesday and 2,000 Wednesday, the value for Tuesday would be 100 and the value for Wednesday 67 — two-thirds of the peak. The day-by-day results can be volatile, so I used a seven-day average of that index. I decided to compare that to the seven-day average of new cases as a function of population, mostly because this helped make comparisons between regions easier.

Without further ado, here is the result for the United States overall.

Just a remarkable overlap. Except, of course, for that first spike.

But that’s easily explained: The United States was barely doing any testing in March and April, as documented by the COVID Tracking Project. There were fewer confirmed cases back then because there were fewer conducted tests.

That void is less apparent when we look at states which were among those hardest hit at the outset of the pandemic. In New York, for example, there was a spike in searches for “loss of smell” or “loss of taste” right before the first surge in confirmed cases. The peak in searches came about 12 days before the peak in case totals.

In New Jersey, which was also among the hardest-hit early states, the pattern is the same. You can also see that the recent uptick in the state was matched by a slight increase in searches.

The first state in which community spread of the virus was confirmed was California. That state has seen two surges, a small one in the spring and another as part of the summer’s surge focused on Sun Belt states. Again, there’s an increase in searches for loss of taste and smell shortly before the increase in cases. The Google data even seems to reflect the double-peak the state went through, though that may be an artifact of the data.

One of the states hit hardest over the summer was Arizona. The state hit its peak in searches for loss of taste and smell shortly before the state hit its peak in new cases. More recently, an increase in new cases was preceded by an increase in searches for those terms.

Notice that Arizona had a spike in searches for sensory loss in the spring, mirroring the national data. (States which were harder hit at the outset of the pandemic were also ones where more testing was deployed, helping to explain why there’s less difference between Google searches and cases in New York and New Jersey.)

So did Florida, where the summer’s surge was also preceded by a spike in Google searches.

In Texas, both the summer surge and its more recent increase in new cases trails increases in searches for losses of taste and smell.

This pattern doesn’t hold universally, it’s important to note. In Wisconsin, the recent surge in cases actually preceded an uptick in searches about sensory loss.

In other states, the Google search data are too spotty to pick out any trends.

In larger states and nationally, though, the correlation is striking. We’ve repeatedly seen increases in searches for information about losing one’s sense of taste or smell shortly before states saw surges in new coronavirus cases.

It’s not necessarily causation, but it’s hard to believe that it isn’t.

Leave a Reply