It's a estimation made with of a model with data from an electoral polling company using phone call to a total of 140'000 individu over two years.

Let's look at the affiliations of the authors : save for an economist in India, two thirds of the team are from a Canadian think tank, the rest work for a big data company in Washington, almost none have published anything ever, and if they did, it was not medecine...

IMHO : A lot of doubt, if only due to the data source.

This smell like a lobbyist's made to order report. To me, it's strange it's been allowed by the Science editorial board.