How NPR used multivariate testing to learn which design element produced more conversions

NPR recently ran controlled experiments live on its site, testing whether a small design tweak would prompt more users to “take action” on a Support NPR button. Tyler Fisher and Livia Labate write about how this worked: in one case they constructed two versions of the page, one of which was the “control” and placed the Support NPR prompt at the story’s end, another which first probed whether the user cared about the story. The “care question” converted more users to the desired action, with statistical significance.