I think we also kind of backfit our expectations to the results. I think if you write this headline either way, you'd have people who would say "yeah, that makes sense". Because if you told us that autistic children with imaginary friends have worse social skills, you'd have people saying "Yeah, that makes sense, imaginary friends don't respond like real people, they're just reinforcing bad habits."

But we should be testing our assumptions far more than we do. Otherwise, we're just cutting off the ends of the roast.