In part, AFAIA (but I've only limited statistical training) this is a well known feature of voluntary datasets. But IIRC they've also done user studies and compared those.