To set educational policy, physicists have to do some social science. Here I look at the most publicized paper on the most prominent current issue: the use of GREs in admission. The paper makes numerous elementary statistics errors: variance inflation by unnecessary collinearity and by unnecessary stratification, collider stratification bias, bias from improper imputation of missing data, conflating failure to reject a null with confirmation of the null, eccentric use of subsamples, invention of a new way to inflate error bars by an order of magnitude, and omission of needed data. The sign of each error tends to support the authors' prior hypothesis. I draw two morals: we need stats education and we need to recommit to scientific honesty.
