It's not quite random, the goal is to bias the errors towards false negatives (bad rejection) than false positives because bad hires are so expensive to recover from.