Punishing potential mothers? Evidence for statistical employer discrimination from a natural experiment