This paper demonstrates a hitherto overlooked problem with the differences-in-differences approach where it is applied to panel data and with overlapping differences.