On Simpson's Paradox in Economic Statistics