A critique of the Kelvin method of measuring work functions