On the measurement of bias in geographically weighted regression models