On a possible bias in estimates of hedonic price functions