Some observations on fitting a straight line to data