On estimating the magnetocaloric effect from magnetization measurements