The linear no-threshold response: why not linearity?