Why behavioural responses may not reflect the population consequences of human disturbance