On the perils of categorizing responses