My point was that this "variation" is not big enough to alter the data, and can be counted as false-positives. I don't think the comparison to a religious select input without atheism is fair since atheists represent a fairly large chunk of the population.