Here is the most informed thoughtout post, made by Slappy, I've seen on the subject taken from the official boards:
this is a straight repost froma similar topic in the arthas form
-------------
this information is misleading.
First, the data this site computes total characters, or attempts to. It takes population snapshots, then compares that with what information it has, adds new people if applicable and then updates existing people if applicable.
Flaws in this system is that, if more people are taking horde snapshots (since they have horde characters and thus cannot take an alliance snapshot), then horde side will show more people since they have more snapshots.
While computing a total character population, this is fine. But in no way can this be used to convey an idea of population imbalance. The only way todo this, is to analyze the snapshots individually and not merge them.
What do I mean?
Every time I have run a census and received a snapshot of the current population at that time (for arthas), it has been alliance 60% horde 40% (give or take less than 1%). Now if at noon, 5pm, 10pm and midnight everyday, the alliance has 60% of population: we can safely assume that there is a population imbalance and that the alliance has more people (at least, for the sake arguement, during those "peak" times).
----------
Now as to computing "total" character population, which this/these sites are trying todo, their accuracy is directly tied to the amount of snapshots, and the variety of said snapshots. If they only snapshot the population on tuesdays at 8pm, their numbers will be skewed. Or for that matter, if they only have had 2 or 3 snapshots, then their numbers will be quite skewed.
However, given a large amount of snapshots, all over varying times, the accuracy of these sites will be more accurate. However, at the moment -- as we can see -- these sites are not very accurate for many servers.