Thanks guys - yes, work in progress. As we've been discussing in this thread (and some people obviously missed ), the statistical sample is flawed. And WN8 uses those for calculations. And the percentiles are used in the tank comparisons.

I have today begun work on getting better averages.

Here is my current thinking:

Fundamentally, a stats site using it's own 'user' database to determine averages is flawed. As I've mentioned in previous posts, people going to a stats site are going to be overwhelmingly in the 'better than average' category. This means, to get more accurate averages, a site would have to compensate by adding worse players in to the database. But as soon as you start artificially injecting players in the the mix, the legitimacy of the stats and averages goes out the window. Basically two main things will happen:

- A lot of 'below' average players are added to 'balance' out the bias of the type of people who visit the site (I particularly dislike this as I am a big fan of percentiles)
- A mix of players are added, helping to balance the averages, but still leaving a 'top heavy' bias.

Whatever happens, the results are less than optimal. They might be 'okay' if you have a very high total number of players recorded versus users, and if you are only interested in 'mean' averages. But neither of those things applies to BlitzStars and neither appeal to me.

So - thanks to some of the positive contributions in this thread - I have decided to test/implement entirely separate stats to use for the averages. And, again thanks to chatter in this thread, I have devised a method to get those stats.

The current selection process is this:

5-10k randomly selected accounts from the 4 regions, so 20-40k total accounts (aka players).

Each selected account must:

- Have a minimum of 1,500 battles played on the account
- Have had at least one battle in the last 30 days

For the individual tanks there is just one consideration:

- A minimum of 100 battles

I hope that these numbers will be sufficient to provide a true and accurate reflection of the playerbase.

I have run an initial test and getting this number of players per region looks like it will be no problem - which is fantastic.

So now what remains is to decide if the above qualifications (battles played, minimum battles in tank, etc) are good to run with.

__If you guys have any thoughts, please share.__