Sample size of tank statistics
Great work with the site - it has become my favourite stats site!
I do wonder about one thing though: What is the sample size of the tank statistics?
E.g. this one: http://www.vbaddict.net/tankstats/usa/heavy/t34-11
There are quite some "jumps" in the charts that shouldn't quite be possible with large sample sizes.
Even my own tank statistics, based on only my own games (low sample size), are much more constant than that.
The charts are just as jumpy, but the scale is different, showing only a few percentage points deviation. E.g. the average XP chart goes from 1006 to 1014 here, based on just a few hundred games.
On the global T34 stats the average XP varies between 350 and 450. That kind of deviation just shouldn't happen with a significant sample size of thousands of games.
we are talking here about ~1 million battles recorded for the T34. Sample size is not the problem, but the code to gather and print the data to the overall tank charts. It's a known issue, which I didn't had the time yet to look after.
Cool, thanks for the info!
Are the total numbers reliable and only the charts affected? E.g. the winrate of the soviet T34 indicates that the total stats may be affected as well.
38.1% global WR seems far too low.
T34 or T-34?
Data for both is reliable.
The soviet one
To be honest, i find that hard to believe. One would have to teamkill in order to drag the team down to 38.1% win rate. Even AFKers or bots achieve around 44% after all.
It's similar for the Cruiser Mk3:
As far as i know that thing is considered somewhat OP, a sealclubbing machine just like the hellcat. Having only 42.4% global WR on it doesn't seem plausible.
Even the CW workhorse T110E5 is listed as 46.9% global WR. That just doesn't sound right.
There are also tanks with 63% winRate like the Pz.Kpfw. B2 740 (f). In the end, the average winRate is 48.8%, which is correct.
Only public games influenced the statistics, no CW.
Keep in mind, not only sealclubbing platoon guys are uploading their data, and the statistics is driven by all of the 30 players of a game.
True, but achieving >60% with a rare (less likely to have an identical tank in the enemy team to balance this out) OP sealclubbing tank with super preferred matchmaking (always top tier) is much more reasonable than a common tank with normal matchmaking and without any particular good or bad reputation performing far worse than AFKers.
The former is a matter of the performance ceiling. A single excellent player in an OP tank can perform as well as two or three or even more average players in average tanks.
The T34 issue however is a matter of the performance floor. It's one thing to perform as good as two or three average players. It's much more "difficult" to perform worse than an AFK player, to be worse than useless. One would have to actually be a teamkiller for that.
Unless half of all T34 drivers are teamkillers or tend to come in 3 player platoons that then go AFK (thus being an instant loss of three tanks for their team) i don't see how it could perform that bad. Even if *all* T34 drivers would just go AFK, it should have a much better global WR than that, being carried by the team to the usual AFK WR of 43-44%.
And then there is the chance that the enemy team has its share of T34s too after all...those things are common.
A tank having a global WR of worse than being AFK is a safe sign that something is wrong, IMHO.
Last edited by jwu.eu; 2014-03-20 at 09:18.