The statistics have been calculated from a data set with very high number of players (700,000+) and battles (in total 6bn+). That is maybe 4 times the users Blitzstars.com has. However, there is still sampling bias as there is any Blitz stats tracker site. This is evidenced by the overall WR distribution and the share of all wins out of all the battles in the database being above 50%. This is a topic worth of a longer post - stay tuned.
Out of pure interest into analysizing Blitz game statistics and to learn data analysis and R.
Relative WR the average of player WR in a tank minus their Average WR in all the same tier tanks. If the tank is perfectly balanced, its Relative WR is 0 %. Large positive Relative WR indicates the tank is “very good” (read “OP”). Negative Relative WR indicate the tank performce worse than other same tier tanks. Relative WR is one of the better (if not the best) best measures for tank performance since it takes the skill differences in tanks' playerbases into account, IMHO.
rWR = Relative WR
stats[DT_filter_battles_grind(all.battles.new, tier, is_premium), WR.tier := sum(all.wins)/sum(all.battles), by=.(account_id, tier)]; stats[DT_filter_battles_grind(all.battles.new, tier, is_premium), tanks.tier := .N, by=.(account_id, tier)]; stats[ , rWR := 0]; stats[ (tanks.tier > 1) & DT_filter_battles_tank(all.battles, tier), rWR := WR - WR.tier];
The bit cryptic syntax is R’s data.table - one of the fastest software packages for data analysis.
Yes, a bit. At the moment Blitzstars.com calculate the Relative WR as WR difference to the players' average WR over all their tanks. BlitzAnalysiz considers only the other same tier tanks. The real differences are likely rather small.