More statistics: Random performance doesn’t predict rank

After the statistics post last week an anonymous donor who just sent the message “Maxim is an idiot” (I have no idea what that means) sent me some data grabs from the WoWs leaderboards. Update: he has sent more, but I had no time analyzing that yet, so that will come tomorrow.

The data contained all EU players who played randoms and contained winrate, damage, battles played and rank. Since it had 141419 players, plotting the linear regression took forever, so we have to settle with calculating the linearcoefficents and the R2 values:


What does this tables mean? The Winrate-Battle field is 0.00037. This means that 1000 battles increase winrate by 0.37%. The same field below shows that the R2 value of this fit is 0.035, which is next to nothing. So the amount of battles played doesn’t really predict the winrate. Go figure, mindlessly jumping into games won’t make you better.

The best correlation (0.579 which is still low) is between Damage and Winrate. 1000 damage increase predict 0.2% winrate increase. That doesn’t look much, so I sorted players into 5 groups by increasing damage:


You can see how huge the damage difference between the groups. The lowest damage group has 18.5K, the highest has 53.1K. The winrate of the groups is also significantly (by the size of their standard deviation) differ.

The number of battles somewhat (0.148 R2 is very low) predicts the chance to play ranked at all. About 100 more random games means +1% chance that the player steps into ranked.

However rank itself depends on random battle count barely. So playing a lot doesn’t mean getting good rank. More surprisingly it barely depends on random winrate too. Not only the R2 is very low, but the linearcoefficient says that every % winrate decrease rank only by 0.383, meaning that even if we believe the trendline (we absolutely shouldn’t with that laughable R2), it says that +60% winrate would predict ranking out, which is also nonsense. So those who are good at playing random are not at advantage in getting good rank.

Rank correlates best with random damage (0.2 R2 is low), the trendline says that +10K damage predict -1.7 rank. That’s also not much, but at least gives one hint: being able to farm damage in random predicts rank better than winning in random. This “proves” (remember, horrible R2) that random and ranked battles have completely different metas and strategies from one shouldn’t be carried over to the other, while the ability to do damage somewhat carries over.

Author: Gevlon

My blog:

2 thoughts on “More statistics: Random performance doesn’t predict rank”

  1. Damage is a better predictor as XP?
    Maybe damage is an indication of the ability to aim or predict the enemies movement that carry over better as the skills/gameplay style that grants more XP.

    If you watch King of the Sea tournaments on youtube you notice a totally different playstyle as ranked or random. A lot more patience.


  2. Damage is also correlated with Tier – playing higher tier games means higher damage on average. Also, raw damage is correlated to ship class played – BBs get more but most of their dealt damage is on other BBs so they provide less %HP damage.
    Case in point, myself, I had below average damage in ranked YY but 62% win rate. That’s because I played as DD hunter so my 20k damage meant a kill but somebody who tried farming BBs with artillery and torps would need 100k damage to get a kill. As a result, my Personal Rating (an artificial stat maintained by stat websites, dependant on mostly damage and win rate) was terrible at <1000 points but I had 60th best frag average on the server.

    Don't read too much into the stats. They can be gamed in Randoms and they are heavily dependant on time and rank in Ranked.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s