I'm berserking!

Before reading this post

The games in this analysis were lichess-annotated 3+0 blitz games from June 2023. More about the data can be found at the lichess open database.

Why focus on 3+0 blitz games?

This post didn't have to focus on 3+0 blitz games necessarily. However, there are computational reasons for choosing specific time controls over other time controls. I decided on games in which only one side berserk. If both sides are berserk, there is less meaning in understanding the effect of berserk. I also got the player elo variance for 10+ games berserk and assumed the rest of the players to have the rating variance of the mean of my sample. This might cause some generalization issues but it simplifies the problem.

How do games end given berserk?

The following is the distribution of how games end given one side berserk. Winning or losing normally (this includes checkmate + resignation) was the most likely result. A draw was the least likely result of a berserk-involved game.

How does Elo change things?

We can also start looking at how Elo is related to berserk decisions. We expect people to berserk when they are higher in elo and trying to gain more points. The bin size for this analysis was 25 elos (the width of each bar = 25).

In the exact lumps we see in the graph above, we get a very steep downfall in around +300 elo. This seems to relate to the number of games that happened in the range. Since June 2023 is the only dataset we are working with, I will be working on an explanation for this steep downfall by analyzing some more datasets. The peak in the -500 Elo range is untrustworthy due to the low sample size.

Explaining the weird peaks

The weird peak at +300 Elo needs some investigation. The biggest phenomenon seems to be berserk and losing on time. The percentage of games that end in Time forfeit increased around the 300+ mark.

Furthermore, from the distribution of categories, we can see that the amount of "berserker and lost on time" increase by a lot in that certain elo range and the "normal wins" significantly decrease.

When should I berserk?

When you berserk in the arena tournament, you gain +1 point for either a win or a draw. So, we can calculate the expected points you can get given some elo difference. You might think that the -500 mark might be good to berserk but the sample size was low on this part. The safest berserk window is probably when you are +500 Elo from the opponent (This might not happen that often).

Future posts

I haven't done a great job explaining the steep downfall at +300 Elo range. I will be studying some other time controls and maybe other timeframes. If there is a new finding, I will write another post about it. If there is anything wrong with this post, feel free to correct me in the comments.