One of the biggest problems with professional Rocket League is the lack of a clear way to determine who the best players in the game are. For some, just looking at the scoreboards is enough to get a good idea of player performance. Casters and analysts will use GPG (Goals per game), shooting percentage, or other per-game measures. But more casual fans may not have the time to research what statistics are most important when determining player performance.
This is where PER, or Player Efficiency Rating comes in. PER was created by John Hollinger in the 1990’s as a way to directly compare players in the NBA. PER is a single number that accounts for all major statistics and gives a good understanding of a player’s performance in a season, or across a span of games. Below is the formula I have created to calculate uPER (Unadjusted PER) and PER. I will go into detail regarding my methodology for each section and finish with a short note on issues with PER.
In the NBA PER calculations, uPER is multiplied by [aPER * (15/lg_aPER)] in order to get the league average in every season to 15 and this final number is the PER rating. Because aPER involves the calculation of the pace of the team in question, it is impossible to calculate for Rocket League (Pace involves possessions. Read more here). Instead, after calculating the uPER using the formula above, I multiplied by the RegionalFactor. The RegionalFactor is just a number used to get the average PER to 15.
In RLCS League Play, teams from different regions do not play each other, so I used a different RegionalFactor for each region, for each season. This same principle will be used moving forward. Also, a different RegionalFactor will have to be used to set the average PER to 15 at the World Championship in November. Eventually I plan to collect enough data about all players in all regions that I can create a substitute for RegionalFactor that can be used. But because there is a limited amount of events where international teams mix, it is hard to get a good amount of data.
Weights
Weights were chosen on a 0–1 scale and I tried to use data to back up my reasoning. But again, this can be a bit suggestive depending on who you ask.
Goals — Goals were given a weight of 1 because they are the single most important thing in a game. The goal of Rocket League is to score more goals than the opponent.
Assists — Assists were given a weight of 0.75 because, again, scoring goals is the most important thing in the game and an assists means you are helping those around you score goals. Assists are tracked poorly in the game, which reduces the importance of them.
Saves — Saves were given a weight of 0.6 because, although there is a weak negative correlation between saves and winning, saving potential goals keeps your team in the game. Saves also prevent the other team from doing the most important thing in the game: scoring goals. Please note, that while saves are a defensive statistic, they are a poor indicator of defensive success. I go into this a bit further at the end of this document.
Shots — Shots were given a weight of 0.4. This is because shots are not tracked well in the game and often balls that are counted as “shots” are not anywhere close to going into net, even if no defenders touch it. But I found a strong correlation between shots per game and winning, which kept shots from having a lower weight.
Shooting Percentage — Shooting percentage was given a weight of 0.5 because it shows the scoring efficiency of a player. Does the player just shove the ball towards the net or can they actually put their shots past the defense and into net? But shots are tracked inaccurately, which keeps the weight of this down a bit.
Adjusted Score — Adjusted Score was given a weight of 0.25 because although score is often thought to have minimal importance in a match, it does show some qualitative properties of a player, like touches on the ball or clears.
Issues with PER
One number is good for making comparisons easier, but it is still subjective and does not tell the whole story of a player’s efficiency or performance. This is a a big issue with the NBA PER system. I could link you to hundreds of thousands of articles that discuss everything wrong about PER and similar efficiency rating systems (here is one) and I agree with them on a lot of points. But at this stage in Rocket League, from an esports perspective, there just is not enough deep analysis done on players and teams. Maybe this is one step in that direction and can spark more analytical discussion.
Another issue is that this value does not account for defensive statistics. This is a major issue with the NBA’s calculation of PER as well, but their issue stems from the creator of the formula not accounting for certain things. In Rocket League there just simply are no good in-game statistics that show how good a player is on defense. Saves are not a good indicator because good defense starts at the midfield, not at the goal-line. Let’s suppose that Player A is amazing at defense because he never lets the opponents get shots on net. Where is that measured? Nowhere. You do not get points for good challenges in the mid-field. Defense is fairly qualitative, not quantitative, in Rocket League.
The final issue I ran into when working on this was that it breaks down with long overtimes. Because of the “golden-goal” nature of OT in Rocket League, combined with PER being a per-minute statistic, long overtimes tend to decrease all player PER’s for that match. This should even out over the course of a season, but when looking at series PER or individual matches played, it can distort the actual performance of a player in relation to other matches/players. NBA’s PER system does not suffer from this, as OT in the NBA is just an extra period of playtime — not first to score wins.
I understand there may be flaws in the system I have created, but this system is open for changes going forward and I will constantly work to make improvements to weights or calculations based on legitimate feedback.
