Written byon Aug. 17, 2018 .
Released in 2017, HLTV’s Rating 2.0 is the second iteration of the only advanced metric designed to sum up player performance into a single value. However, in a previous blog post, Chris demonstrated its shortcomings and the necessity for a new player rating system.
While Rating 2.0 does incorporate many other ways a player can contribute to a game besides kills and deaths, we feel like there are more important aspects of a game that currently are not accounted for; aspects that many in the community, including Astralis’ in-game leader, have been asking for.
Stats/Rating which is based on how much money the opponent spent LETS GO 😁.— Lukas Rossander (@gla1ve_csgo) July 30, 2018
It is a little crazy that it is almost impossible to have a rating below 1 if you make an ace against glocks/usp. But I believe you guys have been trying to figure something out about that already right
So today, we at SixteenZero are introducing Player Round Impact (PRI), an early-stage metric that measures player performance by incorporating the context of each player's kills and deaths.
Consider the following rounds with virtually identical stats from ESL One New York 2017.
Rating 2.0 and KDR would give similarly high scores to all three of these rounds. Watching the clips, though, it is clear the context of each round is very different. But how do we quantify that context?
The first step is to develop a win probability model: a tool utilized by analysts in other competitive fields to track how likely each side is to win a game/match, given a game state.1 Our initial model predicts which side is likely to win a round given the map and the weaponry brought into the round, then adjust its estimate each time a player is killed.
For example, entering a gun round on Cobblestone, the CT side is 53% likely to win the round. If the T side gets the opening kill, the CT side is only 31% likely to win (a -22% change), however, if the CT side should get the opening kill, they would be 69% likely to win (a +16% change).
So, how do we use this model to get PRI? By assigning the attacker and the victim the change in win probability of each kill.
Now, let’s revisit the aforementioned ace rounds through the lens of our win probability model.
|Team Liquid vs. Astralis, Mirage, Anti-eco Round||Event||Attacker||Victim||Liquid Win Probability||Win Probability Change|
|5 v 4||EliGE||Kjaerbye||94%||5%|
|5 v 3||EliGE||device||99%||4%|
|5 v 2||EliGE||dupreeh||>99%||1%|
|5 v 1||EliGE||Xyp9x||>99%||0%|
EliGE PRI Calculation: 5% + 4% + 1% + 0% + 0% = 11% or 11 PRI
Based on the weaponry each team brings into the round, Team Liquid has a significant advantage, and with each kill, Astralis' chances only get worse. As a result, each of EliGE's kills become increasingly less valuable, to the point that each of the last two are virtually unimpactful. Let's see how EliGE's PRI of 11 compares with the other rounds.
|Natus Vincere vs. EnVyUs, Cobblestone, Pistol Round||Event||Attacker||Victim||Na`Vi Win Probability||Win Probability Change|
|4 v 5||ScreaM||Edward||34%||-16%|
|3 v 5||SIXER||seized||15%||-19%|
|2 v 5||RpK||flamie||4%||-11%|
|2 v 4||s1mple||RpK||8%||4%|
|2 v 3||s1mple||xms||24%||16%|
|2 v 2||s1mple||SIXER||50%||26%|
|2 v 1||s1mple||ScreaM||87%||37%|
s1mple PRI Calculation: 4% + 16% + 26% + 37% + 13% = 96% or 96 PRI
Given that it is a pistol round, neither side have an initial advantage. Meaning, not only does Na`Vi enter the round at worse odds than Liquid did in EliGE's ace round, but their odds continue to decrease until s1mple and Zeus are left in a 2-vs-5 situation. s1mple brings Na`Vi back from a 4% win probability by eliminating all of the EnVyUs players and netting himself 96 PRI for the round. That is almost nine times the PRI value of EliGE’s ace. Now, where does nitr0's ace fit in?
|Team Liquid vs. Virtus.pro, Nuke, Full-buy Round||Event||Attacker||Victim||Liquid Win Probability||Win Probability Change|
|5 v 4||nitr0||NEO||72%||15%|
|5 v 3||nitr0||byali||86%||14%|
|5 v 2||nitr0||pashaBiceps||96%||10%|
|5 v 1||nitr0||TaZ||>99%||4%|
nitr0 PRI Calculation: 15% + 14% + 10% + 4% + 0% = 43% or 43 PRI
Even though both teams are on a full buy, Team Liquid have a slight advantage entering the round because they are on the CT side of the notoriously CT-sided Nuke. Since nitr0 never loses a teammate, there is not as much win probability to make up for compared to s1mple. Therefore, nitr0's PRI of 43 places his round somewhere between EliGE's and s1mple's.
Let's also look at rounds with more commonly seen kill and death totals. Focusing on tarik, EliGE, and olofmeister, respectively, each player in the following examples finish their round with two kills and one death.
|Cloud9 vs. FaZe Clan, Mirage, Full-buy Round2||Event||Attacker||Victim||Cloud9 Win Probability||Win Probability Change|
|5 v 4||tarik||karrigan||68%||22%|
|4 v 4||olofmeister||Skadoodle||51%||-17%|
|3 v 4||GuardiaN||Stewie2K||28%||-23%|
|3 v 3||tarik||NiKo||53%||25%|
|3 v 2||RUSH||GuardiaN||81%||28%|
|2 v 2||rain||tarik||54%||-27%|
|1 v 2||olofmeister||RUSH||17%||-37%|
|1 v 1||>||autimatic||olofmeister||55%||38%|
tarik PRI Calculation: 22% + 25% - 27% = 19% or 19 PRI
tarik opens up this back-and-forth round, later gets another kill to bring it to a 3-vs-3 situation, and ultimately dies to bring it back even at 2-vs-2. This is a great example of a round where every kill and death a player took part in had a significant impact. If every round were like that, there would not be significant differences between PRI, Rating 2.0, and KDR. But that is not usually the case, as we can see in the next two examples.
|Team Liquid vs. FaZe Clan, Inferno, Full-buy Round3||Event||Attacker||Victim||Liquid Win Probability||Win Probability Change|
|5 v 4||EliGE||olofmeister||72%||23%|
|4 v 4||GuardiaN||stanislaw||56%||-16%|
|3 v 4||rain||nitr0||34%||-22%|
|2 v 4||karrigan||Twistzz||13%||-21%|
|1 v 4||karrigan||jdm64||1%||-12%|
|1 v 3||EliGE||rain||6%||5%|
EliGE PRI Calculation: 23% + 5% - 6% = 22% or 22 PRI
EliGE opens the round with an impactful kill but eventually finds himself in a 1-vs-3 situation. He manages one more kill but ultimately dies, losing the round and cancelling out the impact of his second kill.
|FaZe Clan vs. Team Liquid, Inferno, Eco Round4||Event||Attacker||Victim||FaZe Win Probability||Win Probability Change|
|4 v 5||stanislaw||NiKo||8%||-7%|
|3 v 5||Twistzz||karrigan||3%||-5%|
|2 v 5||Twistzz||rain||<1%||-3%|
|1 v 5||Twistzz||GuardiaN||<1%||-<1%|
|1 v 4||olofmeister||nitr0||<1%||<1%|
|1 v 3||olofmeister||EliGE||1%||1%|
olofmeister PRI Calculation: 0% + 1% - 1% = 0% or 0 PRI
olofmeister is not involved any kills before he finds himself in a nearly impossible 1-vs-5 situation, so the only way for him to have any impact on the outcome of the round is to pull off the ace clutch. Therefore, since he does not, his two kills and one death are worth virtually 0 PRI.
PRI for a map, match, or an entire event is calculated by taking the average5 of the player’s impact on their team’s win probability over that span. Below is a table listing EliGE's stats from every round of the map on which he got the aforementioned ace against Astralis.
|Team Liquid vs. Astralis, Mirage, EliGE PRI||Round||Round Type||Liquid Side||Kills||Deaths||PRI|
The following table of aggregated player statistics showcases the differences between PRI and the other metrics. Looking at Twistzz’s numbers, we can see a case where Rating 2.0 is able to differentiate from KDR. However, both metrics still struggle to properly weight EliGE's ace. EliGE goes from being the top performer in Rating 2.0 to the fifth-best in PRI.
|Team Liquid vs. Astralis, Mirage||Player||Kills||Deaths||KDR||Rating 2.0||PRI|
PRI can also identify over and underrated performances at a tournament level. Future blog posts will likely dive into why some of these PRI values vary so greatly from the other metrics, but for now we'll just post the overall numbers for ESL One New York 2017, sorting by Rating 2.0:
|ESL One New York 20176||Player||Team||Rounds||Kills||Deaths||KDR||Rating 2.0||PRI|
Our initial win probability model is rudimentary and does not yet incorporate basic, important factors such as time remaining in the round, bomb factors, damage, and changes in equipped weaponry throughout a round. It also does not account for more complex factors like map control or the skill level of players and teams involved. All of these factors are on our radar and development to incorporate some of them is already underway.
However, even with these known gaps, we are still excited to release this initial version of PRI because it offers the community a unique perspective on player performance. Feel free to share any questions, comments, or suggestions you have with us in our reddit post.