8 replies to this topic

[Mawai]

Hi All,

There seems to be a lot of confusion about PSR. It is called Player Skill Rating ... the problem is that it is not a rating at all and some of the complaints about PSR are based on people expecting it to be a rating.

PSR rises and falls based on a fixed value of performance based on your match score. The rest of the team performance (other than win/loss) doesn't matter to the change in YOUR number. Only a high match score, which is primarily based on damage done, and whether you win or lose, matters.

In addition, the system is not conservative and is biased towards a constant increase in the PSR even with a 1:1 win/loss record assuming average player performance. This means that the PSR is NOT a rating. It can not be used to compare the relative skill of two players since the number is dependent on the number of games played. Assuming you play decently then two players both with a 1:1 W/L ratio, one with 100 matches and one with 1000 matches ... the one with 1000 matches will have a much higher PSR even if they have identical skills.

In addition, since match score derives a large fraction of its number from the damage done ... and high match scores are in general more easily attained in specific mechs that are "better" (Stormcrow/Timberwolf/Dire Wolf/Hellbringer/Arctic Cheetah ... maybe Thunderbolt and Stalker-4N in certain situations) and by the heavier weight classes (heavy and assault) since they tend to have more weapons and survive longer. These mechs will get higher match scores on average and will proportionally affect the "PSR" number of the player using them. Thus "Player Skill Rating" again does not reflect actual player skill but is also biased by the selection of mech that the player commonly plays. There may be compensating factors used so that match scores in lights receive a boost or the match score is weighted differently in calculating PSR for each weight class but that is not known.

So, what is PSR actually doing? Over time and on average it is dividing the player base into players that are active and play many games and those that play less. Players who have played 1000 games in the last 3 months are probably on average better that a new player or someone who has only played 50 games. Thus, some of the more active players may well see better games out of it because they are playing with folks who have played MORE and and are more experienced on average.

Over time, more and more players will reach whatever the maximum value of PSR is and the match quality may well decline at that point as many people end up in the same bin. The ones to reach the glass ceiling first are those who get high match scores and win more often than lose.

Anyway, PGI could likely have come up with a much simpler way to rank folks if they had just used the number of games played in the last 6 months and dumped the entire PSR system.

Someone suggested plotting PSR against Elo to see if there is any correlation between an actual conservative ranking system based on winning your matches vs. PSR which is based primarily on damage and increases indefinitely over time unless limited by a cap. I'd also like to see a plot of PSR against games played since Jan 1, 2015 which was the date used for the original PSR seeding. Both of these would be useful to get some idea of whether PSR is doing something like it was intended to do or whether it is hopelessly broken.

TL;DR PSR is NOT a ranking system. Comparing PSR does not give a strong indication of which player actually plays the game better.

[Monkey Lover]

I only read posts from tier 1 players

[LordNothing]

its neither a rating nor is it personal. im not even entirely sure it has anything to do with skill either.

why does everyone in the community, even the devs, have to misname everything. stop it, all of you.

[Idealsuspect]

Isn't a skill rating of course we all know this ( pgi not yet ) but with this PSR you get a tier rank and also MM put you in specific queue based on same "skills rating" ( which aren't because PSR is just a win/lose rating and have nothing to do with your personnal skill.

[Alistair Winter]

Not sure if argument is valid...

[Ex Atlas Overlord]

Mawai, on 02 October 2015 - 07:05 PM, said:

It can be used to compare the relative skill of two players since the number is dependent on the number of games won.

Fixed that for you.

[Chuck Jager]

Basically after 10,000 games the worst players in the worst mech in the worst builds will have a lower score than the best players in the best builds.

Every one will fall in between. So even if every single player makes it to T1. That Tier should still have enough separation in values to divide up the team.

Computers do not need 5 simple tiers. Simple players do.

[MoonUnitBeta]

Again, this is why we need to bring relevancy into match score.
Match score must be relevant to the match that it came from if we hope to have any kind of accuracy, and results within a realistic time frame

Compare your performance with the rest of your team, and compare your team with the other team.
http://mwomercs.com/...heck-something/

It will factor in rolfstomps, by taking into account that you had less of a chance to meet the Pauls requirements about what makes a good player.

Right now we have dozens of assumptions being made with PSR that hurts accuracy and makes it difficult to make any large adjustments.

Assumptions such as:
- All 24 people live the same amount of time, and die at the same time. Each with their equal chances and opportunities.
- The match is perfectly balanced.
- Your team always plays as a team, and you are the odd one out who is being rated.
- The match outcome is irrelevant (such as 12-0, or 12-11)
- Match quality (a combination of match outcome, team composition, and pairing accuracy) is disregarded (read: handicaps if one side has 6 tier 1s, and 6 tier 3's, then a team of 12 tier 2s).
- etc.
These can all have an effect on how drastic PSR adjusts your placement. Instead, skill is dictated to us by a fixed value, and failing to meet those credentials under any circumstance is no fault but your own.

And there's also flaws with match score. There is the lack of acknowledgement for how well players work together, how fast people shoot down UAV's. It cannot track how effective they are at communication, and so on.

I find it ironic that you acquire a team based Match Score, yet the score itself is disconnected from the entire match.

Edited by MoonUnitBeta, 02 October 2015 - 09:58 PM.

[El Bandito]

Right now these are the players I have to face-off after reaching T1 yesterday. Their skill is real deal. Most of them anyway.


.

Edited by El Bandito, 02 October 2015 - 09:49 PM.