I thought it would be a good idea to start collecting information about the new Arena point system implemented in 3.0.8 and start developing some equations and calculations to determine how exactly the new system works and more particularly how the new Ghost Rating works.

As of right now, Arenas are disabled worldwide as a result of either a bug in the points awarded calculations or an unforeseen problem with implementing the system mid season.

To start off I figured I would open the thread with a series of questions to start some discussion.

General Questions
1. Can a ranking system work/judge skill if the winning team gets more/less points than the losing team loses?
2. What was the purpose of the change in the Arena System?

Ghost Rating Questions
1. Is Ghost Rating influenced by your gear?
2. Do your games in one arena type influence the Ghost Rating of your other arenas?
3. Is it possible to lower/reset your Ghost Rating?
4. How does you and your team's Ghost Rating effect the points awarded or lost for an Arena match?

Facts
1. Points awarded from a match are now based off of a third invisible rating (Ghost Rating) instead of Team Rating. (Source)
2. You have a separate Ghost Rating for each bracket. (Source)

Please try to avoid responding with purely anecdotal evidence. Try and include as much information as possible, such as what is your current/highest achieved arena rating is and what your current gear looks like.

Linking to appropriate Blue posts is highly supported and I will try to add them to the OP as appropriate.

Obviously it is not quite possible, yet, to determine exact specifics. Sticking to theory and speculation is the main purpose of this thread at the moment.

My current theory is that Ghost Rating for a player is based off of two things.
1. Your highest achieved Personal Rating in an arena type, (this will not go down ever, even after losing 1000 matches). Possibly it will reset at the end of a season.
2. Your average gear quality.

When calculating the gain in Team Rating after an arena match, it compares the average Ghost Rating of one team against the average Ghost Rating of the other team.

When calculating the gain in Personal Rating after an arena match, it compares your personal Ghost Rating to the average Ghost Rating of the opposing team.

Having a pairing system that pairs not only on rating but on "ghost rating" or more appropriately "expected rating" seems lika a good idea to me. Gladiators bashing noobs for 3 games instead of 20 when starting a new team benefits almost everyone. What this "expected rating" should be based upon can be debated. Seems to me like highest personal ratings and gear are a good way to get it started.

Two points are very important though:
1. Once a team has played enough games to determine their "real rating", ratings of other teams or their gear should in any way influence pairing.
2. This system should almost certainly not affect point gains and losses significantly. It makes no sense to punish a players points gain for having good gear or previous rating. If anything, the opposite should be the case, since good gear suggests the player has a potential to reach a high ranking. The only way points changes should be affected is if a steady 2000 team playes against a 10-0 1700 team. Since the 1700 team is paired on their expected 2000 rating, it would not be fair to take points from the 2000 team equal to what they would lose against a 1700 team. On the other hand, beating the 2000 team should award the 1700 team lots of points so that they can reach their real rating faster. In this case it is ok to turn away from a "zero sum" system. It can be discussed if this gain could even be allowed to exceed 32 points.

This type of system has worked good in warcraft 3 for quite some time. It's quite clear that the system is currently very buggy, and that it's not just an unforeseen results of a midseason patch.

Does anyone have a more in depth explanation of how the WC3 point system works? I quit before the release of the expansion so I am not too familiar with it. From what I have heard it has a similar system (in theory).

As far as I can recall (it's been over 4 years since I last played WC3 ladder):

A player would start at level 1. Winning a match gains you "experience", and eventually "level-ups", and you would be matched against people of the same level. This is generally just like arena rating, with a different expression.

1v1, 2v2, 3v3, 5v5, FFA and each individual Arranged Team grouping would have their own levels.

Just like in TBC, the early WC3 ladders were very prone to "smurfing": Creating a new account that would start at level 1 and playing a lot of games against level 1s, which would either be smurfs themselves or really easy to beat newbie players.

A late Frozen Patch changed the ladder system such that leveling up was vastly accelerated, allowing players to hit their plateau in a dozen games instead of 30.

So I did some arena today after servers came. My friend and I started a new 1500 rated 2v2 and played some games. Our first game was agaist a 2000 rated team, we won 29, they lost 10. We played them again, they won 5 and we lost 2. we only played 7 games, but we finished 3-4. Our team rating was 1560, but the thing thats really odd is that my partners personal rating is 1540 and mine is only 1514. So The system is either bugged or the rating system is done a personal level and not a team level.

Gear wise I had 545 resilience he had 450. I was full pvp disc, he was full pve frostfire. Not sure if these things affected it.

Pure speculation on my part, but this seems the place for it.

Given the arena rating calculations are based on zero sum maths, could the new ghost rating be a characters 'true' rating for a season in a bracket?

That is to say, your ghost rating is calculated from all your games played in a particular season in a particular bracket, irrespective of how many times you join new teams, artificially inflate or deflate your rating, sell arena points, etc.

The personal ratings are based on your gear, I'm running mostly deadly and my teammate has more hateful and Naxx 10 gear mixed in, and my PR was about 20 higher than him after about 6 games. I assume the system thinks you are carrying(or being carried by) your partner based on simple iLvl of gear, and gives you the rating accordingly.

Yeah, I don't understand this system, but I'm really choked over it currently, my team was about 50 more PR than I do. We made this team about 3 weeks ago and only the three of us have played. The other member on the team have a 1800 2s team and i have 1788 2s team. Our 5s are both hovering around 1600ish. I think outgear my team mates by a smidge if I wore all my pvp gear but being a DK I can escape with pve gear in an arena. Both of their PR's are identical and I'm constantly trailing behind but only on wins. When we loose a game I loose equal points too. Its horribly frustrating as taking a loss at a 1850 team rating knocks them down to 1840ish but it knocks me down under 1800.

So here's some data that might help you guys figure things out...
Pre-3.0.8.. My arena teams were aweful... I think i had 3x teams go into the 1300s in 2v2.
After 3.0.8 i finally found a decent teammate and we started winning games (This is my teammates first time playing arena in S5. His max rating last season was 1710, mine was 1920 (in 3v3) and 1790ish in 2v2.
after 13 matches, we were 9 and 4. His personal rating was ~1580, but mine was 1450ish... team rating was 1503 I was losing a shit ton per loss, and wining just a little bit per win.
I decided to quit the team and rejoin to reset my personal rating at this time. We ended up 19 and 13, team ratin 1480.. his personal rating is 1634, mine is 1430ish...

Anyone have any theories as to why my rating is so low? even though i should potentially have the lower 'ghost rating' because of the terrible start i had?

Yeah, my guess is that your poor start seeded you as a "bad player", thus, the system assumes you are being carried, and tries to send you to that rating. I imagine it put your hidden rating at something like 1350, and while the hidden rating is probably going up, it still tries to accelerate your personal rating to what your hidden rating is.

Once I get around to doing some arenas this week, I'll see where it sends me. Should be interesting at least.

In theory, a system like this should be pretty good, both in matching people up and giving people the rating they should have. Now the question is how it actually is implemented. We'll probably need a few weeks to see the real effects.

The main novelty of the current arena system is the fact that it's fully (Bayesian) probabilistic.
For those not familiar with Bayesian statistics, the most important formula is the Bayes' Theorem, in WoW-Terms:

Where P(A|B) is a short form for P(A=a|B=b), and the vertical Bar | denoting conditional probability. .
In the example above, "Skill" is a synonyme for "hidden" rating, as the rating system is designed to pair players with equal skill, whereas "outcome" stands for the possible outcomes of a arena game.
The tricky part in Bayesian statistics is, that usually you are not that much interested in exact values, but you rather calculate with the whole distribution. So, you basically take into consideration all possible values a random variable might take. In the above example,

The random variable Outcome has two possible states, win and loss, with a fixed probability p. Outcome is said to have a Bernoulli distribution, or short Outcome ~ Be(p).

The random variable Skill is continuous, and probably be assumed to be normally distributed, since this allows for easier calculation. So, the skill measure of a new arena player might be normally distributed with mean 1500 and standard deviation 200 (arbitrary values), which can be abbreviated in the following notation: Skill ~ N(mean, standard deviation), so in our example Skill ~ N(1500,200).

With this, it is straight forward to use above formula:
P(Outcome): This denotes the probability of a given outcome, when you don't take into account the skill of the participating players. Since arena games should be fair, this should be P(Outcome=win) = P(Outcome=loss) = 0.5.
P(Skill): This is the probability a player has a given skill, before the outcome of the game is known. This is the rating of the player before the game, with some uncertainity about it.
P(Skill|Outcome): This is the skill you have, when the outcome of the game is taken into account. In other words, it is your new rating. So when you queue for the next game, this becomes the P(Skill) for your next game, which is then updated accordingly to the outcome of this game, and so on.
P(Outcome|Skill): This is the tricky part. This is the probability you win a game, given your and your opponents rating. Its calcualtion is the critical (and tricky) part, but with normally distributed variables its relatively easy. A numerical example: Say, Team A has a Skill s ~ N(1500,200), and Team B has a Skill t ~ N(1700,200). P(Team A wins|s,t) is the probability that a number drawn at random from s is > than a number drawn at random from t. This probability is about 15%. Since a "rare" event happended, the system will guess its skill estimates were wrong and thus closen the gap between the teams, Team A will gain rating, teamp B will lose some points.

How this works exactly, as well as the exact numbers, and how this hidden rating related to team/personal ratings, I don't know, and probably won't find out.
The whole talk about strange rating changes comes from the fact that the system above requires a given skill for every player to begin with. This initial value seemed to be calculated in a strange way, with gear, rating and other stuff well mixed with some bugs to get very strange initial ratings. And as they influence the perceptible rating gains, this resulted in strange changes. Even with the bugs removed, no one has a clue wher his hidden rating is, and thus changes might seem strange.

Sources: MMO-Champion BlueTracker - AJ Repost #2
http://research.microsoft.com/pubs/74419/TR-2006-80.pdf of a thread on forums.worldofwarcraft.com
and admittedly a lot of interpretation and guesswork.

The changes indeed do seem very strange. Today we scrambled together our 5 vs 5 team, to play some matches, and see where we could get.

Now it really feels as if we're hitting a brick wall. Our average gear level is a mix of Tier 7 for the dps, and +- 600 resil on our 2 healers. What striked me odd was that we faced a couple of teams in high ratings, expectedly 1800/1900, and got stomped to the ground.
I expect us to get defeated by high rated teams that have done quite some matches together when starting a fresh team. We're not in line and can use some coordination. However, it feels terrible to only be fighting high rated teams, not allowing us to get any competition going.

In our current setup and experience together as a team, I expect us to get to 1600 rating, and hover there for a while while getting to work together better. However with how it looks now, our ratings will go down for a while, and get us to a long hill to climb before we get to the actual rating where we could be playing. This will disallow us to get any items from our percieved rating, and points as if we have been afking 10 matches each week. In the end we will get more even matchups, but as it looks now, we wont be getting any fair match for a while.

I initially came here to post something else (IE: General crying about how stupid this system) , but as this is the only thread on the subject, figured I might as well help out

I only recorded the rating lost/other team gained today during a streak of 4 losses to the same team.

They were 2031 according to armory.
We were 1730 at the time.

-Character 1 on our team has been 80 for 1 week. This is his first ever arena team
-Character 2 has hit 1800 in 2v2, then recently transferred. This is his first team on his new server

As our PR was exactly equal the entire time, and we've played every 2v2 game together since my partner transferred, I believe Gear ilevel, previous experience (pre wotlk, and pre transfer) etc are completely irrelevent. My partner is a fully 25 man geared PvE MT with a full ilevel 213 set of fury gear and near full deadly/hateful pvp gear. I'm in nearly full ilevel 187 blues with a few ilevel 200 heroic items/hateful items/badge items.

Our games vs the previously mentioned team (all lost)
Lose 5, Gain 4
Lose 4, Gain 4
Lose 5, Gain 4 (Can anyone explain how exactly it reverts to this?)
Lose 5, Gain 4

Also to note: we were ONLY being matched with teams around or above 2000, I armoried all of them. With how new my character is, and the fact we were only around 1750 in Team rating & lower in Personal... I really don't think the system works with any sort of consistency or logic right now.

He was, I beleive around 1800, the highest iv ever seen was 1650ish. Our 3v3 team is the same -1811, and either are on a 5v5 team. Previous seasons we topped at duelist.

Pretty sure blizzard said that each bracket would have a unique rating also.

I also agree with above poster in terms of gear having nothing to do with it. I out geared my mage partner by about 100 or so resilience and he was gaining the PR faster. But I could see how his previous 2v2 team rating would affect his PR now. He had a higher rated 2v2 before joining my team, therefore he is "carrying" me. But I also feel thats unfair in the sense that if I wish to start a season pvping with a friend who might not be amazing, im hurting myself later down the road if I decide to find a new partner towards the end. But I think this will just take time to pan out to see how crazy the swings can be.


A little offtopic, but I also wish that blizzard would put something in so you can only face the same team a few times within the hour or two, it sucks facing the same team 5-6 times within 10 games. (and sucks even worse if they are a counter team or just better players)

I want to input some of my own experience of arena with this new system.

PREPATCH

2v2 = 2008

3v3 = 1930

5v5 = 1824

POST PATCH

2v2 = 2022

3v3 = 2001

5v5 = 1824


We DID NOT play any 5v5s this week.
I play 2v2 and 3v3 with the same partner, a holy paladin.


Holy paladin
2v2 PR = 2021
3v3 PR = 2030
5v5 PR = 1824

Warrior
No 2v2
3v3 PR = 1978
No 5v5

Death Knight (me)
2v2 PR = 2018
3v3 PR = 2000
5v5 PR = 1824


In terms of ilevel, the paladin has the greatest, me the second most, and the warrior the least.

In terms of actions done in arena, the warrior does less damage on average in most games.

IF relevant at all, the paladin is a gladiator, and the warrior does not have ANY pvp titles.

I'm not convinced gear/iLvl has anything to do with the ghost rating...nonetheless I'll supply my own data for the week, including PR and what not.

2's: N/A (made a team, didn't play)

3's: 1954 team rating, 107 games played total, 60 this week (under the new system).
-1998, 1972, 1947 personal ratings for paladin(myself)/death knight/mage respectively
-None of us have the "Epic" achievement; I have an ilvl 226 weapon, the mage and dk have 213 weapons.
-I have a duelist title, the dk is a gladiator reroll, the mage is also a duelist
-For most games played, the dk does the most damage and the mage does the least; I have negligible amounts of healing done (~40k for every 200k damage dealt).

5's: 1860 team rating, 75 games played total, 40 this week
-1904, 1887, 1882, 1867, 1862 PR for paladin/death knight/holy paladin/mage/shaman respectively
-death knight and mage are the same ones from my 3's
-holy paladin has both an iLvl 226 weapon and shield, as well as full 7.5
-shaman has an iLvl 226 weapon and over 1000 resil (double set, deadly and hateful)
-holy paladin was on a gladiator team in s1 but did not receive a title due to not enough games played; shaman is a gladiator reroll
-Similar to 3's, the dk is almost always first in damage dealt, and I do even less healing in 5's than in 3's. The holy paladin does the most healing in 5's.


From that data... item level doesn't seem to have an adverse effect on the ghost rating, as the shaman is actually more geared than our mage. Past titles and such can't have an effect on this new system, because otherwise how would deathknights get anywhere? Damage and healing done during a match may have effects on your PR; perhaps the system calculates your DPS/HPS based on match length and damage dealt?

What is most frustrating for me is the sheer amount of games we are playing and literally not getting anywhere; I've played about 100 games under this new system and in both 3's and 5's and the most points we have won is 12, the most we have lost is 10. My 3's for example was 1933 before the new system, and we all had equal PR... it seems outright stupid that we played 60 games for a mere 20 team rating, while watching our personal ratings fluctuate wildly (+/- 60 of the team rating). It's going to make gear purchases very difficult since the items require both team AND personal rating, they should probably change the items to only require a personal rating...

Some information from the lower ratings.

2v2: Team rating is like 1581, we haven't played any games in two weeks though so I can't give you any information there. Our highest rating from a few weeks ago was around 1605.

3v3: Started out at a team rating of 1492 I believe. We've played about 15 games so far. Only been matched against teams with 1700-1800 rating this week. A win was netting us between 15-20 points, while they lost <10 points (usually 4-7 points, lost for them). When we lost we only lost around 4-7 points ourselves and the other team was never granted more then 10 points for beating us. Our team rating is now 1601, and my PR is 1555. Out highest team rating this week was like 1610, and highest PR was around 1605. We basically got to 1600 and tanked in PR.

We all kind of have shitty PvP gear. I'm running with 2 savage, 2-3 hateful pieces and the rest is raiding gear from Naxx. The ret paladin is running purely all raiding gear with a few pvp pieces (IE: trinket). The mage is running a similar set up to me, few pvp pieces but mostly all PvE gear. There is definitely a lot of ilevel 213 items being used, but almost none of them are PVP pieces - I think maybe one deadly piece on all three of us.

I didn't really keep track of damage dealt/healing dealt so it'd all be pure speculation anyway. Hope this is enough info.

We started at 1913 this week, went 11-12. Ended up at 1903. But my PR, which was at around 1920, ended up in the 1800s, while my partners (Holy Paladin, no other spec) PR ends up above the TR when he started out lower.

We were getting matched up with teams in the 2000s and sometimes 1900s and two or three in the 1800s. We lost one to a team at 1813 and we lost 9 pts, while they gained 8. But when we beat a team in the 2100, they lost 5 and we gained 6.

We're not especially well geared either. I'm still using my Guardian's Neck piece, and my best item is probably one of my daggers from Naxx (normal).

It seems pretty clear that Personal Ratings aren't functioning correctly.

Here is the data from games that my 2v2 played last night:

Total games played: 18.

Beginning Team Rating: 1725
Ending Team Rating: 1723
Team Rating Change: -2

Beginning Personal Rating: 1599
Ending Personal Rating: 1606
Personal Rating Change: +7

That's a 117 point difference between my team rating and my personal rating. Yes, it has slowly gone up from 1500 to this point, but if it continues at this rate it will take ~300 games before my personal rating and my team rating are the same. That's 30+ hours of arena time.

That can't be intended.

(On another note, we were basically being matched with teams that seemed appropriate for our skill level. I looked up the team rating and gear level of the teams we faced, and the vast majority basically seemed appropriate, with just a couple outliers >150 rating difference.)

Just found this image posted on my guild's forums: http://www.imagedumpwitheasy.com/uploads/7c0eb1c4b6.png

If that's correct, stuff's even more wrong than I thought.

What's really weird about that one - other than the fact that they won negative points - is that the other team lost a high value. You'd think if they were so underrated that the other team actually lost a point for being teamed against them, that they would lose a very few points.

On 2's on Saturday, we split two games with a team

We Lost: 16, They Won: 4
We Won: 4, They Lost: 16

It looks like for whatever reason, we were the heavily favored team on the first match, and then they were the heavily favored team. I admit zero sums seemed to make more sense to me.