21 replies to this topic

[Tranzience]

I have browsed the AJ forums for at least a good 5 minutes looking for an answer and even tried a few Google searches. I gave up quick so I apologize in advance if someone has asked this already and I just failed to find it. Anyways, I watched the movie Moneyball again recently and it made me think about the statistical aspect of arena. If you haven’t seen it, some nerd uses a bunch of math to create a baseball team with a small amount of money that goes 20-0. At the end of the movie, the Red Sox guy tells Brad Pitt that his team won more games than the Yankees, but paid only ~270k for each win when they paid 1.2 million for each win. Instead of relying on traditional intuition from the scouts, they use statistics to select their players.
Before I start, I haven’t played retail since s9. I am currently active on Arena-Tournament, and I rarely come to AJ.

I do not think I have ever really seen an in-depth analysis on an arena ladder. Of course you may see “warriors are in 30% of top teams” Therefore warriors are op plz nerf. The question really is this: Is it possible to utilize statistics to achieve a higher team rating? People have for a long time had theories on how the rating system works and how it should be changed, but I don’t think anyone fully understands it. Another question that should be answered is: what is the algorithm behind rating calculation? Everyone knows roughly what to expect as the outcome of each arena. A 2000 team beats another 2000 team, so they gain 15 points and lose 15 points respectively. The team is now 2015 and beats the team that is 1985, they gain 14 points and lose 14 points respectively. Something like this happens, and there is also MMR to factor in, but is it possible to find out the precise equation that gives us rating? We have thousands of different outputs but no one seems to know the input. Knowing this information would be crucial for a number of reasons.

To illustrate why someone should consider this approach, just look at season 9 for example. Blizzard made a change to rating so your team rating and personal match making rating would go up way faster. The use of exploiting the mmr system did not run rampant until they made the change so you saw your personal mmr rather than the team mmr. If you don’t remember how the exploit worked, 2 people tank mmr to 0. One person has mmr of 1500, therefore the team mmr is 500. The team mmr is what determines the points you gain, so when this team got to 1500 the one guy now has 2500 mmr. If the team got to 2k, then the one person has 3k mmr and so on. Using this you could get a team to 4k+ mmr really easily and that’s how tons of people cheated in s9. Also would like to point out a team in full blues in Europe got to 3800 or some shit. The interesting thing is, you could have used this exploit as soon as they introduced match making rating. Most people just didn’t realize it until blizzard unknowingly shoved it into everyone’s face. To prove that, go try it on a private server that runs season 8 like arena-tournament or molten pvp. It would take a long time but it is possible. You can probably still find buried in the forums here, how Erase (I think was his name) has jesus mmr season 8 and used it to boost a ton of people to rank one.

The “Moneyball” theory may be harder on retail, but on arena-tournament you have some more data available. The armory on retail for some reason still does not show match history, and I’m sure you still don’t see personal mmr. Anyways, it would be interesting to see a discussion and actual thought put into this. I suck at math so I can’t do very much analysis on my own. Just to list a few questions that may be worth investigating:

What is the equation that gives arena rating?
If you play druid/lock for example, what is the likelihood you will beat certain comps? Second part of this question is at what point do you lose to that same comp enough that it isn’t worth risking the team rating?
How accurate is mmr?
What would you need to change in the way you queue arena/dodge teams to win more?
Are there inherent flaws with the rating system?
What is the best comp?
Which players matched together would win the most?
What is statistically the best way to get highest potential team rating?

[Gambino6]

do you got a tldr version i can't comprehend this much this late

[originn]

Tranzience, on 08 December 2012 - 05:20 AM, said:

I have browsed the AJ forums for at least a good 5 minutes looking for an answer and even tried a few Google searches. I gave up quick so I apologize in advance if someone has asked this already and I just failed to find it. Anyways, I watched the movie Moneyball again recently and it made me think about the statistical aspect of arena. If you haven’t seen it, some nerd uses a bunch of math to create a baseball team with a small amount of money that goes 20-0. At the end of the movie, the Red Sox guy tells Brad Pitt that his team won more games than the Yankees, but paid only ~270k for each win when they paid 1.2 million for each win. Instead of relying on traditional intuition from the scouts, they use statistics to select their players.
Before I start, I haven’t played retail since s9. I am currently active on Arena-Tournament, and I rarely come to AJ.

I do not think I have ever really seen an in-depth analysis on an arena ladder. Of course you may see “warriors are in 30% of top teams” Therefore warriors are op plz nerf. The question really is this: Is it possible to utilize statistics to achieve a higher team rating? People have for a long time had theories on how the rating system works and how it should be changed, but I don’t think anyone fully understands it. Another question that should be answered is: what is the algorithm behind rating calculation? Everyone knows roughly what to expect as the outcome of each arena. A 2000 team beats another 2000 team, so they gain 15 points and lose 15 points respectively. The team is now 2015 and beats the team that is 1985, they gain 14 points and lose 14 points respectively. Something like this happens, and there is also MMR to factor in, but is it possible to find out the precise equation that gives us rating? We have thousands of different outputs but no one seems to know the input. Knowing this information would be crucial for a number of reasons.

To illustrate why someone should consider this approach, just look at season 9 for example. Blizzard made a change to rating so your team rating and personal match making rating would go up way faster. The use of exploiting the mmr system did not run rampant until they made the change so you saw your personal mmr rather than the team mmr. If you don’t remember how the exploit worked, 2 people tank mmr to 0. One person has mmr of 1500, therefore the team mmr is 500. The team mmr is what determines the points you gain, so when this team got to 1500 the one guy now has 2500 mmr. If the team got to 2k, then the one person has 3k mmr and so on. Using this you could get a team to 4k+ mmr really easily and that’s how tons of people cheated in s9. Also would like to point out a team in full blues in Europe got to 3800 or some shit. The interesting thing is, you could have used this exploit as soon as they introduced match making rating. Most people just didn’t realize it until blizzard unknowingly shoved it into everyone’s face. To prove that, go try it on a private server that runs season 8 like arena-tournament or molten pvp. It would take a long time but it is possible. You can probably still find buried in the forums here, how Erase (I think was his name) has jesus mmr season 8 and used it to boost a ton of people to rank one.

The “Moneyball” theory may be harder on retail, but on arena-tournament you have some more data available. The armory on retail for some reason still does not show match history, and I’m sure you still don’t see personal mmr. Anyways, it would be interesting to see a discussion and actual thought put into this. I suck at math so I can’t do very much analysis on my own. Just to list a few questions that may be worth investigating:

What is the equation that gives arena rating?
If you play druid/lock for example, what is the likelihood you will beat certain comps? Second part of this question is at what point do you lose to that same comp enough that it isn’t worth risking the team rating?
How accurate is mmr?
What would you need to change in the way you queue arena/dodge teams to win more?
Are there inherent flaws with the rating system?
What is the best comp?
Which players matched together would win the most?
What is statistically the best way to get highest potential team rating?

I dont know about all the rest but im gonna go watch moneyball now thanks

[Tranzience]

Gambino6, on 08 December 2012 - 05:25 AM, said:

do you got a tldr version i can't comprehend this much this late

Sure, just read the questions at the bottom.

What is the equation that gives arena rating?
If you play druid/lock for example, what is the likelihood you will beat certain comps? Second part of this question is at what point do you lose to that same comp enough that it isn’t worth risking the team rating?
How accurate is mmr?
What would you need to change in the way you queue arena/dodge teams to win more?
Are there inherent flaws with the rating system?
What is the best comp?
Which players matched together would win the most?
What is statistically the best way to get highest potential team rating?

[Tranzience]

originn, on 08 December 2012 - 05:28 AM, said:

I dont know about all the rest but im gonna go watch moneyball now thanks

Really good movie lol.

[Mity]

gonna remove this since a lot of people haven't seen it apparently

Edited by Mity, 08 December 2012 - 06:06 AM.

[Jacquelol]

G_G_G__G_G_G_G_G_G_G_G_G gnna google moneyball now.

[Breadstick]

reminder that the athletics lost in the playoffs and the movie failed to mention anything about eric chavez, miguel tejada, or their amazing pitching staff. what billy beane and his assistant did was smart, but they only filled in small roles using unknown players when the movie made it look like they completely rebuilt a team which is simply not true

Quote

What is the equation that gives arena rating?
If you play druid/lock for example, what is the likelihood you will beat certain comps? Second part of this question is at what point do you lose to that same comp enough that it isn’t worth risking the team rating?
How accurate is mmr?
What would you need to change in the way you queue arena/dodge teams to win more?
Are there inherent flaws with the rating system?
What is the best comp?
Which players matched together would win the most?
What is statistically the best way to get highest potential team rating?

i don't really see how these are relevant to a moneyball comparison, honestly

[Mity]

Breadstick, on 08 December 2012 - 06:32 AM, said:

reminder that the athletics lost in the playoffs and the movie failed to mention anything about eric chavez, miguel tejada, or their amazing pitching staff. what billy beane and his assistant did was smart, but they only filled in small roles using unknown players when the movie made it look like they completely rebuilt a team which is simply not true



i don't really see how these are relevant to a moneyball comparison, honestly

go to hell yankee fan

[Breadstick]

Mity, on 08 December 2012 - 06:39 AM, said:

go to hell yankee fan

[Mity]

Breadstick, on 08 December 2012 - 06:41 AM, said:

sry i'm still very salty when i saw raul ibombez decide to take the game over vs the o's

[Tranzience]

Breadstick, on 08 December 2012 - 06:32 AM, said:

i don't really see how these are relevant to a moneyball comparison, honestly

I just want to have a discussion on how statistics could be used to improve one's arena rarting. I think it something that has never been looked at enough by the community.

[Breadstick]

Tranzience, on 08 December 2012 - 06:44 AM, said:

I just want to have a discussion on how statistics could be used to improve one's arena rarting. I think it something that has never been looked at enough by the community.

well, people are using statistics when choosing which comps to play, like warrior/hunter/paladin. people wanna play the best comps to get the highest ratings so they look at representation stats and see what high rated players are playing.

it's sort of different in baseball and wow, because if the athletics had the ability to get the same players as rich teams, they would have. instead they had to settle for an alternate way of playing the game. in wow, teammates don't cost anything, so there's no real need to play an innovative style

[Tranzience]

I know baseball is completely different. I'm not necessarly looking at what happens inside an arena but what happens before an arena. Why queue into a team if you know you're going to lose? How about this basic probability example. My win rate vs pally/warrior is 75%. I know that if I queue I will most likely face pally/warrior. Now, 75% is considered by most a pretty decent win rate, but what if I'm going for rank one or something where that 25% could really hurt you. I know I would lose 20 points per lose, but gain only 4 upon wins. Play 4 games and the probability says you win 3 and lose 1. Therefore net loss of 8 points. Why queue if only to lose 8 points?

I did really bad in statistics in high school, but there are things called standard deviation, margin of error, and some other stuff I failed at. Someone better than me could possibly expand upon that example to acheive a rating that is massively affected by how good you are at queuing.

[ceiyoan]

Tranzience, on 08 December 2012 - 07:27 AM, said:

I know baseball is completely different. I'm not necessarly looking at what happens inside an arena but what happens before an arena. Why queue into a team if you know you're going to lose? How about this basic probability example. My win rate vs pally/warrior is 75%. I know that if I queue I will most likely face pally/warrior. Now, 75% is considered by most a pretty decent win rate, but what if I'm going for rank one or something where that 25% could really hurt you. I know I would lose 20 points per lose, but gain only 4 upon wins. Play 4 games and the probability says you win 3 and lose 1. Therefore net loss of 8 points. Why queue if only to lose 8 points?

I did really bad in statistics in high school, but there are things called standard deviation, margin of error, and some other stuff I failed at. Someone better than me could possibly expand upon that example to acheive a rating that is massively affected by how good you are at queuing.

What you're saying in theory would work, that selective queuing would benefit teams and allow them to potentially bypass unwanted teams and face teams that they had a higher probability to win against.

However it's nearly impossible to accurately predict what is currently queuing, not to mention blizzard considers "selective queuing" to be an exploit. As a third point if everyone selectively queued, or even a large chunk, it would probably slow the ladders down as not as many teams would be consistently queuing on a daily"ish" basis.

[Tranzience]

On retail I am sure this is true, but on arena-tournament it is very easy. Also there is the question of what exactly gives us arena rating. There's probably one asian at blizzard who made the whole thing. The community could probably work backwards and find out what it is.

Edited by Tranzience, 08 December 2012 - 05:24 PM.

[Snuggli]

I think it doesn't really transfer to WoW because in Arena, there is clear counters, which, even if you are the same skill level you are likely to lose.

Obviously you could create a massive list and gather huge amounts of data and eventually come to exactly the same conclusion. If both teams play flawlessly (at a R1 level), the team who is at the comp disadvantage should lose. The only thing worthwhile you could get from all of it is a statistical list of which teams you are more likely to beat / lose too, but most people know that anyway.

An interesting thing would be seeing which comp is statistically likely to win against the largest amount of comps... although even this wouldn't be very helpful because you wouldn't want a team that beats 99% of teams but loses to kittycleave, mls and shatterplay.

[Bloobungle]

Reverse mathematics on the current system wouldn't be all that difficult but, it also wouldn't be all that useful.

There are a number of factors that play into your teams success and luckily for us, most of it is common sense and instinctive. Off the top of my head...

• your composition
  
• enemy composition
  
• your team MMR
  
• enemy team MMR
  
• time of day
  
• internet stability
  
• item level
  
• your composition

are the main factors driving your ability to gain rating.

Want to farm teams 700 MMR below your team and have no worry in the world of losing? Queue at 9am on Thursday.
Want to maintain the upper advantage based on class composition? Snipe teams by checking the competition on alts.

Any team of good players can pull it off if they are invested. Ever see teams that are 100-10 at 2300? Beautiful system, am I right? Someone who maintains 2300 at a 30-30 record is actually going into competitive games.

[Jamal]

I personally find this pretty interesting however you'd have to collect a very high amount of data to put it to use.

[Korzul]

That films great, i can imagine it now:

"But man he's got one fucking arm and he clicks.."
"But what does he do?!"
"He gets on bas... i mean he interupts casts "
"And interrupting casts lands kills! Buy him!"

In truth, you can "play the system" quite easilly. Get 2 different comps running in the same team which cover to a certain extent boths comps weaknesses. We did thug/rls last season in the same team on alts and just thug cleaved every trip dps/mage team we met and rls'd the rest. There wasn't a single team we faced we couldn't have an advantage against providing all 4 of us were online when we queue'd.

The big trick is knowing who you're queueing against, using an alt team etc to queue initially and see who's playing is always a good way to protect rating etc. You can selectively queue (dodge/snipe)/ multicomp etc or on the TR just roll random crap with your 3 characters and snipe down anyone who's a potential threat.

I doubt you can quantify to the same extent in moneyball because there's not recorded games of everyone. You can always check the ladder and work out whats "strong" vs the majority of the high teams too which is a good start but really, just multicomp your main team and use an alt team to check whats queueing if you still care about the title. Thats if you have the luxury of being on a server with enough people to do this.