Jump to content

Elo Is A Ladder Not A Scale


99 replies to this topic

#21 Jman5

    Member

  • PipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 4,914 posts

Posted 19 September 2013 - 01:08 PM

View PostThe Boz, on 19 September 2013 - 10:44 AM, said:

The problem is that Elo was designed for 1v1. You and you alone are responsible for how well your team plays, and you and you alone get the Elo gain or loss after the match has ended.
In MWO, your input is mathematically 1/12, but regardless of your personal performance, you gain or lose the same Elo as every other team member. Regardless of their performance.
That is way I do not see a bright future for this system in MWO.

Play your hardest and work on personally improving and you Elo will trend upward. Play like dog poo and blame your team every time you lose, and your Elo will trend downward.

You will never win every game you play well in, you will never lose every game you play terrible in, but your individual effort will make a difference (for better or worse) over many games.

#22 AntiCitizenJuan

    Member

  • PipPipPipPipPipPipPipPip
  • 1,440 posts
  • LocationIn your base, killing your dudes

Posted 19 September 2013 - 01:11 PM

Why they dont just do WoW style Arena and Battleground ratings is beyond me.

WHOA LOOK HOW HARD IT IS PGI

#23 The Boz

    Member

  • PipPipPipPipPipPipPipPip
  • 1,317 posts

Posted 19 September 2013 - 02:10 PM

View PostJman5, on 19 September 2013 - 01:08 PM, said:

Play your hardest and work on personally improving and you Elo will trend upward. Play like dog poo and blame your team every time you lose, and your Elo will trend downward.

You will never win every game you play well in, you will never lose every game you play terrible in, but your individual effort will make a difference (for better or worse) over many games.

Yes, this does work. Statistically. Meaning, after a number of games.
If you want, you can mathlab an Elo simulator with some random numbers and see how efficient it is in multiplayer environments.

#24 Deathlike

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 29,240 posts
  • Location#NOToTaterBalance #BadBalanceOverlordIsBad

Posted 19 September 2013 - 02:15 PM

View PostJman5, on 19 September 2013 - 01:08 PM, said:

Play your hardest and work on personally improving and you Elo will trend upward. Play like dog poo and blame your team every time you lose, and your Elo will trend downward.

You will never win every game you play well in, you will never lose every game you play terrible in, but your individual effort will make a difference (for better or worse) over many games.


I'm kinda thinking the current iteration of the MM is actually placing you with actual people at your own level.

The previous versions of the MM trolled me hard. Spotting the newbie doesn't take too much effort when I look through the trial mech list... it's unrealistic to always be carrying people that potentially lead you to your own demise while those players having no malicious intent to do so.

Edited by Deathlike, 19 September 2013 - 02:16 PM.


#25 WVAnonymous

    Member

  • PipPipPipPipPipPipPipPip
  • Wrath
  • Wrath
  • 1,691 posts
  • LocationEvery world has a South Bay. That's where I am.

Posted 19 September 2013 - 02:37 PM

View PostDeathlike, on 19 September 2013 - 02:15 PM, said:


I'm kinda thinking the current iteration of the MM is actually placing you with actual people at your own level.

The previous versions of the MM trolled me hard. Spotting the newbie doesn't take too much effort when I look through the trial mech list... it's unrealistic to always be carrying people that potentially lead you to your own demise while those players having no malicious intent to do so.


I logged into my 8 year old's account last night, with her 0.04 KDR and she has a 1.0 W/L ratio. How?

I dropped with PEEFSmash on my team twice last night. I can only assume she is averaging him down, and he saves her W/L ratio. I think MM is blending the top and bottom 10% to get the team averages.

(I'm getting her a 24 medallion, not trolling, BTW)

And if you see Baby Voldemort, be nice. She's playing her own account tonight.

Edited by WVAnonymous, 19 September 2013 - 07:26 PM.


#26 Deathlike

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 29,240 posts
  • Location#NOToTaterBalance #BadBalanceOverlordIsBad

Posted 19 September 2013 - 02:41 PM

View PostWVAnonymous, on 19 September 2013 - 02:37 PM, said:


I logged into my 8 year old's account last night, with her 0.04 KDR and she has a 1.0 W/L ratio. How?

I dropped with PEEFSmash on my team twice last night. I can only assume she is averaging him down, and he saves her W/L ratio. I think MM is blending the top and bottom 10% to get the team averages.

(I'm getting her a 24 medallion, not trolling, BTW)


I should just say "it's a bit better more often than not". They are still actively tweaking it and... there will be outliers.

It is funny though. :(

#27 Zerberus

    Member

  • PipPipPipPipPipPipPipPipPip
  • Overlord
  • Overlord
  • 3,488 posts
  • LocationUnder the floorboards looking for the Owner`s Manual

Posted 19 September 2013 - 03:16 PM

View PostMaddMaxx, on 19 September 2013 - 09:14 AM, said:


Yes, but now I have to figure where the hell I fit in that Jungle. :) LOL!

P.S. Oh wait a minute. I am there already. It is Me and Kaffe. :(


No, that clearly says Me and Kaffe :)

To the topic at hand: After earning a nice income raising people`s Elos in LOL and watching them drop back down within weeks or even days, I have to agree with the OP 100%.

What I always find hilarious is when the same people that complain about game objectives (in MWO that would be capping as a tactic to WIN instead of hulk smash robots) "costing them fun" are also complaining about always getting put in with the scrubs despite their awesome game performance... It`s really eerie, almost like the 2 have something to do with each other.... :( ;)

Edited by Zerberus, 19 September 2013 - 03:17 PM.


#28 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 September 2013 - 03:43 PM

View PostHomeless Bill, on 19 September 2013 - 10:18 AM, said:

I'm actually kind of curious as to what the bell curve of win-loss ratios look like. It'd be even better if they could separate it out by 4-man wins and solo wins (I'd really be curious to see the difference between those two ratios).

Elo distribution from the Last matchmaking Command Chair, darker portions indicate Elo score for more than 50 games played.
Posted Image

View PostOriginalTibs, on 19 September 2013 - 10:02 AM, said:

When you have played in excess of 8600 matches and your wins are more than 400 greater than your losses you can bet it will be a while before your ELO changes much.

It actually doesn't take much to move your Elo score. Remember you're betting your Elo points against a lower ranked player. So if your team craps the bed and you lose to a mediocre team you'll lose a lot of your Elo points. If your elo matchmaking doesn't put you against enough players with similar or higher Elo scores it can be difficult to raise your scores. If however, you get locked into a bracket of players with similar Elo scores, then your score won't change much win or lose.

My guess is your win rate is 52% 4500 wins, 4100 losses, at 8600 games, or you win 13/25, or 2.6/5. So to raise your win rate 1% to 53% by playing at a win rate of 53%. You need to win 27 out every 50 games, or 2.7 out of every 5 games you play for the next 1700+ games. You're right moving you win rate might be difficult but your Elo score, though based on win rate, is independent from it in a non linear relation, and therefore can fluctuate much more rapidly.

View PostHomeless Bill, on 19 September 2013 - 11:08 AM, said:

If you're sitting below a 1.0 WLR after a considerable amount of matches (probably upwards of 500), you're less helpful to your team than the average player; if it's above 1.0, you're more helpful.

Or maybe you like light mechs, which while important don't play a crucial role in the outcome of the match. The impact a player can make on the outcome of the match is also varied with mech selection. While it's true if you pick bad mechs you are hurting your team. It does indicated that contribtion to outcome is not uniform and therefore subject to statistical variation. I would say if your win to lose ratio is less than 45% you're a detriment to your team. If it's between 49-and 45% you could probably use a hand with mech selection and fitting.

Edited by Grits N Gravy, 19 September 2013 - 03:45 PM.


#29 Suko

    Member

  • PipPipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 1,226 posts
  • LocationPacific Northwest

Posted 19 September 2013 - 03:47 PM

I agree that elo must be working, at least somewhat. I used to have a ~2.0 W/L ratio, but with the introduction of elo, I've noticed my W/L slowly creeping steadily downward. Not by much, but it is heading to 1.6 and will probably keep going even lower.

My question is; I'll play games by myself (or with one buddy) and we usually do fair or well. However, if I'm in a group with two friend (3 of us total), the odds of us winning go to sh*t. If we win 1/3 of our games, we'll be lucky. This just seems really strange to me.

Edited by ShadowVFX, 19 September 2013 - 03:52 PM.


#30 Homeless Bill

    Member

  • PipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 1,968 posts
  • LocationA Box Near You

Posted 19 September 2013 - 03:59 PM

View PostGrits N Gravy, on 19 September 2013 - 03:43 PM, said:

Elo distribution from the Last matchmaking Command Chair, darker portions indicate Elo score for more than 50 games played.
Posted Image


But we don't know how Elo is calculated. I'm curious as to approximately what WLR correlates with a 1400 Elo score. Or 700. Or 2000. I know it's not a direct correlation, but it's probably extremely close.

#31 The Boz

    Member

  • PipPipPipPipPipPipPipPip
  • 1,317 posts

Posted 19 September 2013 - 04:04 PM

There is no WLR correlation to a given Elo. Elo strives for a 50:50 win:loss.
You win a game, you gain Elo.
You lose a game, you lose Elo.
If your opponents were of a higher Elo than you, you gain more and lose less.
If your opponents were of a lower Elo than you, you gain less and lose more.
Typically, a game will rarely shift your Elo by more than 30, most often 12-20.
The game will try and set you up with matches that has everyone within a given Elo range of one another.

#32 WVAnonymous

    Member

  • PipPipPipPipPipPipPipPip
  • Wrath
  • Wrath
  • 1,691 posts
  • LocationEvery world has a South Bay. That's where I am.

Posted 19 September 2013 - 04:07 PM

View PostHomeless Bill, on 19 September 2013 - 03:59 PM, said:

But we don't know how Elo is calculated. I'm curious as to approximately what WLR correlates with a 1400 Elo score. Or 700. Or 2000. I know it's not a direct correlation, but it's probably extremely close.


Other way round. What does your Elo score need to be to maintain a W/L of 1.0 in this pool of players?

Ooh, I was too slow.

#33 MadCat02

    Member

  • PipPipPipPipPipPipPip
  • The 1 Percent
  • 668 posts

Posted 19 September 2013 - 04:14 PM

View PostNarcissistic Martyr, on 19 September 2013 - 07:45 AM, said:

So... my 51% win rate means I don't suck as bad as I think I do? I can live with that.


I said 55% is good , 51% means you never really done anything sorry XD

View PostShadowVFX, on 19 September 2013 - 03:47 PM, said:

I agree that elo must be working, at least somewhat. I used to have a ~2.0 W/L ratio, but with the introduction of elo, I've noticed my W/L slowly creeping steadily downward. Not by much, but it is heading to 1.6 and will probably keep going even lower.

My question is; I'll play games by myself (or with one buddy) and we usually do fair or well. However, if I'm in a group with two friend (3 of us total), the odds of us winning go to sh*t. If we win 1/3 of our games, we'll be lucky. This just seems really strange to me.


This usually means you are druged down by your inferior teammates no offense . If your teammates are lower elo you might be playing at low elo bracket which means unprediactable numbaer of new players

Edited by MadCat02, 19 September 2013 - 04:14 PM.


#34 The Boz

    Member

  • PipPipPipPipPipPipPipPip
  • 1,317 posts

Posted 19 September 2013 - 04:14 PM

It is realistically possible to have an Elo of 1400 or above, and a WLR of 45%.
It's also possible to have an Elo of under a thousand, and a WLR of 55%.

#35 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 September 2013 - 04:28 PM

View PostHomeless Bill, on 19 September 2013 - 03:59 PM, said:

But we don't know how Elo is calculated. I'm curious as to approximately what WLR correlates with a 1400 Elo score. Or 700. Or 2000. I know it's not a direct correlation, but it's probably extremely close.


boz is correct there is no correlation. We do know how Elo's are calculated. http://mwomercs.com/...79-matchmaking/

Everyone starts at and Elo of 1300. As a person plays his score is adjusted as described in the link, we also apply a K factor to standard adjustments in scores for a fixed period of time. The K factor is designed it compensate for the uncertainty of the initial score. It does this by peanlizling you more for losses to lower ranked players and rewarding you greater for wins against higher ranked players. The idea is speeds the process of finding your natural Elo bracket where your would expect a 50/50 win rate.

What the K factor and how long it applies for are unknown in MWO at the moment.

#36 The Boz

    Member

  • PipPipPipPipPipPipPipPip
  • 1,317 posts

Posted 19 September 2013 - 04:35 PM

The problem here is that it does so right out he gate, per account. It doesn't give you a chance to "learn" the game, and if you fall too low in your first few matches, even learning how to play the game will be hard. If you really are new to the game, it is likely that you will fail because you have no idea what the terms "alpha strike", "ghost heat" and "convergence" mean,

#37 Homeless Bill

    Member

  • PipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 1,968 posts
  • LocationA Box Near You

Posted 19 September 2013 - 04:44 PM

View PostGrits N Gravy, on 19 September 2013 - 04:28 PM, said:

boz is correct there is no correlation. We do know how Elo's are calculated. http://mwomercs.com/...79-matchmaking/

Everyone starts at and Elo of 1300. As a person plays his score is adjusted as described in the link, we also apply a K factor to standard adjustments in scores for a fixed period of time. The K factor is designed it compensate for the uncertainty of the initial score. It does this by peanlizling you more for losses to lower ranked players and rewarding you greater for wins against higher ranked players. The idea is speeds the process of finding your natural Elo bracket where your would expect a 50/50 win rate.

What the K factor and how long it applies for are unknown in MWO at the moment.

My point is that we don't know any of the specifics about it's calculation. Again, there is no direct correlation due to the fact that there's one team that's expected to win.

All the same, there's almost surely a very telling average WLR for a given Elo bracket, and I'd be extremely curious to see some of those numbers. I have a very hard time believing two people with a decent number of rounds and the same Elo rating would have wildly different win-loss ratios.

#38 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 September 2013 - 04:46 PM

View PostThe Boz, on 19 September 2013 - 04:35 PM, said:

The problem here is that it does so right out he gate, per account. It doesn't give you a chance to "learn" the game, and if you fall too low in your first few matches, even learning how to play the game will be hard. If you really are new to the game, it is likely that you will fail because you have no idea what the terms "alpha strike", "ghost heat" and "convergence" mean,

The counter argument is that the K factor puts you in low tier matches more quickly were you'll be facing other new players. Does that really happen, no because the pool is so limited that you're forced to allow matches of great Elo variance. So the kiddie pool never really gets roped off. Its why you see the bulk of the players with an Elo score below 1300 (the initial score) in that graph i posted. If that graph tells you anything it's the K factor could use tweaking, and the distribution is asymmetric.

#39 Jonathan Paine

    Member

  • PipPipPipPipPipPipPipPip
  • Survivor
  • Survivor
  • 1,197 posts

Posted 19 September 2013 - 04:56 PM

In an ideal world, we would have a pretty good idea of where we are ranking-wise. However, given the somewhat shoddy programming record of PGI, the matchmaker is unlikely to put you up against equally skilled players.

What does that mean? You might, conceivably, always play against players with a lower rating than yourself. If that discrepancy is large enough, even a 4/1 w/l rating COULD mean that your actual rating is <1100. Every win is worthless, every loss a heartache :(

Of course, if the matchmaker does put you in with players of equal skill, then your rating should be:
(wins-losses) x 25 based on the
http://mwomercs.com/...65#entry1626065

#40 Grits N Gravy

    Member

  • PipPipPipPipPipPip
  • 287 posts

Posted 19 September 2013 - 05:10 PM

View PostHomeless Bill, on 19 September 2013 - 04:44 PM, said:

My point is that we don't know any of the specifics about it's calculation. Again, there is no direct correlation due to the fact that there's one team that's expected to win.

All the same, there's almost surely a very telling average WLR for a given Elo bracket, and I'd be extremely curious to see some of those numbers. I have a very hard time believing two people with a decent number of rounds and the same Elo rating would have wildly different win-loss ratios.

Actually the specifics are in there.
Spoiler

There's no direct correlation to win rate and Elo scores not because one team is expected. In fact the system takes that into account and rewards you more points if you upset a better team. But because the Elo scoring has nothing to do with win and loss ratios. Elo does not equal win loss.

A person can end up with an Elo ranking of 2000 if they beat enough people with high ranks early in his career, due to the K factor. His K factor will eventually go away and he'll sit on a 50/50 win ratio at 2000 elo over a course of another 50 games. Meanwhile a player at an 1400 Elo, maybe he didn't quite pick up on things early on, but he's really developed a lot of skill. Thus allowing to gain some elo points. So he might have a win/loss of 3/5. Because he's at the top of his bracket, he needs to keep up that win loss to stay there as his losses cost him more points than his wins generate, aka elo hell.

Can you start to see how there is no strong correlation to win loss and Elo rankings? Even if we looked at Elo after 20 games when K factors are still in play; People with the same win and loss will have wildly different Elo scores based on the competition they faced, as their Elo scores are derived directly from their opponents. Within the Elo scores it's reasonable to assume a totally random distribution of win loss, that trends to 50/50 the longer the person stays at that score.

Edited by Grits N Gravy, 19 September 2013 - 05:21 PM.






1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users