69 replies to this topic

[Roadkill]

Sandpit, on 11 December 2013 - 04:02 PM, said:

You're going well above the common understanding of statistics. (No offense to anyone, not intended to be a slight)

Basically:

Team 1 = ELO 500
Team 2 = ELO 500
The predictive component of the MM says both teams are equal, the winner is irelevant, and ELO won't change much if at all regardless of which team wins or how badly a team loses

Not quite. The Elo rating will change by the k-value in this kind of match. K-value is essentially just number Elo uses as the "default" score change. When the match is a toss-up, as in this example, the winner's rating goes up by the k-value while the loser's rating goes down by the k-value. This helps teams/players converge on an accurate Elo rating over time.

Quote

Team 1 = ELO 1000
Team 2 = ELO 500
The predictive component of the MM says team 1 will win. If team 2 wins by any means even a 12-11 then ELO will shift dramatically for both teams. If team 1 wins then ELO won't shift much if at all.

As the Elo ratings diverge, a formula adjusts the delta for a win or a loss. In this example (and assuming a fairly normal Elo setup), we can be pretty sure that Team 1 is going to win. The chances are likely well over 90%. Again, this depends on exactly how the Elo ratings are set up, but generally a 400-point difference in ratings is practically a lock.

So since we're almost certain that Team 1 is going to win, it doesn't make much sense to reward them for winning. The delta is likely 1 rating point, or if the divergence is significant enough it can even be 0.

However, if the unthinkable happens and Team 2 somehow pulls out a victory, they should be greatly rewarded for their superhuman effort. In this case, the delta is likely double the k-value.

Note that Elo ratings aren't static and aren't meant to be. Your rating will move around a bit every time you play due to the k-value. But if you're at your appropriate rating, it should essentially hover around that value... you'll win one and move up, then you'll lose one and move back down, etc.

In chess, Elo ratings are disturbingly accurate. Chess is based entirely on the knowledge and skill of the individual players, so the ratings aren't affected by random chance, network lag, or some derp on the team pulling a Leroy Jenkins. MWO Elo ratings will fluctuate significantly because of those things, but will always auto-correct back toward a more appropriate value as you continue to play.

[Thedrelle]

Balanor ElMarr, on 11 December 2013 - 05:42 PM, said:

Is ELO working correctly? Perhaps since all my win/loss over the last month are close to 50/50 (closer to 0.9 wins to 1 loss but who's counting). During this time, however, it seems that either we roll the other team or we get rolled ourselves. Do the matches seem balanced?

Absolutely not. If they were balanced more matches would go down to the wire.

not true.

There are so many variable in a match that affect how a match plays out, and, sadly, more often than not, a small difference in skill almost always creates a snowball effect where it seems to players that they are being rolled over. and it all could be just because the scouts got wiped early from a miscalculation of where the enemy is, so they can't support the group,

or 1 or 2 pugs getting themselves killed away from the group, weakening the lance they are with.

or just the difference in weight affecting a match.

2 teams with similar ELO are not going to fight to the bitter end in a perfect war of attrition. even 1 loss in a big engagement can roll a match.

[Roadkill]

OneEyed Jack, on 12 December 2013 - 03:14 AM, said:

No, what I say is entirely true, even if the scores are very close. With very close scores the amount of movement is very small. With larger differences, the scores will move not at all (or close to it) if the match follows the prediction, and by a significant amount if there's an upset.

I suppose that depends on what you mean by "very small." If the Elo ratings are very close, the amount of movement is equal to the k-value. That's usually something like 30 points.

If the Elo ratings are wildly different, the amount of movement will either be very close to 0 (if the predicted team won) or very close to double the k-value (if an upset occurred). That would normally be a maximum of something like 60 points.

To me, half of "significant" is not "very small." The k-value ensures that Elo ratings fluctuate pretty much constantly when evenly rated players/teams play one another.

[focuspark]

pre-ELO my WLR was well below 1.0 (~0.33) while my KDR was slightly above 1.0 (~1.25).

after ELO, aka now (remember stats include pre-ELO times), my WLR is climbing towards 1.0 and my KDR is sliding towards 1.0

For all the complains about MM and ELO, it does in fact seem to be working; at least on the macro scale.

[Roadkill]

Kay Wolf, on 12 December 2013 - 09:36 AM, said:

Yes, and no. The matchmaker USES ELO, but ELO does not work to improve player's positions in the game. So, if you match players all off together, based on their ELO range, and Sandpit has actually laid this out, before, there will be no ELO movement for the individual. So, everyone in your ELO bracket, in the range of players who have ELOs similar to you, and because the matchmaker tries to match the teams by exact, or inconsequentially separated ELO, is everyone you will continue to play with throughout your MWO career, and the likelihood of real improvement is little to none.

Does that help?

That's incorrect.

Quote

If not, I'll sort of lay out what Sandpit has expressed before, but in the terms I understand it in...

MWO Matchmaker
ELO Team 1 = 500
ELO Team 2 = 500 (+/- 10%, inconsequential)

In this instance, there will be little movement, if any at all, not only for the teams, but for the individuals, since individual score adjustments are tied directly to the team win, in accordance with PGIs own statements in the past concerning ELO.

Now, if one team ROFLstomps another, when they were perfectly matched, the way ELO works there is NO movement, because the teams were evenly matched.

This is incorrect. When the teams are evenly matched, their Elo ratings adjust up and down based on the k-value. In a typical Elo system, the k-value is around 30. So the winner in your example would end up at 530 and the loser would end up at 470.

Quote

The way it SHOULD work
ELO Team 1 = 1000
ELO Team 2 = 500

If Team 1 wins the match, because the way ELO works there will be no score adjustment to either team, because ELO predicted that Team 1 would win, so it's nothing special, no change.

If Team 2 wins, even by the slightest of margins, it's a major scoring change up for the winning team, and a minor scoring change down for the losing team.

This is correct.

Quote

The matchmaker DOES use ELO, but it is ineffective, because the ELO used is for the overall team, not the individuals playing, and there is never an ELO boost or drop because both teams playing have a 50:50 chance of winning against the other.

Again, this is incorrect. The aggregate rating of the two teams is used to compute the Elo delta, but that delta is then applied to each player's Elo rating individually. Your Elo rating moves after every match, typically by an amount determined by the k-value.

The only time your Elo rating does not move after a match is when your team VASTLY out-rated the other team and your team won. Since your team should win that match, your rating shouldn't be adjusted. The ratings correctly predicted the winner, so no adjustment is necessary.

[focuspark]

... I do believe, after reading more of this thread, those bitching about ELO basically do not understand ELO and were expecting magic.

[Aethon]

/signed

To clarify, I want my w/l and k/d ratios to vary based on my efforts, not artificially regulated by a piece of software. I want my efforts to actually matter.

Edited by Aethon, 12 December 2013 - 11:51 AM.

[Roadkill]

Aethon, on 12 December 2013 - 11:31 AM, said:

/signed

To clarify, I want my w/l and k/d ratios to vary based on my efforts, not artificially regulated by a piece of software. I want my efforts to actually matter.

That's how it currently works.

The matchmaker uses Elo ratings to attempt to put together an evenly matched game.

Your skill (or lack thereof) and that of your teammates then determines who wins that match. The results of the match are used to modify the participants' Elo ratings in the hopes that doing so will make them more accurate for the next time they're used.

The incorrect perception that Elo is used to "drive your win-loss ratio to 1:1" comes from the fact that, if everyone's Elo rating is correct, you should have a 50-50 chance to win any given match. If you truly have a 50-50 chance to win any given match, then your win-loss ratio will settle on 1:1. That's just math. The goal is for the matches to be even... if successful, then the side effect is that your win-loss ratio will trend toward 1:1.

KDR is completely unrelated to, and unaffected by, Elo.

[Aethon]

Roadkill, on 12 December 2013 - 12:06 PM, said:

KDR is completely unrelated to, and unaffected by, Elo.

Then why did my W/L and K/D ratios change when ELO took effect? My skills certainly did not get any worse; if anything, my skills were still improving, as were those of the 2 people with whom I usually dropped (I do not claim to be the best gamer out there or anything, but I am not bad at MWO, and my KDR dropped from just shy of 4:1 down to 2.83:1).

I am ok with being wrong, if you can explain what happened; I am not, however, a fan of any computer regulation of my team composition, aside from that of the machines they bring. Players themselves, IMHO, should never be 'graded' as part of matchmaking. Thus, I cannot see myself ever supporting any sort of ELO system in this game.

Chess =/= Mechwarrior. Not even close.

[Kekkone]

Aethon, on 12 December 2013 - 12:37 PM, said:

Then why did my W/L and K/D ratios change when ELO took effect? My skills certainly did not get any worse; if anything, my skills were still improving, as were those of the 2 people with whom I usually dropped (I do not claim to be the best gamer out there or anything, but I am not bad at MWO, and my KDR dropped from just shy of 4:1 down to 2.83:1).

I am ok with being wrong, if you can explain what happened; I am not, however, a fan of any computer regulation of my team composition, aside from that of the machines they bring. Players themselves, IMHO, should never be 'graded' as part of matchmaking. Thus, I cannot see myself ever supporting any sort of ELO system in this game.

Chess =/= Mechwarrior. Not even close.

Short theoretical answer: You got matched against opponents of the same skill level.

Long answer: Too lazy to write, i'm hoping that Roadkill will pitch in

Edited by Kekkone, 12 December 2013 - 01:05 PM.

[Aethon]

Kekkone, on 12 December 2013 - 01:01 PM, said:

Short theoretical answer: You got matched against opponents of the same skill level.

Long answer: Too lazy to write, i'm hoping that Roadkill will pitch in

So, in other words, statements like, 'KDR is completely unrelated to, and unaffected by, Elo' are incorrect.

Edited by Aethon, 12 December 2013 - 01:31 PM.

[MadcatX]

Aethon, on 12 December 2013 - 01:29 PM, said:

So, in other words, statements like, 'KDR is completely unrelated to, and unaffected by, Elo' are incorrect.

Time to start the new religion of ELOlogy. I know so little as to if it does or does not work, at least this way when matches go well, I can thank Elo, and when I get stomped, well, Elo works in mysterious ways.

[Threat Doc]

For those of you who've tried to bring some semblance of sensible thought to this process, trying to explain that which is, although not a mystery, close to a mystery, I thank you. However, I'm going to have to bow out of this conversation, as it's giving me a headache similar to when I discuss time paradox, and I just got my blood pressure back down, so I'm done here, guys.

[Roadkill]

asdf

Aethon, on 12 December 2013 - 01:29 PM, said:

So, in other words, statements like, 'KDR is completely unrelated to, and unaffected by, Elo' are incorrect.

Nope.

Elo is a rating of your skill level. Nothing more, nothing less. It has nothing to do with your KDR. It doesn't affect anything, truth be told, and is itself not affected by your KDR. It's just a number.

The matchmaker, in creating more evenly matched games for you to play, indirectly affects your win-loss record. In theory that still has nothing to do with your KDR, though playing against opponents whose skill level is different than the ones you've played against in the past could affect your KDR.

You are confusing (perceived) correlation with causality. Your Elo rating does not cause an effect to your KDR.

[Roadkill]

Aethon, on 12 December 2013 - 12:37 PM, said:

Then why did my W/L and K/D ratios change when ELO took effect?

Because, as Kekkone said, you started playing against other players who were the same skill level as you. The matchmaker was able to find others to pair you against who were equally as good (or bad).

If your W/L ratio started dropping, it's because you used to be playing (randomly) against people who really weren't as good as you. You were pugstomping without even knowing it. Now you're playing against your peers, so naturally you're going to lose more often than before.

The same is true of your K/D ratio. If it's dropping, it's because your opponents are better than before, either at killing you or at surviving themselves.

However, Elo does not "regulate" either of those things. How the matchmaker uses Elo can affect them indirectly, but Elo itself is unrelated to them. It's just a rating of your skill level.

Don't blame Elo, blame the matchmaker.

Edited by Roadkill, 12 December 2013 - 02:34 PM.

[Roadkill]

MW4 also had Elo ratings, but it didn't use them to create matches.

[Jabilo]

Wins / Losses 1,772 / 1,609
I pug solo and the above are my stats. Make of it what you will, but as you can see its 'broadly' 50 / 50.

Edited by Jabilo, 12 December 2013 - 02:44 PM.

[Mechteric]

Roadkill, on 12 December 2013 - 02:32 PM, said:

If your W/L ratio started dropping, it's because you used to be playing (randomly) against people who really weren't as good as you. You were pugstomping without even knowing it. Now you're playing against your peers, so naturally you're going to lose more often than before.

Or in my case, pugstomping and I knew it I can appreciate Elo for changing that at least.

The only thing I wish is that they would maybe make separate Elo brackets (maybe 3 brackets?), that way I'll never be put in the same match as someone at the very bottom (its certainly not fair to them to ever be under my crosshairs even if others on their team are of my skill level). That mixed with tonnage limits of some kind would pretty much make every match more enjoyable.

[Joseph Mallan]

Elo makes it so I cannot drop against certain players cause they are better than me. That is something I hate. I don't care if you are the newest player to the game or the best Veteran player the game has. I want a shot at you! Elo prevents me from having this random draw. I'm a sad Panda because of it!

[Ghogiel]

CapperDeluxe, on 13 December 2013 - 06:54 AM, said:

Or in my case, pugstomping and I knew it I can appreciate Elo for changing that at least.

The only thing I wish is that they would maybe make separate Elo brackets (maybe 3 brackets?), that way I'll never be put in the same match as someone at the very bottom (its certainly not fair to them to ever be under my crosshairs even if others on their team are of my skill level). That mixed with tonnage limits of some kind would pretty much make every match more enjoyable.

We know there is a maximum (expanding) threshold from which the MM can grab players closest matches target Elo.

What we don't know is what range that threshold starts at and what is the maximum it is allowed to expand to before failing to find a match.

We know they have tightened that threshold. But some what loosened it again due to the fact it would take 45mins of failing to find match before ever getting one for players in my Elo and I assume those higher too. Now it's tolerable but there are still a good deal more failed to find matches than there used to be. With all the failed to find matches there definitely seems to be a clear cut off.

I think the reason why people still see a trial mech on the odd occasion is probably because you and the trial mech player are not the target Elo, but are at either end of the expanding range, the target Elo for the match could be right between you 2. If the MM picked a high Elo rated player to be the matches target Elo, then it probably wouldn't expand enough to grab a mid range player for that match before failing.