Further Evidence that Elo needs to be completely reworked

[Aristore]

So I've been playing more ranked arenas on my sin and I've actually managed to win 4 more arenas than I've lost since my last thread about elo. Good for me right!?! OH wait except since then my elo has DROPPED by 17 points.

 

http://i.imgur.com/1G7Y7mr.jpg

 

http://i.imgur.com/danvXkY.jpg

 

Let me repeat that. I have an even better record than originally, and my elo has dropped by 17 points.

 

So I win more, and lose more elo?

 

BW ***. This should be basic mathematics. More wins = more elo, regardless. It makes me feel like I shouldn't even bother with a game that doesn't award credit where credit is due. Will WS? Hopefully. After this they will definitely be given a shot.

[Rocanon]

While ELO is not an optimal system for matchmaking of randomized groups, you still have agency in your rating. You probably also need to realize two things.

 

First, ELO isn't just aggregate win%, the relative rating of the people you're playing with/against matters. Its possible to net wins on the day but break even or lose rating if your wins are against people your rating is significantly higher than but your losses are as well.

 

Second, your win% is lower in the second pic, 53% to 55%. So while you won a higher physical number of matches, you lost a larger percentage than you won. Your win % excluding your first 109 matches (the total in the first pic) was only 51%, meaning you did substantially worse over the interval of time between the two cap than the total win% in the second would otherwise indicate.

 

so, y'know, cry moar.

[DarrelK]

Imo each win should be 10 points, each loss 10. Why even bother with this broken ELO ****...

[Aristore]

While ELO is not an optimal system for matchmaking of randomized groups, you still have agency in your rating. You probably also need to realize two things.

 

First, ELO isn't just aggregate win%, the relative rating of the people you're playing with/against matters. Its possible to net wins on the day but break even or lose rating if your wins are against people your rating is significantly higher than but your losses are as well.

 

Second, your win% is lower in the second pic, 53% to 55%. So while you won a higher physical number of matches, you lost a larger percentage than you won. Your win % excluding your first 109 matches (the total in the first pic) was only 51%, meaning you did substantially worse over the interval of time between the two cap than the total win% in the second would otherwise indicate.

 

so, y'know, cry moar.

 

a) Your dick bagishness seems to have skewed your vision because you've missed the point of the thread. On the other hand you have managed to prove you're kind of an elitist tool who looks for opportunities to validate his own ego by digging up a thread which had received 0 replies in 3 days. You also tend to be wrong.

 

The entire point of the thread is that the weights applied to how elo is awarded for wins and loses needs to be re-calibrated. IMO They need to narrow the gap between points awarded for a win against a team with an elo average of 1000 and an elo average of 1600. The first will get you 7 elo, the latter will get you 17. This is especially true for players below 1200. Since their elo is low, they tend to be grouped with low rated players, making it exceedingly difficult to scrape out wins. To add insult to injury their opponents are most likely also low rated, and they will therefore earn a minimal amount of elo for a win.

 

Example: There is a major difference between a player with an elo of 1000 who is say 50-62, and a player who is 0-12. Yet both of them have an elo of 1000. Both are viewed the same in the eyes of the elo gods. But both are clearly not of the same value. If the system was a little more in depth in its view of players I believe fewer players would feel cheated out of elo.

 

They also need to reduce the significance of the first 10 matches. Everyone has experienced the uncanny swings your elo takes in the first 10 arenas. I started 8-2 on my sorc and was at 1420 right off the bat. That's just ridiculous.

 

b) Your argument about win/loss ratio would be valid if that's how ELO worked. But it's not. A player with a record of 51-30 will generally have a similar rating to a player who is 101-80, assuming they had the same elo after their first 10 matches. Let me explain:

Average points for a win- 12.

Average points for a loss: 12

(51 * 12) - (30 * 12) = 612 - 360 = 252

(101 * 12) - (80*12) = 1212 - 960 = 252

 

 

I'm glad you have mastered the art of condescension. I'm sure it makes you feel good. I know I've learned a lot about being condescending just from reading your posts. Thanks Bro!

[Aristore]

Imo each win should be 10 points, each loss 10. Why even bother with this broken ELO ****...

Exactly.

[Rocanon]

a) Your dick bagishness seems to have skewed your vision because you've missed the point of the thread. On the other hand you have managed to prove you're kind of an elitist tool who looks for opportunities to validate his own ego by digging up a thread which had received 0 replies in 3 days. You also tend to be wrong.

 

Didn't notice how old it was. Regardless, I apologize for trying to engage the things you said. That was clearly an error on my part.

 

The entire point of the thread is that the weights applied to how elo is awarded for wins and loses needs to be re-calibrated. IMO They need to narrow the gap between points awarded for a win against a team with an elo average of 1000 and an elo average of 1600. The first will get you 7 elo, the latter will get you 17. This is especially true for players below 1200. Since their elo is low, they tend to be grouped with low rated players, making it exceedingly difficult to scrape out wins. To add insult to injury their opponents are most likely also low rated, and they will therefore earn a minimal amount of elo for a win.

 

This would be referred to as a K value, and should it be altered would have effect of squishing the rating, which would necessitate a squishing of the reward structure to compensate. If you lower gains losses all you do is lower the maximum and raise the minimum, you don't really change the average. Numbers being fiat, the guy with 2700 now only has 1800 but he's still crushing everyone else. You now have 1100 instead of almost 1300 but it doesn't matter because the rating gulf between the two of you is equally massive in practice despite the numbers not being as far away on the number line.

 

Example: There is a major difference between a player with an elo of 1000 who is say 50-62, and a player who is 0-12. Yet both of them have an elo of 1000.

 

The primary difference being that the person who was 0-12 probably just stopped. Also at 0-12 they're probably a bit lower than that, to say nothing of the fact that someone's rating after 12 games is essentially meaningless.

 

Both are viewed the same in the eyes of the elo gods.

 

Its an algorithm and doesn't view anything. All it does it take relative values and do math based on the outcome of their meeting.

 

But both are clearly not of the same value. If the system was a little more in depth in its view of players I believe fewer players would feel cheated out of elo.

 

Which is why League of Legends still uses it right? Its not a great system for group matchmaking since it gets odd with averages and the like, but changing to a system where you get flat points for wins is hardly any better. It isn't truly fair for either party if an ELO 1000 calibre player gets the same 10 points for beating an ELO 2000 calibre player that they would get for beating a 1000.

 

They also need to reduce the significance of the first 10 matches. Everyone has experienced the uncanny swings your elo takes in the first 10 arenas. I started 8-2 on my sorc and was at 1420 right off the bat. That's just ridiculous.

 

No, they need to start you at a baseline ELO that is average (either an objective one such as 1100 or based on the avg. from the prior season.)

 

Even with the funny rating from your first ten, you're still going to overall end up approximately where you would otherwise end up. Where is your sorc now? I'd bet they've gone down since you got out of the initial 10 as you approach a more even win-loss ratio.

 

b) Your argument about win/loss ratio would be valid if that's how ELO worked. But it's not. A player with a record of 51-30 will generally have a similar rating to a player who is 101-80, assuming they had the same elo after their first 10 matches. Let me explain:

Average points for a win- 12.

Average points for a loss: 12

(51 * 12) - (30 * 12) = 612 - 360 = 252

(101 * 12) - (80*12) = 1212 - 960 = 252

 

I'm uncertain if I was unclear or if you just don't grok it.

 

You have a *worse* record in the second pic. Your total game wins were higher but the percentage of wins overall was lower. How exactly is that not functioning? It makes perfect sense that your rating would be lower.

 

Your math is also meaningless, although the arithmetic itself looks valid. Declaring an average point for wins/losses via fiat is illogical and makes the rest of it meaningless because the average is based off of variable factors. If we assume that you're *always* matched up against even teams it wouldn't be quite so, but your rating shifts are a factor of your teams aggregate rating and theirs. The pool isn't large enough for those to always approximately match, and as such your shifts can be wildly variable from match to match (I've gone from winning 5 to 25 on consecutive matches).

 

Consequently, while you can certainly take a *large* pool of match results and arrive at your average, just picking some number for it (and picking the same number for wins and losses) is at best misguided and more likely intellectually dishonest in the extreme.

 

I'm glad you have mastered the art of condescension. I'm sure it makes you feel good. I know I've learned a lot about being condescending just from reading your posts. Thanks Bro!

 

Its not my fault you're wrong and whiny.