Guild Wars 2

How the leaderboards ranking currently is or should be

(MMR) = A
MMR should be governed by how many times the match predicted your team’s points. Losing by 300, 400, winning, etc.
Example: A person who constantly losses with his team’s end score being 300, will have a lower MMR than a player who constantly loses with an end score of 400.
Two people with the same 50% win rate does not mean they have the same MMR.

(Current Leaderboard) = B
Rank is tallied by how many points you earned in a match. The issue with this, is that there’s no cap on the amount of games you play. People with a 50% win rate will naturally be a higher rank if they play more games than others. They’re guaranteed to earn more points, regardless of MMR. If he has a 50% win rate, he just needs to play twice as much to have a positive earning in points. If he’s less than 50%, he needs to play even more games (more than twice as much) to have a positive growth in Rank points.
Or the player simply needs to break even and not lose any points.

There needs to be an equation where
A (MMR)
Directly effects
B (Points earned in match)
= Rank points.

I’m no mathematician or savvy in statistics, but if there was a proper equation, the leader board will show up like this.

Just as an example
Current Leaderboard
Rank____Player______Won/Loss___Win %
600————Joe.123———50W 50L———- 50%
999———- Bob.123———30W 0L———-100%

New Leaderboard
Rank____Player______Won/Loss___Win %
600———- Bob.123——30W 0L——100%
700————Joe.123——50W 50L——50%
Below is what the recommended ratio should be

As you can see, personal MMR takes a bigger effect in the new leaderboard. If a proper equation between MMR and Points Earned in a game showed this effect type, we’ll see people with high win % (or not high, but high MMR people) on the top 25 leaderboards.

The ratio should be this
100 games played at 100% win rate
(obviously extremely high mmr)
=
400 games played at 50% win rate
(Average to below average MMR)

I grabbed that ratio from the end of the last seasonal ladder. I believe Backback ended the season with 97 games at 100% win rate.
I don’t remember how many games the #1 ranked player had this past season, so I used 400 as an example.

My issue with the proposed leaderboard A would be that the number of points earned (and thus games played) would still be the sole determinant of the final leaderboard position. Although it would be fairer than the current one, it still leaves open the situation where a player doesn’t place solely because they played a couple of games less than another player. It also shifts the time burden from being fairly equal for everyone to being mainly on lower ranked (and likely newer) players. For me personally, during the test season I think I only played 6 ranked games, partly because I knew I wouldn’t be able to compete if the determining factor was solely time played. I suspect for many other players that was a factor too, and shifting the time burden to a different populace won’t help much.

Based on their execution, ANet’s idea with the leaderboards is that anyone has a chance to come out on top; not just tournament winners. The way they have achieved this is by having essentially time-played being the determining factor in the leaderboards, instead of MMR. That way, even players with a low MMR can still place highly if they participate a lot. Which thus achieves their second goal of getting more players participating in sPVP. With the main goal of getting more players into sPvP in mind, the current setup doesn’t work as well as it could because a lot of players (like me) will quickly realize that if they’re expected to get a massive amount of points a day just to place at #250 to qualify for a reward, then it’s just not going to happen. Hence the setup is only encouraging a small number of players (who have the most free time) to participate. Ironically, this is probably the same number of people who think they can place in the top 250 in an MMR based leaderboard. I don’t have a problem with time played being a factor, but it can’t be the determining factor.

=============
With the goal that lower MMR players can potentially place in mind, my suggestion from a previous post is:

A points based leaderboard with a moving season-cap as the season progresses, and MMR is used as the tie-breaker for players with equal points.

For instance, if the moving cap was based on 4 pts per day 28 points per week for a 28 day season:
Day 1 in season, max points you can have is 4.
Day 7 in season, max points you can have is 28.
Day 28 in season, max points you can have is 112.
By the end of a 28 day season, the max points you can have is 112. If multiple players have 112 points, MMR is used as the tie-breaker. Many players may not have time to accumulate 112 points in a season (150+ games in a month), leaving room for dedicated PvPers with lower MMR a chance to gain rewards. Also, if a player only plays on weekends, they still have an opportunity to catch up in points because they’re not locked out. Finally, since its a moving cap instead of a fixed cap, it encourages player participation throughout the season instead of just a single week of grinding at the start; and players must also play occasionally to avoid MMR decay.

EDIT
For the example, have the moving cap set at 28 points per week instead of 4 points a day; so that players don’t have to login on the last day(s) to qualify for the top spot.

Win Ratio != Skill

It can’t be emphasized enough.

In a system where players are matched with and against other players of roughly equal skill, win ratio should be close to 50%. If a player has a win ratio significantly higher than 50%, then that player is constantly fighting much lesser skilled players and is probably below his or her proper rating. The exception to this is the very top of the rating distribution. Those players have few opponents of equal or higher skill and they end up fighting weaker opponents more often.

Basing a system on win ratio further assumes that across all games played by a given player, the average outcome or value of a game was 50% of the possible points (winning every time). However, if a players has more matches which he or she is expected to lose than matches which he or she is expected to win, the expected win ratio is below 50%. It doesn’t mean that player is worse than what their rating says.

It’s as simple as this: Player A is pretty bad, but he plays against other pretty players and wins 50% of the time. Player B is above average and plays against other above average players and wins 50% of time. They have the same win ratio, so based on your logic, they must be equally skilled. That’s the same flaw in the current leaderboard.

So your argument is extremely confusing because you’re mis-using the term “rating” or “match-making rating”. Rating is a value for each player that reflects their skill level. It adjusts up or down as a function of the player’s rating in comparison to their opponents’ rating, with “upsets” yielding a larger rating adjustment (Glicko2 is more sophisticated, but has the same basic idea).

What you maybe want is a system that adjusts the current leaderboard point matrix with the rating of the players; games between higher rated players award more points for a win and deduct less points for a loss. In theory, this allows both time and skill to be valuable.

In practice, it really depends on how correct ratings are and how good match prediction is. You need to balance the top players who play less frequently with above average players who form a team and farm in off-peak hours. I can tell you now that both of those have issues: rating fluctuates too much for unknown players (RD’s initial value is probably too high); the prediction outcome uses average team rating when in reality a significantly worse player has a much larger negative influence on victory compared to his decrease of the average team rating.

Exedore.6320:

Win Ratio != Skill
The exception to this is the very top of the rating distribution. Those players have few opponents of equal or higher skill and they end up fighting weaker opponents more often.

If you scroll through leaderboards the vast majority of 65%+ win ratios are such players

I agree with roy, i gave the same advice about cap matches during the season and mantain a points system.

I also really like to use mmr like tie break.

I absolutely agree with you Roy. I just don’t think Anet wants to go this route because they think it’s a “restriction” to players with unlimited time on their hands.
The system I propose caters to both extremely high Rated players playing fewer games, and players with incredibly high amounts of games played. It actually forces high Rated players to play a couple more games; in order to, keep up with players who have unlimited amounts of time on their hands. Except their Rating is so high, they don’t need to grind.

Assuming the system goes well, it promotes players, with unlimited time on their hands to grind in more games and keep up with the Johnsons; players with high rating.

I just used win rate as an example of some one’s MMR level (which is really rating). It may have been confusing when I used the words rate, or added win rat/ratio, in my example leaderboard. I know win % doesn’t mean a person has high MMR (rating)… i’m not entirely sure how MMR or Personal Rating, whether they’re different or the same, is calculated.

Exedore.6320:

Win Ratio != Skill

It can’t be emphasized enough.

In a system where players are matched with and against other players of roughly equal skill, win ratio should be close to 50%. If a player has a win ratio significantly higher than 50%, then that player is constantly fighting much lesser skilled players and is probably below his or her proper rating. The exception to this is the very top of the rating distribution. Those players have few opponents of equal or higher skill and they end up fighting weaker opponents more often.

Basing a system on win ratio further assumes that across all games played by a given player, the average outcome or value of a game was 50% of the possible points (winning every time). However, if a players has more matches which he or she is expected to lose than matches which he or she is expected to win, the expected win ratio is below 50%. It doesn’t mean that player is worse than what their rating says.

It’s as simple as this: (1) Player A is pretty bad, but he plays against other pretty players and wins 50% of the time. Player B is above average and plays against other above average players and wins 50% of time. They have the same win ratio, so based on your logic, they must be equally skilled. That’s the same flaw in the current leaderboard.

---

So your argument is extremely confusing because *(2) you’re mis-using the term “rating” or “match-making rating”. * Rating is a value for each player that reflects their skill level. It adjusts up or down as a function of the player’s rating in comparison to their opponents’ rating, with “upsets” yielding a larger rating adjustment (Glicko2 is more sophisticated, but has the same basic idea).

(3) What you maybe want is a system that adjusts the current leaderboard point matrix with the rating of the players; games between higher rated players award more points for a win and deduct less points for a loss. In theory, this allows both time and skill to be valuable.

In practice, it really (3) depends on how correct ratings are and how good match prediction is. You need to balance the top players who play less frequently with above average players who form a team and farm in off-peak hours. I can tell you now that both of those have issues: rating fluctuates too much for unknown players (RD’s initial value is probably too high); the prediction outcome uses average team rating when in reality a significantly worse player has a much larger negative influence on victory compared to his decrease of the average team rating.

I absolutely wasn’t basing a system on win ratio not at all. Just a player’s own personal rating that identifies a person’s “skill level” best.

About Bolded highlight (1)
My statement from my original post, “Two people with the same 50% win rate does not mean they have the same MMR.” I meant they don’t have the same skill level or personal rating. I agree with you.
From your example, my logic suppose to state that player A and player B are Not equally skilled.

About Bolden highlight (2) & (3)
You’re right i’m completely confusing the two ratings, which probaby made my original post even more confusing… my bad.
But you hit the nail on the head on interpreting what i’m trying to say. I agree completely about how correct ratings needs to be, as well as matchmaking predictions. That was the thought I had when making my original post. It’s what we’ve experienced with the last test ladder.

Again sorry for the confusion. That last part you just said, is what I’m trying to say :P

I still personnaly fail to understand how the current leaderboard is working. I would like to be explained why :
- i am, quite often, teamed with and against people from 1st leaderboard page, although i’m only #650
- people on 1st page get +3 loses, but no negative rating point..

If i’m having the same MMR than front page players, then this means basically anyone can be teamed with anyone. This would explain why i’m getting +1 point for my wins, but still…
In my opinion people should be teamed with people having a similar leaderboard rating.

Abazigal.3679:

I still personnaly fail to understand how the current leaderboard is working. I would like to be explained why :
- i am, quite often, teamed with and against people from 1st leaderboard page, although i’m only #650
- people on 1st page get +3 loses, but no negative rating point..

If i’m having the same MMR than front page players, then this means basically anyone can be teamed with anyone. This would explain why i’m getting +1 point for my wins, but still…
In my opinion people should be teamed with people having a similar leaderboard rating.

You’re matched with players of similar rating. That rating value is determined by a Glicko2 algorithm and hidden to you. The rating goes up or down based on your rating relative to your opponents’ ratings and the match result. Beating higher rated players makes your rating increase more and losing to lower rated players makes it decrease more.

Leaderboard rank has no correlation whatsoever to rating; it’s just an ordering of players by leaderboard points. Leaderboard points are awarded or deducted based on the outcome of each game compared to its predicted outcome. More Info.

The problem with leaderboard points is that they don’t reflect your rating at all. Winning an even match involving all low rated players has the same award as winning a match involving all high rated players. So leaderboard points end up being based purely on time with an advantage given to players and teams who farm in off-hours when they’re likely to be matched against much lesser skilled players.

Saiyan.1704:

I just don’t think Anet wants to go this route because they think it’s a “restriction” to players with unlimited time on their hands.

I see your point with the unlimited time. Of course the leaderboard shouldn’t be determined by those with the most time because the “most free time” folks make up a small percentage of the population; although that it was the current setup does. To resolve their issue however, the following could be implemented:
Set a cap to the number of points that would affect leaderboard ranking. Points can be earned beyond that cap, but they do not affect leaderboard ranking. Everyone earns rewards (see below) based on the number of points acquired. Visually, the current leaderboards site could be modified to indicate that a player has reached the cap; or a new column could be added called “Total Points” to see their total points earned; not just the season cap points. Using a moving-cap while allowing points to spill over the season cap would be pretty complicated; a set cap is easier to implement and understand.

Since there wouldn’t be a moving cap to require players to play weekly, a new system would be needed to encourage them to play often. This can be achieved by adding a rewards system that’s active only during the season.
At the end of a season, all players are rewarded for either:
A) For every X (ex.: 25) points they have earned. Ex: Reward track boosters.
B) For each point earned. Ex: Some silver for each point.
C) Both. Rewards from (A) and (B) might have to reduced.
Note: The reward has to be distributed at the end of the season so that players can’t abuse going above and below the benchmark (25pts) for repeat rewards. Also, with every player getting at least a small reward for participating, it encourages participation throughout the season for everyone even if someone knows they could never place highly.
_________________________________________________________________
So to summarize:
There is a season cap towards the number of points that affect leaderboard ranking. Additional points can be earned, but it will not affect your leaderboard ranking. If multiple players have the same number of points, MMR is used as a tiebreaker.

At the end of the season, all players are rewarded based on the number of points earned. Reward setup can be:
A) Large rewards at benchmark values.
B) Small rewards for each point earned.
C) Both. Adjust rewards accordingly.

There are two systems that work to keep people playing throughout the season:

• Reward players (including those with lots of free time) by rewarding them for each point earned.
• Discourage players from only playing at the beginning of the season via MMR decay.

Finally, that does seem like a lot of rewards on top of the reward tracks, dailies, match win rewards, and top 250 rewards; but that’s what it takes for a competetive points-based ladder to allow infinite points without punishing anyone. The best/unique rewards are still reserved for those at the top of the ladder.
________________________________________________________
Example numbers:
For a 30 day season with a 120 point cap (which is 4 points per day average):

• Using my points per game avg of .67 and assuming that stays constant, that is 180 games in a month, or 6 games a day average.
• Using the leaderboard leader’s values (.65 pts per game), that is 184 games in a month.