So here's the announcement today that bonus pool is coming to WoW arenas: http://us.battle.net/wow/en/blog/839...yond-1_17_2013
And as I've argued months ago (in fact even before that), the SC2's bonus pool system is flawed and counterproductive. There are better ways to reward activity correctly, without distorting the ladder with a bonus pool system. The full post is here: http://www.teamliquid.net/forum/view...ic_id=389449#1
I've also made a post explaining this in the context of WoW (it quotes all the relevant sections from the thread linked above, and "OP" refers to that thread):
While Riot decided to copy parts of SC2's league system a few days ago, they were smart enough to not copy the most counterproductive and distortionary part of the ladder system: the bonus pool. Yet, WoW has.
But what's even worse is that Blizzard continues to use the same arguments, debunked in the OP, to defend this really bad system.
Players getting into PvP mid-season face a very significant challenge to gear up and become competitive because players that started earlier in the season are so far ahead in terms of gear. We want to make it more reasonable for a player to join in on organized PvP mid-season, while still rewarding the commitment of players that have remained invested in PvP throughout the season.
http://us.battle.net/wow/en/blog/839...yond-1_17_2013Blizzard claims that they want to help people who start midseason. Yet the bonus pool punishes these players, by requiring them to play more games than is necessary to stabilize MMR, because they must also spend all accrued bonus pool to be placed correctly. As explained in the OP:Team Rating Inflation
As we’ve mentioned previously, we want to see the ladder rankings decided toward the end of a season, and not a foregone conclusion dictated by what happens at the start. As discussed in the PvP in Mists of Pandaria Dev Watercooler, as of patch 5.2 Team Rating will gradually increase as players participate in PvP over the course of the season. We expect the new system to help ensure that the ladder remains active at all ratings and that the competition stays fierce from start to finish.
They say that they want to encourage activity, instead of camping on a rating. But there's no reason to reward activity for the sake of rewarding activity, as the above example shows. More active players should be rewarded for activity only to the extent that it leads to a lower uncertainty about MMR. But there's no reason to use bonus pool as a proxy to measure this uncertainty, as it is already explicitly measured by sigma, and sigma increases with inactivity and should instead be explicitly taken into account, as suggested in section 3.2 (quoted below) of the OP. In fact, section 3.3 explains how bonus pool can be revamped and done correctly. Unfortunately for WoW, it seems that they've fallen into the same flawed system as SC2.Suppose it takes about 30 games for the ladder system to calculate a player’s MMR to within an acceptably small uncertainty of 50, any more games would just cause uncertainty to fluctuate a little around 50. Alice joins the ladder on Jan 6. By Feb 20 she has played 90 games, just enough to consume her bonus pool, giving her 1500 MMR with an uncertainty of 50. Bob joins the ladder on Feb 20 and plays 30 games that day, ending with 1500 MMR and the uncertainty about his MMR would also be 50. Bob will have fewer points than Alice because he is 60 games short of consuming his bonus pool. But for the purposes of accurate ranking, there is absolutely no reason why Alice should have more points, since they both have equal MMR and equal uncertainty about MMR. Note that a decay system does not face this problem. Therefore, activity as measured by consumed bonus pool can be a bad proxy to uncertainty about MMR.
They further say:
Now they make the treadmill argument, while failing to see that the bonus pool creates the treadmill by causing your rank to continuously fall each day you log in. However, unlike SC2, WoW does not have a crazed and fanatic obsession with ranks. So in that sense, the suggestion in section 3.3 of the OP to deemphasize ranks because of the bonus pool is already in WoW.We believe this adjustment to the Team Rating formula will have a similar benefit to the “rating decay” that some of you have been asking for, but will feel more positive – rather than feeling like you must keep running just to stay in place (i.e. keep your current score), players that continue participating will be rewarded with higher Team Ratings. As a season wears on, this should also make upper brackets more active as well.
And again, they make the completely fallacious claim that the bonus pool is a psychological feel-good decay system. This is an argument already debunked in section 2.3 of the OP. Conveniently, they leave out the fact that the bonus pool distorts ranks (explained in section 2.2 of the OP).
They've also made the claim that this is a good way to encourage activity. But as explained in 2.4, it's a suboptimal way of rewarding activity. In fact, all the sections in the OP on bonus pool deserve to be requoted, because it directly counters every argument they've now made for adding the bonus pool to WoW.
And how the fix the bonus pool:2. The Flaws of the Points and Bonus Pool System
In this section, I will explain how the bonus pool system fails as a positive psychological gimmick and distorts points and ranks.
2.1 Bonus Pool is not a catch up mechanism
Blizzard claims that the bonus pool is to help casuals keep up on the ladder.
This is completely wrong. The bonus pool perpetually inflates the points of more active players, which causes your rank to fall and continually requires you to play on a daily basis, even to maintain the same position. This creates a "treadmill effect”. If Alice and Bob are equally skilled and Alice becomes less active, then her points will diverge from Bob's. The bonus pool doesn't help Alice close this divergence, it’s the cause of this divergence in the first place. It’s the problem, not the solution.Q. What is the Bonus Pool and how are bonus points acquired?
A. The Bonus Pool is an accumulation of points that every player receives whether they're online and playing or not. They're essentially used as a means to help give a player a catch-up boost if they haven't played in a while. The pool does have a cap, but it increases slowly until the end of a season.
If there were no bonus pool, after approximately 30 games, players will hit their MMR and fluctuate around this point unless there’s a legitimate change in their skill. This is what happens when bonus pool is used up anyway. In this case, players would stagnate at their true skill level, meaning that there would be no need to catch up. Thus, having no bonus pool system would be far more helpful to players catching up, since they won’t have to chase a moving target. So Blizzard is wrong about bonus pool and their justification for it is completely fallacious.
2.2 Bonus Pool distorts ranks and increases the time taken for points to self-correct
The bonus pool system causes points to inflate until the season lock, so that the ladder never really stabilizes. Suppose that Alice and Bob have used up their bonus pool and are ranked 10 and 15 respectively. Alice is more skilled than Bob. The next day, Alice doesn’t play, so she falls to rank 14 because other players have used their bonus pool. Bob uses his bonus pool moving him to rank 12. Bob is now erroneously ranked higher than Alice, until Alice and the other players use their bonus pool to increase their points and bump Bob down. Until the season lock, this situation is perpetual.
The bonus pool also obstructs and hinders MMR from self-correcting, because a player’s points cannot decrease until bonus pool is used, even when skill legitimately does. For example, suppose Alice plays actively, she has true MMR 1000, and bonus pool is given at the rate of 50 per week. In week 5, Alice will correctly have 1250 points. Now suppose that she takes a week off and her true MMR has dropped to 800 (e.g. she had a real life injury, or forgot how to play, etc.). Then in week 6, her correct points would be 1100 (1300 minus 200 MMR for loss skill). But, she’ll still be stuck on 1250 points, which wouldn’t decrease until the bonus pool is used up. Even without the HotS change where lost points are absorbed by the bonus pool, Alice's points will on average change very little, until the bonus pool is spent. Hence, Alice's points have been distorted to be erroneously higher than is correct, with adjustment only happening after the bonus pool is used up. In contrast, adjustment would be immediate had there not been a bonus pool system.
These distortions mess up ladder ranks.
2.3 Bonus pool is not a “feel good” decay system, nor a psychological reward
Some apologists of the bonus pool system claim that it’s all about positive psychology. Bonus pool prevents stagnation by letting points increase even if skill and MMR plateaus. A decay system is defined as one that deducts points at the end of each week where the player has not played enough games. Instead of a decay system where players are punished for not playing, the bonus pool system "rewards" players for playing.
At least that's what it tries to be. Bonus pool was seemingly designed with the same philosophy as WoW’s rested XP system. Back in WoW's beta, instead of punishing players by reducing 50% of XP gain when they've played too much, Blizzard doubled all XP and made rested XP a psychological reward by having it always give a 100% bonus.
But such logic cannot be applied to a ranking system, where one player's gain is another player’s loss. Every day you log in, you see your rank fall because of the treadmill effect. Accounting for the fact that other player’s bonus pool causes your rank to continually fall, obviously the reward of increased points is symmetric to the punishment of falling ranks, it’s self-defeating, it cancels itself out.
Thus, these positive psychology arguments are also completely wrong. However, the bonus pool system has replaced the traditional decay system. In this section, I’ve shown why bonus pool is a flawed decay system for the purposes of positive psychology. In 2.4, I show why it fails as a decay system for the purposes of accurate ranking.
2.4 Bonus pool rewards activity in a needless and suboptimal way
Another common argument is that the bonus pool allows for the ladder to reward activity without rewarding mass gaming. While it’s a good idea to encourage activity for the sake of getting people to play the game, this should be done with a levelling system, not a bonus pool system, because the latter distorts ranks as explained in 2.2.
The bonus pool tends to increase the points of active players. But for the purpose of accurate ranking, why should activity even matter?
If we were psychic and simply knew the skill of each player at a given moment, without needing any games to be played, then we would only use this knowledge for ranking, i.e. in an ideal world ranking will be 100% skill based. However, we don't completely know someone's skill at a given moment, unless they play. This is the only reason to consider factoring activity into points and ranks, as higher activity is usually a good proxy to a higher probability that the player's MMR is correct. To the extent that we have good knowledge of a player's current skill, activity should not matter for the purpose of ranking.
This means that ideally, we want to minimize the weight given to activity as a factor, subject to the constraint that the player is active enough to give a reasonably good estimate of his current skill. For example, to have accurate ranks, decay systems that penalize players after a week of inactivity are superior to the current bonus pool system, because they reduce the weight given to activity as it doesn't matter as long as you play a little each week. The bonus pool system, however, requires that you be active always, every single day, so does not satisfy the above-mentioned criteria. Therefore, decay systems result in significantly less distortions than bonus pool systems, particularly for active players.
Suppose it takes about 30 games for the ladder system to calculate a player’s MMR to within an acceptably small uncertainty of 50, any more games would just cause uncertainty to fluctuate a little around 50. Alice joins the ladder on Jan 6. By Feb 20 she has played 90 games, just enough to consume her bonus pool, giving her 1500 MMR with an uncertainty of 50. Bob joins the ladder on Feb 20 and plays 30 games that day, ending with 1500 MMR and the uncertainty about his MMR would also be 50. Bob will have fewer points than Alice because he is 60 games short of consuming his bonus pool. But for the purposes of accurate ranking, there is absolutely no reason why Alice should have more points, since they both have equal MMR and equal uncertainty about MMR. Note that a decay system does not face this problem. Therefore, activity as measured by consumed bonus pool can be a bad proxy to uncertainty about MMR.
In fact, it’s completely unnecessary to use activity as this proxy, because the system already measures it directly and it’s called sigma. Therefore, bonus pool is flawed because it factors activity into points and ranks, when there is no reason for activity to matter since what we ultimately want from it is uncertainty about the player’s MMR, which is a number the system already knows.
3. Fixing the Points and Bonus Pool System
Above we have identified 4 goals of bonus pool.
(1) As a catch up mechanism.
(2) To prevent points from stagnating.
(3) To encourage playing more games.
(4) To allow activity to be factored into points and ranks.
It is logically impossible to achieve (1) for any serious ranking system, as explained in 2.1, so this goal will be ignored. So far I have shown that the bonus pool fails at all of these goals, except (2). But worse than failing, I have shown that bonus pool distorts points and ranks thereby screwing up the ladder. In this section, I suggest how to design a ladder system that achieves all of these goals, while only distorting points and ranks to the smallest possible extent.
3.1 Only encourage playing and reward activity through the leveling system.
3.2 Make a more accurate ranking system by ignoring activity and explicitly including sigma
Note that goals (2) and (3) already address psychology. Therefore, in addressing (4), we are purely concerned with accurate ranking. To explicitly account for uncertainty about MMR, instead of indirectly using consumed bonus pool as an imperfect proxy, points should simply be set to MMR-1.96*sigma (possibly scaled so that the numbers fall into a reasonable range), which is the lower bound of the 95% confidence interval for MMR. Then the current bonus pool system becomes unnecessary. This is essentially what is done in TrueSkill (it uses 3 instead of 1.96).
Here sigma is the uncertainty about MMR, which is initially chosen so that points are equal to 0 for new accounts. It should increase at the end of each week if the player has not played enough games, reflecting the fact that we are less certain about a player’s current skill if he has not recently played.
Currently, the ladder system ranks by points that converge to MMR as long as the player is active enough to spend bonus pool. The proposed system converges to MMR as long as the system gets increasingly sure about the player’s MMR.
3.3 Do the bonus pool correctly: make points converge to MMR-1.96*sigma significantly reduce bonus pool, make bonus pool consume at a faster rate the more bonus pool a player has, give out bonus pool weekly not hourly, update the ladder weekly not in real time, deemphasize ranks and emphasize points
So far we have addressed (3) and (4) without needing the distortionary bonus pool system. The only way to address stagnation without some sort of bonus pool is to increase every player’s points every hour, regardless of their activity. This is not a completely terrible idea. However, this section explains how bonus pool can be redesigned to address (2), while minimizing the distortionary and treadmill effects that are caused by the current system.
The only remaining legitimate reason for bonus pool is to prevent stagnation. Firstly, the suggestion in 3.2 should be slightly amended so that points converge to MMR-1.96*sigma instead of being precisely that as having both a decay system and bonus pool system doesn’t make sense. Next, bonus pool can be significantly reduce, from about 110 per week in HotS, to 20 per week. Additionally, the more bonus pool a player has the faster it should be consume. For example, if you have 100 bonus pool, getting 12 points for winning should use, say, 24 bonus pool, if you have 200 bonus pool, it should use 84 bonus pool. Note that these changes only make sense when implemented together with the suggestions in 3.1 and 3.2 that have already addressed the need to encourage activity and account for uncertainty about MMR. Hence, these changes to trivialize bonus pool have only the purpose of preventing stagnation and nothing more.
These are positive changes because significantly reducing the bonus pool would significantly reduce the distortionary and treadmill effects it creates. Allowing bonus pool to be consumed faster when players have large bonus pools partly addresses the problems in the second example in 2.2 and the “Jan 3 vs Feb 20” example in 2.4. It also partly addresses (1), but no changes in any serious ranking system can (nor should) entirely fix (1).
In addition, bonus pool should be given in bulk, once weekly, instead of in small amounts each hour, and the ladder should only be updated at this time, instead of in real time. Updating the ladder once a week will significantly reduce the treadmill effects since players will no longer see their rank perpetually fall due to other player’s bonus pools. But more importantly, these changes will mostly eliminate the distortionary effects that bonus pool has on ranks as explained in the first example of 2.2. In that example, Alice is more skilled than Bob. She doesn't play for a day and falls below Bob's rank as a result of Bob's bonus pool. So the ladder ranks have become wrong. Now if the ladder were to update only once weekly and bonus pool were changed as I've suggested, then Alice would be able to get back ahead of Bob, before the next ladder snapshot. If she didn't, it would be because she was inactive for the week, so it could be justified that her rank should fall as a small penalty for the chance that her skill has decreased due to prolonged inactivity. However, such an argument cannot be applied to the current bonus pool system because Alice would not lose any skill due to not having played for one hour or one day. The skill lost for 2 weeks of inactivity is far more than 14 times the skill lost in 1 day of inactivity.
Lastly, to further reduce the treadmill effect and to maximize the benefits of having prevented points from stagnating, ranks need to be removed from the matchmaking page and the score screen. Instead, points should be emphasized, as they no longer stagnate. Ranks should be kept in the ladder page in the profile. They are critically important for competition in a competitive game.