Upsets from random external causes in even games between players with a very wide rating gap, would be a source of rating noise that compresses the overall rating range over time.
It also depends on how strongly the rating system rewards/punishes such highly unexpected game results, but if the frequency of such events is low enough, this overall rating range compressing will be small enough to not matter much.
Still, the effect would be bigger at the high and low ends of the rating range. Players with very high or very low ratings may play such lopsided even games more frequently, because there are fewer suitable opponents. So the lowest rated players would have their ratings increased and the highest rated players would have their ratings decreased from this.
Using handicap would avoid this effect.
If handicap were used in most lopsided games (I don’t know if this is already the case or not), it would automatically ensure that overall, ranks are spaced properly.
That being said, I’m not in favour of forcing handicap in ranked play. Handicap should be a choice, but it’s good that is the default now.
I don’t have a strong opinion on this. On a teaching game against a considerably stronger player I ask (and receive) for a 3 stone advantage. Which is, in my very noobish opinion, a natural cutting point. The conclusion of the game is still not really in doubt, but it takes longer for the stronger player to reach a decisive advantage (giving more review material).
Of course, it is very different with ranked games, but I still sense that for the handicap not to distort the game too much, it could max out at around 3 stones. Or may be halve the nominal strength difference and use a halved handicap as a default? I’m nowhere near there, but I imagine that, say an aspiring 2-kyu, really wants to take on a 1-dan without a handicap, but a 6-kyu balks at the thought of doing so without at least 3 stones (unless dictated to do so by the tournament rules).
I do realize that this idea is kinda broken, because I can’t imagine how to make the necessary changes to rank re-evaluation algorithms. Just diluting any changes by some factor is doable, but kinda random.
It’s not just via the (relatively minor effect of) non-skill based results that handicaps help ensure ranks are spaced properly. The very definition of the gap between a 5k vs 1k is that if they play with a 4 stones handicap the game is about* fair. So if they play with a handicap you have good data to calibrate the ranking system. However, if they are only playing even then you end up needing to use winrates, often across multiple steps, to try to recreate that rank spacing, things like 1k beats a 2k 60% (or whatever) of the time, that 2k beats 3k 60% of the the time, 3k beats 4k 60% of the time, 4k beats 5k 60% of the time which has much more variables and sources of error.
* It’s a 3.5 stone difference so white slightly favoured
Do you have any data to back up this claim that this is “very low probability” and “a blip”?
I think that internet disconnection, or a phone call, or someone ringing my doorbell, are pretty much the only reasons why I have lost live games on time on OGS. I don’t know how to filter my game history to get the ratio (live games lost on time) / (total live games), but I definitely would not call it “very low probability” or “a blip”.
if rank system actually works very good and players are normal, then forced 4 stones handicap game between 5k and 1k may be OK
if rank system has ±1 rank problems, then players that need 2 stones handicap in their game, may sometimes get 4 stones
(system may think that they are 5k and 1k, when they actually are 4k and 2k)
always even is better than sometimes too many handicap stones
rank system not necessarily has the ability to fix itself if everyone plays handicap. It may keep working slightly wrong forever even with handicaps.
players themselves may be unusual
someone may have high rank because memorized 10 000 joseki, but bad at other things
and another may be expert at tsumego, but bad at other things.
They would play good even games, but they may have a lot of troubles in handicap games. So on server with handicap stones everywhere they would have low rank and play like a sandbagger each time “even” opponent appears. So forced handicap server would be unfriendly place for players like that.
In France, in irl tournaments, we almost always use “handicap minus 1”, which means that a 1k player would play an even game against a 2k player, and a 1k player would only give 3 stones to a 5k player.
I think such a system, maybe handicap minus 1 or handicap minus 2, would work well for 19x19 OGS ladders.
What’s the problem? How would I even know? All I know is that I sometimes win and sometimes lose. Without handicap I often need 9 stones and get none. Or my opponent does. Sure, there are fringe cases in which handicap will games more uneven. But not on average.
I am against enforcing handicap though, but that’s because my dislike for enforcing is stronger than my liking of handicap.
Since extraneous rare events have been dragged into this discussion, another extraneous factor, far less rare, should be considered: sandbaggers. Although rank manipulation is comparatively rare, sandbagging by the creation of new, non-transparent accounts is rife on OGS. Handicaps help the sandbagger by giving the assistance to the wrong person.
More important is the point made by @stone_defender awhile back, which got some “likes” but no discussion. Using handicaps and starting beginners at the pseudo-rank of 6k or the humble rank of 12k, as the case may be, are absolutely antithetical ideas. In a game without a handicap, the beginner is annoyed by getting crushed over and over, but he may at least realize that this is the trial by fire that any beginner in any game goes through. In a game with a handicap, where he is forced to give a handicap to his 15k (real rank) opponent, for example, he is also crushed, but his annoyance is doubled. Why, he asks himself, do I have to undergo a trial by fire that is abetted by insanity—by being forced to give my experienced opponent a handicap. He understandably leaves OGS in disgust. This is a problem even for handicap as a default, because beginners are unlikely to know how the system works and thus get victimized by the mismatch of handicap and starting rank.
If handicaps are forced, or even defaulted, new players should start at 25k or at a self-selected rank.
Some arguments were made against having very lopsided ranked even games.
Some arguments were made against forcing handicap in games between players that are only a few stones apart.
Perhaps a compromise is possible:
Suppose that ranked games should have an expected winning probability between (say) 15% and 85%. If an even game between 2 players has an expected winning probability outside this range, handicap would be automatically added until the expected winning probability is within this range.
If 9 stones handicap is not enough to bring the game in the ranked winrate range, the game would be unranked.
There would be no opt out for this rule, other than playing the game unranked.
Not sure what you mean. I have no devotion to the current system of starting ranks. dragon-devourer in a thread a year or two ago (which I can’t find now) showed that the current starting point was not required by Glicko-2, and I never saw a contradiction of his assertion and analysis.
Thanks for clarifying the actual AGA rules. I was referring in particular to the OGS implementation. In our Club play on OGS, we use AGA rules by default (on the basis that they are closest to our otb standard of BGA rules). In fact, it was only when someone specified Chinese rules in a handicap challenge without me noticing, that I found out about free placement of handicap stones under those rules on OGS.