I’m not a believer of the match-maker being rigged, but isn’t the premise “SR closely chases your MMR up and down” inherent nonsense?
If so, why bother wasting the god equation on a game when you could just hit the stock market or lotto with it because that is essentially what this statement equates to.
The only way SR could chase MMR is by the match-maker knowing the future.
Prove me wrong.
PS. Isn’t more correct to say that SR lies somewhere between the MMR from the last match and the MMR of your most current match?
SR and MMR are only confirmed to disagree significantly during decay and DCs. Even though MMR is hard to get a hold on (because it is invisible) we can see SR getting pushed towards a particular value, not by selecting opponents and allies to rig matches, but by changing how much SR is won/lost after a match: 3000+ Skill Rating Data and Analysis (now including DCs).
It is the wording more than anything that bothers me and the preposition implies a reversal of cause and effect.
For SR to chase MMR, as this is currently worded, MMR would be the cause. But how could it be the cause when the outcome of the match is yet to be determined?
It can’t ever be any other way. For MMR to be chased it must always be in a position of N-1 in the series with SR always being at N. (where N is the number of matches)
The wording of the premise implies that the match maker always makes a correct prediction for the current match.
After the match MMR is updated based on the results of the match (win or loss) and The AlgorithmTM. Then SR is upated to move closer to MMR. No need to mess with causality or have super human levels of prediction.
Ah ha! I finally figured out why a vid I have mentioned before was no where to be found on the internet, it’s from a Blizzcon panel 2016 and I’m guessing gets scrubbed off sharing sites. So the best one can do is a transcript.
but when Jeff talks about miscalibrated skill ratings and “more of a chase that needs to happen” then odds are that chase system is in place at all times. Sure in the example he is talking about it’s a placement whoops. Yet at the same time they are talking about knowing from the MMR a player is low skill and just not giving them as much SR, being “because we [the system] know that they don’t actually belong there”.
So I’m guessing that system is in place and running for everyone.
If you put in huge work and the system thinks your MMR is higher, it chases with more SR. If your MMR is trash, your SR chases backwards by giving less for wins.
I thinking that’s why people who get boosted/carried or one trick a hero that falls out of it’s power level end up crashing so hard back down earth after a patch.
Sure, that explains the mechanism, but does not address the fault in the initial wording and its corresponding implication.
Chase would imply there is a noticeable delay between the 2 events occurring.
Using the same terminology it would essentially be just as valid to suggest the average of a series chases the last value in that series.
Chase really isn’t the appropriate term to use here as it implies an independent agent and a first pass would seem to imply the only variable with non-deterministic behavior in the equation is the result of the match itself.
Well that’s the why some of us have the theory that MMR moves slowly, SR moves quickly. The MMR number acting like an anchor point to either higher or lower levels of play. Helping to stop a player from being punished too much for huge losing streaks or gaming the system with stacking for easy wins.
Also why so many say “Well play 100+ games and then we will how you do on your new account” when a low SR players buys and places higher.
but again, the best we can do is nerd reverse engineer all of this from random scrap of data online. Don’t get too hung up on certain words.
No one but blizzard really knows the details for sure.
If a win results in an increase and a loss in a decrease and for SR to be closer to MMR which everything seems to indicate. How could SR increase with an MMR anchor having a lower value?
If anything it would seem to indicate MMR would have to vary with every match (and not slowly) else you would have cases where you lose SR even if you won?
‘Anchor’ is perhaps the wrong term to use here, but if it were to vary at a slower rate then surely it is an anchor at which point the whole system is flawed?
Isn’t the true metric of the capability of an Overwatch player, simply how often they win? (Not all qualities can be easily measured with only a subset of the data eg. charismatic but mechanically retarded leader?)
I’m glad you were able to find that transcript. The only “new” piece of information is that MMR is a number that looks “weird and mathematical”. I don’t know why Jeff feels the need to mystify it. It’s probably just a floating point.
I still consider the jury to be out on whether or not MMR moves slower than SR in all cases, or just after decay/DC.
He talks a lot about MMR being different than SR due to miscalibration, but I suspect that was more of an issue during seasons 1 to 3 than it is now.
This a word salad, and the terms are not being used properly. Series’ don’t chase their averages, they converge to their sums.
But then, no one ever accused Jeff of being a mathematician or a programmer. When I say “I don’t have that problem” I mean that I strive to understand people as they mean to be understood, not that they need to talk like a textbook.
“We know by their MMR they are actually a lower skilled player.”
This is the strongest quote I’ve seen that MMR can vary strongly from SR outside of decay (as decay would require the opposite statement).
I agree though with Kaawumba on both of these points:
I would agree MMR for most players is now consistent with SR, and that after 13 seasons the “calibration” certainty assists this. This quote doesn’t change the idea that SR and MMR can go out of whack, or my theory on how that causes the bizarre win/loss streaks.
That being said, this transcript doesn’t quite read right for me. For some reason it reads more like a memory of what someone thought was said.
Keep in mind that 3000+ Skill Rating Data and Analysis (now including DCs) puts limits on how much SR can diverge from MMR without it being obvious. From the decay plot, the difference in points gained on victory vs points lost on defeat is (SR - hidden SR) * 0.1274, or 6.37 if SR - hidden SR = 50.
If you read through all of it, not just the snippet about MMR, it looks pretty authentic to me. There is some fuzzy language, but that’s probably because it was originally spoken rather than written.
That is really awesome to see it quantified - the benefit of 3000+ being without PBSR allowed for some good work there. I’ll take some time to read through that thread in full.
I think it’s more that the MMR is just more volitile, while SR is limited to a specific range.
Like, MMR updates every match based on how you performed, regardless of the outcome.
SR updates only in one direction based on the outcome.
So if you perform well, your MMR will go up, but if you lose regardless your SR will still drop, so you’ll lose less SR that game to sort of mitigate that difference.
LIkewise, in the second game, say you perform well again, but this time you win, so your MMR not only rises, but the SR gains get a little boost as it tries to play catchup.
That’s what I feel like they mean when they say it tries to chase your MMR, that if there’s a difference, the system will try to boost you up to where around where your MMR should be, but SR will still naturally “lag behind” simply by how the system is set up.