Skill rating change
Having had recent bad luck with the match making system, I’ve ended up going from Aurodium 1 down to 4. I’m a little under 6M GP, and every player I’ve gone up against in the last 2 events has been 7.5-9.5M. I’m not quibbling about that, that’s just the way it works now.
What I am wondering though, is why GP difference doesn’t seem to factor into skill rating changes? It seems fairly obvious to me that if you win against someone with much higher GP, you should get a little more bonus skill rating. And if you lose to them you should be less penalised. But the data doesn’t support this, and I can’t find any pattern on how it’s worked out.
I lost 42 rating to someone 9.4M. I lost 37 to someone at 7.3M. In my last match of this event, I clawed back only 36 to someone with 8.9M. On top of that, I also lost 10 rating in between seasons, how did that happen?
What I am wondering though, is why GP difference doesn’t seem to factor into skill rating changes? It seems fairly obvious to me that if you win against someone with much higher GP, you should get a little more bonus skill rating. And if you lose to them you should be less penalised. But the data doesn’t support this, and I can’t find any pattern on how it’s worked out.
I lost 42 rating to someone 9.4M. I lost 37 to someone at 7.3M. In my last match of this event, I clawed back only 36 to someone with 8.9M. On top of that, I also lost 10 rating in between seasons, how did that happen?
0
Replies
The skill rating change is highest in the first match of each GAC season, and reduces each match as the season progresses.
I don’t see why GP should be taken into account in rating changes.
I don’t see how you don’t see this.
I do "see" the underdog argument, I just don't agree with it.
One thing I DID come to realize about this system is that it helps prevent rematches. If you beat an opponent in one round, and then you lose the next while your previous opponent wins, you will still not have the same SR. So in order to get a rematch with an opponent, you need to have a specific sequence of wins and losses for both of you, or else just some how wind up in a very small pool of SR ranges.
As for having SR adjusted based on power differences, the measurement nerd in me would be fascinated by trying to come up with a latent variable model that would take into account GP differences-- but there are so many other variables to include, such as participation and mods. You could say something like mods is an artifact of skill, but roster management skill can also be said to be different from (albeit correlated with) battle skill. So now maybe you are thinking of having two skill ratings, and are looking at multidimensional models just to try and disentangle some of these variables you are using. But ultimately, setting up some of these models (because they would have to be estimated from observed data first) might prove to be beyond the scope of what the company wants to do for this. It would be a model specific to this one game, and thus probably not appealing enough to commit resources to.
Whatever GP-based model they MAY want to implement might just end up with some drawbacks that are similar in magnitude to those of the current system. Namely, one issue with using a raw GP difference is that not all of your GP is used due to the limitations on teams. I have relic toons that don't see the light of day in GAC even with full clears on my part. So GP is an imperfect measurement of the difficulty of a battle. It's one of the reasons why when GP WAS being used for matchmaking, it was only a selection of the top units based on division. Maybe you could say the SR adjustment could also be based on top 80 or whatever, but again, they would still have to do the research into how that affects the shuffling of rosters. Whatever they have going on now (aside from shrinking K1) seems to be mostly what they want. I've found that my own longterm win-loss ratios seem to fall within a reasonable range of the touted 50%.
The game mode DOES reward you more banners for undersized wins, so in a sense, you could say it already does reward you for taking down your opponent while using less GP. Underdogging IS rewarded, it is just at the individual battle level, and not the whole-roster level.
So while I like the idea of a model estimation procedure that rewards you more for difficult matches, GP alone is not an indicator of a difficult match. Sometimes your higher GP matches are your easier ones, so you might actually be rewarded LESS for beating them under a system like that. And once the model converges, there should in theory be no such thing as more or less difficult matches, as you always have a 50% win rate.
And as a business model, CG is simply not going to reward smaller rosters. One of the major points of the redesign was to prevent people from sandbagging their rosters. Ultimately, roster growth is what they want to encourage, and is the platform for the entire game.
As I suggested few month ago it’s better to reorganize it to Year based tournament with current system where at the beginning players sorted by GP and then climbing higher as the can to win great year prize. And then it starts again with sorting by GP and year of new tournament.
In this case no reason to lowering GP because it matters only for 1/12 matches. And every year will gave some new experience than just 50% winrate. Also this will remove GAC squish questions.
Like if you 5M GP - you couldn't be dropped from Bro Ligue, 8M GP - you couldn't be dropped from Chrom Ligue, etc.
Such matches in Carbonite is just slap in face for starter accounts: