Rising Veteran

Re: Lets Talk About 'Skill-Based Matchmaking'?

I'll post here my reply from another thread:

The problem, as I see it, not in the SBMM as it implemented now, but in how the statistical distribution of player skill works in the Apex legends game combined with gameplay balance itself.

First, let's remind ourselves how do the normal Gaussian distribution looks like:

It's the way it works in the ideal case - "average" is in the big middle, with relatively even distribution of skill and probability of success, and on the sides - the best and the worst cases. The problem is, that in apex legends case the spline can (and probably do) look different:

So in our case, the player base can look like that (cause the Gaussian distribution is not the only one normal distribution possible):

Look on the bright red spline. In this case, your skill moves you into the far right side of distribution in insane hyperbolic progression. Why does it happen? Probably the problem is in the game balance itself. There is a huge difference between skill levels. People who can put headshots and who can't. The long TTK makes it quite impossible to outsmart a good shooter - so the skill variations are very limited and leading you to be a mediocrity or a champion - from the statistical point of view.

And if our player base actually looks like this, there is no possible way the statistical algorithm can make it fair for good and bad players alike - we need some fundamental balance changes that would redistribute skill levels among players.

The question of why the distribution favours new twink accounts remain, but possibly "lifetime" achievements have more impact on the final algorithm than the limited sample of fights - which is understandable from the probability theory's point of view - player definitely can have "unprobable" (in our perception) chain of victories or defeats - the overall statistics still prevails. And the old main accounts have a long story of highly successful fights in the random lobby.

1 Reply

JiboTV
6 years ago
@ilya_lizard wrote:
I'll post here my reply from another thread:
The problem, as I see it, not in the SBMM as it implemented now, but in how the statistical distribution of player skill works in the Apex legends game combined with gameplay balance itself.
First, let's remind ourselves how do the normal Gaussian distribution looks like:

It's the way it works in the ideal case - "average" is in the big middle, with relatively even distribution of skill and probability of success, and on the sides - the best and the worst cases. The problem is, that in apex legends case the spline can (and probably do) look different:
So in our case, the player base can look like that (cause the Gaussian distribution is not the only one normal distribution possible):
Look on the bright red spline. In this case, your skill moves you into the far right side of distribution in insane hyperbolic progression. Why does it happen? Probably the problem is in the game balance itself. There is a huge difference between skill levels. People who can put headshots and who can't. The long TTK makes it quite impossible to outsmart a good shooter - so the skill variations are very limited and leading you to be a mediocrity or a champion - from the statistical point of view.
And if our player base actually looks like this, there is no possible way the statistical algorithm can make it fair for good and bad players alike - we need some fundamental balance changes that would redistribute skill levels among players.
The question of why the distribution favours new twink accounts remain, but possibly "lifetime" achievements have more impact on the final algorithm than the limited sample of fights - which is understandable from the probability theory's point of view - player definitely can have "unprobable" (in our perception) chain of victories or defeats - the overall statistics still prevails. And the old main accounts have a long story of highly successful fights in the random lobby.
Sorry again for taking so long to respond to everyone, I figure taking forum time in doses is good. This is an interesting perspective I had only partially considered. I am aware of general distribution theories and have applied some to the problems I perceive as well. I am unsure whether I could go into a deeper discussion on this topic because I have limit data. With the data I do have and the experiences I have collected (Myself and others.), I am still inclined to believe SBMM is the source for at least some of the issues I describe. I think the TTK in relation to the number of bullets in a stock magazine feels fine and not underwhelming or overbearing for players on average with fairness to the ceiling those weapons could reach. I am not sure I understand how your are applying your distribution theories to suggest SBMM may not be the source of some of my issues. I will say I cannot deny that if the skill curve was rounder that SBMM may not be as bad as I perceive it in some areas. That being said, disparity between and within varying skill levels is to be expected and has a larger impact dependent on the averages within each skill level. For perspective, if the average 'predator level' skilled player has a k/d of 1-4, the higher end of that spectrum sits with 4-8, but their overall damage differences on the higher end would likely tend higher in percentage differences when compared to kills.
I feel like I have been distracted from the points I was making though, it would be appreciated if you would humor me and explain a little more clearly how what you said refers and applies to the points I made in my original post. Trying not to get lost in the sauce here.