How Competitive Skill Rating Works (Season 11)

In addition, previous season rank is considered even during placement matches. If you dropped due to playing solo past season, you are stuck at low ranks - cannot join friends or the new groups (vicious circle).

Overwatch’s competitive matchmaking can be considered to be broken especially in comparison to Heroes of the Storm (which has solo ranked in addition). It makes no sense to consider previous seasons.

But you’re in his thread? If Kaawumba’s not allowed to respond to you, who is?

Believing it and doing it are 2 different thing.

I was tell by a GM that i can be low master but i’m not. Overwatch is not just mechanical skill. You need game sense, communication, ult management, team play, strategy.

1vs1 in OW are situational and not the main point of this game. If you are really better than your current rank, you will clim because you will give an advantage to your team and so have more chance to win.

With my current level, i win around 23.7 point for a win and -24.3 for a lose with a winrate at 54%, i will clim arround 150 sr per 100 match. But my winrate will lower as i clim since the level of game will be harder for me.
Soo i need to work on my weak point : team com, target priority, shield management (as a tank) and some other point.

If you cann’t analyse and critic yourself you won’t clim and always think that the system is forcing your rank.

It’s a question of mind set.

1 Like

I still keep recording game data. It’s third season now. I solo queue only, no grouping up, no LFG.
And I think, I’ve found something interesting.

There are lots of posts on this forum, telling the same story - once you win several games in a row, you will start getting complete morons on your team, and become unable to win games. Well, I’ve discovered proof.

Here are two graphs - difference between teams SR and resulting SR movement. Unfortunately, I wasn’t absolutely consistent in recording teams SR, so there are gaps in the graph.

https://imgur.com/a/Wwb56Fd

Note the oscillation of team SR difference. Some irregularities are in place due to randomness of player’s availability, but the tendency cannot be overlooked. The tendency is:

SR team difference function over game number is almost a sinus function. The amplitude can vary, but the function is still the same.
There are huge jumps of amplitude, and these jumps are exactly the times when you lose, and can do nothing about it, or you win and the enemy can do nothing about it.

It has been 68 games, and the pattern is there. It does not seem random. It looks like matchmaker actively constructs these oscillations. I’ll keep recording and see if there are any changes.

It is also interesting, that in this season I got 33 games, when my team had less SR, and only 19, when my team had more SR. If I was more consistent in recording, I probably could’ve got correlation between teams SR difference and winrate.

All data is here: https://docs.google.com/spreadsheets/d/1TBdpG3ahtD31QZ0Xn6HMygxruuIM1285n2cHCNHsNwg/edit#gid=1950589399
Kaawumba, could you please add data from this season to your model?

The standard test to check for periodicity is to take a Fourier transform of the data and look for peaks, but that is difficult because of the missing values. If I interpolate the missing data as you did, I see periodicity. If I replace the missing values with zeroes, I see no periodicity. But since we are trying to determine whether or not there is periodicity, we want to be careful to not use analysis methods that can distort the answer. From this test, I learn that interpolation is dangerous in this case, and may make you see things that aren’t there.

A different test is to use an autocorrelation function (which measures whether the differential at game A is correlated to the differential at game A + x), which is forgiving of missing data. But in this case there is not nearly enough good data, and the error bars came out way bigger than any putative signal.

If you want to know more, I can go through this more carefully and show my work, but I want to be sure that you are interested before I do the significant amount of work required.

As a highly relevant aside, note:

This says that even if the game was buggy (or it is hard to find fair matches in your SR/platform/region) and fat fingering the win/loss scale, you would get more/less SR to compensate and there should be no net bias against you.

Lack of data is evident. I wish there were more people to contribute.
I will continue to gather data, although the frustration of having winrate of 38% interferes. What will be sufficient sample size for autocorellation function?

I revel in my losses, i dropped from diamond to low gold the last two days just playing nonstop and playing casually. Been fun, alot of salt in this game, everybody hates everyone with a passion.

Its kind of a new game for me, reach diamond and then drop to bronze as fast as possible to restart on my great trek back to diamond.

For the signal to be greater than the error bars (for any amount of data) there actually needs to be a signal. Since I highly doubt that the effect you describe is real, I don’t expect you to ever see a signal.

If you mean that you no longer care about SR and just play for fun. Sure, you do you.

If you mean that you intentionally throw down to bronze so that you can stomp baddies and get a false sense of progression, then I hope you get banned.

1 Like

There isnt much else to do on the ladder than do what i just said, once you get above diamond the salt really kicks in. Catch a bad losing streak and just follow it down then work your way back up. I feel its the way the game is intended to be played.

1 Like

I still wonder - do you notice the regularities I speak about? Does it look like a pattern to you?

It looks like a random walk to me. But that’s what I’m expecting to see. I’ve taken too many science classes to believe a pattern without a significance analysis.

1 Like

Speaking of biases :slight_smile:

I’m not sure why you made the chart that you are referring to the way you did. You made a bar graph, but then put a moving average line on top of it. It’s no wonder that a moving average will look like it’s oscillating around random ups and downs.

If you take that chart and change it to a “smooth line chart” and remove the trendline you lose all “sinusoidal” appearance while looking at the raw data rather than a moving average of the data.

There was a post awhile ago that listed a number of games with a 50% win rate and exclaimed “Look at the streaks!”

I went to www.random.org and put the same number of W and L as he had.

The random result was actually MORE streaky. Had the longest streak beat by 2. I didn’t farm for a streak, it was the first one up. Unfortunately, I did all the work and didn’t actually post it for some reason…stupid phone.

WLWWLLWWLWLWLLLLLLLWLLLWWLWWWLLLLLWWLWWLWWLLLWLLWLWLLLLLLLWLLWWLWLLWLWLLWLLLLLWLLLW

WWLWLWWWLLLWLLLLWWLLWLLWWWLLLLWLLWLLLWLLLLLWLLWLWLLWWWLLWWLLLWLLLWLLLLLWLWLLLWLLWWL

Without looking at your data, would you be able to determine which was random and which you had an effect on?

I think the what you see in your data can be best described by this section:

Platinums, % 20.43%
Throwers, % 6.45%
Smurfs/newbies, % 11.83%
Trolls, % 0.00%
Idiots, % 18.28%
Unloosable, % 5.38%
Unwinnable, % 18.28%
Silvers, % 21.51%
Nice 1
Intense 0
Total great games 1
Great games, % 1.08%

You have 1% “Great Games”. Since what a constitutes a “Great Game” is highly subjective the fact that you track all the many ways you can have a bad game but allow yourself to think you have a great game in so rare of an occasion says a lot about how you feel about any data you’re going to be getting.

You know, what the real problem is? No one bothers to collect data. Everyone just trusts - blizzard, his or her point of view, it does not matter.
Gather some data, and let’s analyze it. My sample is flawed, because I am not always able to record it properly.
But. Do you find it random, that teams sr difference changes up and down? Srsly? The graph is not power curve, not straight line, not hyperbole. It is periodic.
As for different problematic categories, well, yes, there are way more ways to spoil a match, than to make it nice. The only subjective category among mine is “idiots”. And I myself am that idiot to other players. It’s an opinion.

Periodic means that it goes up and down on a repeated, timed (where the “period” comes from") cycle. I don’t remotely see periodicity visually without the moving average. I mean, it goes up and down, but I wouldn’t look at it and think it’s regular. It’s actually closer to a straight line, if you scale it to a meaningful scale. 40 SR is essentially the same skill level and you’re well within that.

Let me interpret the following quote for you and how it relates to what I said:

I had you remove the moving average line. That line was “interpolating” the data, basically putting data in where there actually was none. What it also does (at least on your graph) is it REMOVES the numbers that are more random.

Kaawumba’s computer needs to fill in the data with something, but our eyes don’t. Where his analysis can’t handle the gaps your brain can, but of course your brain can’t filter out the noise as well either.

If you stop creating and removing the data with your moving average (and look at the graph as I described above, without any added lines) then what you see LOOKS a lot less periodic and starts to look a lot more random. I posted it here, raw and scaled for perspective: https://imgur.com/LgHvJTS where the altered graph looked like this: https://imgur.com/FKr3fQA

I think you’ll note that none of the “waves” are on the same “schedule”.

If there was a LOT more data maybe there would be some periodicity that a computer could find in there, but if that was the case then your eyes CERTAINLY wouldn’t be able to see it. Rather than LOOKING periodic and BEING periodic, it would LOOK random but BE periodic.

I guess it could LOOK periodic but BE random, but there’s a word for that, apophenia, like seeing shapes in clouds, etc. It’s not real. You should always do do due diligence to make sure the pattern you’re seeing is real.

TL;DR the only reason it appears periodic at all is because you were adding and removing data with the moving average. It may still be periodic, but it would never LOOK more periodic than it currently looks to a person.

Diagnosing by forum post is rather questionable at best, and seeing patterns where there are none is common in perfectly healthy humans.

1 Like

Lol…that wasn’t really my intent. We all see shapes in clouds. Apophenia isn’t a psychological condition, that’s just what it’s called. We all do it, of course!

Severe versions of it can be bad. One experimental early test for schizophrenia is to have people listen to people incoherently talking. If you “hear” sense out of the random noise you are (apparently, I’m not a pro in this) more likely to later become schizophrenic.

The sound they use is…unnerving. Not white noise, just random chatter.

https://www.sciencedaily.com/releases/2007/10/071024115306.htm

Anyway’s, I’ll make it more clear that pattern seeking isn’t necessarily abnormal behavior.

so how long do you have to play without leaving to go down one tier. If I leave once, I’ll get a 10 minute ban. How long, or how many games do I have to play for my next ban to also be 10 minutes? Same with 8 hour ban?

Once I get an 8 hour ban, how long or how many games do I have to play for my next ban to be 8 hours and not 24?

Blizzard hasn’t said, and testing would be painful and probably against the terms of service.

1 Like