Salvaging the Current Color Scale

[Enaris 215]

Being part of a Training Clan, I'm well aware of the fact that people outside the 10% are well aware of their numbers, and their place in the "pecking order".  Now, you may well argue that M-I-T tends to attract a certain subset of lower end players (the upwardly mobile), but I've seen any number of times the excitement that people get when they manage to pass some number which causes their color to change. It becomes something of a intermediate goal as they go along in the process.  I do think that's valuable, and why I think that a system which recognizes differences in the bottom 75% is of overall value to the entire WOT community.

[bjshnog 4,447]

I can't believe you just compared World of Tanks to architecture 101.

 

Yep, I did. Average player in WoT compared to a purple is pretty much the same as an average person trying to build a house compared to a professional with decades of experience.

 

If nobody above those average people existed, people would say they would be pretty good at building, since nobody else does it, but because there are "purple" architects, there are resources for learning that everyone can use, and the average person just doesn't do that.

 

So, more or less, you can say that both of these average classes are not qualified, or incompetent, in education terms.

 

EDIT: By the way, I don't care if people take offense to this. All I care about is that it's a logical comparison, which as far as I can tell at this point, it is.

[Xelos 62]

Ill recommend ROYGBIV spectrum based on cumulative bell curve, understanding the distribution is skewed. It is the most intuitive in my opinion.

R:lowest 2.49% cumulative percentage

O:2.5% to 15.8% " "

Y:15.9 to 49.9% " "

G:50% to 84% " "

B:84.1 to 97.6 " "

I: 97.7 to 99.8 " "

V: highest, > 99.8% " "

[MaxL_1023 1,295]

Yellow and Green would be too wide in terms of actual performance. The bottom of the category would do less then half the damage/kills of the top. The idea is for people in the same color grade to perform somewhat similarly in battle, facilitating the use of XVM.

[Angry_Ray 114]

Ill recommend ROYGBIV spectrum based on cumulative bell curve, understanding the distribution is skewed. It is the most intuitive in my opinion.

R:lowest 2.49% cumulative percentage

O:2.5% to 15.8% " "

Y:15.9 to 49.9% " "

G:50% to 84% " "

B:84.1 to 97.6 " "

I: 97.7 to 99.8 " "

V: highest, > 99.8% " "

Holy shit..... Flashback of Kindergarden and 1st grade initiated.

[Cunicularius 4,543]

 

 Man you found the words I couldn't find when trying to describe what I wanted, thank you.

 

bad@forum when deprived of sleep.

Edited October 1, 2014 by Cunicularius

[Eladamri 25]

I'm generally fine with the current color scheme, but here is my comment for what it is worth.

 

I think where it currently falls down is in regards to newer players.  Several of the newer players that I've discussed the colors with find it really disheartening to be labeled "bad" and lumped in with the bots especially as they get abused in game due to their "bad" color, which they are likely to have for quite some time, even if they are actively trying to improve.  These players are inexperienced, not bad at the game. 

 

If the colors are to be used by the overall community to mark skill and improvement (over normal progression through the tiers), I'd actually suggest splitting black into two colors (possibly around 100 wn8?) so that newer players can see that first mark of improvement that says "yea!, I'm now better than a bot!"

 

Edit: or just have red start at a lower number.

[pauli 229]

If the colors are to be used by the overall community to mark skill and improvement (over normal progression through the tiers), I'd actually suggest splitting black into two colors (possibly around 100 wn8?) so that newer players can see that first mark of improvement that says "yea!, I'm now better than a bot!"

The problem is that wn8 is literally meaningless that low.

[sgtguppie 4]

Is Winning Good? 

Is Losing Bad?

If I LOSE more than I WIN... am I a GOOD player or a BAD player?

Do not concern yourself with the average level of performance among the population for a moment... if I have LOST more games than I have WON, has my performance been more GOOD or BAD?

I don't know about you... but if I played 100 games, and I lost one more battle than I won and you told me that was above average, I'd feel compelled to slap you. 

I would not CARE what the average is, MY concern would be MY performance and MY average will have been MORE BAD THAN GOOD!

 

"Above an average, you say? The only average I care about is my average, and it is NOT good. I am NOT a good player, I lost more games than I won. My performance, even being 'above average' is still poor, isn't it? Then why did you bother to tell me that I am above average? Am I supposed to be PROUD of that? I want to IMPROVE, and you tell me I'm above average! That's not anything to be proud of! I LOSE more than I WIN!"

 

What is the Consideration? What is Average? What is Good? What is Bad?

Average to me is a 1 to 1 Win/Loss ratio... 

Average is less than Good, Bad is Less than Average... but not always.

I don't like this, I think its wrong.

It feels more like pandering to a player base that doesn't want to admit that their achievements are on average more bad than good. 

Most importantly, I think its dishonest, and I won't support it.

 

Does anyone know the average for draws across server?  Mine is 1.8%, which leads me to believe that 48%, by your logic, would be average.

[Dlur 2,350]

@Cunicularious: It's 49% because of draws...

[Never 7,713]

Is Winning Good? 

Is Losing Bad?

If I LOSE more than I WIN... am I a GOOD player or a BAD player?

Do not concern yourself with the average level of performance among the population for a moment... if I have LOST more games than I have WON, has my performance been more GOOD or BAD?

I don't know about you... but if I played 100 games, and I lost one more battle than I won and you told me that was above average, I'd feel compelled to slap you. 

I would not CARE what the average is, MY concern would be MY performance and MY average will have been MORE BAD THAN GOOD!

 

"Above an average, you say? The only average I care about is my average, and it is NOT good. I am NOT a good player, I lost more games than I won. My performance, even being 'above average' is still poor, isn't it? Then why did you bother to tell me that I am above average? Am I supposed to be PROUD of that? I want to IMPROVE, and you tell me I'm above average! That's not anything to be proud of! I LOSE more than I WIN!"

 

What is the Consideration? What is Average? What is Good? What is Bad?

Average to me is a 1 to 1 Win/Loss ratio... 

Average is less than Good, Bad is Less than Average... but not always.

I don't like this, I think its wrong.

It feels more like pandering to a player base that doesn't want to admit that their achievements are on average more bad than good. 

Most importantly, I think its dishonest, and I won't support it.

 

How you judge your own performance is also subjective, and I think you're failing to take draws into account

[Eladamri 25]

The problem is that wn8 is literally meaningless that low.

 

For matches above tier 4 I would absolutely agree.  But for low level matches, where these new players are, there is a difference between a bot/afker/permatroll, and someone that is trying to win and actively working to develop skills and experience through play.  New players are not going to be throwing up huge wn8 numbers (especially at low tiers expected values), and the gap to get from 0 to 450 wn8, assuming normal progression through the tiers, is pretty huge. 

 

I started playing in February of this year and tracking my wn8 around 200 wn8/500 games played. To be brutally honest I think it was personally harder for me to progress from 0 to 200 wn8 than it was for me to go from 200 to 1069 wn8 (my current, which is still way lower than I'd like).  This game comes with a huge learning curve for new players.  From 200 wn8 on I was adding layers to a base level of knowledge, from 0 to 200 I was trying to figure out what that base level of knowledge should be.

 

TLDR: I don't know where the cut-off is, but 450 wn8 seems too high to set as the first obtainable benchmark for new players.

[TheMarine0341 4,972]

TL:DR

We're taking this too personally, instead of a bland objective standpoint. MaxL is being the most objective and focused

 

Full Post:

Average is just that: Average. The last discussion we had as a community on this really opened my eyes as to what percentile Im in (95+) vs. the vast majority of players, which average to around 650 WN8 (and frankly, this helped my rage playing by realizing that those dumbasses were simply doing what the average pubbie would do). And, its bugged me that the color we determined to be *average* was instead assigned to "above average" esq players. Frankly, that really does make *us* look like collective pricks by *personally* deciding that the Average color is assigned to Above, while instead what is indeed average as red, or "Bad." The majority of the users here are WELL above average players, either by recent or overall which skews our view of the scale. 

 

I think the current 11 step can be alright with better scaling, but cutting it down to 7 would be better. But with 11 steps, whatever the average WN8 is determined to have been, it should be Yellow regardless of your personal opinions of how bad those players are. 

 

The bad players dont realize how bad they are, while the good players dont realize how someone can just be that bad. Its a two way street in this regard and we do need to recognize that. 

[MaxL_1023 1,295]

Is Winning Good? 

Is Losing Bad?

If I LOSE more than I WIN... am I a GOOD player or a BAD player?

Do not concern yourself with the average level of performance among the population for a moment... if I have LOST more games than I have WON, has my performance been more GOOD or BAD?

I don't know about you... but if I played 100 games, and I lost one more battle than I won and you told me that was above average, I'd feel compelled to slap you. 

I would not CARE what the average is, MY concern would be MY performance and MY average will have been MORE BAD THAN GOOD!

 

"Above an average, you say? The only average I care about is my average, and it is NOT good. I am NOT a good player, I lost more games than I won. My performance, even being 'above average' is still poor, isn't it? Then why did you bother to tell me that I am above average? Am I supposed to be PROUD of that? I want to IMPROVE, and you tell me I'm above average! That's not anything to be proud of! I LOSE more than I WIN!"

 

What is the Consideration? What is Average? What is Good? What is Bad?

Average to me is a 1 to 1 Win/Loss ratio... 

Average is less than Good, Bad is Less than Average... but not always.

I don't like this, I think its wrong.

It feels more like pandering to a player base that doesn't want to admit that their achievements are on average more bad than good. 

Most importantly, I think its dishonest, and I won't support it.

 

Summary:

 

- Skill and WR are related by some equation which we can't calculate

- The median skill results in the median WR

- The median skill = average skill

- However, the average(median) skill does not result in the average WR

- This is because the relation is not linear

- Therefore, the average player will not have the average WR, as elite players pull the WR up more then bad players pull it down.

- 48% WR is not red, it is yellow. Yellow = average skill = MEDIAN WR = 50th percentile

 

 

No, you still don't understand how the statistics work. In the case of WoT, the average win rate has to be equal to the average loss rate, with draws filling the gap. This is where 49-49-2 comes in. However, the distribution of players is not a pure gaussian. What we are defining as "average" is not average in terms of wins, but average in terms of performance. We are defining this as the 50th percentile, or the MEDIAN win rate. This is the point where half the players are better, and half are worse. This turns out to be slightly lower, near 48%. 

 

We are assuming that skill is a gaussian - by the central limit theorem it has to be at least fairly close. The skill gaussian is not measured in absolute units - skill is a subjective quality which can only be measured objectively as a relative value. This means that the skill gaussian can be accurately approximated as a perfect bell curve centered around the average player's skill. As this gaussian is pure, the average player skill is also the median skill level. Now, apply this to the distorted win rate gaussian.

 

The result proves that the function relating player performance or win rate to skill is not linear. It can only be modeled as either a polynomial, exponential, logarithm or likely something ridiculous which can't easily be defined. I chose a modified exponential model to fit it to WN8 and to maintain a fairly constant relative damage ratio, but based on the win rate categories it is obvious that the actual relation is a lot more complex then anything we can figure out off the top of our head. However, one thing we can be certain of is that a player of average skill will end up having not the mean win rate, but the MEDIAN win rate. Half the players are worse and will have a lower win rate - how much lower depends on the exact relation. Half are higher, but how much higher is also indeterminate at the moment.

 

The reason WN8 and WR have a stretched right tail is because of the nature of the skill relation. Stretching it means that at the high end, an extra 20% of relative skill represents a larger jump in win rate then what you see lower down. Going from 2.0 to 2.2 means more then 1.0 to 1.2. The tail clips off at extreme values due to diminishing returns (no matter how good you are, you can't win them all) but is also stretched again due to platooning, like the WN8 graph is stretched due to tank selection. Basically, it is impossible to actually directly model a true skill vs WR/WN8 relation because it is an inherently chaotic system.

 

However, you can make statistical inferences which allow a metric to be used in a general fashion, especially away from either extreme. That is what our model does. It is not perfect, but it is getting better all the time.

[Dlur 2,350]

I'm only about 80% comprehending what Max just said, but I say we let him pick out how many categories and where the breakoff points should be, and then let a female pick the colors.

[Rexxie 10,788]

I'm really disappointed with people in this topic. I'm disappointed with the sentiment OP taps into ("we worked rly hard so dont change it!"), I'm disappointed Bobi got negbombed, I'm disappointed that a lot of people here are showing their elitist colors, and I'm disappointed that some people have put almost no thought into why things should be the way they are (cunic, angry_ray, bj, I'm looking at you). I'm sad to see MaxL has to explain the things he does, because it stems from a total lack of understanding of the use and purpose of the color scale among the general playerbase.

 

Hopefully we can just drop this entire topic and get back to the other color scale topic, because I honestly don't want to even have to skim over some of the posts here. As Bobi said:

 

The simple fact is that the average player in this game has already been defined as a poor player (worthy of red color status).

 

Stopped reading after this. (oh who am I kidding, I didn't have the willpower to stop here)

 

Will never open this thread again.

[Aquavolt 1,587]

I'm really disappointed with people in this topic. I'm disappointed with the sentiment OP taps into ("we worked rly hard so dont change it!"), I'm disappointed Bobi got negbombed, I'm disappointed that a lot of people here are showing their elitist colors, and I'm disappointed that some people have put almost no thought into why things should be the way they are (cunic, angry_ray, bj, I'm looking at you). I'm sad to see MaxL has to explain the things he does, because it stems from a total lack of understanding of the use and purpose of the color scale among the general playerbase.

 

WHAT ABOUT ME???

[Rexxie 10,788]

you're a flower

[Servios 467]

I'm disappointed Bobi got negbombed

So you support a person, who couldn't bother to take the time to read the entire thread, before forming an opinion, which he felt the need to post immediately. Regardless of how wrong you feel this thread is, the threadstarter took time writing it, so the least people could do would be to take the time to read it, before saying anything.

[RichardNixon 835]

No, you still don't understand how the statistics work. In the case of WoT, the average win rate has to be equal to the average loss rate, with draws filling the gap. This is where 49-49-2 comes in. However, the distribution of players is not a pure gaussian. What we are defining as "average" is not average in terms of wins, but average in terms of performance. We are defining this as the 50th percentile, or the MEDIAN win rate. This is the point where half the players are better, and half are worse. This turns out to be slightly lower, near 48%.

 

The primary reason that the mean and median account win-rate are lower than (1-draws)/2 is simply that good players play more than bad players. For a trivial drawless example, consider a population where each 60% player with 10k battles is balanced by two 40% players with 5k battles. Total wins/battles would be exactly 50%, but the mean and median account win-rates would be lower.

 

Note that this applies even if the skill distribution is symmetrical: If good players play more than bad players, you end up with a right-skew winrate distribution. Average players are more likely to see good players than bad players, and hence their winrate drops below the (1-draws)/2 mark. The effective average skill that players see in games is much higher than the account average.

 

 

We are assuming that skill is a gaussian - by the law of large numbers it has to be at least fairly close.

 

You can define "skill" however you like, but I have no idea why you'd define it as a gaussian given that contribution clearly isn't. I assume you're thinking of the central limit theorem rather than the law of large numbers, but that only defines the distribution of the sample mean, not the population.

 

WN8 exaggerates the contribution skew because the input data wasn't platoon-filtered. Eureqa was understandably confused that the 65% winrate players weren't doing much more damage than the 60% players.

[Angry_Ray 114]

I'm disappointed that a lot of people here are showing their elitist colors, and I'm disappointed that some people have put almost no thought into why things should be the way they are (cunic, angry_ray, bj, I'm looking at you)

How much thought should I really have to put into a system that uses colors to convey skill Rexxie? Are a bunch of colors really that big of an issue? It sounds like some people are putting far too much thought into what things should be. Should a person be called an elitist because they refer to someone that is red/black as bad or even terrible? Please, suggest a term that should be used instead.

 

Sorry, I don't see a reason to get uppity about colors or classification. I merely post my opinion and that's what it is. I'm not going to use my time to ponder a response to something that I sincerely believe is completely functional.

[MaxL_1023 1,295]

The primary reason that the mean and median account win-rate are lower than (1-draws)/2 is simply that good players play more than bad players. For a trivial drawless example, consider a population where each 60% player with 10k battles is balanced by two 40% players with 5k battles. Total wins/battles would be exactly 50%, but the mean and median account win-rates would be lower.

 

Note that this applies even if the skill distribution is symmetrical: If good players play more than bad players, you end up with a right-skew winrate distribution. Average players are more likely to see good players than bad players, and hence their winrate drops below the (1-draws)/2 mark. The effective average skill that players see in games is much higher than the account average.

 

 

 

You can define "skill" however you like, but I have no idea why you'd define it as a gaussian given that contribution clearly isn't. I assume you're thinking of the central limit theorem rather than the law of large numbers, but that only defines the distribution of the sample mean, not the population.

 

WN8 exaggerates the contribution skew because the input data wasn't platoon-filtered. Eureqa was understandably confused that the 65% winrate players weren't doing much more damage than the 60% players.

 

Almost everything related to large human populations falls under a Gaussian. Intelligence, Body Size, Proportions - as skill is a human factor it should follow suit.

[Never 7,713]

Richard, do you have any issue with the scale as defined

[Angry_Ray 114]

Richard, do you have any issue with the scale as defined

In my HONEST opinion (not that it matters), the old scale was better than the one we currently have. The old one gives more leeway in terms of skill curves, and doesn't reduce the average to red.

 

Both scales are functional, but I think that the one we had before was "better". Just my opinion.

 

Edit: I do like the one in that thread though

[RichardNixon 835]

Almost everything related to large human populations falls under a Gaussian. Intelligence, Body Size, Proportions - as skill is a human factor it should follow suit.

 

Intelligence isn't directly measurable, any more than tank skill. IQ tests map their highly non-gaussian results onto a gaussian.

 

I don't know if physical parameters are gaussian. Given that they're highly dependent on nutrition, I'd be surprised if they were truly gaussian given that wealth is not.

 

 

Richard, do you have any issue with the scale as defined

 

I have issues with using overall win-rate for anything at all, but I don't suppose rWinC will ever catch on. An issue with the current scaling is that high-tier solo players get attacked for having winrate "lower" than their WN8, so I'd personally aim on the low side for winrate. 60% is much higher than an average tier 10 solo 2350 WN8 player will achieve.

 

Secondly, if you use the term "average" in the table, you should define exactly which average you're referring to. I can invent one if you need it.