Win/Loss is back! yay or nay! - Page 3

On November 28 2012 00:56 monkybone wrote:


If you are improving your W/L will be slightly more than 50%. As time goes on and your MMR gets more stable your W/L will approach 50%. Is this the information that you want to see?

The problem was that a large amount of people took the statistic as a measure of their skill. When it doesn't really say anything interesting at all.

IMO, Hairy is right and W/L is in principle an entirely useless statistic. He made no false claims.

On November 28 2012 02:19 korona wrote:

I presume you did not read the thread carefully? If people use certain statistics for wrong purposes, such as for a measure of skill, it does not mean there are no other valid uses. Some uses are explained earlier in this thread. And basic progress statistics don't necessary even need special uses as their main purpose is e.g. to count games played, games won, etc (yes there are other win-loss statistics than just the winratio that often is uninteresting). And that itself is valid information.

Yes Hairy and now you made some false claims. Please read the thread before commenting and don't fixate on winratio (win percentage) too much.

On November 28 2012 02:45 Hairy wrote:


I'd like to hear some of those "valid uses" for overall winrate, korona; the only thing I can see overall winrate being used for is:
"Since my data sample began I have a [win/loss] difference of [X]. This means I have [risen/dropped] [impossible to determine] amount of MMR/points!"

It can't even be assumed that those wins and losses are weighted evenly; your opponents' skill will rarely exactly match your own, so the amount of MMR/points you gain/lose will frequently differ between games. Just yesterday I beat my first masters opponent and won 44 (!) points, where normally against my typical (diamond) opponent I would gain half as much for a win.

On November 28 2012 02:45 Hairy wrote:


I'd like to hear some of those "valid uses" for overall winrate, korona;

On November 28 2012 02:45 Hairy wrote:
the only thing I can see overall winrate being used for is:
"Since my data sample began I have a [win/loss] difference of [X]. This means I have [risen/dropped] [impossible to determine] amount of MMR/points!"

It can't even be assumed that those wins and losses are weighted evenly; your opponents' skill will rarely exactly match your own, so the amount of MMR/points you gain/loss will frequently differ between games. Just yesterday I beat my first masters opponent and won 44 (!) points, where normally against my typical (diamond) opponent I would gain half as much for a win.

On November 24 2012 23:59 Hairy wrote:
Plus, you can get a plugin for SC2Gears called 'MMR Stats':
http://www.teamliquid.net/forum/viewmessage.php?topic_id=334561

It provides an estimate your current MMR based upon various factors. Here is a screenshot of my progress over the last 60 games:


The green line is the platinum promotion line, the light blue line is the diamond promotion line, and the dark blue line is the masters promotion line.

On November 23 2012 18:13 Hairy wrote:
If you want to see how many wins you have right now, just take your total matches played and halve it and you'll be pretty much spot on.

On November 28 2012 02:45 Hairy wrote:
I'd like to hear some of those "valid uses" for overall winrate, korona;

On November 29 2012 03:43 korona wrote:

Why are you still fixating on winrate? Didn't we agree that winrate (percentage) is often uninteresting and provides less information the more you play? In many cases when you have reached your typical MMR range + are stable (your MMR is fairly stable without large swings) + your winrate approaches 50%, you cannot deduce much else than that the matchmaking system is doing its job. Thought if the ratio differs much from 50% it is interesting to check why. Actually this is one useful usage itself - to identify deviations (you can spot deviations easier from W-L difference than ratio). Even based on the last sentence claim that 'win-lose statistics are absolutely useless' is false.

Let's see. Basic uses: E.g. if I want to know how many 1v1 ladder games I have played, I can check my games played count. If I want to know how many games I have won, I can check that statistic and so on. These are basic progress statistics that are useful as they provide the information they claim to provide. Even if there were no other uses for these numbers, they still provide progress information such as: 'I have played 70 games, won 38, lost 32 and there were no ties'(btw, these are real numbers for one of my accounts for this season and there are logical explanations why I have won more than lost. And it is not that I have 'improved').

If the numbers are available people can set personal goals based on them, such as 'my goal is to play 100 games during this period of time' (to improve it often requires regular practice. If you do not play, your skill may start to decline). One can simply write the starting numbers down and in the end of that time period count how many games he played. Likewise the other numbers can be used for similar purposes such as 'my goal is to win 100 games during a certain period' (Hmm, actually I should set one of these myself to get 1k wins with every race before HotS with one of my accounts So close now as lowest wins are 921 -> 79 wins needed for that race. So I will need to play roughly 79 * 2 * 3 = 474 games (not taking account my recent results regarding that race) to achieve this as I play random with that account). Of course many goals may not necessary be rational, but if you understand how the ladder & matchmaking system functions you can avoid setting such goals.

By setting goals people can motivate themselves to play more. For example if I in a certain period play worse than usual and thus fall into loss streak that results me a negative win-lose difference of e.g. -15 for that period. I know that my MMR has dropped considerably and that the system tries to match me with 'equal level opponents' based on MMR. This means that my opponents will generally be 'easier' than before the loss streak. If the level from where I dropped was my 'typical MMR range', it should not be too hard for me to climb back. Now I can set a goal to reach '0 win-lose difference' that I will eventually reach (especially due to how the ladder system works - I will win more until I reach my typical level again) and by setting the goal I will be motivated to play more. Different people find different kinds of goals motivational.


I am sure you will agree that if you win a match your MMR increases and if you lose a match your MMR decreases? Thus e.g. if your starting W-L was 123-123 and after the match you have 124-123 your MMR has increased. But based on this we do not know how much the MMR increased as we do not know what the MMR difference originally was between you and your opponent. But as the matchmaker tries to pair you with even level opponents, you can _assume_ that the opponent was such and based on this you can get _rough_ results. Of course this method would not usable if the opponents were 'all over the board' regarding MMR. But the matchmaker seems to generally do a good job and most of the time only pair close MMR range opponents. And as it's fairly consistent, generalizations can be made to get _rough_ results (I never claimed this was exact or even near exact method - Only one possible use for win-loss statistics). I will give you an example by data provided by you, but first lets discuss what that master opponent you faced means.

Max ladder points you can get from a match is 24 and rest are bonus points. Thus you got either 22, 23 or 24 ladder points and rest were bonus points. The ladder points you get are affected by opponent's MMR and your adjusted points. Also there are special rules such as the 'minimum division MMR' that 'artificially' increase the ladder points gained and decreases ladder points lost (this boost is visible especially during the start of the season). Difference of yours and your opponent's MMR did not affect how many ladder points you got. The MMR's were used to match you and to determine how much the MMR's changed after the match. Excalibur_Z:s guide quite nicely explains this both textually and graphically: http://www.teamliquid.net/forum/viewmessage.php?topic_id=195273

It is likely that there was larger than typical MMR difference between you and your opponent, but it cannot be estimated only based on ladder points gained (more data is needed). But just like you did now, you can always check your opponents' profiles and try to spot if there was a large difference (analysis may take time) and take this info into account when doing the estimates. For such matches (case: if opponent's MMR is much larger) your MMR will increase more if you win and decrease less if you lose than in a match against an equal opponent. But matches with large MMR difference are probably quite rare in most leagues at least up to low master MMR range (likely higher. Not sure in which point the population scarcity starts to have noticeable effects and thus the results given by the method become more inaccurate) (on Sea server the population is considerably smaller than on other servers. One could speculate that the matchmaker has to widen its search more often there.)


Let's take your MMR graph as an example (at the moment the accuracy of the relative ratings given by the MMR tool seems generally good). In the graph white (loss) and blue (win) circles are 'good results' and grey circles are estimates (grey ones basically use the same method that I describe). In MMR tool scale you get/lose 16 MMR points when you win/lose against equal opponent. In the tool (for NA & EU) lower border of platinum is 1050, lower border of diamond is 1350 and lower border of master is 1650. Here is a picture of Hairy's post in case he deletes the picture: http://imgur.com/fZZ44)

Now let's calculate how many wins & losses you have in the picture. I counted 38 wins & 22 losses, but these counts may be little incorrect to either direction. So the W-L difference would be +16. According to tool your starting MMR was ~1270. Now let us add 16 * 16 = 256 to it. We get 1526 as result (my recalculation gave W-L of 37-23, which would result difference of +14 and ending MMR of 1494). Now let's see what your ending MMR in the tool: ~1510. And look, the result was roughly the same even if there might be a slight error in win & loss counts. And the method knew nothing about your opponents. It just generalized that all were equal. This also likely (but not necessary) means that the matchmaker made fairly good job during this period and opponents were quite close and/or variations regarding MMR difference balanced each others out regarding the results.

I find the results accurate enough (in many cases you don't need exact results, which cannot be calculated based on the data Blizzard provides especially as we don't know the exact formulas. But a _rough_ estimate is often enough). If you knew your starting MMR level was high plat, you could have estimated based on the W-L difference that your MMR to be mid diamond in the end. And 'low-mid-high-border' accuracy is good enough also in case where you had analyzed your relative starting MMR level without the tool. At the moment most leagues (except bronze & master) are ~18.75 wins wide (against equal opponents).

In platinum & diamond range there are plenty of players to choose from. Thus the matchmaker usually has plenty of 'equal opponents' to pick from (may need to widen the search during 'silent hours'). But somewhere like in the GM range this method would likely give inaccurate results as population is too thin at that level.

Also win-lose difference method is a fast method as you get results immediately (you can do this all the time). E.g If you deduced your relative MMR level by hand based on your profile (with the info blizzard provides), opponent's profiles & opponent's opponent's profiles it would take much much more time (you don't want to do this between every game). Also regarding W-L method keep in mind the longer the time frame (number of matches) the accuracy decreases. These are _rough_ methods that often provide 'accurate enough' estimates. (Of course Not_That:s method could be used to get more accurate estimates. MMR tool is based on it. http://www.teamliquid.net/forum/viewmessage.php?topic_id=332391)


In one of your earlier post you suggested that you can any time 'divide your games played by 2 and get accurate results'. Now let's see how your method works for this time period that is shown in your picture. The W-L in the picture is 38-22 (ratio ~63.3 %) or 37-23 (ratio ~61.7 %) or something very close. Your method would count 60/2 => 30 wins. But this was not the case as you have much more wins for that period. Also you are not in the either extremes of the ladder (exception set by you).


So to sum-up some uses for win-lose statistics (Some give exact values. Some give rough estimates. Some just use the values for some other purposes. There may be more than just these):
- offer basic statistics
- allow setting personal motivational goals (goals themselves may not always be rational, but it is still a valid use)
- allow identifying deviations
- allow more advanced deductions/estimates in combination of other data

It is often easier to give extreme statements such as 'is absolutely useless' and ignore other possible uses & viewpoints. These 'black and white' views are easier to grasp on and it takes less time to think/write about than considering is there other possibilities or viewpoints (for example this post took some time to write, but probably have not spent long enough for proofreading it --> likelihood of errors due carelessness rises). Also even if something is meaningful for one person, another person might not care about it. People tend to ignore things that do not mean anything for them. Also people tend to agree with statements easier if the statement is short and clear, even if it could be false in some situations/viewpoints or even totally false. For example many viewers of this thread will simply skip what I have written, because my explanations are 'too long' and are not in perfect English.

But if extreme statements such as 'is absolutely useless' are made, it is easy prove them false - You need to find only one situation where they are false. Now Hairy as you asked me to provide often very simple uses that anyone should be able think of, can you provide me explanations why do you still consider win-lose statistics 'absolutely useless' or have you changed your mind (games played count, win count, loss count, tie count and anything you can count based on them. Please don't talk once again only about the win ratio). Why for example win count that provides the number of wins is not meaningful (is it e.g. calculated wrong or has blizzard accidentally switched the numbers again)? Why people can set motivational & reachable goals based on the numbers (even if sometimes irrational) if the values don't mean anything? If the numbers are meaningless why people can spot deviations based on them and e.g. identify MMR abusers or deduce fairly accurately how some have e.g. progressed through leagues just by looking at the numbers & other available data? The answer is you cannot invalidate these uses --> Win-lose statistics have meaningful uses especially if you understand how the ladder & matchmaking system functions.

On November 23 2012 17:34 KapsyL wrote:
I am not sure how I feel about this. I dont want other people to look at my win/lose ratio. If only I can see my own then YAY

On December 01 2012 00:19 DuncanIdaho wrote:
As a statistician, who is absolutely IN LOVE with data, the more data, the better, and to this I say, "YAaaAAAaaAAY! FINALLY!!!!!!!1!!1!1"

Those who want to be lied to, take the blue pill to go back to sleep in the Matrix world + Show Spoiler +, or in otherwords, prefer not to get the sad truth, they're ruining it for the rest of us, by pushing their pathetic needs for self-deception on the rest of us who CAN take contructive criticism. Perhaps rather than catering to the n00b crowd and ruining the experience of learning for serious gamers, I say Blizz should either ignore those guys in favor of people like us who don't go crying to mommy when we find out our win/loss ratio isn't spectacular, as is now finally the case again, or perhaps, at worst, give those people an option to turn it off if it truly makes you upset.

Although, all joking and name-calling aside, I think you're doing yourself a diservice by wanting to NOT know your losses. The game gets more exciting as your skills progress, and all new subtle nuances of gameplay open up to you as you play against better and better opponents. AND, in order to improve, you need to know where you're falling short. Personally, I'd like it if Blizzard not only told us our loss ratio, but in addition, win/loss data over the different matchups, over different maps (I also thuroughly enjoy the new addition of the win% on each map, even though I refuse to veto any maps, as this lets me know where to focus for improvement), such that we can even better tailor our practice needs for the goal of getting better.

TL;DR: Don't lie to yourself, embrace criticism, it WILL make you stronger. Wake up, and take control of your life!