SBMM DOESN'T WORK. KILLER STATS FROM 100 MATCHES TO PROVE.
Comments
-
You didn't say "Otz proved what we already know."? Weird that I can scroll up and see that you said it. Okay, keep being upset and defensive if that's what you like.
0 -
I did not say that I knew its possible to win almost every game with SBMM.
Jesus dude, I'm not being defensive, you're having a conversation with you, yourself, and Irene.
0 -
But he did prove in his video that it's possible to win almost every game. To which you replied that he "proved what we already knew", but then said that you didn't know it's possible to win almost every game with SBMM. So he did show something that you didn't know. Dude, it's obviously getting too hard for you to track your own thoughts. Just give it a rest until you sober up.
1 -
1 game as each killer does not a sample size make. I am not the first person to tell you that, it shouldn't have to be said again. I am also the very same person that had to correct you and point out that Otz does get steamrolled by survivors enough that most people can recall those matches. You know, except for you.
and again, I did not say: "but then said that you didn't know it's possible to win almost every game with SBMM" - I have no clue why you keep harping on this.
So again, I have no clue why you are so desperate to drive home fictitious points that you are arguing with yourself.
0 -
Look, I don't know why you are so upset, but further conversation with you would be against forum rules (see thread crapping part) as all you are doing is getting argumentative and personal, instead of address ng anything to the topic of discussion. Have a nice day and I'm sorry, but I don't have patience to explain simple things multiple times.
1 -
So long as you understand why your data is faulty so that you can do better the next time around, you can bow out anytime you like :) - Best of luck.
0 -
My data is not faulty. I explained what "statistically significant" means at least 3 times. It's not my problem if it makes you so upset. Enjoy the rest of your day.
0 -
And all we see is someone climbing the MMR ladder on several characters, without understanding or direction, or even enough data points to justify their confirmation bias.
Enjoy the rest of your day as well!
0 -
You really need to reconsider calling yourself "We". I thought "statistically significant" was a simple concept. Apparently not for everyone. Oh well.
0 -
You did not play enough games to infer that your victories are attributable to MMR not working. You did not control enough variables. You did not demonstrate a solid enough understanding of even basic concepts like killers having individual MMRs. While almost all of us already know MMR is not working, because it's clearly valuing queue times as a more significant factor above all else, that does not make your data accurate or useful.
And by we, I mean you can look at the first two pages of this thread where several people point out these flaws, I'm trying not to rehash them in depth because if you didn't read them the first time then, well, you're not going to the second time around either.
We have to stop meeting like this :)
Edit: You can preach statistical significance until the cows come home, but just because you can broken record a false claim, doesn't mean 'we' are going to let you get away with it. 2 + 2 won't equal 5 just because I shouted it enough in the town square.
0 -
Oh yes, you right on this one: I can preach statistical significance until the cows come home, but if you still have troubles understanding it, than the further conversation is pointless. Can't discuss science of statistics, or any other science as a matter of fact with flat earthers.
You can refer to yourself as 'we' all you want. It's just comical. But by all means, if you identify as a Gollum, enjoy it: “We like goblinses, batses and fishes. But we hasn’t tried Hobbitses before. Is it soft? Is it juicy?”
0 -
I will use we so long as the predominant opinion, again, as demonstrated by the other people who have chimed in to say the very same thing, is that you have a shallow understanding of statistical significance.
You're throwing a lot of insults for someone who was accusing me of being the upset one. Also, I thought you were done with this thread? :)
0 -
I primarily agree with you but I will give these guys a fair point and say that 1-10 games on each killer doesn't really work when we all know there is individual mmr
In order to see "accurate" data we really need like 25-50 games played on each killer and look at the averages from that
Would you not agree that to be a more "accurate" representation of the data?
0 -
- I never said I'm done with the thread. I said I'm done with entertaining your threadcrapping, for the simple reason that I respect forum rules:
- "PLEASE AVOID POSTING THE FOLLOWING: Things that do not contribute to the discussion (e.g. baits, etc.)"
- If you have anything to say on topic, we can continue indefinitely.
- There is no predominant opinion. It's just your opinion about what is predominant. Anyway, back to the topic.
You failed to demonstrate any proof "shallow understanding of statistical significance." Feel free to post one after you familiarize yourself with the concept.
"You're throwing a lot of insults" - Am I? When? Are you able to stay on topic or are you going to get upset again and start feeling insulted?
0 -
I think this Otz video just proved that the community doesn't really know what way the game is balanced. Maybe it isn't survivor sided, or killer sided.
When you look at this video, the tournament he referenced, other streamers, your own experience.. you'll soon see that there is too many factors to determine what the balance of the game is.
Maybe the devs are the only ones with the proper stats to truely know the balance of the game.
1 -
Yes, of course I agree that the more data you have the more accurate it is. I plan adding to it. But, what people forget is that individual killer MMR varies only by maximum 100 points. Devs explained reasoning for that before and it makes perfect sense. The variance is so tiny on the scale of MMR (starting value 1800 I think, I could be wrong), that it could not possibly explain my results. Also, as I said before, it's not even about me. Watch Otz experiment video.
My previous was from 50 games. There is barely any change in average kill between 50 and 100 match results. My interpretation is that my MMR is stable.
Average kills at 50 games were 3.02
Average kills at 100 games are 3.01
0 -
Your interpretation of the data is counfounded by a few issues.
Firstly you have a heavy skew toward one end which will inflate total avergae in this case, median kill rate is probably a better measure for the overall kill rate rather than mean. There is also a lot of internal variance (especially sample size) amongst your descrete variable (killers) which will also increase error around the mean and reduce the accuracy of average.
For error reduced mean you'd want to log transform the data and take the transformed mean as this will give a more accurate measure of mid range. Looking at the numbers it will probably still fall on the killer side but likely not so inflated.
This is the problem with interpreting raw data at face value there are often caveates that mar conclusions. Its why statitics, while brutally boring, plays an essential part in data intpretation.
2 -
Exactly! We don't have data to talk about balance. But we can talk about SBMM. I believe good matchmaking can compensate for balancing flaws.
0 -
I've maintained a 60% escape rate as a solo survivor for over 450 games.
0 -
Holy crap I had to wait for 4 pages of replies for this! You are amazing! This is exactly the advice I needed. I'll research how to calculate error reduced mean in Google Sheets. Thank you very much!
1 -
- You don't know what threadcrapping is, well, not if you truly believe that's what I'm doing. Anyone can count the opinions on the first two pages, whether or not you are able to. Yes, you are throwing a lot of insults, including assuming I'm upset, gollum references, and flat earther references. I don't know why you're trying so hard when it seems that you have little interest in anything beyond your own confirmation bias. The beauty of confirmation bias is you do not need our opinion to go on being wrong. You simply go on.
0 -
I also don't get your W/L column. If its a win loss ratio given a win % then values should fall between 0 and 1. Also the raw data may not transform well given the sample size you'd probably want to transformn your win/loss percentage data for overall assessment.
Also forgive typing errors I have a broken hand.
1 -
"Making off-topic posts in a thread on a message board."
Determining that someone is upset when someone is acting like he is upset is not an insult. There are plenty of threads where people discuss how anything DBD related makes them feel. Feel free to create another one, but this thread is not about your feeling.
Please, just follow the forum rules. Do you have anything to add on topic other than discussing your feeling?
0 -
No matter how much you keep parroting it, my posts have always been about you using faulty interpretations to draw incorrect conclusions, and that will always be on topic in this thread.
0 -
I'd say that sounds about right and also that you're a trustworthy source
0 -
Thank you. I converted win/loss column into percentages and also added a chart of matches with given number of kills. You right, percentages are better than ratios. I'm still amazed that we have such helpful and knowledgeable people here. I hope your hand health soon.
1 -
Why don't you follow your own advice and stop parroting how much you are upset about my interpretation and conclusions. I'm pretty sure everyone got it after the first 5 times you said it and you are not adding anything new, other than adding more details to description of your feeling.
0 -
I only had to go so far as until you stopped defending your faulty conclusion (If that's what you truly mean by your last response, I'm sure further posts will tell). I have a kink for not letting people get away with misinformation and bad-faith statistics :)
0 -
There are plenty of forums where you can discuss your kinks and feeling. This is not one of them. This is my last response to you. I gave you plenty of chances to stay on topic. Since you clearly have troubles understanding forum rules, I will not encourage your threadcrapping by further replies. Now, feel free to type another (or a few) parrot reply providing more details about how you are upset and how you feel about me.
0 -
Well, this is only the second time you were done with me. You do you! Just know, if you push the false narratives again I'll be back :)
Fun fact: To test a D20 you'd probably have to throw it about 100 times to get some statistical significance. If you think I'm implying something about the multitude of variables in your test when compared against the clearly insufficient sample data, you'd be right.
0 -
I wish people who do these win/lose logs posted their build.
0 -
Not a problem! But it's pretty standard no imagination "sweat" build for most killers. BBQ for BP on everyone. Most of time I experiment with something fun, I get punished for it. Which one do you want the build for?
My main: (Cannibal) BBQ, Bamboozle, POP, Pain Resonance
Newly favorite Twins: BBQ, Pain Resonance, Pentimento, Plaything
Fun Nurse: BBQ, Ruin, Pentimento, Plaything
Mean pickaboo Spirit: BBQ, Discordance, POP, Pain Resonance
Anti-SWF Nemesis: BBQ, STBFL, Pop, Tinkerer
and so on. As you can see it's all plenty ordinary and boring. One of perks is swapped to Franklins if more than 2 medkits.
1 -
Solo Q or swf?
0 -
The other thing to be cautious about with this type of data is how its interpreted with respect to SBMM.
As a measure of SBMM there is only one consistant variable, you. There is no other data available about your opponents SBMM or even accurate measure of your own given the value is hidden. This creates a good data set with respect to your games but is a form of pseudo-replication on a larger scale because its confined by your personal SBMM and skill.
That means with respect to SBMM there is an N of 1 or a single consistent variable being measured which is you. So basically a single data point is unrealiable for broader interpretation.
What we can say from data like this though is that specifically in your match ups there is an over perfomance of killer. With the un-transformed numbers just looking at it I'd say you are sitting about 0.75 from a basic average calculation (Wins 4k+3K / Total games).
Its probably a significant trend given the numbers, how strong it is will be attenuated by sample size.
What can be gained from data like this,
- You do well as killer for a good range of killers, you've got the empirical data to show it too so Kudos to you. With respect to SBMM its probably not matching you as well as it could given the winning skew. However given the asymetric nature of the game and the reliance on teamwork from one side, a killer victory skew should be expected as the mechanics of killer are more consistant when compared to the variable nature of teamwork required on the survivor side. This is purely theorising but its what I'd expect so I wouldn't call a killer win skew unusual unless it was really skewed (>0.75).
- Unfortunately we don't have whether or not your opponents were teams or solos which is a confounding factor on perfomance. If you can get that info post game then its very valuable as I'd predict teams will positively/negatively drive win/loss ratio by a significant amount. The interaction co-efficient between these measures would be interesting as BHVR reported a 15% increase in wins by SWF.
- The game is not survivor sided, again however this doesn't mean its killer sided either, just with respect to you at your measured SBMM killer overperforms. You can draw conclusions from this but its very narrow and any suggestions on side bias, or buffs/nerfs, are spurious at best with this limitation.
- It adds value to your personal experience. For example "I get more kills than expected the SBMM may not be matching me well" here is my raw numbers to illustrate why I think this. That is valuable because end user experience is a big part of gaming.
What would make this data more valuable for wider interpretation.
100 players tracking 100 games each. Now you remove the pseudo-replication and have a nested measure of win/loss across killers, across players, across SBMM this is the minimum required to make any real suggestions about broader game balance. Taking into account relative SBMM rating and SWF would make it even more valuable.
How could you analyse this data to demostrate you are significantly overperforming as killer and make a solid point about your gameplay?
If we have an expected outcome of 2K/2E or 0.5 then you would compare your observed outcomes win/loss ratio against an expected 0.5 for each killer. A chi square analysis of observed vs expected would give you this. It can be calculated by hand if you are that commited and there are free online calculators where you just punch in the numbers and it will spit out your result, not the most accurate but hey its just for fun. This will tell you if your observed values are signifcantly greater than an expect balanced outcome based on sample size.
I know this is an essay but I just finished pulling apart my behavioural ecology invertebrate data consisting of about 2500 cells of data points nested across 7 field sites. A think tank exercise on game data interpretation was a bit of fun.
Cheers its my mouse hand so its really gimping my gaming.
2 -
I feel like you have low MMR though, due to your experiment anyway.
I'm not trying to trash on you as a player but the MMR only tracks Escapes / Deaths for survivor, so as you said in your comments on your stats a lot of the 1Ks from your stats were from you dying trying to get a totem, if you're the ONLY player dying every other match your MMR is absolutely going to tank.
0 -
Meanwhile 75% of my killermatches are wins for the survivors and plain suffering on my end. The system is definitely broken, I'm far off from this 50-50 balance they're aiming for.
2 -
Thank you for another informative post Professor Pseudechis. I didn't expect to learn about Chi test today. So, I followed this description of test here https://support.google.com/docs/answer/7004263?hl=en
I don't pretend I fully understand it. But it looks like it needs 2 arrays: observed data and expected data. So I compared J column with 0.5. I had to add fake data in blue rows that had no data because I was getting error, so I added fake 4k and fake 0k in each. And the number I get is not what I expected. If I understand correctly, 1 is fair and 0 is not fair. I think I'm doing something wrong, but not sure what.
1 -
If you can, please post a link to video of your average suffering match. I'm sure plenty of people here could give you some tips. I would surely watch it and would try to help you as much as I can.
0 -
I don't think my rig could deal with that very well, the game has become very hard to play for me from a technical point of view since my rig is quite outdated. All I got is a graphics card with 2 GB VRAM, meanwhile you need twice as much just to play games nowadays. And I can't even upgrade since my mainboard is a DDR3 one. I do have ample of regular RAM, though, (16 GB) but that doesn't really make a difference.
The consequence are several stutters and freezes which seriously hinder my ability to keep up with survivors. My framerate caps out at 30 pretty much, most of the time it's below that. I already tried all the tricks with lowering the graphic settings and going as far to diddle around with the config file itself but it didn't help. The constant visual updates are making the game almost unplayable for me.
0 -
Yes, framerate is super helpful in getting easier wins. One trick you may try is to lower resolution below your monitor native, for example to 720p. Change it in Windows settings before starting the game.
0 -
I quit playing Killer
0 -
That comment was mostly meant to be funny.
However, I'm not a great Survivor. I know my job, but I'm nothing special. I'd say I'm painfully average.
0 -
Both, but mostly Solo
0 -
I'd omit any killers you don't have results, chi squares don't deal well with 0 values and a high number of expected values that match will throw off the calculation. They also don't deal well with count data less than 5 in each sample so you need more games with each killer.
Yeah your two factor is going to be your expected outcome of 0.50 and your observed outcome.
Although I may have misled you a little here as the expected outcome is 2K/2E or draw. So you expect a peak in 2K/2E declining with 3-4K or 1-0K. What we have though is the inverse of that with wins 78, Draw 8, loss 24. To fit this against the expected 2k you are going to wind up with a huge chi square value given the inverse nature of the distribution compared to your expected. With that in mind chi square probably won't work very well here.
If we assume a normalized distribution based around a 2k/2e as the mean you could expect a frequency of win 0.25, draw 0.5, loss 0.25, the majority of the games are draws with a balanced probability of win/loss around that.
you'd expect to see an absolute frequncy of... loss 27.5, draw 55, win 27.5 from 110 games or thereabouts.
Observed though is 24, 8, 78 hugely skewed to win with a decline in the middle.
You could say that just looking at distribution of relative frequency that your game isn't balanced around 2k/2e from this sample set of data, or more likely SBMM corrects causing swings between wins and losses with a few "balanced" draws in between and a significant killer bias in outcome.
Post edited by pseudechis on1 -
So minimum 44 matches since he played 22 of the 26 killers and then plus however many matches you'd wish to watch of your mains?
1 -
So a system that counts only kills and escapes doesnt take real skill into consideration? I am shocked
0 -
To be fair, they are basically only nerfing killers. So there you go. It is no secret that mediocre killers eviscerate mediocre survivors. They have been trying to address if for awhile now.
0 -
That of course depends if he would be good with them, I like to learn new stuff. I would not watch many if he would show only basics.
I am able to tell only minimum what I would watch, maximum depends on a player an how interesting those games are.
0 -
Please show us your data.
0 -
Here:
0
