|
On April 08 2013 09:59 dsjoerg wrote:Someone on Reddit asked: Show nested quote +I'd be interested to see winrates vs T once there are large numbers of widow mines out. Watching GM streams, it feels like the rate is around 10%, and the mines are allowing (relatively) mediocre players to beat Top8 Zergs nearly every time..... I thought the answer would be of interest to you all. GGTracker has 961 Masters TvZ HotS 1v1 Ladder replays. In 616 of those, the Terran produces 10 or fewer widow mines, and wins 49% of the time. In the other 345 games, the Terran produces more than 10 widow mines, and wins 59% of the time. Of course this doesn't exactly prove that Widow Mines are OP, but at the very least it suggests that getting/preventing mass widow mines is an important part of TvZ. More stats: + Show Spoiler +Interesting findings from the above stats: - mass widow mines is much more prevalent in TvZ than in TvT or TvP
- in TvZ, mass widow mine usage increases with league
Here are Masters 1v1 TvZ gamesAnd here are Masters 1v1 Ladder TvZs with at least 10 widow mines at 15 minutes
flaws in this analysis:
sample size too small- 59 games at gm level? laughable
sample doesn't represent the community- do you actually know how many different people played those games? out of those 650ish games Idra could have played like 60 games, who has very particular skill characteristics that distinguish him from the other players. statistics don't work that way. this was just an example to prove why it's flawed. a player could build 100000 thousand widow mines if he can't use them well he won't win more games.
sample not homogene- the games are very varied in terms of length, player skill, etc.
we don't actually know if they kill anything at all with those mines, they could be just for muta defense etc. in a significant portion of these games widow mines probably didn't achieve anything at all.
bottom line, what does this little research prove? that in the sample players who build more than 10 widow mines, they won more games. in that sample and nowhere else.
does it prove that widow mines make you win more? NO
|
On April 08 2013 14:53 tenklavir wrote:
But that is exactly what happens (win % decreases as WM% increase) using the data that you just posted.
Overall win% is 58%. The win% at the high end is not significantly different from that.
Regarding the regression, did you do a weighted regression? There are many more observations at the low end, a weighted regression should be used to take that into account. Excel won't do it for me, and I don't appear to have the right python modules installed on my laptop here 
|
On April 08 2013 14:59 Novacute wrote:
Is OP going to have a heart attack when he does this exact thing with Marines?
LOL
Master TvZ Marine % of Active Army @ 15:00
WM% # games # wins win %
0%-20% 228 128 56%
20%-40% 254 150 59%
40%-60% 230 131 57%
60%-80% 35 21 60%
Sorry for crap formatting, can't format right now as I'm having a heart attack  OK time to go to bed.
|
On April 08 2013 15:03 dsjoerg wrote:Show nested quote +On April 08 2013 14:53 tenklavir wrote:
But that is exactly what happens (win % decreases as WM% increase) using the data that you just posted.
Overall win% is 58%. The win% at the high end is not significantly different from that. Regarding the regression, did you do a weighted regression? There are many more observations at the low end, a weighted regression should be used to take that into account. Excel won't do it for me, and I don't appear to have the right python modules installed on my laptop here
I confused as to why you are even interested in a weighted regression when the data you posted is so lousy to begin with? Especially when in the same breath you post
If widow mines sucked badly enough, then win% would decrease as WM% increased.
Just let this die. You know the data means nothing and a WLS will change very little in your favor.
|
I think its safe to say that there is insufficient evidence to make any statistical based balance claims on the Widow Mine.
|
On April 08 2013 14:10 dsjoerg wrote:Show nested quote +On April 08 2013 12:47 petered wrote:
A much better metric would be either percentage of army resources in mines Great idea, thanks! ![[image loading]](http://i.imgur.com/8Y75XGl.png) So here we are looking only at Master level TvZ games that made it to at least the 15:00 mark. And then we measure the resource value of widow mines in their army at that time, compared to the total resource value of their Active Army. Mostly this data reinforces things that "everybody knows" about widow mines -- they don't suck, they're not necessarily OP, T loves em vs Z, especially at the high level. That 10-15% row is interesting and we should keep an eye on it as more data rolls in. If I were a betting man (and I am!), I'd bet on that row remaining the highest winrate, but for it to move closer to the rest.
I think you should put this data OP, it is revealing.
|
I found the statistics interesting. Thank you.
That he found a difference means it could be worth looking into (doing more extensive research/statistical collection).
|
I am pretty sure we can do this with ravens and difference after 10 ravens will be much higher(RAVEN OP?)
P.S:I am not being sarcastic, i would like to see if this is the case.
|
I never build any widow mines in TvZ and have a 67% winratio in the matchup (EU mid master). Is Terran now generally OP?
I would really be careful with such stats.
|
|
|
Of all the TvZ's where Zerg got more than 2 Zerglings, they win ~50%.
Zerglings OP. Nerf pls.
|
Nice work dsjoerg, for showing what can be done with GGtracker. Interesting stuff, and the thread is a pretty fun read too.
For me the next big one to test is the siege tank in all three MUs ^^ (don't use the 15 minute marker though, as there are siege timings at the 10 minute mark, the 160 push, and then there's the ''get rid of sieges, broods are out'' after the 20 minute mark).
|
On April 08 2013 16:12 TeeTS wrote:
I never build any widow mines in TvZ and have a 67% winratio in the matchup (EU mid master). Is Terran now generally OP?
I would really be careful with such stats.
I don't think that Terran is overpowered. I would say that Zerg got used to the passive playstyle too much, and don't know how to deal with the new stuff. I am pretty sure that even if Blizzard doesn't do anything, we will see different games in next few months.
On April 08 2013 16:23 DidYuhim wrote:
Of all the TvZ's where Zerg got more than 2 Zerglings, they win ~50%.
Zerglings OP. Nerf pls.
That Logic is quite flawed, even though I know you were sarcastic, you could at least try.
|
i wish moderation was a bit harder on shit replies to an op who put effort into trying something new with ggtracker which can be used to research interesting stats
|
He received shit replies because his study is biased and was only conducted to give credence to balance complaints.
|
Did you think that maybe when Terran players get ahead, they experiment more with Widow Mines? I know that I tend to play around with new HotS units a lot more when I feel that I'm sufficiently ahead.
|
On April 08 2013 16:12 TeeTS wrote:
I never build any widow mines in TvZ and have a 67% winratio in the matchup (EU mid master). Is Terran now generally OP?
I would really be careful with such stats.
well, that's why we want to use bigger sample sizes than "1". However flawed they may be, the OP uses the best he can to achieve that.
|
On April 08 2013 15:33 BronzeKnee wrote:Show nested quote +On April 08 2013 14:10 dsjoerg wrote:On April 08 2013 12:47 petered wrote:
A much better metric would be either percentage of army resources in mines Great idea, thanks! So here we are looking only at Master level TvZ games that made it to at least the 15:00 mark. And then we measure the resource value of widow mines in their army at that time, compared to the total resource value of their Active Army. Mostly this data reinforces things that "everybody knows" about widow mines -- they don't suck, they're not necessarily OP, T loves em vs Z, especially at the high level. That 10-15% row is interesting and we should keep an eye on it as more data rolls in. If I were a betting man (and I am!), I'd bet on that row remaining the highest winrate, but for it to move closer to the rest. I think you should put this data OP, it is revealing.
such a magnificient display of ignorance.
On April 08 2013 14:28 Entirety wrote:Show nested quote +On April 08 2013 14:22 tenklavir wrote:
A quick basic linear regression analysis of Master TvZ Widow Mine % of Active Army @ 15:00, testing Win % as a function of WM%, yields an Rsq = 0.097.
Any of you geniuses arguing for the validity of this kind of nonsense want to guess what that means? Merely 9.7% of the variation in winrate can be attributed to a variation in the % of resources devoted to Widow Mines. In other words, the correlation is lousy.
:p
|
On April 08 2013 11:25 BronzeKnee wrote:Show nested quote +On April 08 2013 11:19 opterown wrote:On April 08 2013 11:17 BronzeKnee wrote:On April 08 2013 11:14 opterown wrote:On April 08 2013 11:13 BronzeKnee wrote:On April 08 2013 11:06 opterown wrote:On April 08 2013 11:04 BronzeKnee wrote:On April 08 2013 11:00 opterown wrote:On April 08 2013 10:57 dsjoerg wrote:On April 08 2013 10:48 Emzeeshady wrote:
Also, we already knew Widow Mines are really good vs Zerg. This isn't all that informative. Well there ya go. OP is either wrong or obvious. Or both! In other news, here's the >10 mutas stats: well with the muta analysis, have we not reached a similar conclusion for ZvP? 49% with less than 10 mutas, 63% with more than 10 mutas for master level. the GM level is even greater (but with less sample size, of course). so we should also discuss how mutas are very strong in ZvP, stronger than widow mines are in TvZ nice stats, btw :D commend you on that! So, you just admitted that his analysis of Widow Mines was correct. And this also explains why every Zerg build Mutalisks in ZvP. However, the reasoning is pretty simple. In WoL, Storm would eat away at Mutalisk packs, while in HotS a single storm isn't crippling because they can move off and regen. Thus Protoss players are forced more and more into Phoenixes, which is similar to when a Protoss player forces a Zerg into Corrupters, you can switch quickly into ground units and just win. when did i say it was correct? i said: if you claim that widow mines are strong against zerg, i expect a similar claim for zvp. no such claim has come out. the fact that mutas, another tier 2 unit has even a greater differential in stats should be a bigger concern. therefore by following your line of reasoning, we need to nerf mutas before we nerf widow mines :D i commend his work and effort, not his analysis or conclusion I'm going try and explain this as simply as possible. Reading the OP, we find that people who make 10+ Widow Mines in TvZ have a 10% high winrate based on GGtracker statistics. That is the analysis and the conclusion. Though the OP states "Of course this doesn't exactly prove that Widow Mines are OP, but at the very least it suggests that getting/preventing mass widow mines is an important part of TvZ" It basically says nothing. So now you state: On April 08 2013 10:30 opterown wrote:
i bet if you compared games with over/under ten mutalisks, or over/under ten high templar, etc you would get similar stats And then someone states such statistics. You commend the work and effort. By commending the work and effort, you are basically saying "hey look you did a good job collecting statistics" and that is opposed to doing a bad job, in which the statistics are basically wrong. Thus, since the analysis and conclusion were based entirely on the collection of statistics and he makes no inference on whether or not Widow Mines are OP, and then you commend the very same process of statistical collection, you have agreed to the conclusion of the OP which is: "that people who make 10+ Widow Mines in TvZ have a 10% high winrate based on GGtracker statistics." cool, so you get some winrates. i'm not attacking those numbers, i'm saying the conclusions that a lot of people are going to make are invalid. EXACTLY! Thank you! But remember this, those conclusions that argue that Widow Mines are not overpowered now have to be independent of the statistic presented! Otherwise they are invalid! and whenever did i say the statistics were wrong? i claimed that a) analysis of over/under 10 of other units would get similar stats b) you cannot conclude anything significant from this so i don't quite get why you are specifically talking to me about this I caught you in an logical inconsistency and I'm pressing it to prove my point, that statistics don't lie, only those who abuse them. You stated the conclusion of the OP was wrong "when did i say it was correct?... i commend his work and effort, not his analysis or conclusion", but the conclusion was the statistics. Then you backtracked "i'm not attacking those numbers, i'm saying the conclusions that a lot of people are going to make are invalid" As I stated, the numbers were the statistics, which was the conclusion. Then you say that other people are going to make invalid conclusions, which is exactly what I'm fighting against.
Your ignorance about statistics is unbelievable. It's not just a question of interpretation of numbers but also how to present them and how to generate them. You can lie with numbers and they are not unquestionable facts which just need to to interpreted correctly. In this case clearly the numbers themselves are flawed because of how they were generated and not just the interpretation.
|
Even IF we we start to treat this data as statistically viable (quite a long shot) you still have to include the natural noise in the form of sqrt(N), where N ist your sample size. So in 300 games you have a base noise of roughly 17 which coverts to 6% variance in winrates. With your 100 sample size it is even 10% variance in winrates (on a basis of 60 it could be 50, or 70 within assumable gaussian distribution ). So, someone stop this thread please  .
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/JIgXeQb.png)
