<!--quoteo(post=2058802:date=Jan 10 2013, 09:37 PM:name=Brony)--><div class='quotetop'>QUOTE (Brony @ Jan 10 2013, 09:37 PM) <a href="index.php?act=findpost&pid=2058802"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->Salt, try this thought experiment:
Flip a coin 1000 times. Note how many "heads" you get the first 250 flips. Then note how many "heads" you get the last 750 flips. Do you think its reasonable to assume that your percentage of "heads" will be any different in the first sample than the last? And what if you repeat this experiment 100 times, wouldn't you agree that the percentage of "heads" will be very close to 50% in all the samples?
The reason most people argue so strongly against you is not because they are defending NS2Stats, its because what you are doing is questioning a very fundamental concept in statistics and the scientific method. Sampling has been used for centuries and a lot of research is based on it, which has been performed by folks a lot smarter than you and me.<!--QuoteEnd--></div><!--QuoteEEnd--> No I would not expect that at all. Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips. What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page. A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.
<!--quoteo(post=2058806:date=Jan 10 2013, 04:53 AM:name=hakenspit)--><div class='quotetop'>QUOTE (hakenspit @ Jan 10 2013, 04:53 AM) <a href="index.php?act=findpost&pid=2058806"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->No I would not expect that at all. Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips. What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page. A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.<!--QuoteEnd--></div><!--QuoteEEnd-->
This would be my answer as far as it goes for the thought experiment of Brony. The variance factoring in ranging from skill to luck to chance, knowledge and teamwork all pitch in on the random factor. On top of that comes a small thing i'll point out.
Not questioning your thought experiment, because it is validly true, none the less. The factors pitching in to form game win results are not as simple as just a coin toss. Rather than a full 50/50 chance of tails or heads landing, the results of games are not randomly polled, but fixed. consistently. What is happening with the server polling here, is not the first 25% taken of all results taken in is being polled. But randomly across all the results. which are already random. As such, it keeps on increasing the chance that the servers being polled have a more strongly favored winrate for Aliens, rather than marines. Since it's not 'completely' random but fixed, Only a certain amount of servers have ns2stats installed.
it's certainly gnawing my mind though. results are results none the less, but it feels as though when you factor in chance, upon chance upon chance, that polling results randomly across all servers, at a random pace you can get skewered statistics.
If ns2stats keeps spreading though, and getting bigger and bigger, they'll cover more servers which would more accurately set the average. at times i checked out ns2stats it kept telling me i only played like 15 and now 19 games in total. When really i've played more like 200+ rounds. I might be one of the very few players experiencing this, but it feels wrong.
<!--quoteo(post=2058812:date=Jan 10 2013, 01:19 PM:name=Salt)--><div class='quotetop'>QUOTE (Salt @ Jan 10 2013, 01:19 PM) <a href="index.php?act=findpost&pid=2058812"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->What is happening with the server polling here, is not the first 25% taken of all results taken in is being polled. But randomly across all the results. which are already random. As such, it keeps on increasing the chance that the servers being polled have a more strongly favored winrate for Aliens, rather than marines.<!--QuoteEnd--></div><!--QuoteEEnd-->
Why?
If we stick with the coin toss example:
If you have tossed 100 times and got a sequence consisting of 52 Heads and 48 Tails(not 52 Heads followed by 48 Tails but any given random pattern) and then randomly select 25 throws the chance to hit a Heads-throw is 52% each time and the chance to hit a Tails-throw is 48% each time.
Thats like if you have a town with 45% of the people being female, each time you randomly select a person theres a 45% chance to pick a female. So if you randomly pick 500 people chances are you get around 225 females and about 275 males.
Of course a game of NS2 is more complex than a coin toss, but we are not talking about a sample size of 100 but several thousands. And please consider why a sample size of 1000 is enough to accurately foretell the election results in a country with 62 mio. IF your arguments would be right this would be downright impossible.
And for the raindrop example: Given there are 1000 cm² of which 60% are hit by a raindrop and 40% are not hit by a rain drop. If I now randomly select 100 cm² and look at wheter they were hit by a raindrop, the ratio betwenn hit and not-hit will be around 60-40. The pattern of raindrop-hits in this case is completely irrelevant.
<!--quoteo(post=2058802:date=Jan 10 2013, 03:37 AM:name=Brony)--><div class='quotetop'>QUOTE (Brony @ Jan 10 2013, 03:37 AM) <a href="index.php?act=findpost&pid=2058802"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->The reason most people argue so strongly against you is not because they are defending NS2Stats, its because what you are doing is <b>questioning a very fundamental concept in statistics and the scientific method. Sampling has been used for centuries and a lot of research is based on it </b>, which has been performed by folks a lot smarter than you and me.<!--QuoteEnd--></div><!--QuoteEEnd-->
<!--quoteo(post=2058840:date=Jan 10 2013, 02:34 PM:name=VittuLima)--><div class='quotetop'>QUOTE (VittuLima @ Jan 10 2013, 02:34 PM) <a href="index.php?act=findpost&pid=2058840"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->I have 100 hours in NS2. 0 games played in NS2stats. Those "statistics" are worthless.<!--QuoteEnd--></div><!--QuoteEEnd-->
I lived in Germany for more than 25 yrs and was never asked in a pre-election poll.... these polls have to be worthless.
If you have tossed 100 times and got a sequence consisting of 52 Heads and 48 Tails(not 52 Heads followed by 48 Tails but any given random pattern) and then randomly select 25 throws the chance to hit a Heads-throw is 52% each time and the chance to hit a Tails-throw is 48% each time.
Thats like if you have a town with 45% of the people being female, each time you randomly select a person theres a 45% chance to pick a female. So if you randomly pick 500 people chances are you get around 225 females and about 275 males.
Of course a game of NS2 is more complex than a coin toss, but we are not talking about a sample size of 100 but several thousands. And please consider why a sample size of 1000 is enough to accurately foretell the election results in a country with 62 mio. IF your arguments would be right this would be downright impossible.
And for the raindrop example: Given there are 1000 cm² of which 60% are hit by a raindrop and 40% are not hit by a rain drop. If I now randomly select 100 cm² and look at wheter they were hit by a raindrop, the ratio betwenn hit and not-hit will be around 60-40. The pattern of raindrop-hits in this case is completely irrelevant.<!--QuoteEnd--></div><!--QuoteEEnd--> Firstly your comparing statistics effectiveness in predicting a 2 or maybe 3 possible outcome situation in the case of election polls. Similar to a coin toss though you get outliers in polling data...heck the way most electoral systems run they clearly show pockets of support based on geographical location. This can be traced down even further to streets within a suburb (based on socio-economic groupings), most reputable statisticians use a wide sample base to help avoid these occuring though. Its the fact there are only limited actions (toss coin - heads/tails or cast votes - election won/lost), compare this to a game that can be lost tens if not hundreds of thousand different ways, with slightly different actions undertaken. This could be by not using the right starting hive upgrade, or upgrading to exo's before weapons or armour upgrades. Or any of the other countless other decisions.
Have you ever heard of a game called two up? Look into it...might help you see how many variables are in the results of two consecutive tosses of coins.
<!--quoteo(post=2058848:date=Jan 10 2013, 03:13 PM:name=hakenspit)--><div class='quotetop'>QUOTE (hakenspit @ Jan 10 2013, 03:13 PM) <a href="index.php?act=findpost&pid=2058848"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->Firstly your comparing statistics effectiveness in predicting a 2 or maybe 3 possible outcome situation in the case of election polls. Similar to a coin toss though you get outliers in polling data...heck the way most electoral systems run they clearly show pockets of support based on geographical location. This can be traced down even further to streets within a suburb (based on socio-economic groupings), most reputable statisticians use a wide sample base to help avoid these occuring though. Its the fact there are only limited actions (toss coin - heads/tails or cast votes - election won/lost), compare this to a game that can be lost tens if not hundreds of thousand different ways, with slightly different actions undertaken. This could be by not using the right starting hive upgrade, or upgrading to exo's before weapons or armour upgrades. Or any of the other countless other decisions.
Have you ever heard of a game called two up? Look into it...might help you see how many variables are in the results of two consecutive tosses of coins.<!--QuoteEnd--></div><!--QuoteEEnd-->
I'm not talking about "Party A will loose while Party B wins" I'm talking about
Party A: 38% PArty B: 32% Party C: 14% Party D: 7% Party E: 5% Others: 4%
So according to you with geographical, social and what not influences such a statistic HAS to be completely worthless if based on WAY LESS than 0,1% sample size. Still these results are shockingly correct in most cases and never are fundamentally wrong.
The whole point and use of statistics is that you can get a very accurate image of what the big picture looks like on the basis of only a small sample size. The whole point of statistics is to simplify these overcomplicated situations you decribe so that these details no longer are <b>relevant</b>.
I let's not forget that a NS2 match may be kind of complicated, but there are FAR more complicated problems out there that really can't be proven/solved by math but given a VERY accurate account for by statistics.
Take your basic model of a Atom with is electrons. I really can't say where these electrons are. Possibly the electron of one of the atoms within my computer actually is on Mars in this instance - there literally is no way to prove that it's not. BUT statistically speaking I can be quite sure it's right where it should be very close to it's own core. I can even say the same for ALL the electrons doing ther electron thing in my computer. I don't need to know the least bit about any of them. I dont even have to know anything about physics and still this statement is right with a probability of more than 99,9999%. So now the big question is: How big of a sample size was examined before it was concluded, that electrons with a probability far greater than 99,999% are very close to their coresponding core despite them theoretically being able to be more 10000000000 km away from it? I don't think they really looked at 25% of all electrons within our universe, do you?
And I think we can agree on quantum physics being more complicated than NS2, can we not?
<!--quoteo(post=2058806:date=Jan 10 2013, 06:53 AM:name=hakenspit)--><div class='quotetop'>QUOTE (hakenspit @ Jan 10 2013, 06:53 AM) <a href="index.php?act=findpost&pid=2058806"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->No I would not expect that at all. Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips. What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page. A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.<!--QuoteEnd--></div><!--QuoteEEnd-->
Oh dear God. Please please please please go and read an introduction to Statistics, or go on a course, or something.
I'm a crystallographer. I solve the 3D structure of proteins. One of the key features we use to determine whether or not the protein model we create is 'too detailed' to be explained by our experimental data is the use of a free-R statistic. Put simply, we take approximately 5% of our data points (let's say that's 1000 data points, for example). We do nothing with these data at all. They are separated from the 95% of data, on which we base the protein structure model. Then - and this is the cunning bit - we compare the model we've built with these so-called 'free' data, and we expect to have approximately 5% worse correlation of the model to these data. When you compare the model with the free data, it turns out that the 5% of the full dataset correlates extremely well with the model.
This may be a little more advanced stuff than you need to understand how many samples are required to predict with an acceptable confidence limit the outcome of a random-or-otherwise event, but hopefully it triggers you to RTFM on statistics...
The only question here is 'how well does NS2stats sample the full dataset?' I suspect the answer is 'not terribly well,' but anything is normally better than nothing. I suspect the predictive power of even a fairly small sample such as NS2stats is deceptively high.
<a href="http://bobhall.tamu.edu/FiniteMath/Module8/Introduction.html" target="_blank">Here is a randomly picked online tutorial for statistics which you may find helpful</a>.
<!--quoteo(post=2058848:date=Jan 10 2013, 03:13 PM:name=hakenspit)--><div class='quotetop'>QUOTE (hakenspit @ Jan 10 2013, 03:13 PM) <a href="index.php?act=findpost&pid=2058848"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->...heck the way most electoral systems run they clearly show pockets of support based on geographical location...
...a game that can be lost tens if not hundreds of thousand different ways...<!--QuoteEnd--></div><!--QuoteEEnd-->
To hakenspit and Salt,
I realize this is proably like talking to a wall, but I can't stop myself, I need to stand up for science and logical thinking. And even if you guys wont take it in, some other readers may.
1. The only NS2 statistic that has been discussed in this thread is Alien/Marine win/loss ratio. So yes, coin flipping or elections are both good analogies. True, the reasons for winning a game are complex. But you need to compare apples to apples. The reason for how a person decides to vote in an election is equally complex. What we are discussing here are the different outcomes.
2. Your arguments are based on the assumption that the servers running NS2Stats are in some statistical way skewed, for example: "Good alien players and bad marine players tend to play on servers that use NS2Stats more often that not." You may not realize that your arguments are based on this, but they are. This assumption may be true or not, but so far, nobody has presented any arguments or data supporting this assumption and the null hypothesis (google it) is that those servers are no different from other servers.
<!--quoteo(post=2058776:date=Jan 10 2013, 06:59 AM:name=hakenspit)--><div class='quotetop'>QUOTE (hakenspit @ Jan 10 2013, 06:59 AM) <a href="index.php?act=findpost&pid=2058776"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->UWE, unlike most people here, are not saying the win:loss rates indicate any issues with the fundamental balance of the intended design goals. In fact they are setting about tweaking aliens (some say buffing) to make them more fun to play.<!--QuoteEnd--></div><!--QuoteEEnd-->I don't know which UWE you're talking about, but perhaps you better read the blogs and posts. Unknown Worlds Entertainment does in fact acknowledge a *fundamental* balance issue in the game right now. The changes to aliens are actually old changes that have been in the pipe for some time. If you think the devs are going to ignore the win/loss rate and just tell people to L2P, then please pass over some of what you're smoking, because that's some good sh!t.
UWE will be altering balance to change the win/loss ratio. They already did it with the Onos egg drop getting tied to 3 hives, and there are other balance changes coming down the pipe - like doubling the amount of time an alien has to spend in the spawn queue. They have said they want the win:loss rate to be closer, and that they will be making changes to achieve that end. Read the blogs.
I'll let the rest bicker about statistics, since it really is a moot point. UWE know the game favours the aliens right now. They have *SAID* the game favours the aliens right now. They also know that an imbalanced game is not a fun game. Expect changes, and expect QQ from people like yourself who can't bear to hear that pub games might actually become fun to play.
ScardyBobScardyBobJoin Date: 2009-11-25Member: 69528Forum Admins, Forum Moderators, NS2 Playtester, Squad Five Blue, Reinforced - Shadow, WC 2013 - Shadow
Probably the biggest legitimate criticism of ns2stats that can be made is that its a form of convenience or <a href="http://en.wikipedia.org/wiki/Sampling_(statistics)#Accidental_sampling" target="_blank">accidental sampling</a> which no one has confirmed whether it is a representative sample of all NS2 matches. I personally suspect that it is, since I see no big reason why servers running ns2stats would have a substantially different playerbase than those that aren't.
I did attempt to tackle <a href="http://www.unknownworlds.com/forums/index.php?showtopic=114175" target="_blank">analyzing NS2 balance</a> using the then publically available official stats back in the beta, which covers more detail about what I view as the most important factors when using this info. Looking at the B235 public ns2stats matches, the precision looks pretty good (roughly 7500 matches which would be a roughly +/- 1.5% for the 95% CI assuming a binomial distribution), but the accuracy may be poor depending on what question you want answered.
<!--quoteo(post=2058877:date=Jan 10 2013, 11:24 AM:name=Savant)--><div class='quotetop'>QUOTE (Savant @ Jan 10 2013, 11:24 AM) <a href="index.php?act=findpost&pid=2058877"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->UWE will be altering balance to change the win/loss ratio. They already did it with the Onos egg drop getting tied to 3 hives, and there are other balance changes coming down the pipe - like doubling the amount of time an alien has to spend in the spawn queue. They have said they want the win:loss rate to be closer, and that they will be making changes to achieve that end. Read the blogs.<!--QuoteEnd--></div><!--QuoteEEnd--> I really hope that spawn queue change won't make it to patch. Whose bright idea is it to remove devour but add an alien spawn system that can have the entire team staring at a timer for 2 minutes?
<!--quoteo(post=2058877:date=Jan 10 2013, 10:24 AM:name=Savant)--><div class='quotetop'>QUOTE (Savant @ Jan 10 2013, 10:24 AM) <a href="index.php?act=findpost&pid=2058877"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->and there are other balance changes coming down the pipe - like doubling the amount of time an alien has to spend in the spawn queue.<!--QuoteEnd--></div><!--QuoteEEnd-->
I'm going to privately laugh my ass off when that turns out to be incorrect. What an asinine idea.
Since I love statistics, I actually read all posts. What we need is Yuuki.
Dear Yuuki, please come and explain that it doesnt matter how good ns2stats samples the full population. The main problem here is ppl thinking that these win ratio numbers is a good measure of game balance from the very beginning.
Google is your friend (evil, but friendly evil) :)
Quoting me quoting me:
<!--quoteo--><div class='quotetop'>QUOTE </div><div class='quotemain'><!--quotec-->Some quotes from me and others:
Win ratio balance is only one of several balance goals that NS2 should strive for. Another is ensuring that different tech trees or path are viable. Additionally, making sure all starting map tech nodes are viable is another worthy balance goal.
*
One of the problem is that fixing as a goal to get 50% winrate is a very ill defined problem. Let me explain.
The winrate depends on a large number of parameters : skulk health, res influx, lmg damage, tech cost, etc.
Taken alone almost every one of this parameters allows to get a 50% winrate. The reason why this is true is the following. Take the lmg damage as an example. Put it to zero, the marine winrate drop to zero. Put it to infinity (one shot everything), the marine winrate goes to 100%. By a continuity assumption there is a lmg damage value where the winrate is 50% (it cannot go from 0 to 100 without passing by 50).
The same is true for almost every other parameter. What this means is that saying "parameter X is responsible for the non-50% winrate" is meaningless since it is trivially true for any parameter.
I don't go into combination of parameters (e.g. for two parameters there is probably a curve of 50% winrate, for three a surface, etc.) but the problem become even more degenerated.
So, what we need right know is more, clearly stated, balance criteria, in addition to the 50% winrate.
*
This is what I wanted to say, the 50% percent win-rate goal is a narrow view of balance, and it's actually an easy problem, because there is lots of solutions to it. The hard balance problem is to get most of the tech tree to be useful, to maximize the number of valid build orders. Starcraft bw reached that by having almost all the units being used at the pro level (even the scout), and different viable tech paths (bio vs mech in TvZ).<!--QuoteEnd--></div><!--QuoteEEnd-->
People on these forums range from amusingly oblivious to completely moronic. Not suprisingly the majority that chooses to post seems to be the latter.
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.
<!--quoteo(post=2059411:date=Jan 11 2013, 04:37 PM:name=xDragon)--><div class='quotetop'>QUOTE (xDragon @ Jan 11 2013, 04:37 PM) <a href="index.php?act=findpost&pid=2059411"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->People on these forums range from amusingly oblivious to completely moronic. Not suprisingly the majority that chooses to post seems to be the latter.
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.<!--QuoteEnd--></div><!--QuoteEEnd-->
I'll openly criticize whatever I want, if any part of that comment is directed at me. Increasing alien spawn queue time would be the most asinine technique they could possibly use to bridge whatever balance gap there is. It's certainly the easiest way to piss off probably everyone who plays on the larger, non-competitive minded servers. And no, I really don't care that balancing is focusing on 8v8. It might be "negligible" there, but I hardly see that as a minimal change on larger servers.
<!--quoteo(post=2059411:date=Jan 11 2013, 05:37 PM:name=xDragon)--><div class='quotetop'>QUOTE (xDragon @ Jan 11 2013, 05:37 PM) <a href="index.php?act=findpost&pid=2059411"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->People on these forums range from amusingly oblivious to completely moronic. Not suprisingly the majority that chooses to post seems to be the latter.
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.<!--QuoteEnd--></div><!--QuoteEEnd--> Can you not read or did you just ignore my quote immediately before the statement regarding doubling the spawn time? People on these forums range from cocky posters to... cocky posters that can't read! Not surprising the majority that chooses to post seems to be the later.
Oh, and since you apparently cannot count, it doesn't even take 27 dead players on the alien team to have a 2 minute spawn right now. 18 will do.
<!--quoteo(post=2059405:date=Jan 11 2013, 02:25 PM:name=Yuuki)--><div class='quotetop'>QUOTE (Yuuki @ Jan 11 2013, 02:25 PM) <a href="index.php?act=findpost&pid=2059405"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->By a continuity assumption<!--QuoteEnd--></div><!--QuoteEEnd-->
I'd be very leery making such an assumption. There could easily be threshold effects that produce discontinuities. In fact, I'd bet there are lots of them.
But that's an incidental point, of course. The rest of your comments are well-said.
I'll note that while it's true that a game should be balanced in more ways than just win ratio, it nevertheless does still need to be balanced in win ratio. Right now it's not.
Comments
Flip a coin 1000 times. Note how many "heads" you get the first 250 flips. Then note how many "heads" you get the last 750 flips. Do you think its reasonable to assume that your percentage of "heads" will be any different in the first sample than the last? And what if you repeat this experiment 100 times, wouldn't you agree that the percentage of "heads" will be very close to 50% in all the samples?
The reason most people argue so strongly against you is not because they are defending NS2Stats, its because what you are doing is questioning a very fundamental concept in statistics and the scientific method. Sampling has been used for centuries and a lot of research is based on it, which has been performed by folks a lot smarter than you and me.<!--QuoteEnd--></div><!--QuoteEEnd-->
No I would not expect that at all.
Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips.
What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page.
A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.
Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips.
What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page.
A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.<!--QuoteEnd--></div><!--QuoteEEnd-->
This would be my answer as far as it goes for the thought experiment of Brony.
The variance factoring in ranging from skill to luck to chance, knowledge and teamwork all pitch in on the random factor.
On top of that comes a small thing i'll point out.
Not questioning your thought experiment, because it is validly true, none the less. The factors pitching in to form game win results are not as simple as just a coin toss.
Rather than a full 50/50 chance of tails or heads landing, the results of games are not randomly polled, but fixed. consistently.
What is happening with the server polling here, is not the first 25% taken of all results taken in is being polled. But randomly across all the results. which are already random.
As such, it keeps on increasing the chance that the servers being polled have a more strongly favored winrate for Aliens, rather than marines.
Since it's not 'completely' random but fixed, Only a certain amount of servers have ns2stats installed.
it's certainly gnawing my mind though.
results are results none the less, but it feels as though when you factor in chance, upon chance upon chance, that polling results randomly across all servers, at a random pace you can get skewered statistics.
If ns2stats keeps spreading though, and getting bigger and bigger, they'll cover more servers which would more accurately set the average.
at times i checked out ns2stats it kept telling me i only played like 15 and now 19 games in total. When really i've played more like 200+ rounds.
I might be one of the very few players experiencing this, but it feels wrong.
As such, it keeps on increasing the chance that the servers being polled have a more strongly favored winrate for Aliens, rather than marines.<!--QuoteEnd--></div><!--QuoteEEnd-->
Why?
If we stick with the coin toss example:
If you have tossed 100 times and got a sequence consisting of 52 Heads and 48 Tails(not 52 Heads followed by 48 Tails but any given random pattern) and then randomly select 25 throws the chance to hit a Heads-throw is 52% each time and the chance to hit a Tails-throw is 48% each time.
Thats like if you have a town with 45% of the people being female, each time you randomly select a person theres a 45% chance to pick a female. So if you randomly pick 500 people chances are you get around 225 females and about 275 males.
Of course a game of NS2 is more complex than a coin toss, but we are not talking about a sample size of 100 but several thousands.
And please consider why a sample size of 1000 is enough to accurately foretell the election results in a country with 62 mio. IF your arguments would be right this would be downright impossible.
And for the raindrop example: Given there are 1000 cm² of which 60% are hit by a raindrop and 40% are not hit by a rain drop. If I now randomly select 100 cm² and look at wheter they were hit by a raindrop, the ratio betwenn hit and not-hit will be around 60-40.
The pattern of raindrop-hits in this case is completely irrelevant.
Quoted for truth.
I lived in Germany for more than 25 yrs and was never asked in a pre-election poll.... these polls have to be worthless.
If we stick with the coin toss example:
If you have tossed 100 times and got a sequence consisting of 52 Heads and 48 Tails(not 52 Heads followed by 48 Tails but any given random pattern) and then randomly select 25 throws the chance to hit a Heads-throw is 52% each time and the chance to hit a Tails-throw is 48% each time.
Thats like if you have a town with 45% of the people being female, each time you randomly select a person theres a 45% chance to pick a female. So if you randomly pick 500 people chances are you get around 225 females and about 275 males.
Of course a game of NS2 is more complex than a coin toss, but we are not talking about a sample size of 100 but several thousands.
And please consider why a sample size of 1000 is enough to accurately foretell the election results in a country with 62 mio. IF your arguments would be right this would be downright impossible.
And for the raindrop example: Given there are 1000 cm² of which 60% are hit by a raindrop and 40% are not hit by a rain drop. If I now randomly select 100 cm² and look at wheter they were hit by a raindrop, the ratio betwenn hit and not-hit will be around 60-40.
The pattern of raindrop-hits in this case is completely irrelevant.<!--QuoteEnd--></div><!--QuoteEEnd-->
Firstly your comparing statistics effectiveness in predicting a 2 or maybe 3 possible outcome situation in the case of election polls.
Similar to a coin toss though you get outliers in polling data...heck the way most electoral systems run they clearly show pockets of support based on geographical location.
This can be traced down even further to streets within a suburb (based on socio-economic groupings), most reputable statisticians use a wide sample base to help avoid these occuring though.
Its the fact there are only limited actions (toss coin - heads/tails or cast votes - election won/lost), compare this to a game that can be lost tens if not hundreds of thousand different ways, with slightly different actions undertaken. This could be by not using the right starting hive upgrade, or upgrading to exo's before weapons or armour upgrades. Or any of the other countless other decisions.
Have you ever heard of a game called two up? Look into it...might help you see how many variables are in the results of two consecutive tosses of coins.
Similar to a coin toss though you get outliers in polling data...heck the way most electoral systems run they clearly show pockets of support based on geographical location.
This can be traced down even further to streets within a suburb (based on socio-economic groupings), most reputable statisticians use a wide sample base to help avoid these occuring though.
Its the fact there are only limited actions (toss coin - heads/tails or cast votes - election won/lost), compare this to a game that can be lost tens if not hundreds of thousand different ways, with slightly different actions undertaken. This could be by not using the right starting hive upgrade, or upgrading to exo's before weapons or armour upgrades. Or any of the other countless other decisions.
Have you ever heard of a game called two up? Look into it...might help you see how many variables are in the results of two consecutive tosses of coins.<!--QuoteEnd--></div><!--QuoteEEnd-->
I'm not talking about "Party A will loose while Party B wins"
I'm talking about
Party A: 38%
PArty B: 32%
Party C: 14%
Party D: 7%
Party E: 5%
Others: 4%
So according to you with geographical, social and what not influences such a statistic HAS to be completely worthless if based on WAY LESS than 0,1% sample size. Still these results are shockingly correct in most cases and never are fundamentally wrong.
The whole point and use of statistics is that you can get a very accurate image of what the big picture looks like on the basis of only a small sample size.
The whole point of statistics is to simplify these overcomplicated situations you decribe so that these details no longer are <b>relevant</b>.
I let's not forget that a NS2 match may be kind of complicated, but there are FAR more complicated problems out there that really can't be proven/solved by math but given a VERY accurate account for by statistics.
Take your basic model of a Atom with is electrons. I really can't say where these electrons are. Possibly the electron of one of the atoms within my computer actually is on Mars in this instance - there literally is no way to prove that it's not.
BUT statistically speaking I can be quite sure it's right where it should be very close to it's own core. I can even say the same for ALL the electrons doing ther electron thing in my computer. I don't need to know the least bit about any of them. I dont even have to know anything about physics and still this statement is right with a probability of more than 99,9999%.
So now the big question is: How big of a sample size was examined before it was concluded, that electrons with a probability far greater than 99,999% are very close to their coresponding core despite them theoretically being able to be more 10000000000 km away from it?
I don't think they really looked at 25% of all electrons within our universe, do you?
And I think we can agree on quantum physics being more complicated than NS2, can we not?
Random occurrences are not as random as people think and as such you often see groups of heads or tails in coin flips.
What you dont see is a consistent head - tail - head - tail pattern. Now given a coin toss has only 2 possible outcomes the statistical average to achieve a 50 - 50 ratio is minute compared to that of a game result analysis. The number of variables that can influence things is staggering.
The best example of how little people really understand the random nature of the world is in trying to pick a natural and man made rain fall pattern on a page.
A clear majority of people pick the man made pattern as natural as they don't believe results can be clustered.
So just because we believe that 1000 games with exact same variables (Map, starting locations, player numbers, comm exp, numbers on going live, tech tree options for each side (each is their own variable) etc) might be sufficient to get an average...you are wrong and you would be drawing conclusions on a sample size too small once you start to take in all the unique instances you have to account for before you can start to get trends within each of these.<!--QuoteEnd--></div><!--QuoteEEnd-->
Oh dear God. Please please please please go and read an introduction to Statistics, or go on a course, or something.
I'm a crystallographer. I solve the 3D structure of proteins. One of the key features we use to determine whether or not the protein model we create is 'too detailed' to be explained by our experimental data is the use of a free-R statistic. Put simply, we take approximately 5% of our data points (let's say that's 1000 data points, for example). We do nothing with these data at all. They are separated from the 95% of data, on which we base the protein structure model. Then - and this is the cunning bit - we compare the model we've built with these so-called 'free' data, and we expect to have approximately 5% worse correlation of the model to these data. When you compare the model with the free data, it turns out that the 5% of the full dataset correlates extremely well with the model.
This may be a little more advanced stuff than you need to understand how many samples are required to predict with an acceptable confidence limit the outcome of a random-or-otherwise event, but hopefully it triggers you to RTFM on statistics...
The only question here is 'how well does NS2stats sample the full dataset?' I suspect the answer is 'not terribly well,' but anything is normally better than nothing. I suspect the predictive power of even a fairly small sample such as NS2stats is deceptively high.
<a href="http://bobhall.tamu.edu/FiniteMath/Module8/Introduction.html" target="_blank">Here is a randomly picked online tutorial for statistics which you may find helpful</a>.
...a game that can be lost tens if not hundreds of thousand different ways...<!--QuoteEnd--></div><!--QuoteEEnd-->
To hakenspit and Salt,
I realize this is proably like talking to a wall, but I can't stop myself, I need to stand up for science and logical thinking. And even if you guys wont take it in, some other readers may.
1. The only NS2 statistic that has been discussed in this thread is Alien/Marine win/loss ratio. So yes, coin flipping or elections are both good analogies. True, the reasons for winning a game are complex. But you need to compare apples to apples. The reason for how a person decides to vote in an election is equally complex. What we are discussing here are the different outcomes.
2. Your arguments are based on the assumption that the servers running NS2Stats are in some statistical way skewed, for example: "Good alien players and bad marine players tend to play on servers that use NS2Stats more often that not." You may not realize that your arguments are based on this, but they are. This assumption may be true or not, but so far, nobody has presented any arguments or data supporting this assumption and the null hypothesis (google it) is that those servers are no different from other servers.
In fact they are setting about tweaking aliens (some say buffing) to make them more fun to play.<!--QuoteEnd--></div><!--QuoteEEnd-->I don't know which UWE you're talking about, but perhaps you better read the blogs and posts. Unknown Worlds Entertainment does in fact acknowledge a *fundamental* balance issue in the game right now. The changes to aliens are actually old changes that have been in the pipe for some time. If you think the devs are going to ignore the win/loss rate and just tell people to L2P, then please pass over some of what you're smoking, because that's some good sh!t.
UWE will be altering balance to change the win/loss ratio. They already did it with the Onos egg drop getting tied to 3 hives, and there are other balance changes coming down the pipe - like doubling the amount of time an alien has to spend in the spawn queue. They have said they want the win:loss rate to be closer, and that they will be making changes to achieve that end. Read the blogs.
I'll let the rest bicker about statistics, since it really is a moot point. UWE know the game favours the aliens right now. They have *SAID* the game favours the aliens right now. They also know that an imbalanced game is not a fun game. Expect changes, and expect QQ from people like yourself who can't bear to hear that pub games might actually become fun to play.
I did attempt to tackle <a href="http://www.unknownworlds.com/forums/index.php?showtopic=114175" target="_blank">analyzing NS2 balance</a> using the then publically available official stats back in the beta, which covers more detail about what I view as the most important factors when using this info. Looking at the B235 public ns2stats matches, the precision looks pretty good (roughly 7500 matches which would be a roughly +/- 1.5% for the 95% CI assuming a binomial distribution), but the accuracy may be poor depending on what question you want answered.
I really hope that spawn queue change won't make it to patch. Whose bright idea is it to remove devour but add an alien spawn system that can have the entire team staring at a timer for 2 minutes?
I'm going to privately laugh my ass off when that turns out to be incorrect. What an asinine idea.
Dear Yuuki, please come and explain that it doesnt matter how good ns2stats samples the full population. The main problem here is ppl thinking that these win ratio numbers is a good measure of game balance from the very beginning.
Quoting me quoting me:
<!--quoteo--><div class='quotetop'>QUOTE </div><div class='quotemain'><!--quotec-->Some quotes from me and others:
Win ratio balance is only one of several balance goals that NS2 should strive for. Another is ensuring that different tech trees or path are viable. Additionally, making sure all starting map tech nodes are viable is another worthy balance goal.
*
One of the problem is that fixing as a goal to get 50% winrate is a very ill defined problem. Let me explain.
The winrate depends on a large number of parameters : skulk health, res influx, lmg damage, tech cost, etc.
Taken alone almost every one of this parameters allows to get a 50% winrate. The reason why this is true is the following. Take the lmg damage as an example. Put it to zero, the marine winrate drop to zero. Put it to infinity (one shot everything), the marine winrate goes to 100%. By a continuity assumption there is a lmg damage value where the winrate is 50% (it cannot go from 0 to 100 without passing by 50).
The same is true for almost every other parameter. What this means is that saying "parameter X is responsible for the non-50% winrate" is meaningless since it is trivially true for any parameter.
I don't go into combination of parameters (e.g. for two parameters there is probably a curve of 50% winrate, for three a surface, etc.) but the problem become even more degenerated.
So, what we need right know is more, clearly stated, balance criteria, in addition to the 50% winrate.
*
This is what I wanted to say, the 50% percent win-rate goal is a narrow view of balance, and it's actually an easy problem, because there is lots of solutions to it. The hard balance problem is to get most of the tech tree to be useful, to maximize the number of valid build orders. Starcraft bw reached that by having almost all the units being used at the pro level (even the scout), and different viable tech paths (bio vs mech in TvZ).<!--QuoteEnd--></div><!--QuoteEEnd-->
<a href="http://www.unknownworlds.com/ns2/forums/index.php?showtopic=121172&view=findpost&p=1980702" target="_blank">http://www.unknownworlds.com/ns2/forums/in...t&p=1980702</a>
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.<!--QuoteEnd--></div><!--QuoteEEnd-->
I'll openly criticize whatever I want, if any part of that comment is directed at me. Increasing alien spawn queue time would be the most asinine technique they could possibly use to bridge whatever balance gap there is. It's certainly the easiest way to piss off probably everyone who plays on the larger, non-competitive minded servers. And no, I really don't care that balancing is focusing on 8v8. It might be "negligible" there, but I hardly see that as a minimal change on larger servers.
Stating that with the currently under testing spawn changes aliens would be sitting in the spawn queue for 2 minutes is completely incorrect, and not even possibly unless you somehow manage to have 27 players on the alien team, and EVERY SINGLE ONE die.
I suggest that before you openly criticize something that you have 0 comprehension of, that you stop and play a couple rounds on the balance mod. From all the games I have played (public or pugs), I can honestly say I have only noticed aliens getting spawn queued longer in a couple, with the differences being negligible. The spawn changes in the patch are a huge step in the right direction, but IMO are still to lenient for aliens. Aliens can still spawn, on one hive, about 2 twice as fast as marines on a single IP.<!--QuoteEnd--></div><!--QuoteEEnd-->
Can you not read or did you just ignore my quote immediately before the statement regarding doubling the spawn time?
People on these forums range from cocky posters to... cocky posters that can't read! Not surprising the majority that chooses to post seems to be the later.
Oh, and since you apparently cannot count, it doesn't even take 27 dead players on the alien team to have a 2 minute spawn right now. 18 will do.
I'd be very leery making such an assumption. There could easily be threshold effects that produce discontinuities. In fact, I'd bet there are lots of them.
But that's an incidental point, of course. The rest of your comments are well-said.
I'll note that while it's true that a game should be balanced in more ways than just win ratio, it nevertheless does still need to be balanced in win ratio. Right now it's not.