Curse

tmakinen · Mar 20, 2009, 03:25 PM // 15:25

Alleji's approach is the only completely objective one, so if we want to settle this once and for all then we should observe map changes over a period of one month or so. About 4 to 6 players from different time zones should be able to cover all the six daily maps and if the observations are pooled together then proper statistics can be derived.

About the sample size - there aren't any 'magic' numbers related to how big the sample must be. It is just that the larger the sample the smaller the relative margin of error, and if you do proper statistic you should estimate the margin of error anyway.

Skyy High · Mar 20, 2009, 05:30 PM // 17:30

Just to throw some small amount of actual statistics into here - I take some measure of offense at the thought that a simple calculation of an average registers as high level statistics with you people - I performed a simple t test on your data.

t{calc} = abs(avg{win ratio} - (expected win ratio)) / stdev{win ratio} * sqrt(# of data points)

t{calc} = (0.68 - 0.50)/0.4688*10 = 3.84
Closest t value in my book's table:
t{99.9% confidence, n = 120} = 3.373

Therefore, we can say that the calculated win ratio does not equal 50% (ie, Luxons do in fact statistically win more than Kurzicks) with 99.9% certainty.

The trouble with the analysis that I have just performed is that win ratio is calculated from binary data: a team can only either win or lose. This means that these numbers could not possibly be normally distributed, which is an assumption the t test makes. I'm trying to think of a better way to do this; so far the best thing I can think of is to calculate the standard deviations and confidence intervals of the actual scores of each game, and use that to determine whether each set of data (the Luxon and Kurzick scores) are statistically different from each other. If you want me to do that, I'll need the actual Excel sheets, because I'm not copying all of your numbers by hand. This will again run into the problem of the data not exactly conforming to a normal probability distribution (because the data sets have a hard cap at 500, as once a team gets to that score they win the game), but it should be more manageable than discrete values.

snaek · Mar 20, 2009, 05:43 PM // 17:43

Quote:

Originally Posted by tmakinen

Alleji's approach is the only completely objective one, so if we want to settle this once and for all then we should observe map changes over a period of one month or so.

seconded

op's approach sounds more like "this is how much i can win in ab" and not "how much do kurz/lux win."

lord of all tyria · Mar 20, 2009, 07:02 PM // 19:02

Quote:

Originally Posted by Darth The Xx

Outliers were any games where either side scored less than 100 points and the scores were not recorded and so do not count towards the final results.

So if you lost a mass of games by 500-10, you don't count them? My quick skimread is probably missing something, I hope so anyway.

Darth The Xx · Mar 20, 2009, 07:27 PM // 19:27

Quote:

Originally Posted by lord of all tyria

So if you lost a mass of games by 500-10, you don't count them? My quick skimread is probably missing something, I hope so anyway.

There was only four

and they were 3 kurz losses 1 luxon loss.

Anyway good input so far guys. I do realise that the sample size isn't terribly large and its not like I'm going to record 1000 AB games as has been suggested.

The data is biased as has been pointed out, however I wanted to observe who won/lost the most when I played (Don't know if I mentioned this at the start) so is it not valid for me personally?

If somebody else (preferrably from the kurz side) wants to do the same thing to compare you are more than welcome

Expherious · Mar 20, 2009, 07:38 PM // 19:38

All the good AB'rs wait until the double faction weekends, I won like 18 in a row on the last x2 weekend.<Kurzick

Bobby2 · Mar 20, 2009, 08:06 PM // 20:06

Quote:

Originally Posted by Songbringer

When I was luxon I always felt like I won more then when I play as a kurzick

This. Does make me very curious about Alleji's results, though.

glacialphoenix · Mar 20, 2009, 09:00 PM // 21:00

Quote:

The data is biased as has been pointed out, however I wanted to observe who won/lost the most when I played (Don't know if I mentioned this at the start) so is it not valid for me personally?

Well, if it's your experience then you're definitely free to say so. Yep, they're valid for you - if you find yourself winning many times when playing Luxon, hey, good for you and all that. Problem is, your first post came across as you presenting conclusive data - which, as has been stated, it is not.

If someone else took various other results I'd be interested in looking at the outcome, of course, but I should think it's pretty hard to get a fair sample.

GourangaPizza · Mar 21, 2009, 02:20 PM // 14:20

Interesting statistics. Guild Wars should take a leaf out of Warhammer Online by displaying team and individual's statistics at end of every PvP match (e.g. GvG, AB, HB)

Songbringer · Mar 21, 2009, 11:16 PM // 23:16

Quote:

Originally Posted by Bobby2

This. Does make me very curious about Alleji's results, though.

Why does it?

Bobby2 · Mar 22, 2009, 03:07 AM // 03:07

Quote:

completely objective

This.

/12chars

MasterSasori · Mar 23, 2009, 11:00 PM // 23:00

By far the most convincing data on the Lux vs. Kurz AB debate. Good job.

But consider this Darth:

To make the data more consistent, you need to have as many of the same players as often as possible. The problem with this is that more than likely, they are your friends/guild members and guild/friend groups generally perform better than PUGs. However, even this is hard to prove.

Still the factor of how much does this matter still applies. So assuming that Lux>Kurz in AB.... so what?

MasterSasori · Mar 23, 2009, 11:01 PM // 23:01

By far the most convincing data on the Lux vs. Kurz AB debate. Good job.

But consider this Darth:

To make the data more consistent, you need to have as many of the same players as often as possible. The problem with this is that more than likely, they are your friends/guild members and guild/friend groups generally perform better than PUGs. However, even this is hard to prove.

Still the factor of how much does this matter still applies. So assuming that Lux>Kurz in AB.... so what?

Darth The Xx · Mar 24, 2009, 04:08 AM // 04:08

Quote:

Originally Posted by MasterSasori

By far the most convincing data on the Lux vs. Kurz AB debate. Good job.

But consider this Darth:

To make the data more consistent, you need to have as many of the same players as often as possible. The problem with this is that more than likely, they are your friends/guild members and guild/friend groups generally perform better than PUGs. However, even this is hard to prove.

Still the factor of how much does this matter still applies. So assuming that Lux>Kurz in AB.... so what?

Its a debate thats been raging since the beginning of factions.

D A R K S O I F O N · Mar 24, 2009, 04:41 AM // 04:41

i would have to say that over the period of time doing jq, luxons at times have more in numbers every 12th match or so. It's a joke to see this but it's happened so many times.

Eg Start of match yellow shrine has 4 luxons, green has 4 and purple had 3. Thats 11 luxons on an 8 on 8 match.

MasterSasori · Mar 24, 2009, 09:00 PM // 21:00

Quote:

Originally Posted by Darth The Xx

Its a debate thats been raging since the beginning of factions.

I'm not disagreeing with whether you are right or wrong - your data is convincing. But really its one of those questions that don't really matter. It's like me finding out that penguins poop close to 90 degrees than 80 degrees.

Besides, people don't seem to listen too much to GW Guru.

If they did, I would see a million Ether Prot Eles while pugging.

PS - this post isn't meant to be rude. I think you did the gw a great service by giving actual data instead of opinion.

Oleg · Apr 05, 2009, 01:39 PM // 13:39

It's going to be virtually impossible to limit the variables to one thing when you are participating in the AB (so I'm actually happy that you did most of them at the same time). But there are a huge number of variables here that seem to have changed. Internal validity problems aside it is interesting.

Did you make any observation on how they won? Did they stay in separate groups or mass and do a clockwise cap? Do Kurzicks and Luxons behave differently? Do they alter strategies per map? Did you see a higher drop out rate on either side when losing? If so, at what point?

Just wondered.

Trinity Fire Angel · Apr 05, 2009, 03:09 PM // 15:09

i have a few points. the outliers should be recorded. you are biasing the results by removing the scores. by participating you are influencing the results. one good team is enough to win most matches. i always seem to win more AB when i have a decent team and mostly when my guild and alliance play as a team with vent. you also should be playing when you are on the kurzick side as well as playing for luxon side. with 50% random pugs on each map and 50% with your guild.

finally, you need to collect data from a range of peolpe doing exactly the same tests at different times and you need to do it from both american disctrict, english, international, german and french, russian... etc. so you get a wide mix of nationalities that influence the scores.

Bobby2 · Apr 05, 2009, 05:09 PM // 17:09

-Every-single-euro-I-know- heads straight to American District.

glacialphoenix · Apr 05, 2009, 05:39 PM // 17:39

Quote:

finally, you need to collect data from a range of peolpe doing exactly the same tests at different times and you need to do it from both american disctrict, english, international, german and french, russian... etc. so you get a wide mix of nationalities that influence the scores.

Out of curiosity. Does that actually affect who you really meet in the battle itself? I mean, I know American districts don't allow you to use things like Chinese characters in team chat. But I've definitely met Japanese, Chinese guild teams etc in the battle itself - and no, they're not talking in English.