|
Edit: Weighted averages across matchups and seasons along with some other stuff at the end.
First, the average game length(all matchups) by tournament, which was the question that started it all:
http://i.imgur.com/WZmGn.png
![[image loading]](http://i.imgur.com/WZmGn.png)
I used the records page on GOMTV to find the average game length on the different GSL tournaments and matchups, to see how that changed as time passed. It was basically inspired by what Day9 said in SotG 42 about how game lengths must have increased as the first seasons were mostly all ins. It doesn't show the ongoing tournaments.
Not all matchups were played enough to be statistically significant in most cases, but they were all used. The 3 first GSTL didn't have that many matches, so each matchup there is of course with a smaller amount of games and more vulnerable to random variances. The only matchups that weren't played at all in a tournament were PvP in GSL Open 3 and ZvZ in GSTL May (which is why you'll see them below the bottom in some of the graphs).
Even though Terran were doing pretty well in the first 3 GSL, there weren't that many mirror matches. The highest was GSL2 with 27 TvT, in comparison there were 48 TvT in GSLJan, 52 TvT in GSL Super Tournament and 57 TvT in GSLMay. I don't know the amount played in the other matchups, because everything but mirror matchups was pulled directly from the GOMTV stats page (they don't show avg game time for mirrors -_-).
Here's the summary of all matchups, all times are in Blizzard minutes, even though time on the GOMTV page are in real minutes. Each matchup is given from the perspective of the winner, so ZvP means a Zerg won in that game time, and PvZ means a Protoss won in that game time.
![[image loading]](http://i.imgur.com/NNJi2.png)
As expected, ZvT, TvZ and TvT games last for a pretty long time, while ZvZ and PvP are a lot shorter. One interesting anomaly was GSL3 ZvZ, which had an average game time of 20 minutes with 6 games played(Nestea vs Sen, Tosyad vs FD and TheWind vs GoldenfOu).
You also have GSL3 being the season with the longest games on average, and GSLMay having the shortest games. In general the game length goes up to a top on GSL3, then get shorter until GSLMay, and have been increasing since then(but still shorter than GSL2).
These are the numbers used for the graphs, the sum row and column aren't weighted by amount of games, but I hope it wouldn't have made the numbers completely different (weighting would mean if there was 10 1 hour PvZ but 1 10 minute PvP and PvT, the weighted average would be 52 minutes for Protoss instead of the 27 minutes that the simple sum column would show).
+ Show Spoiler +
Graph in the PvT matchup
+ Show Spoiler +
PvZ
+ Show Spoiler +
ZvT
+ Show Spoiler +
Protoss wins
+ Show Spoiler +
Zerg wins
+ Show Spoiler +
Terran wins
+ Show Spoiler +
Mirror matchups
+ Show Spoiler +
Another thing I did for fun was a way of calculating the chance of getting different matchups in semifinals and finals of a tournament, given the way the brackets were set up with the different races and that each player had a 50% chance of winning the series. Since it's not 50% chance to reach the next it's not really useful for anything, but was still a bit fun to work it out.
With that spreadsheet the Super Tournament had a 3.7% chance of getting the TvT TvT semifinals it did, with TvT TvZ at 11.2% being most likely and PvP PvP at 0.03% being least likely.
Edit:
All the next are weighted by amount of games. It doesn't mean all the above are wrong, it's just the correct way to show total averages in a tournament.
First, the average game length(all matchups) by tournament, which was the question that started it all:
+ Show Spoiler +
Then, the game length in a matchup put together by tournament, to compare between matchups:
+ Show Spoiler +
Average game length over the seasons so far, sorted by game length and showed by matchup(ZvP is still only for games zergs won, ZvP/PvZ is for both put together):
+ Show Spoiler +
And then some stats used to make the graphs in the end:
+ Show Spoiler +
Some fun facts from that (remember all are in game time):
-There's been 1386 games up to and excluding the current seasons so far, at a total of 432.5 hours.
-There's been 392 ZvT, 358 PvT, 272 TvT, 218 PvZ, 77 PvP and 69 ZvZ.
-134 hours of ZvT, 108 hours of PvT, 92 hours TvT, 65.5 hours PvZ, 17 hours of ZvZ and 15.7 hours of PvP(even with 8 more PvP played).
-The teamleagues had a different matchup that was played most for each one. The first had PvT (18 vs 8 TvT at second place), the second had TvT (13 vs 10 ZvT at second place) and the third had TvZ (17 vs 9 ZvP).
-The average GSL game is at 18:43 in game time.
|
|
|
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs?
|
Very cool someone took the time out to make this.
|
GSL2 was the 2 rax SCV stuff, im pretty sure. That is when I became MKP fan4lyfe.. Even now as Zerg.
Cool data man !
|
|
|
Good work. Glad that you went through the effort. May I ask for the reasoning behind matchups like ZvP and PvZ separated? Is it possible to output graphs that combine them?
|
thats interesting and most of it makes sense, but i wouldnt have thought that gsl may would have the shortest average
|
Very interesting, thank you =D
Really shows the game development (for example Zerg wins a lot of ZvT late-games lately because of infestor/broodlord)
I love the PvP one hahaha
|
What the hell?
I would have bet my life that this graph would look completely different. Interesting.
|
On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs?
I believe GSL2 was 2Rax SCV all-in fest.
|
On July 06 2011 06:24 Primadog wrote:
Good work. Glad that you went through the effort. May I ask for the reasoning behind matchups like ZvP and PvZ separated?
ZvP = zerg wins
PvZ = protoss wins
shows you which race is stronger in lategame
|
Maybe they are Zerg wins and Protoss wins, respectively?
|
Ah, that makes sense. Any chance of outputting graphs that combine them?
|
Haha the pvp graph is amazing.
|
so the average GSL game is almost 18 minutes thats a lot longer than i expected
|
On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs?
ZvT/TvZ in GSL3 were actually really long that season, with the zerg wins over terran barely being longer than TvT that season(and the longest any race had to play to win in any season). 16 games went over 30 game minutes, and only 4 were done in under 10.
|
On July 06 2011 06:23 Ponyo wrote:
GSL2 was the 2 rax SCV stuff, im pretty sure. That is when I became MKP fan4lyfe.. Even now as Zerg.
Cool data man !
GSL 1/2 was jan/feb when the 2rax bullshit was popular on close maps
March/april was when tank/marine vs ling/bane/muta was figured out and games started lasting a long time on big new maps
|
On July 06 2011 06:24 Primadog wrote:
Good work. Glad that you went through the effort. May I ask for the reasoning behind matchups like ZvP and PvZ separated? Is it possible to output graphs that combine them?
It was basically the way GOMTV had the stats in their page, so I preferred to have it that way. Makes it better to see which race was strongest against the other in which part of the game (ZvT in GSL3 was terran winning at average 21 minutes, and zerg winning at average 26 minutes). Getting ZvT and TvZ into one graph would be a bit extra work, but if it's useful it can be done, and I'd probably do weighted average game length of each season at the same time.
|
Great data O_o
Hats off to ya Sir.
|
|
|
On July 06 2011 06:29 Lobo2me wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? ZvT/TvZ in GSL3 were actually really long that season, with the zerg wins over terran barely being longer than TvT that season(and the longest any race had to play to win in any season). 16 games went over 30 game minutes, and only 4 were done in under 10.
GSL 3 was the year of marine/scv all-in against any Zerg that went hatch first (also season of bit by bit). But from what I remember games where the Zerg went pool first played normally and went on pretty long. Also FD and Nestea both went pretty deep that year and mostly defended early pressure well (other than against Rain).
|
Protoss games aren't going to say much for the first few seasons because the only person who was winning games was oGsMC.
IIRC, just January GSL, Protoss only had like a 23% winrate against Terran out of 50games.
EDIT:
Not to mention the race as awhole has been terribly represented up until the last few months
|
These graphs actually provides a little bit of back story to all the QQ people were doing about terran around GSL 1-3. If you look at the Zerg win graph and the protoss win graph they both bottom out somewhere (minimal games one or too short for data to be accurate). Terran however stays far more consistent than the other two races.
|
I think it is totally unimportant who won a ZvP, ZvT or TvP, therefore 3 graphs are dispensable. Besides this I am surprised cuz I thought games would last longer nowadays.
|
On July 06 2011 06:48 TempeRr wrote:
These graphs actually provides a little bit of back story to all the QQ people were doing about terran around GSL 1-3. If you look at the Zerg win graph and the protoss win graph they both bottom out somewhere (minimal games one or too short for data to be accurate). Terran however stays far more consistent than the other two races.
It's not really a win graph, it just shows average game length, and doesn't differ between 1 or 100 games played. The times they bottom out are when there were no games of that matchup played, PvP in GSL3 and ZvZ in GSTL May.
In general TvT was always a matchup more played than the other mirror matchups though. The lowest amount of TvT in a season was 12 in GSL3 with 3 seasons going above 45 Tvt, the highest amount of the other mirrors matchups was 17 PvP in GSL1 and 17 ZvZ in GSLJan, with most seasons having under 15 of PvP or ZvZ. The second highest mirror in GSL1-3 after the 17 PvP in GSL3 that wasn't TvT was 9 ZvZ in GSL2.
|
On July 06 2011 06:54 zul wrote:
I think it is totally unimportant who won a ZvP, ZvT or TvP, therefore 3 graphs are dispensable. Besides this I am surprised cuz I thought games would last longer nowadays.
The graphs are different, how fast a zerg won against a terran says nothing about how fast a terran won against a zerg. Working on a weighted graph now to see if it's any different and to compare the different matchups with each other more easily.
|
Nice job! I am glad you are working on a weighted graph. I am very curious to see if it makes things look different. The math nerd in me was very concerned when you mentioned you hadn't taken that into account in the first set of graphs 
|
This is really really interesting. Thanks for writing this up.
|
Soooo, everyone's perceived view of the 'cheese' seasons with the tiny maps like steppes and blistering actually produced longer games? That baffles me.
|
On July 06 2011 07:06 Micket wrote:
Soooo, everyone's perceived view of the 'cheese' seasons with the tiny maps like steppes and blistering actually produced longer games? That baffles me.
It's a small sample size, if you read carefully you'll see that one of the samples is only 6 games. I bet if you included every single game ever played somehow, smaller maps would always equal shorter games.
|
On July 06 2011 06:57 Lobo2me wrote:Show nested quote +On July 06 2011 06:54 zul wrote:
I think it is totally unimportant who won a ZvP, ZvT or TvP, therefore 3 graphs are dispensable. Besides this I am surprised cuz I thought games would last longer nowadays. The graphs are different, how fast a zerg won against a terran says nothing about how fast a terran won against a zerg. Working on a weighted graph now to see if it's any different and to compare the different matchups with each other more easily.
I get what you saying, but still: the match-up stays the same And according to the title the thread is about the average game length in the GSL - not about how fast a race won vs another (compared through all GSL Seasons). But I dont want to pick on details too much, I respect your work collecting all the data and put them into this graph.
|
Lol surprised by results! =O But fun to watch nevertheless! ^^,
|
On July 06 2011 07:12 zul wrote:Show nested quote +On July 06 2011 06:57 Lobo2me wrote:On July 06 2011 06:54 zul wrote:
I think it is totally unimportant who won a ZvP, ZvT or TvP, therefore 3 graphs are dispensable. Besides this I am surprised cuz I thought games would last longer nowadays. The graphs are different, how fast a zerg won against a terran says nothing about how fast a terran won against a zerg. Working on a weighted graph now to see if it's any different and to compare the different matchups with each other more easily. I get what you saying, but still: the match-up stays the same And according to the title the thread is about the average game length in the GSL - not about how fast a race won vs another (compared through all GSL Seasons). But I dont want to pick on details too much, I respect your work collecting all the data and put them into this graph.
It didn't make that much difference though. The average game length for each season (with each game equally weighted) follows the same pattern with only one difference. Instead of it increasing up to GSL3, decreasing to GSLMay and then up again, it has GSTLMarch going up from GSLMarch and then it follows the same pattern (down to GSLMay before building up again).
|
LOL at the PvP data. I'm so glad I don't have to play that MU.
Thanks for making these charts, really interesting to compare the community's perception of the matchup to the actual data.
|
Hm, this is a good idea, but it's hard to see trends the way the data is organized. I may try to do a more formal statistical analysis. (If I can find my damned stats textbook).
|
On July 06 2011 06:26 LesPhoques wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? I believe GSL2 was 2Rax SCV all-in fest.
It was GSL3 I think. It only started after MKP introduced the 2 rax all-in against Nestea in the GSL2 finals. Then there was the whole Artosis hoo-hah over what the Terrans were doing on ladder just before GSL3. And then we had BitByBit, Rain and a bunch of other Terran all-inners in GSL3.
Not quite sure why the match up length was so long in GSL3.
|
On July 06 2011 06:33 SafeAsCheese wrote:Show nested quote +On July 06 2011 06:23 Ponyo wrote:
GSL2 was the 2 rax SCV stuff, im pretty sure. That is when I became MKP fan4lyfe.. Even now as Zerg.
Cool data man ! GSL 1/2 was jan/feb when the 2rax bullshit was popular on close maps March/april was when tank/marine vs ling/bane/muta was figured out and games started lasting a long time on big new maps
IN GSL 2 MK was the only one using 2 rax. However in gsl 3 many more people used it (like rain).
|
Updated first post with weighted averages and some other stuff.
|
On July 06 2011 06:28 b_unnies wrote:
so the average GSL game is almost 18 minutes thats a lot longer than i expected
That's 18 Blizzard minutes which is just under 13 real minutes
|
Nicce work, interesting, thx
|
How did you gather all this data?
|
On July 06 2011 08:18 decaf wrote:
How did you gather all this data?
http://www.gomtv.net/records/index.gom
Found everything there, I hope it's right. I did find a few errors, for example Jinro vs Rainbow doesn't have a recorded time, but the game lasted 36:28 in game time in the VOD. I also didn't manage to get the correct average for a non mirror matchup in the first season even after triple checking (was off by 20 seconds or so), so I just trusted they were right and used their numbers.
|
Very awesome!
Just as a side question, what is your guy's average game length. I think my TvZ is ~ 30:00, TvT ~ 25:00 & TvP ~20:00. Anyone else?
|
Interesting stuff...
Are you using mean or median for your averages? Given your small sample size I guess mean will have to do but of course this just makes your data way more susceptible to outliers. I wonder if your graphs look any different if you compare mean and median.
Thanks mate!
|
The graphs are generally marred by low sample size. This overall one is this most telling graph, because it collates the full dataset and best answers Day9's proposition that overall gamelength trended upwards.
If you connect the dot between GSL1 and GSL ST, you get an approximate 2-minute (ingame) increase in gameplay time and a generally visible (but high variance) trend upward.
|
On July 06 2011 10:24 Primadog wrote:The graphs are generally marred by low sample size. This overall one is this most telling graph, because it collates the full dataset and best answers Day9's proposition that overall gamelength trended upwards. If you connect the dot between GSL1 and GSL ST, you get an approximate 2-minute (ingame) increase in gameplay time and a generally visible (but high variance) trend upward.
Seems to me that game time stayed about the same if you consider GSLs 1 and 3 as outliers.
|
On July 06 2011 10:38 Cassel_Castle wrote:Show nested quote +On July 06 2011 10:24 Primadog wrote:The graphs are generally marred by low sample size. This overall one is this most telling graph, because it collates the full dataset and best answers Day9's proposition that overall gamelength trended upwards. If you connect the dot between GSL1 and GSL ST, you get an approximate 2-minute (ingame) increase in gameplay time and a generally visible (but high variance) trend upward. Seems to me that game time stayed about the same if you consider GSLs 1 and 3 as outliers.
That is part of the reason I wondered whether these were medians or means. At least you would have some idea of their susceptibility to outliers. In a sample size this small it only takes one or two very long or short games to completely skew the data.
|
so it is statistically proven now that pvp sucks :p
|
How is ZvT and TvZ, PvZ and ZvP different? just curious...
Seems like bad math imo :/
|
On July 06 2011 10:52 Malpractice.248 wrote:
How is ZvT and TvZ, PvZ and ZvP different? just curious...
Seems like bad math imo :/
On July 06 2011 06:26 Binabik wrote:Show nested quote +On July 06 2011 06:24 Primadog wrote:
Good work. Glad that you went through the effort. May I ask for the reasoning behind matchups like ZvP and PvZ separated? ZvP = zerg wins PvZ = protoss wins shows you which race is stronger in lategame
A little reading would not go astray.
As for the math, given the small sample pool there isn't much he can do. At least he weighted them correctly. Give the guy a break.
|
Excellent job! I knew someone would pull through
|
Nice to see all the data compiled. Very interesting
|
Amazing job.
Day9 is gonna be happy .
|
PvP graph quite hilarious
|
Good data, and you clearly took alot of time to gather/document all this data. I still don't think this does justice to how much getting into the late game helps zergies.
|
That. Is one crazy chart.
Mad props :D
|
Dear lord, good work.
I don't remember GSL 3 being that epic, lol.
|
Great data. I'm really surprised though to be honest. I would have expected a dramatic increase in game time from season 1 but that was not the case.
|
On July 06 2011 09:14 Probulous wrote:
Interesting stuff...
Are you using mean or median for your averages? Given your small sample size I guess mean will have to do but of course this just makes your data way more susceptible to outliers. I wonder if your graphs look any different if you compare mean and median.
Thanks mate!
I'm using the mean, and including every match. The reason is that every game counts. I'm not trying to find the weight or length of an item that has a true value, and thus an outlier might be a wrong number. But if the average on ZvP was 5 minutes and the average for PvZ was 20 minutes, they'd all be outliers when compared to the average, but they'd also all be a correct representation when it comes to the average, since I'm trying to find the average of the games played in that tournament, not the true average of that match up.
While the sample size for certain matchups in a season is small, overall the amount of games played in the seasons I'd say are well within statistical significance. 160-170 games each 64 man single elim tourney and 200-210 games each Code S/A is a good amount.
On July 06 2011 10:52 Malpractice.248 wrote:
How is ZvT and TvZ, PvZ and ZvP different? just curious...
Seems like bad math imo :/
Because they all go by the winner. If you can only win by 6 pooling, but you try to stay in the game as long as possible if it doesn't work, then your average win time will be a lot shorter than your average loss time. This isn't win percentages where I can know the other part of the matchup by knowing one part, sure I'll know the time that TvZ losses are if I know ZvT wins, but that doesn't say anything about TvZ wins or ZvT losses.
|
On July 06 2011 08:13 andyrichdale wrote:Show nested quote +On July 06 2011 06:28 b_unnies wrote:
so the average GSL game is almost 18 minutes thats a lot longer than i expected That's 18 Blizzard minutes which is just under 13 real minutes
13 real minutes doesnt matter anyways considering all tournament games is calculated in blizzard minute
|
Sweden33719 Posts
On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs?
GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too?
Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day.
|
Nice work, fun reading through all this
|
Erm... what is the difference between ZvT and TvZ? Why are there two different graphs?
edit: Ah, it's who won the matchup, nvm.
|
Nice work OP! Great info to have.
|
On July 06 2011 16:31 Liquid`Jinro wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day.
Leenock vs Clyde was GSL2 iirc. I can believe that the TvT viking tank stalemates which were hte meta game of the day.
|
Interesting idea to have both sides of a matchup to see if there is some late game advantage, at least statistically, for a given matchup.
|
On July 06 2011 16:31 Liquid`Jinro wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day.
You actually have 2 out of the 3 longest games ever played in GSL, 1st place against Ensnare in GSLJan at 51:30 (1:12:06 game time) and 3rd place against Moon in GSL3 at 39:08 (54:47 game time).
GSL3 did have 2nd to 4th place in game length though(3 in top 10), but GSLJan had 4 in top 10.
Marineking vs Nada was in GSLJan, Nada didn't get into GSL3. Idra vs Ensnare was also GSLJan. Leenock vs Clide was in GSL3, with an average game length of 43 Blizzard minutes in their 3 games.
|
On July 06 2011 16:31 Liquid`Jinro wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day.
GSL3 was the one MC won. The games you mention were all GSL4 (GSLJan) which MVP won.
|
Sweden33719 Posts
On July 06 2011 16:41 Adreme wrote:Show nested quote +On July 06 2011 16:31 Liquid`Jinro wrote:On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day. Leenock vs Clyde was GSL2 iirc. I can believe that the TvT viking tank stalemates which were hte meta game of the day.
Eh, that wasnt at all the meta of TvT in GSL3 - Nada vs Marineking was that long because they had an elimination race that ended with floating buildings all over the map and a really long chase around the map, X2. Me vs Ensnare was that long because we pretty muched mined out the map fighting nonstop.
|
On July 06 2011 17:02 Liquid`Jinro wrote:Show nested quote +On July 06 2011 16:41 Adreme wrote:On July 06 2011 16:31 Liquid`Jinro wrote:On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day. Leenock vs Clyde was GSL2 iirc. I can believe that the TvT viking tank stalemates which were hte meta game of the day. Eh, that wasnt at all the meta of TvT in GSL3 - Nada vs Marineking was that long because they had an elimination race that ended with floating buildings all over the map and a really long chase around the map, X2. Me vs Ensnare was that long because we pretty muched mined out the map fighting nonstop.
Tank lines could be seen as "stronger" back in GSL 3 and Code A early 2011 though. Only mkp used bionic back then and abused the immobility. (like your game against mkp on Xel Naga back then)
Many games were decided by people stupidly running into tank lines and losing because of it, that evolved into tank stalemates and then MvP and MkP excelled in using bionic. highpoint was MvP on Metalopis were he double expanded against a fast expand + siege and just rolled siege tanks with MMM.
|
Wow this is pretty damn interesting thanks for pulling the stats.
|
Its the new long maps that make the all-in games last +2 minutes long on average compared to GSL1.
|
According to data, zergs dominate late zvt. Thoughts?
Is it players' fault or should Blizzard take care of this?
/Protoss player there
|
So, it means that the fact "you need to end the game before 20 in game minutes in TvZ" is true.
|
On July 06 2011 06:47 Dommk wrote:IIRC, just January GSL, Protoss only had like a 23% winrate against Terran out of 50gamess
Slightly exaggerated, it was 34.8% win rate and it was only 23 games.
Anyway I'm slightly surprised by the results, I did kind of expect more of a increase in game length. Though I feel an average 20min game length is probably ideal for a good mix of games.
|
On July 06 2011 17:38 darkness wrote:
According to data, zergs dominate late zvt. Thoughts?
Is it players' fault or should Blizzard take care of this?
/Protoss player there
Only in GSL3 (who cares) and GSL ST (one tournament). Other than that it's fairly even.
eta: well it's actually way more complicated, all it really indicates is what players *try* and do. If a Zerg believes they're best in the late game then most of their wins will come then, regardless of the truth, because they'll play to survive to late then win.
|
Some nice facts here not surprising that PvP and ZvZ are somewhat shorter tho haha
|
On July 06 2011 17:44 Yaotzin wrote:Show nested quote +On July 06 2011 17:38 darkness wrote:
According to data, zergs dominate late zvt. Thoughts?
Is it players' fault or should Blizzard take care of this?
/Protoss player there Only in GSL3 (who cares) and GSL ST (one tournament). Other than that it's fairly even. eta: well it's actually way more complicated, all it really indicates is what players *try* and do. If a Zerg believes they're best in the late game then most of their wins will come then, regardless of the truth, because they'll play to survive to late then win.
That's true, but also Terran has ways to beat Zerg early, while Zerg really only has baneling busts to kill Terran early.
So Zerg wins are naturally going to take longer and Terran wins times are going to be averaged against things like bunker rushes and 2-rax all-ins.
|
On July 06 2011 17:42 tertle wrote:Show nested quote +On July 06 2011 06:47 Dommk wrote:IIRC, just January GSL, Protoss only had like a 23% winrate against Terran out of 50gamess Slightly exaggerated, it was 34.8% win rate and it was only 23 games. Anyway I'm slightly surprised by the results, I did kind of expect more of a increase in game length. Though I feel an average 20min game length is probably ideal for a good mix of games.
http://www.gomtv.net/records/index.gom?searchType=3&race=P&vsrace=T&season=0&leaguetype=20&leagueid=20119&gamever=0&mapid=0
In GSL January, Protoss won 12 out of 44 games against Terran, a 27.3% win rate. For sets it was 9 out of 31, a 29% win rate. ZvP is what you said though.
|
I wonder why the 20 minute mark is the average game length. Any ideas?
I would say that you mine out 2 bases around that time?
|
On July 06 2011 21:21 Koshi wrote:
I wonder why the 20 minute mark is the average game length. Any ideas?
I would say that you mine out 2 bases around that time?
nah it's because it's most likely to reach a really huge army by then (200/200) and if one player fails at controlling or has a worse army he's losing the fight and lost the game
|
The most telling fact was the data on Terran wins.
On average, terran is most devastating if the game goes to the late-mid game.
|
On July 06 2011 21:21 Koshi wrote:
I wonder why the 20 minute mark is the average game length. Any ideas?
I would say that you mine out 2 bases around that time?
Remember that's just the average. You could get 20 minute average with 4 minute 6 pools and 36 minute 5 base epic macro games too. A proper statistical analysis would require writing over every single result (I only had to calculate average on mirrors myself, which was around 400 of the 1368 games) and do some sorting on time brackets in addition to the matchups.
|
272 TvT's
WOW that's a lot of mirror
|
Nice write up, always really cool to see this info compiled together.
/tipofmyhat
|
On July 06 2011 21:32 Lobo2me wrote:Show nested quote +On July 06 2011 21:21 Koshi wrote:
I wonder why the 20 minute mark is the average game length. Any ideas?
I would say that you mine out 2 bases around that time? Remember that's just the average. You could get 20 minute average with 4 minute 6 pools and 36 minute 5 base epic macro games too. A proper statistical analysis would require writing over every single result (I only had to calculate average on mirrors myself, which was around 400 of the 1368 games) and do some sorting on time brackets in addition to the matchups.
But it is around 20 minutes in every match-up. I would say that with these statistics we could argue that there is something around the 20-25 minute mark that is pretty important. Because if the average is around the 20 minute mark I want to conclude that "normal" games end around the 20-25 minute mark.
Not pointing a finger to you of course, but what I find even more annoying than treacherous statistics are the snarfs that in every damn statistic thread say: "this isn't enough data, this is crap". You used data from all the games from the GSL, and we are discussing the length in the GSL games. Data can always be used, if used carefully.
On July 06 2011 21:30 Binabik wrote:
nah it's because it's most likely to reach a really huge army by then (200/200) and if one player fails at controlling or has a worse army he's losing the fight and lost the game
Maybe a partial reason but I don't think this is the main reason. I watched close to every game so far and I don't feel it mostly ends because of that. It just ends because mistakes start to pile up. But why does it end around that 20 minute mark?
|
|
|
I think if it's possible a graph that would be interesting to see is the winning percentage of a match where the x-axis is the time that the game lasted and the y axis is some kind of bar graph (or line graph) like you have here that represents the percentage of a race to win that match up, that way we can see when each race actually wins the most in comparison to how much it loses
|
On July 06 2011 22:18 Kiarip wrote:
I think if it's possible a graph that would be interesting to see is the winning percentage of a match where the x-axis is the time that the game lasted and the y axis is some kind of bar graph (or line graph) like you have here that represents the percentage of a race to win that match up, that way we can see when each race actually wins the most in comparison to how much it loses
That could be done, where each bar was a 1 or 2 minute interval. I reckon it would have to be across all GSL tournaments, or else you'd get a lot of bars that were 100% or 0%, meaning that there was only one game that ended in that time interval.
|
On July 06 2011 16:31 Liquid`Jinro wrote:Show nested quote +On July 06 2011 06:21 ZAiNs wrote:
GSL3 has the longest on average? I thought that was the cheese season of 2 Rax SCV all-inning 90% of TvZs? GSL3 had me vs Ensnare (50 minutes), MarineKing vs NaDa (2x40 min games pretty much), me vs MC game 1 (prob almost 30 min?), Ensnare vs IdrA (prob at least 30 min?)... and so on. Lots of really long games. I think it had Leenock vs Clide too? Anyway, 13 min avg game, eh not too bad - Id imagine BW is about 15, at least thats what my avg length was when I checked my old ladder database back in the day.
Your memory jinro... GSL 3 was the first GSL you qualified for. YOu came to the semis losing to mc.
|
Does this include both Code A and S?
|
On July 06 2011 23:42 Gamegene wrote:
Does this include both Code A and S?
Yes, everything GSL that has been concluded besides the World Championship (because it doesn't show up in the stats).
|
|
|
|
|
Good work.
Question1: Is there a reason you decided against showing standard deviation in your graphs? Introducing RMS deviation might be a quick way to add information value.
Question2: What are your thoughts on the mappool changing over the seasons? Bigger maps = longer games appears to be a reasonable explanation for slightly longer games.
|
On July 07 2011 00:21 xian_ wrote:
Good work.
Question1: Is there a reason you decided against showing standard deviation in your graphs? Introducing RMS deviation might be a quick way to add information value.
Question2: What are your thoughts on the mappool changing over the seasons? Bigger maps = longer games appears to be a reasonable explanation for slightly longer games.
I didn't calculate it at all, and neither did GOMTV.
Their stats page looks like this
http://www.gomtv.net/records/index.gom?searchType=3&race=T&vsrace=P&season=0&leaguetype=10&leagueid=21&gamever=0&mapid=0
where I get average game length in a matchup in a specific tournament. It says on the same page how many games were played, but if I want to find the SD I'd have to write in every single time for every single matchup and calculate that, and since I'm only using OpenOffice in a simple way I didn't want to start with everything at once, especially since there's almost 1400 games.
I did calculate SD for the first two GSTL, not sorted by matchup:
GSTL1: Average 1126.23 seconds, SD 445.14 seconds, n = 40.
GSTL2: Average 1175.5 seconds, SD 487.70 seconds, n = 42.
If we then consider those as independent variables, and calculate the chance of them randomly being that far from each other, I got that there was a 36.8% chance of being within +- 49.27 seconds given that the true average was the same in both tournaments, which means I can't really conclude that it did change.
|
+10 Internets! great job =)
|
On maps, bigger maps means that games will last longer even without the strategies changing in the matchups.
Some quickstats on maps, all are in real time:
Steppes of War had 58 games played in GSL1-GSLJan. Out of those, we had
8 under 6 minutes, 25 between 6 and 12 minutes, 22 between 12 and 18 minutes, and 3 above 18 minutes (25 minutes being the longest and also only above 20 minutes).
In comparison Tal'darim Altar have 108 games total, with 8 games under 6 minutes, 42 between 6 and 12 minutes, 37 between 12 and 18 minutes, and 21 above 18 minutes.
For steppes the percentages in the time brackets were 13.8%/43.1%/37.9%/5.2%, and for Tal'Darim they are 7.4%/38.9%/34.3%/19.4%.
|
I think the fact that you didn't weight by the amount of games might have thrown something off. I was currious about that high average game length for GSL3, and after taking all 162 games i calculated the average length to be 15:08. I might be wrong, but this is what i got.
|
It's really interesting that TvX is consistently longer and PvX is consistently shorter. What does that mean about the game in general? Very interesting. Nice write-up.
|
On July 07 2011 05:21 Slope wrote:
I think the fact that you didn't weight by the amount of games might have thrown something off. I was currious about that high average game length for GSL3, and after taking all 162 games i calculated the average length to be 15:08. I might be wrong, but this is what i got.
That's correct, and the same thing I got.
The first part wasn't weighted, but there it didn't need it because I didn't try to make any graphs for the total average game time, it was just a sum column to get a rough idea of it.
The second part took the average game times, multiplied them by the amount of games it was calculated from, and divided it again by the games to get the correct and weighted average for everything, different matchup over all seasons and different seasons over all matchups.
You got 15:08, which is the number you get by using the times on the gom site. That's the real time. I used that as input, but I multiplied everything by 1,4 to get the Blizzard time. 15:08 is 908 real seconds, multiply that by 1.4 and you get 1271 Blizzard seconds, divide that by 60 and you get 21.187, which is pretty close to the 21.18 that I used in the graphs and that shows in this:
![[image loading]](http://i.imgur.com/fNANx.png)
|
PvP still looking like not much has changed over time ;;
|
Oh, I didn't know that was real time not blizzard time. Now it makes more sense.
|
I wonder why Liquipedia has very different numbers for the open seasons 2 and 3 and GSL January. The rest is pretty close to your numbers though.
GSL2 15,41 vs 18,81
GSL3 15,56 vs 21,18
Jan 17,03 vs 19,81
|
On July 07 2011 06:24 pusmoh wrote:
I wonder why Liquipedia has very different numbers for the open seasons 2 and 3 and GSL January. The rest is pretty close to your numbers though.
GSL2 15,41 vs 18,81
GSL3 15,56 vs 21,18
Jan 17,03 vs 19,81
I did a test on GSL2. Going by the average game length(11:12) and the length/amount on different maps, the average time was calculated correctly from the map averages. But when I calculated the average time for the 9 games on Blistering Sands, I got 12:26 real time instead of the 10:28 they had. So at the moment I'm trusting that my numbers are correct, of course the gom numbers might have changed between the time Liquipedia pulled them and now.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/XRCD4.png)
![[image loading]](http://i.imgur.com/qCP5c.png)
![[image loading]](http://i.imgur.com/Tzo5e.png)
![[image loading]](http://i.imgur.com/6Xq3s.png)
![[image loading]](http://i.imgur.com/sOBXT.png)
![[image loading]](http://i.imgur.com/pRg83.png)
![[image loading]](http://i.imgur.com/OxNdj.png)
![[image loading]](http://i.imgur.com/NEpLZ.png)
![[image loading]](http://i.imgur.com/OGOmE.png)
![[image loading]](http://i.imgur.com/WON3x.png)
![[image loading]](http://i.imgur.com/wt3GH.png)
