|
I wanted to do a version of this post on win rates and game lengths using a larger data set. Thankfully, Playhem has recently made replay packs for all tournaments available as long as you are logged in to your playhem account. I downloaded almost all of the completed NA Playhem daily tournaments. This produced a dataset of over 21,000 replays ... over 10x the size of the MLG Providence dataset. I used SC2 gears and some excel kung-fu to analyze the results based on the length of the game. If you're interested in the raw data, please PM me; the Excel file isn't compatible with Google Docs.
Different Game Lengths in Different Matchups:
Lets first take a look at all 6 matchups to see if games in a certain matchup tend to be shorter or longer than other matchups. The first graph shows the percentage of games in each matchup that end exactly in that time bucket. The second graph shows the cumulative percentage of games that have ended by the time we reach the end of that time bucket. Note that these timings are using Game Time (i.e. the time you see on the in-game timer), not wall clock time.
:![[image loading]](http://farm8.staticflickr.com/7180/6784379544_67f398910f_z.jpg)
Game_Length by niq77174, on Flickr
![[image loading]](http://farm8.staticflickr.com/7062/6930519265_23af0d446f_z.jpg)
Game_Length_Cumulative by niq77174, on Flickr
Analysis:
PvP and ZvZ appear to be very different games from the other four matchups. In those two mirror matches, players face a more than 40% chance of seeing the game end in the early game. In no other matchup does the chance of losing in the early game go above 20%.
PvZ Over Time (5636 games):
Despite the high win rate for Zerg in the early game, the PvZ matchup is quite balanced across all phases of the game:
![[image loading]](http://farm8.staticflickr.com/7179/6930496375_1c3c839d5a_z.jpg)
Playhem PvZ Win Rates by Game Length by niq77174, on Flickr
Yes, the raw PvZ win rate shows an extremely high win rate for Zerg in the early game, followed by a very even matchup. However, PvZ is also the matchup that is least likely to end before the 10 minute mark. This is at least circumstantial evidence that, if an early-game attempts by Zerg to kill of the protoss that fails, it produces longer games, where Protoss is at less of a disadvantage. A closer analysis of early-game PvZ wins, most likely examining build orders, is needed to identify failed Zerg all-ins and determine the win rate in longer games.
Beyond the 10 minute mark, neither race experiences an Win Rate above 55% that suggests serious imbalance until the 30-35 minute mark, when protoss briefly has a win rate just above 55%. However, the number of games that reach the 30 minute mark (324 games total) may be too small to draw any meaningful conclusions.
ZvT Over Time (4532 games):
The ZvT matchup shows a significant advantage for Terran until the 15 minute mark, with Zerg slowly working its way back unto the game, and even to a significant advantage at the 25 minute mark. But then, the Zerg advantage begins to erode; past the 30 minute mark, there are 254 total games, and Zerg has a 52.3% win rate in those games.
![[image loading]](http://farm8.staticflickr.com/7040/6930496419_f8fb7191db_z.jpg)
Playhem ZvT Win Rates by Game Length by niq77174, on Flickr
ZvT is the slowest of all 6 matchups; slower even than TvT (!). As in PvZ, the high early-game win rate against Zerg is slightly deceptive. ZvT has the fewest games which end in the 10-15 minute range and before the 15 minute mark overall.
TvP Over Time (3997 games):
The TvP matchup swings wildly. Protoss appears to be strongly favored in the early game; then Terran takes a big advantage in the midgame, then things swing back into protoss's favor in the lategame. However, in a small sample (24) of extremely late games, the matchup appears to even out.
![[image loading]](http://farm8.staticflickr.com/7192/6784379496_43648f6efb_z.jpg)
Playhem TvP Win Rates by Game Length by niq77174, on Flickr
Unlike Zerg in PvZ or ZvT, each race in TvP faces significant risk of losing during the phase of the early & midgame where the opposing race favored. Excluding ZvZ and ZvP, TvP is 2nd most likely to in before 10 minutes, and is the matchup most likely to end in the 10-20 minute window. This suggests that both races may be trying to take advantage of the the point in the game where their race has the edge.
Do you have an alternate interpretation of these results? Thoughts on what it is about the state of the current metagame and/or balance in general that would produce the results that we see here? Are there other questions you would like to see answered?
|
Seems to be pretty accurate.
|
Seems like very interesting analysis, definitely timings within games that people seem to hit to win. I do wonder though whether knowing data like this exist can help people look at their match ups to sacrifice say winning percentage at certain points to help smooth out the line. That is, if I know toss has the best shot at beating me before 10 minutes, i'm going to be defensive til 10 until i get some sort of timing window.
Though I wonder if there is correlation between the different windows. Like this the fact that toss wins before 10 at a high percentage lead to the fact that they lose after 10 at a high percentage. Be interesting to analyze build orders, but I think that takes a lot more data crunching capability that I imagine few people have.
|
I expected the tvp graph to be honest. Terran has the best opening units and can punish protoss players who are gearing up for the mid to late game, but around when colossus, HT and 3/3 speed lots enter it becomes a very hard fight for terran.
what I wasen't expecting was the pvz graph. Honestly expected protoss to be in a bigger lead by around the 25+ minute mark, yet stays relatively even up until the point where there is insignificant data to draw any conclusions.
|
|
|
I think the PvT graph is quite telling. Deathball = very strong apparently :D
|
Great post. The bigger data set is a huge plus in terms of statistical reliability. I also liked the way you cleaned up your graphs, they're much better this time around :D
Very interesting stuff, seems to fit with what most people seem to think for the time being. Surprised TvZ is longer than TvT though, that seems hard to believe 
|
I'm really interested in what major things occur where the graph lines intersect. Specifically, in ZvT the ~18 minute mark looks like a significant change occurs. I'd have to watch some replays of it but maybe that is when zerg typically gets their brood lords out. Or perhaps the shift is reflective of the cumulative effect of 5-8 minutes of mutalisk harassment?
What about TvP, is that first intersection when terran has their MMM rocking and rolling? What is that second intersection? Does protoss now have templar AND colossi? I think this could be revealing of some very typical timings in the current metagame, and more importantly insight on how to deal with it as each race.
|
The ZvZ winrate makes me sad. Macro ZvZs tend to be some of my favorite games, but everyone just does 9 pool or 14/14 baneling all-in. It's really annoying when the entire game comes down to ling/bling micro -_-
|
On February 26 2012 16:12 synapse wrote:
I think the PvT graph is quite telling. Deathball = very strong apparently :D
strong assuming you survive to that point. From the cumulative graph, it seems something like 80% of tvps end before 20 minutes anyway.
|
Ah! I was hoping that you would get around to using the playhem data. These graphs unfortunately confirm all of my suspicions about everything ever. =(
I guess a game where everything was a flat line would be pretty boring too.
|
Honestly, the early game and late game balance does not look good at all, according to these. As you stated in the op, it would be interesting to see how much of the mid game appears balanced due to failed all ins or cheese.
|
On February 26 2012 16:25 _scout wrote:
I'm really interested in what major things occur where the graph lines intersect. Specifically, in ZvT the ~18 minute mark looks like a significant change occurs. I'd have to watch some replays of it but maybe that is when zerg typically gets their brood lords out. Or perhaps the shift is reflective of the cumulative effect of 5-8 minutes of mutalisk harassment?
What about TvP, is that first intersection when terran has their MMM rocking and rolling? What is that second intersection? Does protoss now have templar AND colossi? I think this could be revealing of some very typical timings in the current metagame, and more importantly insight on how to deal with it as each race.
I think you're right with both, the zergs would be on hive and getting broods out at around that time. Around 10 minutes is when I always have a good number of marine/marauder and medivacs, and that's when you can take control. Toss doesn't have much aoe out, so you can win the game pretty easily from there. Once they get out storms + colossus it gets really hard and that's where P has the advantage.
|
On February 26 2012 16:40 Reborn8u wrote:
Honestly, the early game and late game balance does not look good at all, according to these. As you stated in the op, it would be interesting to see how much of the mid game appears balanced due to failed all ins or cheese.
Early-game winrate means nothing, as a game ended early is a cheese play 99.99% of the time or a very successful rush play that forgoes fast expansion to deal damage into midgame or outright kill an opponent.
That being said, winrates at different times mean different things in game. Terran will seem strongest vs other races more and more in midgame and early game bcause we're moving toward constant harassment gameplay and most late-game wins come from damage dealt in midgame that the P stuck it out OR from a big drop that wasn't handled well/denied bases. Same could be said, increasingly, vs zerg. Playstyle is becoming a contributing factor to winrates, as "macro" terrans will, in general, lose more without doing significant damage with their timings and will do so when protoss AOE is on the field en masse.
While I would love a buff to help Terran win a game if you started with 5-base, split map, all tech'd, and all upgraded with standard production infrastrucutre (ie super late-game/late-game scenarios) I would also love to see playsets change again. Terran has seemingly done this most consistently, and I expect the race to do so again. Meanwhile in winrate land, we'll be racking up wins in early/mid because of aforementioned harassment play that can break opponents to gg if they don't respond well.
|
could you put up % of pros that play each race and how that might affect the stats?
|
The TvP one makes a lot of sense. Protoss gets favored for all ins, terran gets favored for midgame pushes before protoss tech, protoss favored after that with deathball.
Also now that I'm seeing the TvZ one, it reaffirms the epic fail that was the ghost nerf.
And I agree the fact that these are not more even in all three seems like a bad sign, not the best way to balance the game IMO.
|
I can see the problem Blizzard has in balancing with these graphs. Make terrran weaker at start for zerg's benefit and protoss will just roll terran early. But you can't make protoss weaker because they are already a little weaker than zerg early.
Ghost Snipe nerf and EMP nerf was a total fail by this graph. Terran is already weak late, can't tech switch like other race, make them weaker?
|
On February 26 2012 16:12 MercilessMonkey wrote:Great post. The bigger data set is a huge plus in terms of statistical reliability. I also liked the way you cleaned up your graphs, they're much better this time around :D Very interesting stuff, seems to fit with what most people seem to think for the time being. Surprised TvZ is longer than TvT though, that seems hard to believe
When it's fun times goes fast. 
|
winning percentages based on when game ends is super interesting, it sort of confirms my suspicions of certain matchup's powercurve.
|
I think the colors on the top graph are too close its hard to tell them apart, at least for me.
|
|
|
|
|
|
EPT
