|
Part I : Win rates by Matchup & Length of Game
Part II : Win rates for "Pro" players vs "Scrubs" by matchup & Length of Game
For my next big attack on the Playhem data, I want to dig into build orders to see if we can identify "allin" build orders and calculate their effectiveness, independent of the length of the game. But before I do that, which is going to require figuring out , I want to take a brief detour into the results Pro vs Scrub matchups, to help answer the question "How often does the better player win?".
The definition of "pro" used previously isn't really suitable, since I was using the same data set to both determine who was a pro and then analyze pro win rates. Oops. I'd really like to have a defintion of "pro" that's independent of the data. But, that's not exactly easy. So instead I'll do the next best thing; I'll split the data set in half, use one half to decide who's a Pro and who's a Scrub, and then use the second half for analysis. Games played on even days were used only to identify Pros, while games played on odd days were used to analyze matchups. In addition, I strengthened the definition of pro to "any player who playes 7 games in the Ro8 (but ignoring Ro4, Ro2, etc.). In essence this means that during the even day tournaments, you must make the Ro8 at least three times. There are 96 pros in the playhem data set. The Pros make up a similar share of the player pool at all rounds up until the Ro2 in the odd-day tournaments as they do in even-day tournaments.
Let's look at the rate at which pros lose to Scrubs in Pro-vs-Scrub matches, based on the race of the Pro. In this chart, the Pro is the first race listed, which is why there are entries for both PvZ and ZvP. There are 3176 games in the entire data set.
![[image loading]](http://farm8.staticflickr.com/7067/6804589646_e793255ce1_z.jpg)
Playhem: Rate of Pro Losses to Scrubs by niq77174, on Flickr
Consistent conventional wisdom, PvP is the most "random" of the matchups. A Protoss Pro is only a 3-1 favorite against a Protoss Scrub, while a Terran Pro is an 8-1 favorite over a Terran Scrub. ZvZ is also somewhat random; 5.5-1 favorite vs an average 6.5-1 favorite across all matchups.
If you add up the losses to Scrubs across all matchups, then the overall Rate of losses to scrubs for each race looks like this:
Terran: 10.9%
Zerg: 15.7%
Protoss: 17.6%
Conversley, here is the rate at which Scrubs of a given race advance vs Pros:
Terran: 10.9%
Zerg: 16.4%
Protoss: 16.4%
Having seen these numbers, let me extend an olive branch to the Protoss. For while they have a great a-move army in the late game, they seem to face a number of other difficulties.- In mirror matches, PvP is the matchup where the weaker player is most likely to advance. By contrast, in TvT the weaker player is least likely to advance.
- In non-mirror matches, Protoss is overall the weakest race (both Z and T are above 50% vs P), while Terran is the strongest race.
The randomness of PvP (and to a lesser extent ZvZ) suggests that Zerg and Protoss players may want to avoid the mirror match by offracing in tournaments that are not race-locked. It may also explain the difficulties Protoss have had in tournaments. If it's more likely that weaker Protoss players advance, then they will be easy pickings in later rounds.
Why are PvP and ZvZ so random? Why is TvT so not random? Of the non-mirror matches, why is ZvP the most random? I have no idea what the answers to these questions are. Any conjectures?
|
Well I think ZvZ and PvP may have a higher randomness because in both cases build order choice can play a large role, and may simply decide the outcome.
Also in ZvZ, ling/bane wars are quite volatile, and while the player with better control will generally win, one slip up early on could cost you the game
in PvP one force field on the ramp could cost you the game.
I think the only real equivalent in TvT would be if hellions get into your mineral line early on...
In general though, I think PvP and ZvZ are just more volatile, so it is easier for non-pros to take pros out. This is one of the reasons that Nestea's ZvZ undefeated streak was so amazing.
|
pvp is a bad matchup as is some zvz so that makes sense
tvt is the best matchup in game and thus is the best stat wise here
|
On March 04 2012 12:06 orangesunglasses wrote:
pvp is a bad matchup as is some zvz so that makes sense
tvt is the best matchup in game and thus is the best stat wise here
your bias is showing a wee bit.
|
On March 04 2012 12:18 sc14s wrote:Show nested quote +On March 04 2012 12:06 orangesunglasses wrote:
pvp is a bad matchup as is some zvz so that makes sense
tvt is the best matchup in game and thus is the best stat wise here your bias is showing a wee bit.
I'm Protoss and I agree. PvP is god-awfully bad and has an unshakable coinflip feel to it. It's like there's a cycle of builds that blind counter other builds, and you just roll whatever one you're most comfortable playing. Lategame PvP is downright stupid, what with mass Colossi/MS butting heads. Complete a-move.
|
On March 04 2012 12:02 j3cht wrote:
Well I think ZvZ and PvP may have a higher randomness because in both cases build order choice can play a large role, and may simply decide the outcome.
Also in ZvZ, ling/bane wars are quite volatile, and while the player with better control will generally win, one slip up early on could cost you the game
in PvP one force field on the ramp could cost you the game.
I think the only real equivalent in TvT would be if hellions get into your mineral line early on...
In general though, I think PvP and ZvZ are just more volatile, so it is easier for non-pros to take pros out. This is one of the reasons that Nestea's ZvZ undefeated streak was so amazing.
This is pretty much my thoughts on the matter. PvP and ZvZ are extremely volatile matchups with a very high focus on your opening and extremely minor mistakes. PvZ/ZvP games are very often decided by what you do in the midgame in my opinion, with a huge focus on unit composition and the protoss' ability to either secure a third effectively without falling hugely behind, or end the game right there, making it easier in that matchup to beat pros as well. In essence, the more decisions, engagements, and game knowledge a particular game requires in any matchup, the greater the chance is the "better" player will win, while the fewer that the outcome is dependent on the greater the chance the "worse" player will get lucky and come out ahead.
Edit: Interestingly, this also gives somewhat of a vindication to my belief that if you learn how to execute build orders proficiently into the midgame, you can get a 15% winrate against just about anyone. The few who can micro their way out of straight out of straight build order losses against people who can do this are truly exceptional.
|
I think we're seeing these ratios because not all of the MU's are understood. Players like NesTea have the best ZvZ in the world because of their UNDERSTANDING of the matchup. I think that the ratio is simply caused by the fact that at this stage in the game, the people who are in GM and high high masters RIGHT NOW are there because of excellent mechanics and an understanding of what playstyles have been explored SO FAR in the game.
I think in a few years we'll be seeing these ratios go way down (especially in ZvZ, I personally feel like this MU isn't nearly as random as people say it is). The only MU I think could stay volatile is PvP simply because of the build order rock paper cissors we seem to have.
Also, IMO ZvP is the most volatile non-mirror MU because very often it boils down to protoss hitting a specific timing that zerg needs to scout extremely well to defend adequately. The ammount of different timings is quite large so I think we'r seeing good players die to all-ins they've never seen (let's say a zerg not expecting DT's after counting a certain number of chronos on the forge and thinking warpgate pressure). Or protoss players being caught off guard by an inferior zerg player cancelling his 3rd hatch and baneling busting. Mainly the all-ins haven't all been fleshed out and identified by everyone 
|
glorious tvt 
nice data, thanks.
|
I did want to add that my theory for why TvT is so non-volatile can be summed up in one word:
Tanks.
|
On March 04 2012 12:18 sc14s wrote:Show nested quote +On March 04 2012 12:06 orangesunglasses wrote:
pvp is a bad matchup as is some zvz so that makes sense
tvt is the best matchup in game and thus is the best stat wise here your bias is showing a wee bit.
in an MLG interview Huk said pvp is still the worst matchup in the game.
And from a skill wise perspectve (ignoring 'boredom') I'd agree that TvT is the best matchup, followed by tvz.
|
The data is unsurprising but It still is great that you have confirmed what everyone had already suspected.
|
On March 04 2012 12:51 xrapture wrote:in an MLG interview Huk said pvp is still the worst matchup in the game.
And from a skill wise perspectve (ignoring 'boredom') I'd agree that TvT is the best matchup, followed by tvz.
Certainly from a top player's perspective PvP is the worst. In playhem there's a serious skill gap, and the better player still loses 1 in 4. When there is less of a skill gap, the situation gets much worse. I would suspect that once you get to HuK's level, you have close to no control over your PvP win rate against anyone capable of making it into Code A.
If there's a GM Protoss who wants to post their win rates in each matchup that'd be useful .
|
On March 04 2012 13:13 ZeroTalent wrote:Show nested quote +On March 04 2012 12:51 xrapture wrote:in an MLG interview Huk said pvp is still the worst matchup in the game.
And from a skill wise perspectve (ignoring 'boredom') I'd agree that TvT is the best matchup, followed by tvz. Certainly from a top player's perspective PvP is the worst. In playhem there's a serious skill gap, and the better player still loses 1 in 4. When there is less of a skill gap, the situation gets much worse. I would suspect that once you get to HuK's level, you have close to no control over your PvP win rate against anyone capable of making it into Code A. If there's a GM Protoss who wants to post their win rates in each matchup that'd be useful .
Well, even as a master's Terran I find it nearly impossible to lose to a diamond Terran. Is it the same way for master P and Z's with their mirrors?
|
Fucking love this chart.
Shows the real problems with PvP and ZvZ, and the fact that they're extremely volatile matchups.
Thanks for posting!
|
yea god i hate PvP with a passion. sometimes i want to switch races, but i just enjoy playing toss too much. but then i get like 3 PvP's in a row and want to kill myself.
|
I believe this picture explains how mirrow matchups work
+ Show Spoiler +
The defensive building is normally tied to the teching build for Terran and Zerg. Without bunkers or sunkens, ZvZ and TvT would be very similiar to PvP. But because cannons are tied to the forge, we have many build order wins/losses. Starcraft BW got around this problem by having the shield battery, a weaker defensive building tied to the teching building - the gateway. You need to get a defender's advantage from somewhere.
ZvZ really is a knife fight. Even if you're better with a knife you're still on a razor's edge. Plus, the level of execution you need is very high, so among lower level players, it's essentially a luck based matchup.
|
On March 04 2012 13:23 xrapture wrote:Show nested quote +On March 04 2012 13:13 ZeroTalent wrote:On March 04 2012 12:51 xrapture wrote:in an MLG interview Huk said pvp is still the worst matchup in the game.
And from a skill wise perspectve (ignoring 'boredom') I'd agree that TvT is the best matchup, followed by tvz. Certainly from a top player's perspective PvP is the worst. In playhem there's a serious skill gap, and the better player still loses 1 in 4. When there is less of a skill gap, the situation gets much worse. I would suspect that once you get to HuK's level, you have close to no control over your PvP win rate against anyone capable of making it into Code A. If there's a GM Protoss who wants to post their win rates in each matchup that'd be useful . Well, even as a master's Terran I find it nearly impossible to lose to a diamond Terran. Is it the same way for master P and Z's with their mirrors?
Same for me. I played on my friend's diamond league account for 2 days and won every single TvT. When I offrace as Protoss I can sometimes beat master league players in PvP but at the same time can lose to platinum league players aswell, the matchup seems so random when I play it, sure I'm not really a protoss player and don't understand the matchup too well, but in TvT I always feel like the better player wins, in PvP I don't have that same feeling at all.
edit: The numbers are actually really good for all matchups except PvP, when the better players wins around 8-9 out of 10 matches on average then the game is extremely skill-based, and luck is not that much of a factor like some people (namely idra) make it out to be. Sure there are all-ins and there is some luck factor, but when the better players wins 9 out of 10 games I think that you can't really argue that sc2 is very luck-based.
edit2: In most sports upsets happen and there is a chance the underdog wins, that's a good thing, as long as it is based on skill and there only is a small luck factor, which is exactly the case in SC2. It's mainly skill but there is a slight chance for an upset with a little bit of luck. You see that in european football all the time, where top teams sometimes lose to "worse" teams, it doesn't happen very often, but just like in SC2 it happens sometimes and makes everything more interesting. It might not even be luck anyway, just a top player or pro player having a bad game and the "scrub" playing extraordinarily well.
|
Personally I don't like how Playhem is formatted. There are so many best of 1's in the beginning rounds that, if you're not absolutely on your game in those early rounds, you're gonna get knocked out with no chance to redeem yourself. The statics are interesting, but not all that surprising to me.
However, I do hope to see some changes in the PvP and ZvZ match ups and some point to make them less volatile @_@
|
I am really interested in how these numbers pan out when cross referenced with how effective cheese/all-ins are. Although I have no stats to back it up, I always felt they seemed more effective that they should be. Still, these stats do confirm that feeling I get during PvP is totally natural and the match up is sort of a weird mess.
|
On March 07 2012 01:07 ChaosTerran wrote:
edit: The numbers are actually really good for all matchups except PvP, when the better players wins around 8-9 out of 10 matches on average then the game is extremely skill-based, and luck is not that much of a factor like some people (namely idra) make it out to be. Sure there are all-ins and there is some luck factor, but when the better players wins 9 out of 10 games I think that you can't really argue that sc2 is very luck-based.
edit2: In most sports upsets happen and there is a chance the underdog wins, that's a good thing, as long as it is based on skill and there only is a small luck factor, which is exactly the case in SC2. It's mainly skill but there is a slight chance for an upset with a little bit of luck. You see that in european football all the time, where top teams sometimes lose to "worse" teams, it doesn't happen very often, but just like in SC2 it happens sometimes and makes everything more interesting. It might not even be luck anyway, just a top player or pro player having a bad game and the "scrub" playing extraordinarily well.
I think this is right. Though, for the most we're talking about pro level players versus very serious casual players or up-and-coming pros here. It would be like taking the MLB All-Star team and having them play a AAA allstar team. Sure, the AAA team would win a few times (just look at the World Baseball Classic) but it's so rare that it's close to dumb luck.
I may try to re-run these numbers using more buckets for players than "pros" and "scrubs" to see what happens when one player is only a slight favorite. Because being an 8-1 favorite is actually a very uninteresting game from a spectators perspective. Even in the NBA, the best team finishes the season at something like 65-17, which is a 4-1 favorite or so. The '97 Bulls were a 7-1 favorite. Great NFL teams can go 15-1 or 14-2, but that's rare; in the typical season the best team is 13-3 or 12-4.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/YWnCb.jpg)
