|
EDIT 26.12: Thank you for your feedback (especially Kokujin, TheBigO who found inaccurate data in the chart)!
Dear TL,
The ongoing discussions concerning the shape of the SC2 scene, the impact of the recent WCS revamp, progamer salaries and SC2 viewership made me think about progamer careers in general. How long does a career in SC2 last? Does the race matter? When do progamers tend to retire? How often do we see "fresh blood" or, in other words, pros who have notable careers in SC2 even though they have not been present since WoL?
To do it, I imported data from the best source that I know: Liquipedia. The rules were as follows:
Player pool
My general idea was to include all notable SC2 players. I was a bit limited by the quality of data, especially with regard to foreigners who did have substantial playoff runs in premier tournaments. The player pool for which I made my calculations includes almost all of players who participated in the following events:
+ Show Spoiler +Blizzcon 2010
Blizzcon 2011
2012 Battle.net World Championship participants
WCS 2012 Continentals top8 (top4 for Oceania and South America as they gave just 2 seeds and 3 seeds respectively)
WCS 2012 Nationals runner-ups
WCS 2013 Top 64 standings
WCS 2014 Top 64 standings
WCS 2015 Top 64 standings
All top8 NASL
All top8 WCG
All top8 DH
All top8 IEM
All top4 IGN Starleague
All top8 ESW
All top8 Iron Squid
All top8 ASUS ROG
All top8 TSL
All top8 GSL Open
All Code S and OSL ro32 and SSL Challenge participants
All WCS Premier ro32
All players who were on ProLeague teams (even if they were never fielded)
I did not have enough data with regard to most of Air Force ACE players and some other members of KeSPA teams (mostly those who were not fielded in ProLeague games).
Calculation rules
Again, I was limited to what Liquipedia offers. The general idea behind these rules was to err on the side of caution, or in other words to lenghten the career in case of any doubts.
The rules were as follows:
+ Show Spoiler +
Career start
1. Date of first known result in a tournament or date of joining a team for a first time, whichever is earlier
2. If 1) not applicable or impossible to determine, release date of the SC2 expansion in which the player first became active.
3. If 2) not applicable or impossible do determine, release date of WoL.
Career end
1. Date of announced retirement.
2. If 1) not applicable, but retirement confirmed, date of last appearance in a tournament or date of leaving the last team, whichever is later.
3. If 2) not applicable, retirement not confirmed, but there have been no tournament appearances since 31.12.2014, date of last appearance in a tournament or date of leaving the last team, whichever is later (unless the player is still listed as active by an active team).
4. If 1-3) not applicable, the player is considered active and for the purpose of calculation the end date is set as 31.12.2015.
Other rules
+ Show Spoiler +I listed only the race that the given player is most known for or has had most success playing (i.e. Zerg for TLO, even though he spent some time playing Terran).
Caveats
+ Show Spoiler +1. I have no background in statistics. Any criticism from anyone with such background would be vastly appreciated. 2. I did not have time to double check the entire player pool (405 players). I randomly (random.org) selected 30 players for double-checking and found 4 cases of bad data (mostly missed tournaments) which on average resulted in shortening careers by ~40 days. I would say it is safe to assume that the +/- caused by my mistakes would be +/- 30 days. 3. The data is only as good as Liquipedia and have to be interpreted in the context of the rules I assumed. Apart from possible mistakes in Liquipedia data, the following factors are not taken into account: - preparation time before tournaments (Korean players who were first fielded in ProLeague Hybrid season or debuted in GSL Open are mostly affected), - inactivity in 2015 if retirement was not announced and last tournament took place in 2015. 4. As TheBigO correctly pointed out, several players' careers are overestimated because they were not playing consistently full time during their career, which the chart did not take into account. Full discussion here.
Raw data
+ Show Spoiler +
Without further ado, my conclusions:
1. On average, the duration of a career of a SC2 progamer is 3 years and 2 months (1173 days). The median is a bit longer and equal to 3 years and 6 months (1267 days).
2. Korean careers are slightly shorter than foreign careers (3 years for Koreans vs. 3 years and 6 months for foreigners). The difference is a lot smaller if we take median duration into account (Korean: 3 years and 4 months vs. foreign 3 years and 7 months). This could be a result of not taking into account a lot of foreigners who never made it to top8 in premier tournaments or WCS Premier, but taking into account all KeSPA team players, even if they were never fielded.
3. Terran careers are shortest on average (3 years 2 months, 1143 days) and very similar to Protoss careers (1159days). Zerg careers are longest on average (3 years and 4 months, 1224 days).
4. On average, players retire after 2 years and 6 months (902 days) of competitive playing.
5. I don't see any obvious indication of cumulation of retirements in particular periods (for example after HotS or LotV release), although last months of 2013 and 2014 appear to be most "retirement-rich".
![[image loading]](https://i.imgur.com/t6WXzUf.png)
6. Very few notable players began their careers after the release of HotS (although of course some of them never had success in WoL and became notable in HotS). It's probably too early to draw conclusions for LotV yet, but I believe this shows the need for more grassroots action, minor tournaments etc. More descriptively, this seems to confirm that SC2 is, unfortunately, on the decline.
It would be cool if we could somehow include earnings in this discussion, as it would show more or less how viable a career in SC2 can be with regard to actually living off it on a regular basis. However, there seems to be far to few data to do it in a meaningful way (no way to track sponsor money, streaming revenues and, as we know, teams and KeSPA tend to be quite secretive with regard to salaries).
Thanks for reading and I will be grateful for any feedback!
|
Big thanks for compiling this data. I just arrived in Korea from SFO and I'm really curious too on the state of.. things in general.
In 2000, it was not hard to find family/friends to go to a PC bang to find people to play with. Now I have to practically beg my relatives to play with me.. I keep having hope that 2016 will bring back the era.
|
I would have told you just over 2 years if you asked me straight up in terms of remaining somewhat relevant and then it's a continual decline into the abyss.
|
who were the notable players taht began their careers after the release of hots?
|
Bisutopia19338 Posts
End of Year cycles should have the most dense periods. And I think 3 years is long enough to burn out mentally/physically or realize you weren't cut out for the highest level. Thanks for the post OP!
|
On December 25 2015 21:47 EJK wrote:
who were the notable players taht began their careers after the release of hots?
According to my narrow definition of 'notable': GunGFuBanDa, Has, Sora, Arium, Shana, Hitman, Journey, Neeb, Lambo, Elazer, iA and TANGTANG because everybody loves him. That's only 12.
This is the entire list of those whose career starting date was on or after 12.03.2013 (27 players):
+ Show Spoiler +Jieshi
GunGFuBanDa
Sora
Believe
Pure
Barbie
Splendid
Remember
StuN
Hoon
Has
Neeb
Arium
Hitman
Shana
Guilty
Journey
DynaMite
TANGTANG
eins
NaTuRal
iAsonu
Lambo
KassiA
Adios
Elazer
Slam
|
Fiddler's Green42661 Posts
Good read. Kind of curious how this matches up with their prize earnings/salary and how it would break down by country, but that kind of data could only get gatthered by Blizzard.
I'll add this in. Peak play (that is a player at their height of their skills) generally lasts between 3 months to an year before either they drop off or the rest of the world has caught up.
|
Katowice25012 Posts
This is awesome. Thanks for doing the legwork - I often get into this discussion with people but have always been too lazy to actually look into it. Great spreadsheet.
I wish it were easier to look at "peak play" like stu mentions, that's another important element here but quantifying that sounds like a nightmare.
|
Thanks for your replies! 
On December 26 2015 00:43 stuchiu wrote:
I'll add this in. Peak play (that is a player at their height of their skills) generally lasts between 3 months to an year before either they drop off or the rest of the world has caught up.
On December 26 2015 00:48 Heyoka wrote:
I wish it were easier to look at "peak play" like stu mentions, that's another important element here but quantifying that sounds like a nightmare.
Yep, that's a cool idea but the only idea that does not seem to be inherently flawed (i.e. biased as hell) would be trying to take ratings (such as Aligulac stats, which as we know have their own shortcomings) and try to find peaks there. But still that would require a lot of fine-tuning.
My first idea: if X is the maximum score that a given player has achieved, the peak period would be the period before and after the peak, during which the score was - say - 95% X or higher. But this does not seem to be good enough, as for example for sOs this approach with a 95% factors gives the following result:
- peak: 11.11.2015
- peak period: 30.09.2015 onwards
- achievements during the peak period: Blizzcon 2015 championship
- achievements that narrowly miss the peak period: DH Stockholm top4 and all-kill of KT in ProLeague
Compare INnoVation:
- peak: 14.10.2015
- peak period: 19.08.2015 onwards
- achievements during the peak period: GSL championship, MSI top4
- achievements that narrowly miss the peak period: IEM X gamescom championship
- achievements not included: the famous period when every other pro was waking up screaming after INno's hellbats appeared in their dreams.
In other words, this approach does not take into account the general rating inflation.
My second idea: comparing Aligulac ratings relative to the average ratings during a given period. But personally I give up on testing this particular idea. I don't have enough skill to parse that, that's much more than my Excel skills can handle 
|
a good number of players were coming over from their sc1 teams and many who have yet to retire will generally raise that bar for average career length. then again, there were quite a few scandalous events that pretty much pummeled players into retirement.. including slayers drama, prime, startale's situation, zenex, fxo, nasl, and a good number of foreign teams that were playing around with startup money in hopes of going into the green at some point.
it was just naivety on the part of both the players and the organizations involved. if we look at the teams that are still involved and supposedly profiting, most of the foreign ones are narrowed down to EG, millennium, acer, root, TL and a couple more. most players last as long as their respective teams do.
I just believe the times are changing and starcraft 2 is a demanding game in terms of commitment, dedication, drive, and person practice regimen. becoming a fresh pro requires you to think about the game and yourself years ahead and currently the most stable teams are in korea when they provide food, lodging, practice and a good environment to increase your skill as a player.
then again, what other games are out there for you to become a professional at? are there any other RTS?
there are mobas that most closely resemble the gameplay of an RTS, except in a team setting--those are at the peak of popularity right now. there are fps, fighting games in a few different 'genres' or types, hearthstone, a couple mobile games, and perhaps a few more games I'm forgetting at this moment.
sc2 is not a timeless e-sport in the same way that tabletop or conventional sports are timeless as games.
there are niche professional scenes out there that are wide enough and large enough to be self-sustaining, and starcraft is one too, thanks to blizzard support and Korean competition setting the bar. shall we look at other games' average professional careers? they will most certainly be similar to the data you've gathered here.
in my opinion, it's just pretty much the nature of competition wherever you go to find it.
if you want to revive a game like this, you just need to have more people playing it, making it their 'thing'.
in this day and age, everyone is mostly a variety gamer--escaping from one game to the next and only leashing onto the professional coverage as a lingering way of keeping up with the scene and their friends within it.
gone are the days of everyone playing counterstrike, sc:bw, quake, dota1.
the giants have all made their spiritual successors to those games or have failed in trying, and unfortunately none of them have the same exposure as their predecessors (save for LoL or csGO) because the growing age of internet and gaming had such a special influence on people. -people- are 'retiring' from gaming.
it's kinda up to the new age to figure out what they like best.
|
|
|
On December 26 2015 01:36 corydoras wrote:Thanks for your replies! Show nested quote +On December 26 2015 00:43 stuchiu wrote:
I'll add this in. Peak play (that is a player at their height of their skills) generally lasts between 3 months to an year before either they drop off or the rest of the world has caught up. Show nested quote +On December 26 2015 00:48 Heyoka wrote:
I wish it were easier to look at "peak play" like stu mentions, that's another important element here but quantifying that sounds like a nightmare. Yep, that's a cool idea but the only idea that does not seem to be inherently flawed (i.e. biased as hell) would be trying to take ratings (such as Aligulac stats, which as we know have their own shortcomings) and try to find peaks there. But still that would require a lot of fine-tuning. My first idea: if X is the maximum score that a given player has achieved, the peak period would be the period before and after the peak, during which the score was - say - 95% X or higher. But this does not seem to be good enough, as for example for sOs this approach with a 95% factors gives the following result: - peak: 11.11.2015 - peak period: 30.09.2015 onwards - achievements during the peak period: Blizzcon 2015 championship - achievements that narrowly miss the peak period: DH Stockholm top4 and all-kill of KT in ProLeague Compare INnoVation: - peak: 14.10.2015 - peak period: 19.08.2015 onwards - achievements during the peak period: GSL championship, MSI top4 - achievements that narrowly miss the peak period: IEM X gamescom championship - achievements not included: the famous period when every other pro was waking up screaming after INno's hellbats appeared in their dreams. In other words, this approach does not take into account the general rating inflation. My second idea: comparing Aligulac ratings relative to the average ratings during a given period. But personally I give up on testing this particular idea. I don't have enough skill to parse that, that's much more than my Excel skills can handle
Great descriptive stats, and if you did all this on Excel - take a bow!
I think to effectively compare Aligulac ratings relative to the average ratings during a given period you could probably just do a correlation on the two variables (columns), that being said I do not know how your data is structured, some other statistical procedure may be better you might have to do fixed effects, MANOVA,etc... I was interested in your analysis between races, to check the probability levels between races you could do an ANOVA. I am really interested in the effectiveness of what are considered buffs and nerfs. How accurate are the actual patches! For this it would be required to scrape all the patch and code nerfs and buffs. Thereafter, it would require some interesting process to get the information. This would just allow you to make inferences rather than description.
Still, you did some excellent work!!
I think maybe doing these stats on the Aligulac database would be great, I could just never getting the bloody thing loaded on SQL - not enough memory in my computer to load the dump - I think! If anyone can get me a CSV with that data, We could do some analysis.
|
On December 27 2015 02:46 PickyProtoss wrote:
I think to effectively compare Aligulac ratings relative to the average ratings during a given period you could probably just do a correlation on the two variables (columns), that being said I do not know how your data is structured, some other statistical procedure may be better you might have to do fixed effects, MANOVA,etc... I was interested in your analysis between races, to check the probability levels between races you could do an ANOVA. I am really interested in the effectiveness of what are considered buffs and nerfs. How accurate are the actual patches! For this it would be required to scrape all the patch and code nerfs and buffs. Thereafter, it would require some interesting process to get the information. This would just allow you to make inferences rather than description.
Still, you did some excellent work!!
I think maybe doing these stats on the Aligulac database would be great, I could just never getting the bloody thing loaded on SQL - not enough memory in my computer to load the dump - I think! If anyone can get me a CSV with that data, We could do some analysis.
Yep, and there go the ideas that I knew they have to exist but I simply don't have enough background in statistics to come up with
I didn't do much with Aligulac apart from hand-picking some of the best players and throwing their data into a spreadsheet to take a look at how it could work out. So there's no real structure to speak of. Not yet anyway, but unfortunately right now I'm spending Christmas going through my legal stuff instead of working on SC2 stats 
As for the nerfs and buffs, this should be interesting but I intuitively feel that this would not give any significant results in terms of retirements and career durations. Correct me if I'm wrong but periods such as BL/infestor or 2-base blink domination, while obviously affecting tournament results, did not feel (or maybe someone calculated it already?) particularly rich in retirements.
|
Why is Parting is listed as retired on dec 21 2015?
Neeb is also listed as retired
|
On December 27 2015 03:35 Kokujin wrote:
Why is Parting is listed as retired on dec 21 2015?
Neeb is also listed as retired
Thanks for pointing that out!
PartinG - I originally had him listed after Rain, so it must have been the classic 'one copypaste too far'.
Neeb - you are right, I missed his WCS 2015 participation, so he should definitely be listed as active.
I corrected the chart linked in the OP. The impact of both corrections is rather neglibile (net +1 day for average career duration).
|
I was actually pretty interested in this article and your analysis. Thank you for all of your hard work, but I see a major problem with your analysis.
Essentially, you only looked to see when a player joined and when a player retired. You did not look to see if a player retired during their career and then returned several months if not years later. As a result, you most likely overestimated several players' careers because they were not playing consistently full time during their career. I would recommend looking at the major outliers (people with more than 4 years of playing) to see if their careers should be shorter. Kiwikaki, Stephano, and Naniwa all come to mind as individuals who retired and then came back.
Example:
Kiwikaki retired on April 24, 2012 to pursue a career in Poker and not Starcraft 2. Source: http://www.teamliquid.net/forum/fan-clubs/188537-the-kiwikaki-fan-club?page=28#557
Kiwikaki returned to professional SC2 on May 8, 2015. Source: http://www.teamliquid.net/forum/starcraft-2/485008-kiwikaki-returns-to-root-active-roster-part-time
As a result, Kiwikaki's career should be shortened by 1,109 days (subtract the two dates shown above).
|
Also, you included the following players twice in your data. You should remove the duplicate.
Alicia
ByuL
Illusion
LoWely
Oz
Rain
Ryung
Seed
|
On December 27 2015 05:26 TheBigO wrote:
I was actually pretty interested in this article and your analysis. Thank you for all of your hard work, but I see a major problem with your analysis.
Essentially, you only looked to see when a player joined and when a player retired. You did not look to see if a player retired during their career and then returned several months if not years later. As a result, you most likely overestimated several players' careers because they were not playing consistently full time during their career. I would recommend looking at the major outliers (people with more than 4 years of playing) to see if their careers should be shorter. Kiwikaki, Stephano, and Naniwa all come to mind as individuals who retired and then came back.
Yep, this is a shortcoming. More examples could be listed. For instance, should I list MC as active or retired (the difference is +/- 6 months)? He publicly stated that he considers himself retired because he receives no salary. Still, he streams on a regular basis and participated in the last HSC with pretty good results. There were lots of other similar stories of being on and off. I decided against treating each such story individually because there were not a lot of them and to research most of them in depth would require to go beyond Liquipedia (like in Kiwikaki's case) and if I went this way to verify 'semi-retirements' and 'comebacks', why not verify everything else in other sources...and that would be a massive research to which I unfortunately cannot commit. So in the end I decided to simplify, even though admittedly this results in worse data.
On December 27 2015 05:47 TheBigO wrote:
Also, you included the following players twice in your data. You should remove the duplicate.
Oooops, this shouldn't have happened. Many thanks, I corrected the chart. The results in OP had to be slightly adjusted, but we're still well within the originally expected +/- 1 month caused by my mistakes (not inaccurate or inadequate data on Liquipedia), so hopefully it's not too bad
|
Re Dayshi, Millennium is not a country, yet.
I don't understand what the x-axes represents, it seems to be nothing, which I find discombobulating.
|
|
|
I think x-axis might be race?
|
Nope, X-axis does not really represent anything meaningful. I tried to vizualize the retirements in a way that would help see any "dense" moments.
|
What you should be counting is the ratios not the numbers. Obviously if there are 1000 players and then 200 retire at one point and 200 at another point the second wave would be way more significant. Every 4th against every 5th.
|
On December 27 2015 20:26 juvenal wrote:
What you should be counting is the ratios not the numbers. Obviously if there are 1000 players and then 200 retire at one point and 200 at another point the second wave would be way more significant. Every 4th against every 5th.
Sounds reasonable. I'll try when I can spare some time.
|
Terran careers are shortest on average (3 years 2 months, 1143 days)
All that micro and splitting taking its toll...
Interesting insight. So in conclusion Sc2 isn't dead :p
|
I would argue that mid to late Heart of the Swarm was the most poorly designed era for Starcraft 2, and as such the increased retirements during that time is not surprising. The game was simply the least amount of fun to watch or play during that time, due to Swarm Hosts or lack of late-game play in the Protoss match-ups.
|
On January 15 2016 10:57 NMxSardines wrote:
I would argue that mid to late Heart of the Swarm was the most poorly designed era for Starcraft 2, and as such the increased retirements during that time is not surprising. The game was simply the least amount of fun to watch or play during that time, due to Swarm Hosts or lack of late-game play in the Protoss match-ups.
I think PvZ was quite fun in general even before SH was removed, and I don't get what kind of ironic trickery you are trying to achieve with stating that protoss had no lategame
Even if you hated it, it was in general much better than the zerg matchups at the end of WoL
|
On January 20 2016 03:03 neptunusfisk wrote:Show nested quote +On January 15 2016 10:57 NMxSardines wrote:
I would argue that mid to late Heart of the Swarm was the most poorly designed era for Starcraft 2, and as such the increased retirements during that time is not surprising. The game was simply the least amount of fun to watch or play during that time, due to Swarm Hosts or lack of late-game play in the Protoss match-ups. I think PvZ was quite fun in general even before SH was removed, and I don't get what kind of ironic trickery you are trying to achieve with stating that protoss had no lategame Even if you hated it, it was in general much better than the zerg matchups at the end of WoL
I am sorry, you misunderstood, I worded it poorly.
What I meant to say is that Protoss players had the option of successfully cycling early or mid game allins in a Best of match up, instead of ever going into a macro game. Example of players abusing that strategy the most are Has, PrintF, PartinG.
PvT was often a cycle between a variation of 1-base or 2-base blink allin, proxy oracle builds, 2-base colossus allins, sometimes immortal/sentry allins.
PvZ's were 2-base immortal allins, fast 3-base 10-gate blink(most popular and effective), phoenix/warp-gate allins.
PvP was arguably the most well designed Protoss match up with almost always an option to enter late-game.
At some point, there were very few, if no Protosses, who would willfully go for a macro game when there is a plethora of allins for all non-mirror match ups that can be successfully cycled - because all of them require vastly different reactions in order to properly hold.
Protoss players always had the option for a successful macro game, but macro games are significantly harder to execute - and allin builds were just so difficult for other races to beat 100% of the time, and there were so many to choose from.
Reason why Brood War had so many long macro games is because should a professional player choose to play safe - early cheese and allin builds were never consistently effective. For this reason, if you wanted to be one of the best Brood War players - you literally had to learn how to play macro correctly. You would still see aggressive builds once in awhile, but only to punish players who open greedily.
In Heart of the Swarm, you had Protoss players get surprisingly high tournament results - all by cycling 3-4 allins per match up. Unlike Brood War, even if you choose to play safe against a Protoss in Heart of the Swarm - there would be no way to play safe against all the variations of a Protoss allin(as a Terran, you prepare for an Oracle build, you lose to a Blink allin, etc.).
Worst of all, if a Protoss DID choose to play economically - he would immediately gain an advantage as a Terran player(for example) would preemptively attempt to counter an allin, rather than play economically himself.
Despite all of this, by the end of Heart of the Swarm, the match ups were all statistically balanced. Both Zerg and Terran would try to perfect their counters and edge out defensive victories against Protoss with incredible mechanics(Terran also had the SCV pull win condition for awhile), but the ultimate sacrifice is that the Protoss match ups would not be fun to watch. It would literally be the Protoss performing an aggressive build every single time, with the other player under incredible pressure to perform all of the defensive moves correctly. There were very few, if no, late-game macro scenarios in non-mirror Protoss match ups; and sadly, that's what a lot of us learned to love about Starcraft - is crazy late-game mechanical battles.
|
|
|
|
|
|
EPT
