Starcraft Statistics: What Tournament Finals Should Organizers Hope For?

Published on 05/01/2017 11:00 PDT by ROOT Gaming

Tournament organizers across mediums from tournaments at a game shop to traditional professional sports to the biggest eSport events in the world all want the largest audience they can get. They do everything they can to increase viewership but something that is rarely discussed in this community is two aspects of tournaments that greatly impact viewership. It's not time of day the tournament starts or prize pool, rather it's two things that I want to discuss today:

- What matchups are being played?

- What players are on stream?

Today I want to look at the topic I mentioned, some players draw more viewers than others and some matchups have the same impact, what are those? We heard in 2016 when a stronger region lock was put into place that it was done since Korean versus Korean wasn't meeting the goals set for viewership, but was it successful, was more foreigner involvement helpful or did it just hurt viewership? Do mirror matchups really hurt viewership? We are going to try and answer those questions today.

Methodology

Now if we are going to compare players and matchups across tournaments with different time zones, prize pools and formats we first need to create a differential understanding. What this means is we need to create a basic template to compare each player and matchup against. Firstly we need to create a database of tournaments to use. To do this I wanted to focus not just on LAN's but I didn't want to use tournaments that had viewership so low that we saw no change when different players or phases took place. So after running a few limiters to find which data set would make the most logical set, so here are the parameters:

- The tournament needed to last over an hour


- The tournament needed viewership that peaked over 1,000
- If a tournament is single elimination the player pool needs to be eight or more
- If a tournament is double elimination or has a group stage it needs a population of four or more

Secondly we need to make a few basic calculations, the first is we need to make a few factual points:

- Viewership increases the longer a tournament goes on so hour two of a tournament will have higher viewership than hour one of a tournament as social media and other sources draw players in.
- Viewership increases the later into a tournament we go, so the round of 8 will have more viewers on average than the round of 16 or a group stage.
- Viewership is higher when it takes place in a time slot that overlaps most with North America and Europe rather than favoring one too heavily.
- Viewership is higher on weekends.

Once we make these initial conclusions we can start measuring the total population average and then compare it to specific queries. Since this was only made possible due to the amazing Fuzic we can only go back to 2013 but that is still four years of data.

First let's normalize the percentage increase tournaments see by series that day. What this means is that we know the second series of the day will have better viewership than the first, let's examine by how much. When we look at all tournaments in our pool this is what we find. We are looking at series number on that specific day and comparing it's increase in viewership to the previous series. So let's look at this quickly, we have three series in a day:

- Series 1: 100 viewers
- Series 2: 150 viewers (50% growth)
- Series 3: 225 viewers (50% growth from Series 2, 125% growth from series 1)

Now obviously some tournaments have multiple streams with different games running for the same tournament so we'll treat each stream independent, the first series on each stream for example rather than first broadcast series of the day, regardless of stream. Let's jump into these two charts then shall we:

Viewership Growth By Increase From First Series


Viewership Growth By Increase From Previous Series


We obviously see that events that have long days do well but we also see a drop off if an event runs too long, this is likely due to two factors: fatigue and time zone shifts. Either one of the major time zones heads to bed. There are also tournaments that had more series per day but I there were not enough events that streamed 16 series a day on a single stream to collect meaningful data. But we see the biggest growth in series one through six when each series adds 25% or more viewership than the previous series but in the final two series they add only about 6% increase per series.

Big takeaway here is you want a longer tournament but your returns only go so far as the day drags on and time zones begin to be impacted your growth will slow for the most part. The same applies for games within a series, a series that goes to it's final games, preferably in a best of five or seven, is going to have better viewership than a series that is finished in the minimum games possible. A series with comebacks also help. Polt against Hydra in 2015 WCS Season 1 for example had amazing growth as the series progressed while INnoVation mopping the floor with Stats 4-0 at IEM Gyeonggi didn't have the same growth. Again this is about viewership increase as a percentage, not total numbers, so we can avoid comparing tournaments directly since foreign tournaments tend to do better (again, total viewership, not viewership increase) than Korean ones.

With that normalized we can look at individual players, matchups and where players are from. To do this we compare viewership for each player, matchup, etc against the norm. We can apply to that matchups as well, let's start there for our table:

Matchup
Delta Percentage
TvP
7.3%
TvZ
5.9%
TvT
5.7%
ZvP
-3.3%
ZvZ
-6.7%
PvP
-11.1%

Now again, this isn't that PvP loses 11.1% of viewers, rather it is that PvP has a viewership growth curve 11.1% lower than the mean for that given series. The most popular matchups feature Terran with the most popular being Terran versus Protoss being 7.3% more popular than your average series. What that means is that if your average series has 100 viewers then the average TvP series will have 107 viewers while the average PvP series will have 89 viewers. The least popular matchup is the Protoss mirror matchup which means Protoss are in the most popular and least popular matchup. Two mirror matchups occupy the bottom two spots and the Terran mirror is the least popular Terran matchup to watch which points to mirror series being less popular. So if I'm a tournament organizer I'm rooting for a finals with a Terran and preferably against Protoss or Zerg, purely from a viewership perspective.

The next area I wanted to apply this data to was the matchup, but from a foreign and Korean angle. After 2015 the WCS system went through a number of changes and while no official reason was given the most repeated reason by casters and community members was Blizzard's tournament partners ESL and DreamHack felt the Korean only finals was impacting viewership. With the data we can now see if this was actually true, did the "faceless" Koreans crushing foreigners and then meeting in the finals really hurt tournament viewership. Also I treat all Koreans the same so players like Polt, ForGG, TRUE, Hydra, viOLet and such are still considered Korean despite living for periods outside Korea. Shall we dive in?

Matchup
Delta Percentage
EU vs KR
16.9%
Foreigner vs Korean
12.2%
NA vs KR
6.7%
EU vs EU
-0.6%
Foreigner vs Foreigner
-1.3%
NA vs EU
-1.6%
NA vs NA
-2.3%
Korean vs Korean
-4.2%

Starcraft 2 viewers seem to agree with the theory presented that Korean only finals don't see the same increase. Again I want to specify this isn't totals, merely that a Korean only finals has less growth than other matchups. So this won't be impacted by events like GSL being held in bad times zones for North America for example since growth tends to remain the same regardless of time zones.

But we did find some interesting things, viewers aren't huge fans of only foreigners either, they seem to enjoy series most when it's foreigner versus Koreans and preferable Europeans against Koreans. We see that in general Europe is a bit more popular, either due to having better players on average or just having a larger fan base.

The last topic I wanted to explore with this data was specific players. Now since I don't want to be rude I won't be looking at players who drag viewership down but I do want to show you the most "popular" players, meaning the players who bring viewership above what it would normally be. So if we take a player named Topher. On average the third series of the day we see an increase in viewership of 30.13% but during all of Topher's third series of the day viewership increases only 15%, we know he's obviously not bringing in viewership, which we actually see in a few rare players. Now if we see another player named Chris and his third series increase is 45% we can see he brings in an estimated 50% increase than your average player. Please remember that this covers multiple years so a recent phenom like Neeb or ByuN isn't going to rate as high due to them playing in many tournaments prior to this year but hadn't reached the cult following they have now. Also since we want a large sample size players who only played a few tournaments each year obviously won't be included. The table shouldn't hold too many surprises but let's take a gander, here are the top 10 most popular players by how much more viewership increase they see than average. To read the chart it's pretty simple, when a series featuring Polt is playing on stream viewership the boost in viewership is 10.3%, so with that covered, here is the table:

Player
Delta Increase
Polt
10.3%
Snute
9.1%
Life
8.9%
Stephano
8.7%
MC
8.7%
Zest
8.3%
Jaedong
8.1%
PartinG
7.7%
INnoVation
7.6%
Scarlett
7.3%

The names aren't surprising, Polt and Snute's popularity are the best among Korean and foreigners. Life at his peak had a massive following and Stephano's popularity goes back a long ways. The Protoss masters MC, Zest and PartinG were fan favorites and the Tyrant's popularity was without question. Scarlett and INnoVation bringing up the rear may surprise a few but the machine and Canada's terror have followings of their own. The differences between Scarlett than the 20th ranked player is less than 1% so being excluded from this list is no insult, it was very close besides Polt at the top. Also for you Has fans, he's ranked 12th, everyone needs a little cheese in their life.

Conclusion

This has been a very exciting study for me and educational as well. So when you wonder why this year's IEM Championship may not have had the same viewership as last year's, despite both having amazing games? Last year featured two of the most popular players in Starcraft 2's history (Polt and Snute) in the second most popular matchup (Terran versus Zerg) in the best regional matchup (European foreigner against a Korean). 2016's IEM World Championship was the perfect storm of factors. While tournament organizers don't rig brackets and such, it is clear that obviously some matchups and players are preferred. We are seeing that this year in the WCS and Korean leagues. The WCS qualifiers viewership is down by a decent margin and I'd look to the lack of matchups against "foreign" Koreans like Polt, Hydra, and viOLet. While TRUE is still the system he's just one player. Seeing Neeb crush Koreans in qualifers was a huge draw in last year's WCS. The opposite is true in the Korean qualifiers. This year has seen a large influx of foreigners into Korea to try and qualifier for their leagues. While there was always a Chinese contingent of players that numbers has grown as well as American and European players have moved to Korea. In the second season of GSL we even had two foreigners qualify in MajOr and Scarlett and had 11 foreigners attempt to qualify. This is also true for regions like China where players aren't seen often and when they face Koreans or NA/EU players the viewership boost is much higher than normal.

This is important because ESL, DreamHack, Blizzard and other production companies tend to hide Chinese and Korean players on B streams. This is a mistake from a viewership standpoint and viewers also miss the chance to see different storylines. There is a lot to look at here but the biggest take away for me is embrace the "Korea versus the world" theme and don't hide non-NA/EU players on B streams, or worse yet, not stream their games at all. 

This in no way suggests that one matchup or player is better, but merely looks at historical data to see what players prefer. Obviously not everyone things Terran versus Protoss is the most popular matchup and I'm sure some viewers hate watching Polt or Snute, this is just the data.

If you have any questions about the data please let me know and I'll answer them to the best of my ability.

About the author:

Topher is an American football and eSports writer with a focus on statistical metadata research. You can follow Topher on Twitter

Thumbnail image rights to of Jennika Ojala and ESL

Fuzic for the data

ViBE 439
puCK 337
Semper 124