Deep Dive for TL: Challenger ADC Analysis

Deep Dive for TL: Challenger ADC Analysis

Preface

Steve Arhancet of Team Liquid (TL) has been a long time friend and believer in my work since Season 2 when Curse was rocking the legendary Voy, Saint, and Nyjacky lineup. He was the first person to give me an opportunity to showcase some of the research I had been doing and how it could potentially be applied in the Esports field to help increase the team’s performance.

Mobalytics had been blessed with recent success thanks to TechCrunch and Steve and I were already talking about future potential work together. Up until now, Mobalytics had been focused on our own product, with plans to seriously engage the professional gaming analytics world after our open beta release. However, things changed.

TL was in a precarious position and I was absolutely determined to do whatever we could to help them. We didn’t have much time at all. We had to figure out what resources we could divert to provide the most value in the least amount of time possible. The whole team got together and we wracked our brains on how best to approach the problem.

Truth be told, we played a minor role in the selection of the players. Player performance is a multifaceted problem to solve, and there are so many important factors that you must consider before recruiting a player into a 5 man team. In an attempt to simplify our approach, let’s break performance down into 4 major parts: Psychological, Physiological, Long Term Game Performance, and Instantaneous Performance.

For this project, the only part we looked at with our numbers based approach was a small window in the Long Term Performance department. For the rest, the TL team would have to rely on other established techniques, mainly intuitive, to gauge the right fit for them. Now here is what makes the problem infinitely more complex: You have to gauge performance within the team construct, and then evaluate overall team performance. That’s a whole other story that we will go into some other time, but you can see how small the impact of our analysis is in the grand scheme of things.

What Did We Do?

To help TL find the right ADC fit for the team, we had to understand what type of Challenger player(s) they were looking for to fill the hole left by Piglet’s shift to mid. The clearest sentiment voiced was that they wanted someone who could just “do their job in most circumstances”. With an understanding of the needs, we structured our thinking to start from a macro view and move into a more detailed dissection of contextual game elements.

To evaluate players on a macro level, we used our Gamer Performance Index (GPI). The entire design of the GPI’s 8 major skills (and attributes within each skill) revolves precisely around determining a player’s unique play-style, strengths, and weaknesses. Each of the attributes and skills are scored from 0-100 based on a combination of game expertise and sample data gathered across the entire ranked solo queue spectrum.

Even though we are only looking at solo queue, it is important to remember that a player’s style is something consistent across games. Sure they can be more cautious and pace themselves for the slower play style in LCS, but a reckless player’s very blood boils with aggression. He won’t suddenly change to a riskless “do my job” ADC overnight. Since you don’t want to see me ramble about data forever, I am only going to use two distinct examples of what we put together to illustrate the story.

Disclaimer: To protect player privacy, we have de-identified all player data.

skills breakdown league of legends doubleliftskill breakdown league of legends pros doublelift
Case 1: Player A
Takeaways: All-rounder, extremely low deaths, safest bet
Player A is an incredibly sturdy player who is safe, reliable, does his job and stands out for just how infrequently he dies. Trust us when we say a Deaths score of 2.5 is EXTREMELY low. Definitely seems like a guy who can do his job, so we suggested when trying this player out, look to measure his capability of carrying games when given resources and have confidence that he won’t be a lost sheep when someone isn’t holding his hand.

 

skills breakdown league of legends doubleliftskills breakdown league of legends doublelift
Case 2: Player B
Takeaways: Hyper aggression compared to most peers, heavy focus on winning lane and CS’ing, needs to be the carry
Player B is a completely different beast. He is extremely aggressive with a heavy focus on winning his lane (potentially drawing a lot of attention from the enemy jungler as a result) and carrying the game. As such, compared to Player A, Player B dies much more frequently. In our opinion, this style would have made for a much harder and time-consuming transition from solo queue to the LCS. We suggested paying close attention to how he handles himself when he’s not given all the resources during tryouts.

The Devil is in the Details

There are probably a ton of questions being juggled in your head right now if you are the overly analytical type. What about differences in ADC champions played? What about games in different divisions? The questions go on and on, as there are a lot more issues than just numbers in games, it’s easy to misrepresent our findings. In an attempt to isolate the signal from the noise, we took several factors into consideration and worked to eliminate as many variables as possible:

  1. We understood that we were under time and resource constraints, and no matter how much filtering we did, we would not be able to achieve a perfect data set. Either our sample size or our data would suffer somehow.
  2. We understood that different ADC champions would lead to fundamentally different numbers between players, but luckily most players are playing the same carry type in this meta, so we decided to ignore any minor differences because of this.
  3. Looking at all games was too noisy, so we only included games that were longer than 25 minutes in our sample size. In our opinion, short ROFLstomp games polluted our data pool without providing valuable information and needed to go.
  4. We identified the most significant questions we could ask and answer in a short time frame:
    • How do you perform when you are not the main resource focus of your team? Can you hold your own when behind and deprived?
    • How do you perform when you are the main resource focus, the carry? Can you deliver when ahead and fed?
    • What are the key performance time zones you excel at or are weak at?

Let’s take a look at some of the information we presented to TL to help them make a decision. Taking the above constraints into consideration, we looked at a range of 65-85 games for every player. Again, for the sake of brevity, I will only show two complementary graphs to illustrate our point.

performance during low gold situations
Graph A: Player Performance during Low Gold %
We scored how well five different players fared with Low Gold Share (empirically determined threshold by the analytical team) across four different metrics: Damage Per Minute (DPM), Deaths, Kill Participation, and Win %. On the y-axis, scores range from 0 to 1 with 1 being the best and 0 the worst, except for Deaths where the scores are reversed. A few things to note:

  1. The five individual players are being benchmarked against the average Diamond+ player.
  2. Total games played with Low Gold Share can be found to the right of the graph. Players C and D don’t have enough total games played with Low Gold Share to make an informed conclusion.

So what can we tell from this graph?

  1. Player A is relatively awesome at playing with Low Gold Share.
  2. Players C and D almost never have any games with Low Gold Share. They are the carry and will do whatever it takes to be one. This is crucial information in its own right.
  3. Due to conclusion 2, we cannot judge their play with Low Gold Share well. However, we can say that not having experience with Low Gold Share could be detrimental to performance.
  4. Player D seems to just not like playing with Low Gold Share (deaths through the roof, a possible sign of tilting) and Player E doesn’t fare so well either.

There are other things one can discern, but let’s move on to Graph B.

performance during high gold situations
Graph B: Player Performance during High Gold %
Graph B features the same metrics as Graph A but for High Gold Share games. Immediately noteworthy is that there are enough games to make informed conclusions for all the individual players, but far fewer total games for the average Diamond+ player, but still more than enough for our purposes. Keeping in mind that there is another bracket for Medium Gold Share games, here’s what we found:

  1. What immediately jumps out is there are not that many differences in performance between players when they have High Gold Share.
  2. Because of conclusion 1, it is plausible to suggest that when it’s all sunshine and rainbows, unless you’re a stellar team carry (Hi Doublelift!), you are not doing too much better than an OK one. On a star-packed LCS team with resources going to other players, this is very important.
  3. Because of conclusion 1 and 2, we draw a more clear picture of each player’s tendencies and playstyles. Out of 70 games, Player D played over 50% of his games with a High Gold % whereas Player A has less than 20% of his games with a High Gold Share %. This gives teams looking to try out players an idea of what situations they’re most comfortable in.

In the End

While we evaluated as many metrics and players as we could in the above fashion, given the time constraints, we did not get to dive into nearly as many as we would have liked. Ideally, we would have also analyzed champion-only stats and more than a few others to provide more contextual background. However, if we’re keeping everything in perspective, all the metrics in the world regarding an individual’s performance may not accurately reflect how they perform on stage in a team environment. This is where human elements like instinct and resilience come into play to make pro gaming all the more interesting.

For us, this project demonstrated that we could efficiently create insightful analytics for professional teams and provide real value. This was a great litmus test and we are very excited to continue working with TL to push the envelope for player performance assessment in esports.

Thanks again to all the members of TL for the opportunity, and a huge shoutout to our awesome team, especially Data Architect Ryan “Geei” Dean and Lead Analyst Hewitt “Prohibit” Benson for going into crunch mode on this incredible project.

Mobalytics was designed to help every gamer unlock their full potential, not just professional teams. If you’re curious about what we can do for you, sign up for our private beta and follow us on Twitter and Facebook for more thoughtful content like this. We’ll see you on the rift!