Jungler slash lines improve measurement of early-game effectiveness

TL;DR Because of changes to the jungle as part of Season 7, I am proposing a new way of measuring junglers’ early-game effectiveness.

Key Findings

Changes to the jungle for the 2017 season have created some loss of meaning in one of the oldest statistics used for professional play: CS Difference at X minutes. The addition of more small creeps in the raptor and krug camps has created wider variance in the value of a single creep score (CS), leading to inflated CS gaps for junglers without any real difference in the gold or experience gaps being generated.

These changes have some implications for how we should report on junglers’ early-game head-to-heads. Here are the key points:

  • Creep score difference (CSD) statistics do not reflect, and never have reflected, the gains earned from ganking or skirmishing, and can be misleading regarding overall performance.
  • CSD statistics are becoming less accurate and less valuable in measuring jungler performance.
  • Gold difference (GD) and experience difference (XPD) are better indicators of early leads than CSD.
  • GD and XPD don’t directly show whether the leads came from farming or fighting. However, compared to the past, kill+assist leads are becoming noticeably more useful than CS leads in predicting gold advantages, while CS leads remain more important in predicting experience advantages.

For these reasons, I am recommending that analysts, writers, and broadcasts use a combination of GD and XPD to measure junglers’ early-game effectiveness, and limit the use of jungler CSDs to discussions of play style.

When reporting GD and XPD in articles or on-screen, they can be reported as a “slash line” statistic, similar to the way batter statistics are presented in baseball. For example, the week 1 slash line for Echo Fox jungler Matthew “Akaadian” Higginbotham, using 15-minute statistics, would be +811/+297 GD/XP15, meaning that Akaadian averaged a lead of 811 gold and 297 experience over his opponent at the 15-minute mark. If CSDs are still of interest in some cases, they can be included as a third segment of the slash line.

Similar slash line reporting could be adopted for laning roles, but this is less necessary. The value of a single CS has less variance in the lanes, and CSDs remain a bigger part of the story for laners. Even when a laner is “losing lane” and being forced to miss last-hits on creeps, they should still be gaining experience from those creeps’ deaths in most cases, making XPDs less independent of CSDs. For now, CSD statistics remain viable for storytelling around the top, mid, and ADC roles.

Read on for a more detailed breakdown of the numbers behind these findings, as well as a table of NA LCS junglers’ week 1 statistics that demonstrates the slash line approach.

Exploring the Math

In the statistics that follow, summer 2016 data (patches 6.10 through 6.18) includes NA LCS, EU LCS, LCK, LMS, CBLoL, NA CS, EU CS, World Championships, and IWCQ, with a sample size of n = 1,117 games. Spring 2017 data (patches 6.23 through 7.1) includes week 1 of NA LCS, EU LCS, and LCK and IEM Gyeonggi, for a sample size of n = 91 games. These findings will be monitored as and when Riot Games makes further changes to jungle balance, and as the sample size for the spring split grows.

CSDs have always needed more context for junglers than for laners. There is more range in the values of jungle monsters than there is for lane minions. Also, it is possible to steal the opposing jungler’s farm and create bigger advantages compared to the lanes, where both players are receiving the same flow of minions to hit and cannot directly steal from one another.

The jungle rework in the season 7 preseason patches has made the difference between the jungle and the lanes even more measurable. There are more small jungle monsters with very low gold and experience values, and camps spawn less frequently. This both decreases the average value of a single CS, and creates more opportunity for junglers to build large CS gaps by stealing camps. As a result, the average absolute difference in jungler CS at 10 and 15 minutes has grown by 55% or more in pro play since patch 6.23, while gold and experience differences have remained almost entirely unchanged.

This is true for 10-minute CSD/GD/XPDs:

It is also true for 15-minute statistics:

In other words, raw CSD values are becoming less important for predicting how much of an advantage a jungler is getting over his opponent.

The next question that is implied is whether we are seeing any change in the relative value of farming vs. fighting. We can measure farming advantages through CSDs and fighting advantages through differences in kills+assists. The table below shows what is happening so far this split in the correlations between CSDs, kill+assist differences (KADs), and gold/experience differences.

These correlations show that CS leads are now less predictive of gold and experience leads for junglers than they used to be, while kill+assist leads are playing the same role as before in generating gold leads and a slightly larger role in producing experience leads. This is due both to the reduced average value of individual jungle creep kills and an increased focus on kill-oriented junglers.

On balance, early-season data shows that CS leads remain the better predictor of experience advantages, while kill+assist leads have become a relatively more important predictor of gold advantages. This nuance allows us to get some implied findings about the sources of leads when interpreting gold and experience gaps, without requiring us to include CSDs and/or KADs in every comparison.

Background Factors and Future Trends

One background factor that may be influencing the CS vs gold correlation is the removal of lane swaps. This has led to fewer early tower kills. However, since lane swaps almost always produced an equal number of tower kills across the two teams, they should not be having much impact on the comparison. Lane swap games also saw slightly fewer kills in the first 15 minutes of games, which may be leading to slightly higher variance in gold/experience gaps in the standard lanes meta enforced in patch 6.15.

For the future, it must be borne in mind that Riot is planning to make further jungle changes, such as increasing the respawn frequency of jungle camps and possibly lowering some gold or experience values, so these correlations will continue to change over time. Changes in the broader meta may also lead to different champions being played in the jungle, which could affect the value of farming and fighting. These trends will need to be monitored over time.

Jungler Slash Lines

Slash lines are intended to replace CSDs as the most common measurement of a jungler’s early-game performance. In the past, CSDs were commonly reported, sometimes alongside GDs and occasionally first blood rates. However, a combination of GD and XPD is a better measurement of performance because it captures the end product of CS leads while also capturing advantages gained from fighting, which is junglers’ unique early-game role.

How to Present Slash Lines

Slash lines should be written with gold difference first and experience difference second. Neither number requires a decimal place, but both should include a + or – symbol to show direction. On first use within an article, the slash line should be followed by either a full naming phrase, such as “gold difference / experience difference at 15 minutes,” or an acronym-based postscript, such as “GD/XPD15.” (If 10-minute numbers are being reported, they can be reflect in the postscript, e.g. “GD/XPD10” instead of “GD/XPD15.”) If used on-screen for video or in an infographic, the full name or the acronym should be presented as a title or as a footnote.

GD@15 and XPD@15 numbers should typically be used because of a general movement that is starting to push the LoL statistics community towards 15-minute statistics and away from 10-minute numbers. This movement is taking place because the 15-minute timing is a better reflection of the “early game” phase as a whole, especially within the standard lanes meta that has become enforced since the 6.15 balance patch. Changes to this effect will be coming to Oracle’s Elixir’s in the coming months.

An Alternative to Slash Lines

If slash lines prove to be too difficult for fans and/or analysts to read, present, or interpret over time, one alternative is to combine gold and experience differences additively and report them as a single statistic.

This approach may be viable because of similar ranges within both GD and XPD statistics (see the trend tables above), which means that neither gold nor experience would dominate the result too heavily. An additive statistic would lose the ability to infer whether the advantages came from more farming or more fighting, but context stats like CSD could still be used for that purpose.

NA LCS Jungler Slash Lines, Week 1

The table below illustrates the week 1 slash lines for NA LCS junglers, with CSD@15 and KAD@15 included to provide context on where the gold and experience gaps came from.

From these numbers, it’s obvious that Akaadian did very well for himself, as did Chaser of Dignitas and Reignover of Team Liquid. On the other end of the scale, some junglers who struggled early included P1 Inori, TSM Svenskeren, and NV Ninja (who was forced to temporarily role swap due to his jungler’s delayed work visa).

Even without looking at the other columns, we can see from the slash lines that Akaadian’s advantages came more from fighting than farming, since his GD@15 is bigger than his XPD@15. That is confirmed by his CSD@15 and KAD@15 numbers. By contrast, Moon’s XPD being lower than his GD suggests that he suffered more from weak farming that weak fighting.

For evidence of CSD’s potential to be misleading, look at Svenskeren and Xmithie. Svenskeren earned a positive CSD but was still behind in the important gold and experience metrics, due to his opponent earning kills and assists, and possibly tower gold as well. Xmithie, meanwhile, had a below-average CSD but still earned a positive slash line because he kept the kills and assists even.

Oracle’s Elixir releases downloadable match data files for public use

OraclesElixir.com is now providing downloadable match data files from the 2016 season, and will be releasing 2017 match data on a weekly basis going forward.

Files for the spring split, MSI, summer split, and regional qualifiers plus World Championships can be accessed now on the match data download page.

Match data includes game-by-game statistics for each player and team from the NA LCS, EU LCS, LCK, and LMS. Files can be downloaded in .xlsx format and opened with Microsoft Excel or Google Sheets or imported into statistical analysis software. A list of initially included variables is available in the data dictionary.

Statement from Tim “Magic” Sevenhuysen, founder of Oracle’s Elixir

When I created OraclesElixir.com, one of my goals was to help the LoL esports ecosystem grow. Over the past two years, we’ve made some progress. Writers and commentators are drawing on a wider range of stats than ever before, and there’s been great improvement in the general awareness and understanding of statistical analysis in LoL esports.

By releasing disaggregated match data, I hope to kick that growth into much higher gear. Everyone from professional analysts to hobbyists will have a source of high-quality match data, which they can use to improve their analysis or develop new concepts.

I’m very excited about the chance to open up more of my work to the community, and I can’t wait to see what you all do with this data going forward.

About OraclesElixir.com

Oracle’s Elixir is the premier source for League of Legends esports statistics, covering the North American and European League Championship Series (LCS) and Challenger Series (CS), LoL Champions Korea (LCK), the League of Legends Master Series (LMS), the LoL Pro League (LPL), Circuito Brasileiro LoL (CBLoL), and more.

Contact Information

Please direct inquiries to tsevenhuysen@oracleselixir.com.

IEM GyeongGi Stats Delay

IEM GyeongGi stats pages will be delayed while I find a workaround for some issues with the season 7 data streams.

Currently, control wards are not being properly identified in the data, which means that some of the stats reported by Oracle’s Elixir cannot be calculated. I’m hopeful that this will be resolved soon.

If Riot doesn’t fix the issue by the end of the tournament, I will gather and post the IEM stats and leave out any “broken” data, as a workaround.

Sorry for the delay!

League of Legends esports statistics and analytics by Tim Sevenhuysen