We recently introduced our new metric called Gold Shift Events (GSE). I thought it was time to use data from the summer split to dig a little deeper. I will give a short recap of the metric for those who don’t know about it yet. For a full explanation, please read my introductory article here. As the name suggests, a GSE is registered whenever there is a significant shift in gold differences between the teams during a game (think Baron, multiple kills or towers). Such an event can either be won or lost, there can only be one event per one-minute interval and, due to the calculation method, no team can win (or lose) two GSEs in two consecutive minute intervals. By definition, a GSE does not have to be a single event. It evaluates gold difference changes within the last or last two minutes and can be triggered by, for example, a double kill in the bot lane and a tower kill in the top lane 40 seconds later happening between the 22 and 24-minute mark. The general assumptions are that this still represents a valuable piece of information (since the gold result is the same as one resulting from a huge team-fight) and that things happening at different points on the map are often times still connected by some causal effect.
Today I will go over the data from all LMS, LCK, EU LCS, and NA LCS games that were played before Monday (June 13th). After giving an overall summary over GSEs, I will evaluate GSE win percentages at different gold lead thresholds and compare between “good” and “bad” teams. Given the still limited sample sizes, I will use aggregate data from the four regions I have available: LCK, LMS, EU & NA LCS.
Summer split summary
There have been 722 individual gold shift events combined. This is calculated by summing up the number of GSEs and dividing it by two, because every shift is captured twice — once for each of the teams participating. SK Telecom is the most successful team in terms of GSEs, having won a whopping 47 out of the 48 events they have participated in. One great thing about the statistic is that I can easily figure out important moments during games and take a look at them. Here is SKT’s only GSE loss so far.
Remembering that the gold shift threshold needed to trigger a GSE scales linearly with time (5 minutes = 600 gold, 30 = 2000 gold), here is a graph displaying the distribution of events over game time. The graph below does not track GSEs after 40 minutes, because there are too few observations per minute after that point. Even before, there are large differences, most of them also due to the limited sample size per timestamp (unless someone has a good argument for a causal explanation of jumps such as the one between 18 and 19 minutes).
Comparing GSE win percentage at different gold advantages
Starting with our next round of stats updates, we will introduce two GSE refinement statistics, which show the percentage of events won when owning 52% or more and the percentage won when owning 48% or less of the total gold (for other uses of 52%+ and 48%-, see this article). Here, given the richer data we get from aggregating over the regions, I am going to look at an even more refined version. Below you find a Stata output table that sums up the information. Due to the way GSEs are calculated, I use the gold differences two minutes prior to the event triggering. There might be some instances where the gold lead heavily tilts twice within a very short timeframe and thus the actual gold difference right before the event is smaller or larger, but it should be the very rare exception. Since every GSE win is a loss for the other team, I only report the numbers for gold leads; if you want them for deficits, just subtract the means from one.
Teams that are ahead have a GSE win percentage of, on average, roughly 70.3%. Not surprisingly, the higher the gold lead, the higher the average GSE win percentage. What is interesting is how quickly that number accelerates. Overall it paints a pretty nice and clear picture. If a team is ahead by 50% to 51% of the total gold, it wins – on average – around 52.8% of its gold shift events. The number jumps to ~ 67.5% between 51% and 52%, already a sizable advantage. Teams that own more than 54% of the gold (a huge lead) two minutes prior to the trigger came out ahead 96 out of 105 times, with 4 of the 9 teams losing the GSE going on to lose the game. There were only two teams, CJ Entus and ESC Ever, that managed to lose two GSEs when having such a big lead.
Comparing “good” and “bad” teams
To go even further, I thought it would be intriguing to look at how “good” and “bad” teams are doing in GSEs with certain gold leads. I put the descriptions in quotation marks, because I simply make a distinction between teams with an overall win rate this season of over 50% and 50% and lower, which does not necessarily tell us whether a team is good or bad this early in the season (never mind that no professional team is truly bad). Here, the deficit numbers are not the same as one minus the mean, because we split the teams into two categories. I will still only report the numbers for gold leads, as this contains enough information.
win_above50 = 0 stands for teams with an overall win rate of 50% or lower this split, win_above50 = 1 for teams with over 50%
Everything but a clear distinction between the two groups would cause me to doubt our metric, but as we can see, the “good” teams have higher GSE win percentages for each gold lead category — and usually by pretty big margins. Obviously, teams that don’t win too much are less likely to even have a large gold lead, which is why they record a lot less gold shifts for the higher gold lead variables. The most striking difference, at least in my opinion, is the top row — 50% to 51% of the total gold. In fairly close games with a small lead, “bad” teams actually lose around 55.5% of the events while winning teams are much more likely to come out ahead in GSEs. While it is hard to pinpoint the exact importance of performing well when a lot of gold is at stake — making comparisons to slow, methodical gains difficult — it seems obvious that GSE performance is a huge factor in deciding the outcome of a game.
I think especially the last point highlights the value of the metric and I hope I was able to give a bit more insight into the data. If you have any questions, comments or critique, you can voice them in the comments, our contact form or write me at eike[at]league-analytics.com. If you like what we do here and want to get updates on new articles, stats, and more, follow us on twitter. To finish it all off, here is a video from the second largest gold shift event (the largest one was a game ending push with lots of towers falling) in terms of percentage gold difference compared to game time (gold shift threshold to be exact):