Who is Significantly Ahead or Behind? – Introducing Another New Metric

With the launch of our statistics section, we introduced a few new metrics, one of which I talked about in more detail here. Two of the new stats we have not elaborated on are the ones that tell you 1. the percentage of time a team has spent with at least 52% and 2. The percentage of time a team has spent with 48% or less of the total gold. After getting some good feedback from Tim “Mag1c” Sevenhuysen over at Oracles Elixir, I thought it was time to properly explain those creations. I will detail how they are generated and use them to provide an overview and a short analysis for the LCS. Due to the limited number of games this split, I will use the 2016 NA LCS regular season spring split for this. Whenever I only mention the 52% metric, I am doing so because the two are actually the same statistic only from different points of view — the team ahead (52% and above) and the one behind (48% and below).


Significantly ahead / behind


First of all, why do I use the 52% threshold? The important thing I wanted to capture was a significant number that represents a sizable lead, while being straight forward and easy to understand. A 52 to 48 percent (4 percentage points) advantage amounts to a roughly 8% lead. Anything far below the threshold would not allow for a clear interpretation and restricting it to higher margins would limit the observations and reduce the usefulness of the metric. If one team is 8% ahead, I think the statement “the team is significantly ahead” is usually fair. So 52% of the total gold it is.

The generation of the metric is fairly simple. Riot’s match histories give information on the current gold earned for every full minute (timestamps from now on). So for every game and every given timestamp within a game, I sum up the gold for each team individually and divide it by the sum for both teams together. If team A has 33k gold after 20 minutes and team B has 29k, team A owns 33/(33+29) ≈ 53.2% of it at that point in time. Having 52%+ of the total gold or not is a binary question (1 or 0), meaning the 53.2% would “trigger” a 1 in this case. For every team and every timestamp during their season, the metric checks whether the team is significantly ahead or not. After averaging over all recorded timestamps we end up with a number between 0 and 1, representing the percentage of time the team owned 52% or more of the total gold.


Illustration using the NA spring split (regular season)


First, I will provide some basic facts that summarize the spring split so you can get a feel for the statistic. Here is a table that shows how often the teams were significantly ahead, behind and what percentage of game time was “close” (a 7% gold lead might not be very close, but it is the easiest way to specify it here, hence the quotation marks).

Ahead Behind Table

As expected, Immortals, who dominated the regular season, spent almost half of the time ahead by 52% or more and rarely trailed by large amounts. Interesting is that TSM and Renegades had very similar numbers. Further exploration would be necessary to find out more, but since this is a post detailing the ahead / behind metrics, I am not going to do that here. One important aspect that should be considered is that close games usually tend to be longer, meaning they can have a bigger influence on the numbers than blowouts because more timestamps are recorded.

To illustrate my last point and give some more insight, you can find a graph below that displays the average percentage of time either of the two participating teams in a game was significantly ahead for the individual timestamps (as a result, the average over all minutes is twice as high here, when compared to the average generated from the individual team numbers). The stamps are labeled with the number of games that reached the corresponding minute mark. I cut out everything after minute 45, because the sample size becomes very small and the numbers won’t tell us much on an aggregate level.

ahead behind vs minutes

There are two major trends here. At the beginning, games did not see a large discrepancy between the teams very often. The percentage slowly rises until the 28-minute mark, at which point it starts dropping again. It makes a lot of sense that teams don’t get a big lead (in terms of gold percentage) very early on. Usually, there is simply not much interaction going on before 2 minutes and the pace quickly picks up after that. Even then, games were closer early than during the mid-game. While it is obvious that the absolute difference between teams is, on average, lower at the start of the game, the relative difference was also lower during the spring split. This might be due to snowball effects (2% advantage leads to 3% and so on), better opportunities to gain a bigger share of the overall gold or something else (let me know if you have any ideas).

The other trend, later in the game, is easier to explain. As I mentioned before, close games tend to be longer and teams with huge leads are more likely to close out the game early. If you look at the number of games that reached a certain minute mark, you can see it drop of fairly quickly towards the end. My interpretation is that as the games went on, more and more teams were, on average, able to finish off the opponent. The ones that were left are the closer games, which explains the drop off. The mechanism which lead to the rise in 52+% in the beginning might still be at work, reversed or not be present anymore. We cannot tell from looking at the data, because we cannot entangle it from the selection effect concerning close games. The volatility at the end is very likely due to sample size and you can imagine what happens if we go even further, where there are even fewer games left.

I hope I have given some better insight into the statistics and some driving factors. Before I let you dive into our statistics section to look at the numbers for this split, I want to mention a few thoughts and possible improvements to the metric.


Thoughts / further improvement


  • Possibility to only use timestamps up until a certain point (say 40 minutes) to reduce the larger impact of close games.
  • Maybe one could use a different (slightly lower?) threshold.
  • A different method for calculating what significantly ahead / behind means is possible, though it has to be easy to understand and clearly developed to be of much use. The advantage of my current method is that it has very clear boundaries that are easy to work with.
  • Report the numbers for different time intervals (for example: 10-20, 20-30, 30+) for each team.

In closing, I want to say that there a lots of possibilities for further analysis with this statistic and League of Analytics will definitely dive deeper into the topic with more recent stats soon. To find “percentage of time with 52% or more of gold” and “percentage of time with 48% or less of gold”, head over to our stats section and check out the team stats for the individual regions. If you are interested in new metrics or stats in general, you can also check out my article on gold shift events or read Bridgeburner’s post about this split’s LCS rookies.

Check Also

TSM stats summary

LCS Summer 2016 Regular Season Stats Summary: Team Edition

After doing the player stats a few days ago, here is a quick summary of …