Do you ever feel like a team is better (or worse) than their win/loss record suggests? Do you wish you had a way to measure that impression?
Here’s a stat that might scratch your itch: I call it Gold Spent Percentage Difference (GSPD).
In simple terms, GSPD measures the gap in how much gold both teams have spent at the end of the game. A large positive GSPD (+10% or higher) shows that the team had spent a lot more gold than their opponents, while a large negative GSPD (-10% or lower) shows that the team had spent a lot less gold.
Using average GSPD, we can spot the teams who are ranked higher or lower than their performances seem to warrant. For example, we can see that Gambit Gaming, Team Dragon Knights, and SK Gaming may have deserved a better fate, while Roccat, Elements, and Dignitas overachieved. There are even some small shuffles in the LCK standings, with KT Rolster looking better than their current third-place rank.
But before we get too deep into the findings, let’s start by defining our measurement.
Feel free to skip this section if you want to dodge the math. Otherwise, let’s dig in!
The calculation underlying GSPD is percentage difference, which uses the following formula:
In other words, we first take the difference in gold spent between both teams, then divide that by the average of the two teams’ gold spent. The team with the gold lead receives a positive value, and the team with the gold deficit receives a negative value. (Technically, this formula is usually converted into an absolute value to just measure the size of the gap, but we are assigning positive and negative values by changing the order of which team comes first in the equation, depending on which team we are talking about at the time.)
This equation produces mirrored GSPD values for both teams: if one team has +10% GSPD, the other team will always have -10%. Mirrored values are the core advantage of using percentage difference, rather than just dividing Team 1’s gold spent by Team 2’s gold spent. If we just do simple division, we don’t get mirrored values. For example, if Team 1 finished with 20,000 gold spent and Team 2 finished with 15,000 gold spent, simple division would work like this:
Team 1: 20,000 / 15,000 = 133.3% (or +33.3%)
Team 2: 15,000 / 20,000 = 75.0% (or -25.0%)
The team with the deficit will always receive a smaller absolute value, so when we average values across games, the positive values are going to have more influence on the averages than the negative values will. That leads to results that skew towards, or inflate, the positive values.
If we use percentage difference instead, using the same example values, here’s what we get:
This simplifies to:
5,000 / 17,500 = +/- 28.6%
Now we have a value that is pretty close to using the simpler method, but equally weighs being ahead and being behind when we calculate averages across games.
There are likely to be two common questions about why I’ve constructed this stat in this way. First, why are we looking at gold “spent” rather than gold earned? Second, why are we comparing percentages instead of actual gold values?
Why Gold Spent?
Gold earned is a great way to track teams over the course of a game, but gold sitting in someone’s pocket at the end of the game doesn’t provide actual value. Most games end with a final team fight, or at least a final kill or two, leading to the victorious team storming the base and often destroying two or more Towers along the way. Those kills and Towers put a big surge of earned gold into the winning team’s pockets, but the end-game gold is all gravy: it didn’t actually help the team win the game, it was just a consequence of victory.
Gold spent, unlike gold earned, tells us what the true power difference was between the teams at the game’s deciding moment.
Why Use Percentages Instead of Actual Gold Amounts?
Percentages help us to see the actual degree of difference between the two teams. If we only use actual gold spent values (e.g. +5,000 gold or -2,500 gold), we don’t how significant that amount was. A gold lead of 5,000 is huge when both teams have only earned around 30,000 gold each, but if the game goes long and both teams have earned 60,000 gold, that 5,000 gold difference is much less significant. By contrast, a 10% gold lead is always big, no matter what stage the game is at.
Now for the fun part: let’s look at the average GSPD values for teams in the NA LCS, EU LCS, and LCK, and see how well they measure those teams’ performances. In these tables we’re looking at average GSPD across the full 2015 Summer split regular season, including tiebreaker games. (Up-to-date GSPD numbers are available on the NA, EU, and LCK team stats pages, or click the tables below for full-size images.)
The big “winners” are Fnatic (+18.9%) and SK Telecom T1 (+15.5%), both of whom have completely dominated their regions this split. Enemy and SBENU Sonicboom come out looking much worse, both teams in the negative double digits.
Comparing GSPD to league standings, we can see how much better the LCK GSPDs match up with the teams’ actual ranks. Both of the LCS leagues have some big jumps (look at Team Dragon Knights, Gambit Gaming, and SK Gaming, for example). But in the LCK, only KT Rolster and Najin e-mFire have moved, improving by just one slot each.
The simple explanation is that it comes down to number of games played. Both the LCS and the LCK are set up as double round robin regular seasons (each team plays each other team twice), but in the LCK each match is a best-of-three series, while in the LCS each match constitutes only a single game. In other words, LCK teams play about 2.5 times as many games as LCS teams. That leads to more “accurate” standings (if GSPD is any indicator). In a shorter season, there’s more opportunity for a few close games to heavily influence a team’s place in the standings.
The biggest victim of the short season seems to be Gambit Gaming, who finished with the fourth-highest average GSPD in the EU LCS, but suffered several losses in very close games and ultimately finished eighth, forced to play in the Promotion tournament. Gambit had the largest average GSPD in wins (+21.8%) and the smallest average GSPD in losses (-11.2%) in the EU LCS, making them a clear example of a team that apparently deserved a better fate. With a longer season, they would have had more chances to earn a few narrow victories to balance out their narrow losses, and may have climbed back into the top half of the standings.
Team Impulse is another interesting case study with even more drastic numbers: their average GSPD is much higher than Team Liquid’s, but they only finished third in the standings. This is because Impulse tended to win in blowouts (+19.7% GSPD in wins) but lose in close games (a remarkably high -2.5% GSPD in losses). Impulse twice lost games this split where they had gold-spent leads. A deeper interpretation is that Impulse relied on snowballing large leads to close out games, but if their opponents were able to hang around and keep things relatively close, Impulse was vulnerable to giving up comebacks.
Some examples on the other side of the fence are Roccat, Elements and Enemy. These teams usually won by small margins and lost by large margins. Enemy, particularly, had a rough go, finishing the split with only three positive GSPD values, and losing one of those games, but somehow earning two wins with gold-spent deficits. Credit goes to the teams for squeaking out victories when they could, but over a longer season, that kind of play has more and more chances to bite you.
As we head into the LCS playoffs, we can look at GSPD as a way to evaluate teams’ relative strength or weakness. We can compare Counter Logic Gaming and Dignitas and see why CLG should be favored to win. We can contrast Gravity and Team SoloMid and anticipate a very close series, much closer than we may have expected even two weeks ago. We can see further evidence of Fnatic’s dominance, noting that none of their 18 wins came with a gold-spent deficit, even if they found themselves overcoming massive mid-game odds more than once. We can lament GIANTS’ fate, finishing sixth—below two teams with worse average GSPDs—and facing H2K when perhaps they deserved a higher seed and a more manageable first-round playoff opponent.
Of course, GSPD is imperfect, as any statistical measurement is bound to be. It doesn’t account for the power you get from killing Dragons, for example. But GSPD identifies new angles of looking at teams’ performance, and prompts us to ask important questions about where unexpected numbers, like those for Gambit Gaming and Team Impulse, might be coming from.
You can find average GSPD numbers in the Team Stats tables, updated daily.