Are Mountain Drakes Overrated? Statistical analysis of the value and priority of dragon types

Ever since Riot introduced the concept of randomly-spawning elemental drakes, there has been consistent debate around the relative value of each dragon type. Consensus dictates that infernal and mountain drakes are the most valuable, with cloud and ocean as also-rans.

Currently, the trend of opinion among many experts–somewhat unexpectedly–is that mountain drake is the most useful dragon, above even the ever-popular infernal.

Curiously, the data does not bear out this opinion.

Based on statistical analysis of 2017’s games to date, the value of mountain drakes is much lower than popular perception. Inferno drake is statistically the most valuable by far, while cloud and ocean drakes continue to be underrated.

The statistical analysis below measures a) the level of priority teams are placing on securing each dragon type early in the game, and b) the value of each dragon type in generating wins. Data is drawn from the NA LCS and CS, EU LCS and CS, LCK, LMS, CBLoL, TCL, IEM Katowice, and MSI, covering patches 7.1 through 7.8.


One way to measure the relative priority of each dragon type is to see how hard teams work to secure them when they spawn. We can get a partial look at this by assessing the first dragon of each game. This will not tell the whole story, but it will provide a pretty decent view.

The table below shows how early in the game each dragon is taken when it is first to spawn, and how much of a lead the securing team had when taking that dragon (as a way to measure how likely teams are to risk a fight or sacrifice a cross-map objective in order to secure the dragon). The gold lead is measured by looking at the most recent minute-mark prior to the dragon dying, and measuring the percentage of the game’s total gold held by each team. A gold percentage of ~52% is generally a substantial lead. The table also reports the win rate of the team that secures each first dragon type, which is a crude measure and will be expanded on more later.

Dragon Type Avg. Time Secured Avg. Prior Gold% Win%
Infernal 10:41 50.58% 71%
Ocean 12:01 50.77% 66%
Mountain 13:34 50.89% 64%
Cloud 13:59 51.10% 66%
Overall 12:36 50.84% 67%

Infernal drake is clearly the most highly prioritized dragon early in the game. Infernal drakes are killed much sooner than other spawns, and are contested in closer games. In other words, teams are willing to fight for an infernal drake even when the gold is close, while if a cloud drake spawns, teams are more likely to wait until they are in control of the map, have a lead, and can take the cloud drake uncontested.

Mountain drakes are taken later than infernal or ocean drakes. This is because mountain drakes are most valuable in the later stages of the game, especially after the Baron spawns at 20 minutes. It’s not necessarily worthwhile for a team to go after an early mountain drake if it means risking losing a team fight and falling behind in gold and/or vision. Meanwhile, ocean drakes are very valuable in laning and mid-game split pushing. Even so, it’s somewhat surprising that mountain drakes would be taken so much later than ocean drakes, and only taken with larger gold leads.

Even more surprising is that the win rate for taking a first-dragon mountain drake is the lowest of any dragon type, at just 64%. This is especially strange given that mountain drakes are taken from larger gold leads and that they are so useful in extending leads and closing games. In theory, if teams are taking mountain drakes around the 13- or 14-minute mark or later and going in with a decent gold lead in advance, they should be set up very well to win the game, and yet the results don’t seem to bear out that theory.

The Value of a Dragon

To dig deeper into the value of different dragon types, I constructed a statistical model that relates dragons to game outcome, while controlling for gold leads and map side. Similar to the early-game rating (EGR) model, I produced a logistic regression, based on the 20-minute mark. Input variables include:

  • Gold difference at 20 minutes
  • Infernal drake difference at 20 minutes
  • Mountain drake difference at 20 minutes
  • Ocean drake difference at 20 minutes
  • Cloud drake difference at 20 minutes
  • Map side (0 = blue, 1 = red)

The dependent variable is win/loss (0 = loss, 1 = win).

The model shows that holding inferno drakes at 20 minutes is a vastly better predictor of winning games than taking any other type of dragon, while mountain drakes have surprisingly low value.

When interpreting these results, it’s important to remember that the model controls for gold differences and map side: in other words, the model estimates the value of each dragon type regardless of whether a team has a gold lead or deficit, and regardless of which side of the map they’re playing on.


The sheer statistical value of infernal drakes is impressive. All else being equal, having one more infernal drake than your opponent at 20:00 gives a 70% probability to win the game. That’s equivalent to being up 2,000 gold, for what it’s worth. (Of course, if you’re up an infernal and 2,000 gold, your probability jumps to 84.5%, so: Get you a team that can do both.)

Mountain drake is the big loser in this model, with the smallest statistical value at the 20-minute mark. The statistical certainty on mountain drakes is a bit lower, with a 0.035 p-value, but models like this typically assign statistical significance at p-values up to 0.05. Even with the model being relatively robust, there’s clearly a need to break down some of the potential reasons for the huge mismatch between mountain drake’s perceived value and its measured value.

The most important thing to understand about mountain drakes is that if you are in position to use them, you’re probably already winning.

Mountain drakes give a big buff to your damage to neutral objectives, including towers and epic monsters like Baron and dragons. That means you are stronger in both split pushing and Baron control. But “stronger,” in this case, is a somewhat complicated concept.

While split pushing with a mountain drake buff, every auto attack on a tower hits for more damage. But to apply that damage, you need to reach the tower, so if your split pushers can’t stand up to their opponents in 1v1s, can’t shove their lanes as quickly, or can’t push as safely due to a lack of map/vision control, then the mountain drake’s split pushing benefits won’t really come into play.

Mountain drake enhances Baron control by allowing your team to kill it more quickly. In theory, this helps when a Baron is being contested, by making it a bit safer to burst the last bit of health to prevent an enemy Smite steal, or when piling into the pit to try to steal yourself, but both of those scenarios are suboptimal anyways: you should say “No” to 50-50 Barons, even if your mountain drakes make it 60-40. More relevantly, mountain drakes help to sneak the Baron more quickly in fog of war or rush the Baron more quickly when the enemy team has players out of position on the other side of the map. But these scenarios are more reliant on rotational play and vision control than they are on sheer damage numbers, and in many cases team composition will be a larger factor in Baron-killing speed than the mountain drakes will. At core, you should only get to apply mountain drake to the Baron if a) you are in control of the Baron already, b) your opponent makes a big rotational mistake and allows you to pull off a sneak, or c) you are attempting a steal. The mountain drake buff itself does not produce any of these conditions: it does not induce your opponents to make macro mistakes, nor does it help you gain control of vision or minion waves (though it may somewhat help you take down some towers to gain map control).

TL;DR: Mountain drake’s benefits are vastly more useful when a team is already winning. Mountain drakes only really benefit split pushing if your team is in position to control the lanes and reach the towers to apply pressure, and they only really benefit Baron play if your team is winning in macro or if your opponent is making the mistake of leaving Baron in fog.

Compare this to an infernal drake, which grants combat stats that are beneficial in split pushing no matter the game state, by granting more damage to win 1v1s and to push waves faster; that help take towers faster by increasing your attack damage; and that help win team fights, which are an essential component of both closing out leads and making comebacks. Infernal drakes are as good as gold: they always have value, by increasing core combat stats that can be applied to any scenario. Meanwhile, mountain drakes are only useful when the map state allows you to apply them.

As an aside, it’s interesting to note that the best team in the world, SK Telecom T1, only captured 56% of the mountain drakes in their games, even though they had 65% overall elemental drake control. Samsung Galaxy had 61% elemental drake control but only 46% mountain drake control, en route to a second-place LCK regular season. Mountain drake control was higher for KT Rolster, who finished second in the LCK playoffs—they took 58% of the mountain drakes—but KT’s most notable flaw was their lack of macro coordination in the mid and late game, and in the end they were blown out in the LCK finals by SKT. As a counterpoint, Team SoloMid had impressive 74% mountain drake control, and Europe’s best teams sat around 60%, but all things considered, the best teams didn’t necessarily have the highest mountain drake control, relative to how well they controlled the other dragon types.


There’s no need for a massive paradigm shift in how different drakes are being prioritized; mountain drake continues to be a valuable tool, specifically for closing out leads. However, statistical analysis shows that some conceptions around the relative value of mountain drakes are off-base and deserve to be reexamined, while ocean drakes and cloud drakes provide more than enough value to be worth scooping up when they’re available.

As with any analysis, and especially with statistics, context is key and interpretations can vary. If you see an angle on this data that hasn’t been explored here, chime in with a comment and keep the discussion and learning flowing.

11 thoughts on “Are Mountain Drakes Overrated? Statistical analysis of the value and priority of dragon types”

  1. I’ve always thought that Mountain Drake seemed a bit overvalued, especially because in almost all cases the difference between killing a turret/dragon/baron etc. in 9 seconds compared to 10 is inconsequential. It’s nice to see some good data to back this up. Thanks!

    1. It’s actually slightly quicker than that because the additional true damage is calculated from your total damage (i.e. before Armour/MR reductions). You can see this if your UI shows your damage outputs.

  2. Mountain drake is useless because it only really helps kill baron and elder drake. If towers become actually difficult to kill (or, otherwise, do damage) then mountain will be useful.

    I have always, personally, rated mountain as my least favorite for this reason.

    1. If you’re familiar with linear modeling, you can look up logistic regression (often called “logit” modeling). That’s the approach I used. It’s a form of multiple regression with a binary dependent variable.

      1. I was curious in particular about how you are controlling for gold. My interpretation of what is here and my (poor) understanding of logistic regression, is that you are determining the influence of each of the above factors at 20 minutes on a win, and then ‘zeroing out’ the influence of the other factors to try and observe one of them in isolation.

        I was wondering if that would then also remove any influence these factors have on each other before the twenty minute mark? For instance if a gold lead at ten increases your chances of taking a drake, or if ocean drakes help get CS leads in lane, etc?

        As I said I am not overly familiar with these things so I am probably missing something, this is just what jumped out at me while I was reading it.

        If you have time to reply, thanks!

        1. The simple answer is that it’s complicated! :)

          Measuring a “snapshot” at 20 minutes does not remove the influence of these things on each other before the 20:00 mark. It is definitely true that if you have a big gold lead at 10 or 15 minutes (which means you probably have a big map control lead), then you’re more likely to pick up the next dragon that spawns. And your other example holds true, as well.

          This is why I don’t think the numbers from this model alone should dictate our interpretation. There’s still a lot of room for context and nuance.

          1. Thank you for taking the time to reply, I think stats in league are very interesting and hope you, and others, keep looking into things like this.

Join the Discussion

This site uses Akismet to reduce spam. Learn how your comment data is processed.