LoL Side Advantages and What They Suggest About the Radiant Advantage

March 19, 2014

A while back we looked at Radiant vs Dire win rates in 6.74 and 6.77 and found that Radiant had a slight overall advantage that was most pronounced in short games and actually became a disadvantage during long games.  While we don’t have absolute proof, it’s generally accepted that the late game Dire advantage is driven largely by the positioning advantage Dire has for Roshan.  But when it comes to the other side of the dynamic, Radiant’s  advantage during short games, all we have are competing theories with no clear evidence.  Recently in League of Legends, a stat site released some data on the balance between their sides in various modes, and the implications could help explain what’s going on.

For those unfamiliar, the LoL Blue team corresponds to Radiant positionally and Purple corresponds to Dire.  The default LoL map has some asymmetry, but nothing to the extent found in the Dota map.  What the stats show is that on the standard League map, the Blue side has a significant advantage in all modes (Unranked: 56.3%, Ranked: 55.2%, Ranked Teams: 53.5%).

Where things get interesting is that the same advantage exists in ARAM (55.4%) which is a perfectly symmetrical map.  This is the strongest evidence I’ve seen that the Radiant advantage is related to the viewing perspective/camera angle.  The LoL map with the smallest side advantage is Dominion (51.8%), which also happens to be the only map tested without a diagonal tilt, as the map is East/West symmetrical.

It’s still not proof, but I think there’s decent evidence here in support of the camera angle theories.  Dire’s inherent Roshan advantage could actually be rather appropriate.  Roshan’s positioning would likely have the greatest net effect in low level games as low level games have the longest average duration.  These also might be the players least capable of dealing with the camera disadvantage, and therefore most in need of a larger counterbalance.   It’s fairly speculative, but it would be a pretty unique feature if both the Radiant and Dire advantages separately scaled at a similar rate with the overall skill of the players in the game.

Advertisements

[Guest Post]A New and Improved Test for Farm Dependency

March 17, 2014

For a few months now I’ve been in an e-mail correspondence with a reader by the name of Bishop’s Guest on the subject of putting together a better farm dependency test, and this week he provided me with a .pdf writeup on the test and a table of the results.

Given that this has been in the works for several months, the data used is from 6.77, so it’s not the 6.80 Farm Dependency test that people have been asking for.  Think of it as foreshadowing.

6.77 Farm Dependency pdf

Farm Dependency Table in Google Drive

Anyway, I’m going to let the pdf speak for itself.  It explains how the test works, and also comes with visual representations of each hero’s farm dependency, like this comparison between Anti-mage and Faceless Void.

FarmDepAMvsFV


Hero Win Rates by Match Duration: 6.80 Edition

March 10, 2014

The duration test isn’t that heavily requested compared to farm dependency (which is coming), but I needed part of this in preparation for something else.  Turns out there are some interesting trends emerging since we last looked in 6.78.

I should mention in advance that I have changed the definition of Short/Mid/Long divisions.  Previously they were determined by splitting the sample evenly into 3 parts, and the dividers turned out to be 33:20 and 41:46.  To simplify things I’ve changed to using 30:00 and 40:00.  They’re close to the 1/3 split, but it’s much easier to remember that short matches are less than 30 minutes and long matches are longer than 40 minutes.  It also has the added benefit of being directly comparable to how datDota‘s duration filter works for competitive comparisons.

Anyway, let’s start things off with how the 5 new heroes since 6.78 scored:

680DurationNew

Preferring the long end of things we have Earth Spirit and Legion Commander, but I would caution against reading too much into these two.

It’s no secret that Earth Spirit players are largely struggling since the 6.80 nerf, and it’s my suspicion that this might be capable of creating or exaggerating a late skew.  Essentially, if a hero has a proclivity for being blown out, then a match that lasts 40 minutes has better than expected odds of not being a blow out.

As for Legion Commander, a lot of players are still very much dedicated to her (not very good) jungle.  If you have a passive, farm oriented jungler and the match ends in under 30 minutes, chances are the contest did not resolve in your favor.  I think there’s some evidence of this being a general jungler trend, with the junglers that escape from it being those that aim for early objective control (Chen, Enigma, Lycan, and Ursa).  If we had a collection of purely laning Legion Commander games, I suspect she would still have a late skew but that the skew would not be nearly as dramatic as this one.

We also have Phoenix with a moderately late skew, which is somewhat surprising to me as I’d expect his minus attack speed oriented strategy to fall off in late game.  Perhaps the scaling on Sun Ray keeps him relevant in late game.  Alternatively, it could just be that there are a lot of bad Phoenix players in this sample, as this was release week Phoenix, and they might be disproportionately likely to lose games early.

On the short side of things we have Terrorblade.  He has a pretty early skew for a carry, but he does fit the pattern of having a lot of free damage directly packed into his kit in Metamorphosis.

Finally we have Ember Spirit who has no discernible skew in either direction.  This isn’t terribly surprising.  He doesn’t have any of the features you’d expect to see in a short skew hero (support orientation, pushing power, free auto-attack damage), but he also doesn’t have a passive start or dominant item-based late game.  Ember Spirit  is likely better off aiding his teammates’ preferred match tempo than trying to set his own.

With the new heroes addressed, I want to move on to the top 15 heroes preferring short games because that side of the list has changed dramatically since 6.78.

680DurationTop

Push heroes always skew early, but 6.80 has taken it to new extremes.  The new and improved Lycan leads the pack, but 6 of the top 7 are pushing heroes.  An interesting note is that this push mini-meta has had interesting effects on the push heroes that haven’t recently received changes.  Nature’s Prophet and Leshrac have both seen an increased early skew, while former top 15 entry Luna has dropped down to 42nd.  One explanation for this is that Nature’s Prophet and Leshrac can support any push strat, whereas Luna might be in direct competition with Lycan, Pugna, or Death Prophet as the centerpiece.

I also want to mention the former top 3 short game heroes in 6.78, Treant Protector, Spirit Breaker, and Huskar.  All 3 have seen nerfs since then, and correspondingly their early prowess has diminished.

  • Treant Protector:  16.53% -> 9.46%
  • Huskar: 14.28% -> 7.37%
  • Spirit Breaker: 12.82% -> 2.22%

This appears to be the inverse of the earlier Earth Spirit theory.  Being a dominant hero in a patch period seems to exaggerate the heroes short skew.

On that note I’m going to close things with the complete chart.  This time I’ve put the Very High and Normal charts side-by-side.  One interesting thing about the Normal bracket (est. < 3200 MMR) is that it exhibits a much stronger correlation between a hero’s overall win rate and their short skew despite having longer games on average.  So for those of you concerned with digging yourself out of the trench, don’t agonize so much over getting mid.  The simple trick is just learning how to make an impact early, and you can reliably do that from a variety of positions.

680DurationFull


6.80 Win Rates and Meepo: Skill Can’t Fix Everything

March 5, 2014

One of the most common responses to yesterday’s 6.80 hero win rate chart was shock that the hero whose win rate decreases the most in high skill games is Meepo.  This flies in our face of our intuition that the heroes that improve the most in higher skill games are skill intensive heroes, particularly ones that are micro-intensive.  Meepo ought to be the poster boy of this type of hero, so what gives?

First, let me point out that Meepo doing better in low skill games is not a new phenomena, as it’s been true for several patch periods, including ones before the recent buffs to his ultimate.  Second, there’s a common sentiment that this win rate shift is caused by smurfs playing Meepo in the normal brackets.  While I can’t rule this out, I’m skeptical that this effect is strong enough to account for the entire shift.  Dota 2 is reasonably good at frustrating the creation of smurf accounts, and outside of extremely managed cases, successful Normal bracket smurfs should move to High relatively quickly.  So rather than relying on those two explanations, I want to propose that past a certain point of Dota awareness, we begin to overestimate our potential to reach a theoretical skill ceiling while underestimating our opponent’s ability to meet a much lower skill ceiling and disrupt these attempts.

When it comes to deceptively nebulous concept of player skill, we have a tendency to focus on really obvious displays of it.  The ability to last hit and farm is the biggest one when people focus on moving out of Normal ranked games.  Meanwhile, the closer we get to the top end of the bracket we focus on being able to play skill intensive heroes.  It’s no surprise that the list of heroes that see the greatest usage rate increase in Very High games almost always include Invoker, Rubick, and more recently Mirana.  Past a certain point, people start fishing for WoDota moments and merely outfarming your opponent in a pub game becomes the passé way to win.

But what we forget is that higher skill also includes improved map and situational awareness.  I have a surprisingly good example of this coming up that I don’t want to spoil, so for now let’s focus on ganking.  Watch the front page of live games and you’ll see a dramatically greater emphasis on early game aggression (including players who don’t need a tooltip to know that Smoke of Deceit actually exists).  These players are better at spotting early weak points and punishing them, and it’s as significant of a shift as the improvement in last hitting and mechanical skill.

So we have Chen.  He’s a hero that does better in Very High partially because those players have better micromanagement mechanics, but also because he’s now playing on teams that actually know how to take advantage of his creep army and use it to create early kills.  Meepo, on the other hand, is a big squishy gank target.  Or maybe it’s better to say 1 to 5 squishy gank targets.  But either way it’s a vulnerability that you can’t entirely mitigate through better play, and it’s a vulnerability that’s significantly larger than anything other high skill cap heroes like Chen, Invoker, or Rubick have to deal with.

And yeah, we’re totally capable of finding “that one time” when Meepo got out of control, and for most of us, that time will probably involve N0tail.  But take a look at N0tail’s datDota history with the Meepo.  5-5 is a respectable record for a hero most pros won’t dare to touch, but the quality of the teams that he’s beaten with Meepo doesn’t really stack up to the quality of teams that he’s lost to.  Meepo looks unstoppable when a team lets him get out of control, but better teams will invest a lot of resources into making sure that doesn’t happen, and they’ll be more efficient at making these investments pay off with kills that prevent Meepo from ever getting out of control.

You hit a similar scenario with Tinker.  Yes, we’ve all seen a Tinker with a Soul Ring, Bottle, Boots of Travel, Blink Dagger, Force Staff, and Scythe of Vyse that’s capable of completely dominating the map.  But at the same time we forget the incredible amount of investment that took to get him there and the vulnerability he had while he was farming (and the vulnerability the rest of the team has during this period of temporary 4v5 if we want to take things a step further).  Shutting down a solo Tinker’s farm isn’t much harder than shutting down a solo Anti-mage.  The big difference is that Anti-mage is an investment that warrants 4 protects 1 coverage and Tinker does not.

Basically, Tinker and Meepo are examples of heroes where it’s easy to get overly enamored with their best-case scenarios.  From there you then fall into the trap of believing that getting to that best-case scenario is solely a matter of the skill of the player playing the hero and, in the process, forgetting that the opponents are just as capable at exploiting these heroes’ liabilities and preventing that scenario from ever coming to fruition.  This isn’t to say that Meepo and Tinker can never have their day in the sun competitively.  It’s more that if you want to make heroes like this work you’ll have to have a complete team plan on how you’ll compensate for those liabilities without investing more than is warranted in their protection.  While also have a savant playing them.

 


6.80 Hero Stats: Full Win Rate Chart and Spreadsheet

March 5, 2014

It’s a bit late, but I do actually read the comments sections and noticed various people requesting a more complete list of the results used in 6.80 Hero Analysis.  So here you go.

Sortable spreadsheet including sample size data

Chart of VH-H-N Win rates

680winrates