Who is the Most Farm Dependent Hero?

People will often ask “What hero is the strongest with max farm?” and it’s really a pretty pointless question. I don’t have hard statistics on this, but I’d guess less than 1% of games end with even just one person on the team maxing out their inventory. If you just like to ponder silly hypotheticals then knock yourself out, but if you’re trying to learn about Dota there are more practical ways to go about it.

“Who is the strongest carry?” is a little better, but it’s also impossible to answer. Even if we simplify it to ask “Who is stronger, Clinkz or Antimage?” Well, it’s definitely Clinkz if he has a 10-minute Orchid and can concentrate exclusively on making Antimage’s farming life a living hell. Conversely, it’s likely Antimage, provided he get to 30-40 minutes with some strong items, including a Manta Style to nullify Clinkz’ Orchid. So then we have to ask which scenario is more probable…but probable given what? There are 8 other heroes in the game. That makes for a lot of possible team comps, some of which will be favorable to Antimage and others will be favorable to Clinkz. If we’re most concerned with the highest levels of play, we probably only care about Clinkz’ ideal comps vs Antimage’s, but even that is a mess of finding the ideal teams, then finding possible counters to those teams, then counters to those, then eventually tying up the whole thing in some kind of probability map. And even if you pull it off you’ve only compared two carries. To each other. In a vacuum. With ~40 more carries waiting to be addressed.

In short, it’s not a question worth tackling head-on.

I want to approach the question a different way. Forget talking about “best” or “strongest,” I want to instead simply find out which heroes need farm, and which heroes do not.

This is my first attempt.

The idea is that I take a hero and make a list of all their games, ordered by how much CS/min they achieved in the game. I divide the list into 5 equal sections. I calculate the win rate of each section, and then measure how much their win rate increased in the upper sections and decreased in the lower sections. Farm dependent heroes should see their win rate correlate highly with their CS/min. Farm independent heroes should see little correlation at all.

It’s still very much an alpha project. I’m not satisfied with the methodology yet, but the bigger issue is that the sample size I have is not sufficient for every hero. This can be addressed eventually, but in the meanwhile it makes it difficult to judge which methodology is more accurate.

For those who don’t want to traverse the spreadsheet, here’s a summary of the overall rankings:

1 Anti-Mage
2 Luna
3 Faceless Void
4 Drow Ranger
5 Lycanthrope
6 Lone Druid
7 Huskar
8 Outworld Destroyer
9 Phantom Assassin
10 Clinkz
11 Weaver
12 Morphling
13 Bloodseeker
14 Lifestealer
15 Shadow Fiend
16 Storm Spirit
17 Riki
18 Skeleton King
19 Sniper
20 Templar Assassin
21 Juggernaut
22 Ursa
23 Bounty Hunter
24 Viper
25 Chaos Knight

Some other well-known carries barely missed the cut. Dragon Knight and Spectre came in at 26 and 27 respectively.

At the bottom end of things we have

80 Sand King
81 Dark Seer
82 Nyx Assassin
83 Treant Protector
84 Ogre Magi
85 Clockwerk
86 Wisp
87 Shadow Demon
88 Rubick
89 Earthshaker
90 Keeper of the Light

Now for analysis

Most of the top end entries aren’t too surprising, but there are a few curious placement quirks.

Anti-mage, Luna, and Faceless Void as the top 3 might be correct, but it might also reflect that these three have a pretty high potential of being able to farm once the game is already won.

–Speaking of Anti-mage, this is a 6.74 sample. I expect he would drop significantly if this were repeated in 6.76c. Base hero performance in the bracket matters, perhaps more than it should. Alchemist performs poorly in this sample, and should see a much higher placement with the post 6.74 buffs. This probably also plays into why Outworld Destroyer shows a higher farm importance in Normal than he does in the other brackets, as it’s much easier for him to simply out-carry the other team at lower levels of play.

–I expect Lycan to drop even harder than Anti-mage post 6.74.

Huskar scores a lot higher than I feel most people would expect. He’s often painted as a ganker, but I have to wonder if he’s not optimally played as a full-on hard carry.

Bloodseeker seems like a less pronounced version of Huskar, but with much emphasis on “less pronounced.” Bloodseeker’s farm importance falls off hard in the Very High bracket, which I think offers the most accurate picture.

Clinkz being as high as he is isn’t surprising in the least. In the Normal bracket he’s only the 51st most prolific farmer, which is part of why he does so poorly in general matchmaking win%. I see Clinkz as the most extreme case of the time sensitive hard carry, in that his farm at 15 minutes in is way more important than his farm at 50 minutes in. Weaver at one spot lower probably functions similarly.

–Int characters are an interesting study
Outworld Destroyer is clearly the most definite Int Carry in the game.
Storm Spirit is fairly close. This makes sense because of his unique ability to turn excess mana regen into both damage and survivability.
Nature’s Prophet works as a carry in Normal, but falls off hard in High and Very High. In Very High he has the 2nd highest average CS/min, but only the 48th highest farm sensitivity. That pretty much says everything you need to know about Nature’s Profit.
Invoker and QoP stay reasonably high, but they’re beginning to push my definition of carry.
Necrolyte, in my opinion, is not a carry. He’s 68th in the Very High sample. GPM just doesn’t appear to be a priority for him (and for me, dependence on GPM is the most important aspect of a carry. If you want to call Necro a semi-carry, fine, but that’s a very different archetype.)
Silencer appears to me to have a better case for carry-dom, but his farm dependence scores are even a bit lower. Granted, this could reflect player attitudes toward Silencer as much as it does anything about Silencer’s kit.
Tinker also doesn’t see a farm dependence, though it’s possible what’s happening here is that a Boots of Travel-wearing, March of the Machines-spamming Tinker player can actually harm their team if they’re not careful by taking farming opportunities away from the actual carries. Keeper of the Light has a similar issue, and it’s not terrible surprising that he’s far and away the least farm dependent hero in all three samples, despite having the (I cannot believe this. Literally. Did I screw up the programming?) 10th highest CS Average in the Normal bracket.

–Moving on, I’m skeptical about Slardar’s carry potential. I think he makes more sense as an initiator/carry buffer, a la Magnus (who is not in this sample unfortunately). Kinda wonder if he can hang as a suicide solo. This would help him fit into lineups with a hard carry much easier, and might make him an interesting 3rd slot pickup for teams.

–Besides hard supports, initiators score really low on farm dependence, especially junglers. This makes sense because it’s really hard to shut down Dark Seer and Enigma’s farm, but neither of them get much out of farm once you have the basics like Soul Ring/Blink/BKB on Enigma. Earthshaker, Treant Protector, Tidehunter, Sand King, Enchantress, and Omniknight are similar cases, with the one odd exception being Chen. For now I attribute Chen’s performance to sampling error.

Broodmother’s results are all over the board from bracket to bracket. I don’t really care anymore. I’m nearly convinced at this point that she’s the worst character in the game if the enemy team is even half-competent in shutting her down.

Tiny is a weird case because he has two distinctly different playstyles. Carry Tiny appears to be much more of a thing in Very High, but I have to wonder if he’s been eclipsed by the recent Sven buffs. Sven is another case where the 6.74 to 6.76c comparison will be an interesting one.

I think that does it for now. Check out the spreadsheet yourself and come up with some opposing conclusions.

(I don’t know which WordPress Suggested Tag is better: ‘Farming Life’ or ‘Pointless Questions’)

3 Responses to Who is the Most Farm Dependent Hero?

  1. Farm dependency is a really interesting question and I like the way you are approaching it, analyzing cs rather than GPM. Have you considered doing a linear regression for each hero. I think this will give you a much more accurate comparison of farm dependency.

    If you are interested I would be happy to assist you with this. Unfortunately I don’t have match data to run a regression, but if you email me we can work together on it.

    Thanks and keep up the good blogging!

  2. KotL’s Illluminate chews through creep waves. If he’s counter-pushing threatened tier 2 towers he makes a stack of gold. Shame he does so little with it.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: