More TI3 stuff is coming soon, but there will be a slight delay before we hit the actual new stuff. In the meanwhile, with TI3 done I’ve put together a sufficient 6.78 sample (~36k games from Very High), so let’s lead off with the farm dependency readings of the 4 new heroes that weren’t present in 6.77.
![[6.78]FarmDepNewHeroes](https://dotametrics.wordpress.com/wp-content/uploads/2013/08/6-78farmdepnewheroes2.png)
Back when I wrote Abaddon Is Not a Carry, I received a couple complaints. Some were reasonable, others less so. One of the more reasonable ones is that I didn’t have any data supporting my claim. Well now we do, and it turns out that carry Abaddon doesn’t seem to be working too well.
The scoring system I’m using doesn’t mean a lot, so let me break it down how it works with Abaddon. I have about 3200 Abaddon games, and I divide them by the ending CS/min of Abaddon. I divide them into 5 evenly sized groups with the best CS performances in the top group and the worst in the bottom. Carry heroes tend to score really high. Anti-Mage for instance has been top 5 in every iteration of the test and came in #1 in the current test. What determines this is that Anti-Mage’s top 20% CS performances had win rates of near 80%; his bottom 20% CS performances had win rates under 10%. This proves that in top end matchmaking the success of teams with Anti-Mage is highly correlated with how well that team manages to find Anti-Mage farm.
Abaddon’s results are almost the complete opposite, which is why he comes in at 92 out of 102. The average boost a hero receives from being in the top 20% is about +19% to overall win rate. Abaddon’s top 20% is only 14%. On the other end of the spectrum the picture is even more stark. On average, performances in the bottom 20% have a -22% win rate; Abbadon’s is a meager -10%, which is the 5th best performance in this category (or 5th worst if you’re hoping for proof that carry Abaddon is the one true path). What this tells us is that support Abbadon is extremely viable. Semi-carry Abbaddon also appears relatively viable, since his top 20% score is relatively high for someone with such a low score overall. That being said, there is absolutely no evidence that he’s working as a carry. And that’s fine. Puck and Zeus have really low farm dependency scores, and you don’t see people demanding that they only play support. What you should recognize though is that if you’re straight up trading passive farm on an Abaddon, Puck or Zeus for a Anti-Mage, Faceless Void, or Lone Druid, you’re losing that trade hard.
As for the other 3 new heroes, Skywrath Mage‘s profile actually looks pretty similar to Abaddon and Nyx Assassin. I see these as heroes that can either play a supporting role or a low-farm dependency semi-carry role, which leads to their bottom 20% score being relatively high, but their top 20% score also being high relative to the heroes that are near to them in overall score. Elder Titan has really low dependency scores overall, so I’d consider him a straight up farm independent presence regardless of where you run him. Bristleback is much closer to the middle of the pack than the other new heroes. I’d consider his profile similar to Centaur Warrunner in that he won’t win you any games through carry potential, but he makes a good platform for defensive utility items to both support his team while also giving him the survivability he needs to maximize his annoyance.
As for shifts among the rest of the cast between 6.77 and 6.78, it can be difficult to say what’s a genuine shift and what’s just sample noise. But there are 4 relatively big movers.
In confirming some of my previous findings, Huskar‘s farm dependency has gone down significantly with 6.78’s rework. You get a lot more out of him now by focusing on early domination with inexpensive items like Ghost Scepter.
Naga Siren‘s farm dependency also dropped significantly. This suggests to me that support Naga has definitely caught on in pubs and has likely been reasonably successful. It should be mentioned that most of these matches occurred before TI3, so it’s not merely a reflection of imitation. This also points out that these measurements are still a reflection of public trends and not entirely objective statements on something like support viability.
On the more dependent side of things, the biggest positive mover is Omniknight. In terms of ranking it’s only going from 96th to 77th, but I wonder if the new Aghanim’s upgrade has given him a better cash dump than what he previously had available.
The other relatively big mover was Templar Assassin, though it’s at least a percentage point smaller than the other three. I’m curious to see if this was driven by an itemization shift, say more people rushing Yasha instead of a Blink Dagger, and might look into it more later if I have the time.
Anyway, here is the total score list, but as always don’t get too wrapped up in the precise placement. As always it is an estimate, and some heroes just don’t get played all that often in Very High.
![[6.78]FarmDepAll](https://dotametrics.wordpress.com/wp-content/uploads/2013/08/6-78farmdepall.png){kind=small}
DotaMetrics 
The top 7 in the chart are exactly the heroes that fell out of the current meta.
You either play a 5 man game with Gyro or go 4 protect 1 with PL being the best option.
Other options are Lifestealer , Weaver and Alchemist for the obvious reasons.
———————–
AM and Luna are pretty niche picks.
Great article as usual, but would it be possible for you to include a number than represents the variance for farm dependency? For example, Phantom Assassin doesn’t require a whole bunch of farm imo. She can go straight for BKB-basher and wreck some face. I would expect battlefury builds to both achieve and require a lot of farm, but early teamfighting/ganking builds to be less dependent.
I also think split pushers get shifted higher up, not because they “need” farm, but because while winning the game as a split pusher, they are highly likely to receive lots of farm while doing so. For example, I don’t think Luna is much more of a carry than gyro, but I do think she is a stronger split pusher. Lycan doesn’t need much farm imo, but if played with the right team comp, he is going to get lots as he rapidly pushes lanes by himself. However, I don’t know how you might account for this.
On the Phantom Assassin point, I’m not sure that’s a question of variance so much as it is a question of timing. It’s possible that early PA builds wouldn’t score as significantly less farm dependent overall. For example, Clinkz is farm dependent in that he needs Orchid as quickly as possible, but he doesn’t really want the game to go on indefinitely, and his scores are still quite high.
That being said, given a large enough sample size it could be possible to do a score analysis on PA games with a Battlefury and PA games without a Battlefury, and see if they perform differently. This is similar to what I was talking about in comparing TA’s that rush a Blink Dagger to TAs that rush a Yasha. Also, at some point I’m going to try doing some hero analysis on win rates by match duration, and maybe PA will prove to be less end-game dependent than is widely believes.
As for split pushers, it’s tricky to correct for something like that, but it’s also not guaranteed that it’s as clean of a relationship as we might be inclined to believe. Dark Seer is a pretty accomplished split pusher, but he typically comes in around 70th.
I feel like it’s a little off to describe Dark Seer as a split pusher. He usually stops after taking the suicide t1 which isn’t really that hard to take with how far out it is. His presence in most games past that is as a teamfight character who can push a wave or two during a lull between fights, but he really doesn’t have enough of escapes, pushing power, physical damage (for towers, the things you push), or ways to get to an emergency teamfight to make him a really good splitpusher.
Why is KOTL so low?
The regular availability and ease of CSing with Horse-beam means many of his best CS performances are likely poaching CS from more farm dependent heroes.
It’s more of a quirk of pub psychology than a direction to avoid any kind of CS at all costs when playing him.
I’d say because when Kotl gets a lot of cs from illuminate spam (a common situation), his winrate doesn’t change accordingly. The other supports near the end of the list (eg. rubick, ogre, sd, es..) can use effectly at least some of the little farm they get, buying ghost scepter, blink, force staff – all items that are less useful on kotl.
#1 great info break down. i mean where does your brain come from. it’s awesome! able to translate such info into clear brackets.
as well as reflect the current meta n compare.
damn good stuff.
moreover don’t worry about the flame because those who can’t see that Abaddon is not a carry probably can’t comprehend statistics comparisons.
he can be but its like there’s so much that needs to be done and like you mentioned – but at what cost.
Yeah, a lot of it is that people don’t recognize opportunity costs (or don’t recognize that opportunity costs vary drastically depending on the skill level of the game), but a big source of misunderstanding is just the simple linguistics. I’m pushing for a distinct differentiation between Carries and Semi-carries, but for a lot of people everything is either a Carry or a Support. So when I say, “Necrolyte isn’t a Carry,” they interpret it as “Necrolyte should never receive a solo lane.”
wow. you just opened up another pandora’s(lockless) box.
laning strategies colliding with farm/exp requirements.
i guess it gets more wonky due to the fact it depends who (if) a solo lane Necrolyte lines up against. then evaluating possible outcomes at the end of the laning/mid/end(late) game too.
but yea. its like also knowing when to bend the rules of a carry, hard carry, semi-carry. support, greedy support.
slightly separate from this topic but a point that was made during TI3 was that Alliance prioritised their supports getting levels alot more than other teams say dignitas. And in fact became quite a theme in the grand finals itself.
Actually a more accurate example is people playing Naga Siren carry. If the pro scene has Naga playing support most of the time, pubs will follow (before TI3 Naga support was not incredibly rare, and it made sense), so now that Abaddon is in CM, imitation will soon follow. If pros don’t do carry Abaddon, only then will carry Abaddon rightly be condemned (most of the time).
Re: why KotL is so low, there are some obvious reasons — for instance, KotL blasting waves with ponies will take farm away from the carry. But if these are only very-high games, I suspect that’s not the main reason… I think it’s because KotL is going to get farm specifically when things go bad:
Defending rax with ponies, racking up CS in what is likely a losing effort.
Teammates are getting ganked and he’s taking lane farm until they return.
KotL attempting split push to counter opponent’s push.
All of these are far from ideal, and the CS is more of an effect rather than a cause. But this is simply showing the correlation. I think the same reasoning also applies to some of the other lowbies on the list, like Earthshaker. ES can make decent use of farm, but he’s often played as a 5, so ES getting farm = something already went very sideways during the game.
Great catch there! Didn’t think of that.
this brings about a interesting question, why dont you do a exp dependency test too?
It’s in consideration, and maybe I should just stop over-thinking it and try out a simple version. My one point of worry is that a team’s preferred farm profile tends to stay relatively constant throughout the game, but an exp distribution profile often varies. A lot of semi-carries depend on getting a level advantage from a solo lane and then converting into a full-time ganker while giving up a majority of the creep experience to the rest of their team.
In an ideal world, what I’d really like is an exp dependency test at something like the 15th minute, so you could focus specifically on the post-laning distribution of exp and not just the end-game distribution.
What do you think about somehow factoring the CS/min of the rest of the team into the calculations in order to differentiate failed games as a carry and successful games as a support?
Probably the most damning factor for Abbadon actually being a carry is his one steroid: it’s not even a personal one! It benefits everyone attacking the target equally, meaning that it’s more useful to min/max and give farm priority to someone with another steroid on top of that.
His entire “thing” is having a giant list of team utilities, except his ult is really useful at preventing the enemy team from killing him. Like… that’s really standard support play. Hell, his shield is a super good skill and the whole “removes disables” part of it goes to waste if he’s the carry, usually the guy you want to disable lategame anyways.
I don’t even think that AM deserves that no 1 spot. His farm dependency purely results from the fact that most people rush a battlefury even against heroes where it is in no way advised e.g. when you need to fight early with your team. I think that a lot of lost games couldve been won with AM adapting his item build, meaning going for a fast manta and having good fighting potential instead if going for the pure farm item that is battlefury.
This is definitely true. Something that gets ignored a little (not by Phantasmal necessarily, but generally by the people reading these numbers) is that a lot of these numbers are a result of how people are playing the character. The support naga siren mentioned earlier is a good example; as people stopped playing her as a carry, she stopped needing to be a carry to win.
The same is true of AM: he is commonly played as someone who needs to get his farm off the ground quickly and then win by splitpushing and then showing up for a fight or two. Every minute his BFury is slowed down is a big hit to his overall effectiveness with this playstyle, and between poor map visibility and poor last hitting he can easily get ganked and lose a ton of ground, seeing Battle Fury at 22 minutes or even later in a bad game. Now, this is Very High apparently, so you’re not going to be seeing as many Antimages who can only get one CS a wave, but there’s a lot that can go wrong for him and he really needs a coordinated team to be getting the effect he’s trying for.
Last time you posted this chart there was some analysis of the numbers comparing results from normal v high v v.high. Is it possible for you to at least post the chart from each skil bracket separately? I’m interested to see what’s changed, especially with heroes like Necro who had massive differences and have been changed since.
Is there a GDocs link for the raw data like last time? Would be interesting to see