More from The Dotabuff Blog
17 comments
Only carry
    This comment was deleted
    azmarco

      ^nice pfp

      Jaq

        Bruh what the fuck is that picture

        mllcg

          I usually get MMR at the beginning, and middle of the patch. towards the end I give them most back (which is why i play less then) lol..

          RIP spectre next patch. Hero is too busted now

          Topspur

            Make it so you can see heroes shard and Aghs upgrade on their hero page thank you very much.

            VezirOdSemberije

              @kawaiisocks

              Generally more useful in general

              Love you <3

              David Blaine

                What patch are you talking about? I'm confused

                m3me_fr0g

                  "Naturally, sample size should also play a big role in hero evaluation. For example, in the current patch Spectre Spectre is very popular (20%+ pick rate) and very successful (55%+ win rate), meaning that regardless of who picks the hero and regardless of their experience, they will have a good chance of winning the game. Whereas a hero like Meepo Meepo, who has 55%+ Winrate in Divine+ games is only played in 1.9% of all Divine+ games. While the hero can be good if a player knows what to do with them, it is still not a hero one should prioritise off the stats alone."

                  "Sample size" is the wrong term to use here. It isn't an issue of sample size. Low sample size means that the sample is more likely to be contaminated by random noise. The sample of Meepo games is still huge. The difference is just popularity and how pickable the hero is in a variety of games. Meepo is very good in good Meepo games and bad otherwise. Pretty much every game is a good Spectre game.

                  m3me_fr0g

                    "That said, an easy to play hero with high popularity and low winrate is something you should generally discard until the next patch. These are usually quite easy to filter out immediately, unless they get direct buffs in letter mini-patches."

                    What's Juggernaut's winrate when he is last-picked against a team with no BKB-piercing stuns and relatively low physical damage? Because that's going to be very different from his winrate overall. The problem with using winrate figures is that you aren't picking a hero in a vacuum, but against a specific set of enemy heroes, with a specific set of allied heroes. To take an extreme example, every hero's winrate includes the games where someone picks that hero and griefs, but if you don't grief you could immediately eliminate those games from your sample.

                    KawaiiSocks

                      @VezirOdSemberije

                      That's me overediting, fixed now) Sometimes I rewrite/reedit a sentence so many times it just gets turned into a freaking frankenstein mosnter)

                      @Taylor Rapid

                      > every hero's winrate includes the games where someone picks that hero and griefs

                      It also includes every game it is played by a smurf and every game someone outperforms their bracket. It also indirectly includes other hero winrates, since Dota is always a team game, so Juggernaut with 44% win rate had games when he played with WD (55%) for example. I don't know what the resulting expected win rate is, but what I do know for certain is that datapoints in a 5v5 game will push individual hero winrates closer to 50%, since they are mixed and matched in teams. It is a lot more complicated than that since picks aren't random, but the point still stands.

                      > you aren't picking a hero in a vacuum, but against a specific set of enemy heroes, with a specific set of allied heroes

                      I agree that in a 5v5 individual winrates mean less, but when evaluating heroes in a new patch it is still an incredibly important statistic. For heavy negative winrate outliers (like Juggernaut in the current patch) most other statistical factors are helpful in the move towards the mean. There is a lot of nuance, sure, but is this nuance helpful if, for example, you pick in the first phase?

                      I also agree and said many times that "it is Dota, every hero has a game" and yes, there are games where Juggernaut is going to be the absolute best pick despite being bad in the patch, sure. Yet for players with limited time to play and limited time to learn heroes this isn't helpful: we can't all be Yatoro and play all heroes at high level all the time.

                      Playing one of the meta heroes also isn't picking from 3 heroes, Dota patches are usually good enough so that meta for any given role has 5-7 viable heroes with their pronounced strengths and weaknesses and returning to your example:

                      > What's Juggernaut's winrate when he is last-picked against a team with no BKB-piercing stuns and relatively low physical damage?

                      It might be good, but will it be just straight up Chaos Knight or Spectre good? Maybe, I guess. So in a very small subset of games Juggernaut might become as good as the best meta heroes, and will be rubbish otherwise. Again, why not just pick CK, Spectre, WK, PL, Riki, Muerta, Troll, Naga, Void? There is definitely one of them who could also be great with the restrictions you've placed in mind.

                      And just to finish off: Juggernaut this week has a ~44% win rate, but if you look at his Counters page for this week:
                      https://www.dotabuff.com/heroes/juggernaut/counters?date=week
                      The hero he is doing the best against and has an almost 59% win rate versus is Batrider, the poster boy for a BKB-piercing ability. And the reason this win rate is so high is two-fold:

                      1. The sample size is small at only ~1700 matches played this week. So this data has compratively low significance and I'd assume that if more matches were to be played, it would get pushed down. Same would happen to Meepo's individual winrate if there were more games played on him, or in other words if Meepo games had a larger sample size and the data had more statistical significance we could me more sure that thier calculated winrate is reflective of their actual meta strength.

                      2. Batrider is also a bad off-meta hero in the current patch with a ~40% win rate. So Jugg vs Bat is basically two low-tier heroes battling it out and neither should have been picked in the first place. You might argue that Jugg having an almost 60% win rate against Batrider makes him potentially a good pick if you see Bat on the opposite team, but then again: Spectre, CK, WK have a 66%, 65%, 64% wr against Bat respectively.

                      So... why pick Juggernaut in this patch, why go through all these fringe scenarios in your head, why invent the wheel when you can "Spectre Blade Mail go Brrrrrr!" and win most of your games. There will be a patch where Jugg is good by default and in this patch there are going to be good Jugg games as well and then he is going to be the one going "Brrrr!". Maybe it's even going to be the next patch, since Jugg desperately needs massive direct buffs, being one of the most useless carries in 7.34.

                      The advice isn't "don't pick Jugg" it is:

                      > An easy to play hero with high popularity and low winrate is something you should generally discard until the next patch. These are usually quite easy to filter out immediately, unless they get direct buffs in letter mini-patches.

                      Then again, this is for when you want to win games as easily as possible. There are many other valid reasons to play Dota: it is an awesome game to play casually and for fun, getting joy not necessarily from winning, but from self-experssion or execution of jank strategies. I did this in MTG back in the day, though I will point out that Magic: the Gathering is a 1v1 game and when playing jank or trash decks I wasn't frustrating and disrespecting four other human beings on my team.

                      This comment was edited
                      WildeManBeats

                        I’d pick brood here

                        turtok

                          @KawaiiSocks

                          In which context are 1700 data points too small?

                          By most standards and being liberal, we can assume an approximation to the true distribution of a variable if we talk about n >30. When being conservative we can go to arbitrary numbers of about 50/60. As such we get a good idea of the parameters, i.e. mean win rate.

                          Now, an argument against this would be your mentioned 5v5 nature of the game henceforth none of the games will have the same 5v5 pick situation apart from Bat v Jugger but I guess in this context we should be able to agree to regard only the 50 top-played heroes in this conversation as 2 already "off-meta" heroes picked. Yada-yada-yada, thinking about the combinations still possible is a complex mess to grasp quickly.

                          My point being: Even for a week (why not the whole patch? Which would be 40k), 1700 data points aren`t bad.

                          The rest I can more or less agree.

                          Nevertheless, the sample size is indeed not the argument to make for Spectre, but what you say right afterwards against Meepo is how the playtime is split across skill brackets and Spectre easily "go brr win" and Meepo "go brr win if good".

                          KawaiiSocks

                            You are right. For some reason 10% of population got stuck in my head, but in reality for a 95% CI with 5% MoE most of the time you just need like ~400 data points, maybe more in a 5v5, but still.

                            xd

                              Make it so you can see heroes shard and Aghs upgrade on their hero page thank you very much.

                              DANCE WITH ME

                                Hm... no meta changes in 7.34e and a weird text about Timmy and Frostivus on top... That was what we deserved? xD

                                KawaiiSocks

                                  Personally, I am also disappointed to the point of frustration, but we will have stats post for tomorrow so maybe these small changes are going to be more impactful than it looks like

                                  Metsis

                                    The pro meta patch is here. Just tinkering with the numbers and nothing was really nerfed all that much. Little nerfs here and there. Basically 2 buffs and 35 nerfs. The pub meta will not change based on this. Everything excellent is still very good and there were so few buffs that there ain't no raising stars... They could have done this day after TI finished... No one worked on this at all.

                                    This might change the pro meta a little bit and that might be what they wanted to do as there is a tournament coming up again. And Frostivus was announced, but nothing was released for it. I was hoping for more, but expected this. And we got what we got. This is how it usually goes after TI is done, it seems that the team goes on vacation after TI is finished and after they get back to work we start getting some changes. I just hope we get more frequent updates so the meta is kept alive and so that it doesn't get too stagnant.