HS replay has several hundred thousand recorded games. Hearthstone has 70 million players. The data gathered by the HSreplay can't even remotely compare to the data gathered from all games played by all players. Comparing the two is pointless.
As you point out, hsreplay's contributors are self-selected. Also, Blizzard's much larger data set gives them options for analysis that are a lot more interesting than just "what percentage of games where this card is played," particularly when they're looking at a specific card. The question I'd start with would be to compare overall win rates in various match-ups between otherwise identical decks with and without the card in them, which hsreplay doesn't answer.
Now, this won't necessarily tell you the difference between the card's actual power and its perceived power, because it's very possible that if the card is perceived to be very strong, not running it will end up a marker of lower skill even if its actual performance diverges from the community expectation. But, it's a more targeted analysis than that statistic you posted, which just looks at the card in the context of that specific deck.
(One might imagine a quirky card that were not generally useful but set up a particular powerful combo. That card would have really nice numbers on those per-card win rates on hsreplay, but only in the specific context of the deck that relies on that combo to succeed.)
This.
Many others have talked about biases involved in HsReplay data versus Blizzard's data, as well as considering the condition of the game state for statistics like '% played WR', which is central to all of this as well.
As for the statistics, even trying to come up with a valid and actionable measure for questions like 'how good is card X?' can be very hard and time consuming. Ultimately, some sort of setup similar to a randomized control trial is ideal for answering the question, but infeasible.
Let's take Innervate as an example. As Lysenko suggested, looking at the win rate difference between two decks that are identical, outside of the inclusion of Innervate, would get more towards the question of 'how good is Innervate?'. But even then, you must replace Innervate with something, and so you are now comparing Innervate versus a particular alternative. Now, how many alternatives do you think are valid to consider? Pair with that, you need to try and control for the skill level of the players piloting these decks, skill level of opponents they face, ranks the games are played at, classes they face, types of decks they face, etc. Because even if one of those is not balanced between the Innervate vs alternative deck, that could be the cause for the win rate difference, and not the Innervate substitution. And assuming you can even obtain a valid difference, now you have to compare it to other druid cards in a similar fashion, because you need to put Innervate's number within the context of a similar number for other Druid cards. Now, might some sort of matching procedure within Blizzard's data set obtain a set of data you feel is valid to attempt this? Well, I wouldn't be so sure, because so few Druids don't play Innervate! When I looked at HsReplay, 99.6% of Druid decks have Innervate, so you potentially don't even have a valid data set to try and answer this question, even if you are Blizzard!
TLDR: Hsreplay stats and data can be misleading, and these questions are very hard to answer, even if you are Blizzard, but Blizzard is better equipped to answer them, but still might not be able to, and that's still assuming that both Hsreplay and Blizzard have people who know what they are doing with the data...
My question is, what purpose do all the stats from say rank 25-20 serve in terms of card balance or design (no one is shattering the earth there with some magnificent homebrew, or we get statements from team 5 like "hunter the highest winrate in ungoro" (no kappa)? and in the lower ranks there are so many misplays and incorrect/ not optimal uses of cards...(I'm no flawless player for sure, but come on).
IMO they need to put more weight on the top tier stats for these decisions (i.e. those more likely to be generated by HSReplay and VS)...but I digress
I think when Blizzard is analyzing, they aren't looking at decks that don't include and decks that do include the card in question. I think they are looking at the same deck and seeing the difference when that card is played and when it is not. This comes with its own flaws, since for example every time Druid wins before ultimate infestation comes out it looks like that card isn't important, but it is a different comparison than looking at decks that do and don't include the card. I'm not sure this is what they are doomg, but the way they've talked about their analysis made me assume this is how they are comparing.
HS replay has several hundred thousand recorded games. Hearthstone has 70 million players. The data gathered by the HSreplay can't even remotely compare to the data gathered from all games played by all players. Comparing the two is pointless.
You know what statistical sampling is?
Agreed. If you think a sample size of several hundred thousand isn't enough you have no idea how statistics or probability work. Assuming the sample is representative the statistics from the same analysis would almost certainly be nearly identical. However, if you argue instead that the sample is biased because mostly a specific type of subset of players (i.e. serious) use HS replay, then you might have a valid argument.
Actually Drawn WR is better to show how good the card is, since the chance that you draw the card is the same whether the situation is good or bad. UI has the third best Drawn WR, which means that it's a pretty good card, while Spreading Plague is almost at the bottom, means that it's not a very good card for Jade Druid.
The "actually drawn" win rate doesn't tell you how the card compares to other possible alternatives in the same deck, though. It may be that the next best card on the list (that didn't make it into the original deck) is a lot worse.
Agreed. If you think a sample size of several hundred thousand isn't enough you have no idea how statistics or probability work. Assuming the sample is representative the statistics from the same analysis would almost certainly be nearly identical. However, if you argue instead that the sample is biased because mostly a specific type of subset of players (i.e. serious) use HS replay, then you might have a valid argument.
They only have several hundred thousand games if you count all of them across all classes and decks. If you look at individual decks, they may have 10 or 20,000 game sample for a particular deck, made up of a relatively small number of players. Lesser-used decks may only be used by a handful of people for 1-2,000 games.
The much larger pool that Blizzard can examine helps with doing analyses like comparing a set of decks that are otherwise identical except for one card swapped out, because it may bring the number of players using these alternative decks up to the level that can serve as the basis for a reasonable analysis (whereas those decks may not even register on hsreplay because so few people are using them in their smaller pool.)
Agreed. If you think a sample size of several hundred thousand isn't enough you have no idea how statistics or probability work. Assuming the sample is representative the statistics from the same analysis would almost certainly be nearly identical. However, if you argue instead that the sample is biased because mostly a specific type of subset of players (i.e. serious) use HS replay, then you might have a valid argument.
They only have several hundred thousand games if you count all of them across all classes and decks. If you look at individual decks, they may have 10 or 20,000 game sample for a particular deck, made up of a relatively small number of players. Lesser-used decks may only be used by a handful of people for 1-2,000 games.
They have hundread of thousand of games for the popular decks.Just this particular deck list has more than 100000 games after ktf.You can even see it in the picture.Even unpopular decks like mine(guldan the teddie killer) had more than 3500 recorded games.So no,what you claim is false.They have more than enough stats.How to interpret those stats and data though is open to debate and what we are discussing here.
The played statistic is useless. You want the drawn statistic. By that number, for your one deck, aya blackpaw is the best card in the jade druid deck. As other have said, spreading plague sits in your hand a lot vs. slow decks, but it turns what would be a 40% matchup into close to a 50% matchup against aggro decks. Without it, jade druid actually loses to something, and thus can be pushed out of the meta.
You are also looking at data for one specific list, and likely not sorting based on anything like matchup/rank. Blizzard is going to be using much more sophisticated methods of data analysis. And honestly, I don't understand why people don't think spreading plague should be nerfed. Before the nerfs were announced, I saw piles of threads complaining about how spreading plague is too good at covering druid's weaknesses.
I pointed out earlier in the thread a specific metric that Blizzard can probably measure with their dataset that hsreplay doesn't have enough games to support: Comparing performance of a popular deck to performance of a nearly identical deck with a single card swapped out. Do the searches yourself, you'll see they don't have enough data on those variations. Blizzard may not either, but with 300x more games, they'll come a lot closer.
How's about we look at this Spreading Plague card mentioned in the initial post.
"...Spreading Plague sits much further outside of the normal range. It's in a league of its own when it comes to absorbing damage."
"Given how good it is at defense, you could argue the card would see play even if it had zero capacity to deal damage."
"The data suggest that for Spreading Plague to be more in line with the rest of the card pool, the scarabs it generates would need to be closer to 1/2’s (60% lower base stats)."
You know who said all that? Hsreplay themselves, yesterday.
HS replay has several hundred thousand recorded games. Hearthstone has 70 million players. The data gathered by the HSreplay can't even remotely compare to the data gathered from all games played by all players. Comparing the two is pointless.
You know what statistical sampling is?
Assuming the sample is representative.
Pretty sure it isn't given the self selecting nature of these 3rd party data gathering sites, but as has been stated, who really cares about rank 25-15 games?! :)
Rotate the entire core set, that's the only good answer. But then of course, if you do it increments... you sell more packs.
I am so sick of this argument. Jade druid is overall a cheap deck. It smothers more much expensive decks. More expensive decks requires more packs to get, one way or another. Will you all shut up with your dumb conspiracy theories and wait until the nerfs are in the game before you run your mouths about shit you don't understand. Stop defaulting to "omg blizzard didn't do what I wanted, dumb greedy bastards it's all about money."
If you craft jade , you pay and play jade. Fiery win axe would be played in many decks, also Hex, Also Innervate. Also old warsong commander, starving buzzard, rockbitter weapon, and so many others.. those are free.
You tire of an accurate argument - that I can understand, we apologize it is not intended to you. On the other hand we tire of a game that used to be great but keeps worsening due to Business decisions that disregard gameplay
Your logic is horrible. If hearthstone wanted more money they'd keep control decks on top, not let a control killer run rampant as you suggest. You don't understand basic economics. A F2P player could make a jade druid deck in 1/3 to 1/2 the time it would take them to make a strong control deck. Lern2logic
As people pointed out, Spreading Plague is generally played to try and salvage lost games (which it does extremely well). It's generally heal for 15-35 (and sometimes even fight back for board). It's incredibly strong and needed a nerf as well.
That said, in the mirror, whoever casts UI first is at a HUGE advantage. It really shows how powerful UI is. That Blizzard didn't nerf UI says a lot about their attitude towards nerfing expansion cards.
P.S. Spreading Plague needed a stat nerf more than a mana nerf.
As people pointed out, Spreading Plague is generally played to try and salvage lost games (which it does extremely well). It's generally heal for 15-35 (and sometimes even fight back for board). It's incredibly strong and needed a nerf as well.
That said, in the mirror, whoever casts UI first is at a HUGE advantage. It really shows how powerful UI is. That Blizzard didn't nerf UI says a lot about their attitude towards nerfing expansion cards.
P.S. Spreading Plague needed a stat nerf more than a mana nerf.
I don't know, losing innervate and making spreading plague cost one more mana is going to be a huge hit for druid when dealing with aggressive decks. Sure, UI is very strong but I'm not so sure that it's going to be much of a problem card tbh. Innervate has been the main culprit for druid shenanigans since forever and the same is still true in combination with UI. Both innervating into it and innervating HP/wrath to avoid overdraw. Losing innervate is also going to make UI lose a chunk of its power.
You're forgetting that blizzards statistics also factor in the dust value of the card in determining it's statistical performance, so UI isn't nearly as op when considering it costs 400 dust. It's also not a part of basic and classic which is another heavily weighted factor in their really great stats. The best stats...
Hi there mates,as you already know blizzard claimed that:
Spreading Plague ended up being the best performing card in Jade and Taunt Druid, Innervate was in the top three, and Ultimate Infestation was somewhere around the middle
However as you can see in hs replay stats below,the above claim is disproven.In fact shows the exact opposite of what they claim.UI has the highest win ratio,while SP has the lowest.
While these are the stats for this exact deck(the most popular jade druid deck),you can easily see the same pattern in every jade druid list.
Now we to to keep in mind that Hsreplay data are calculated by the games of players who have installed deck tracker and not all the games.However there is a great disparity from the numbers and blizzard's claims.Most importandly when you have such a big sample of games (hundreds of thousands of games per deck) the statistics converge to the actual true results.This means from mathematical point of view knowing the entirety of games won't change the results by much.
So please discuss and more importantly keep it civil and i say that because i wrote an aggresive comment my self on the news thread.
If we are to judge power level of cards by win percent when played Leeroy would be the most OP card in the universe.
If we are to judge power level of cards by win percent when played Leeroy would be the most OP card in the universe.
No,it has already been said on previous comments that cards like leeroy,pyroblast etc are finishers so their winrate can be easily explained.Ui is not a finisher but a value/stabilzer card and still has more winratio than most of finisher cards.Read previous comments for more details.
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
You can't even compare this.
HS replay has several hundred thousand recorded games. Hearthstone has 70 million players. The data gathered by the HSreplay can't even remotely compare to the data gathered from all games played by all players. Comparing the two is pointless.
My question is, what purpose do all the stats from say rank 25-20 serve in terms of card balance or design (no one is shattering the earth there with some magnificent homebrew, or we get statements from team 5 like "hunter the highest winrate in ungoro" (no kappa)? and in the lower ranks there are so many misplays and incorrect/ not optimal uses of cards...(I'm no flawless player for sure, but come on).
IMO they need to put more weight on the top tier stats for these decisions (i.e. those more likely to be generated by HSReplay and VS)...but I digress
I think when Blizzard is analyzing, they aren't looking at decks that don't include and decks that do include the card in question. I think they are looking at the same deck and seeing the difference when that card is played and when it is not. This comes with its own flaws, since for example every time Druid wins before ultimate infestation comes out it looks like that card isn't important, but it is a different comparison than looking at decks that do and don't include the card. I'm not sure this is what they are doomg, but the way they've talked about their analysis made me assume this is how they are comparing.
Actually Drawn WR is better to show how good the card is, since the chance that you draw the card is the same whether the situation is good or bad. UI has the third best Drawn WR, which means that it's a pretty good card, while Spreading Plague is almost at the bottom, means that it's not a very good card for Jade Druid.
The "actually drawn" win rate doesn't tell you how the card compares to other possible alternatives in the same deck, though. It may be that the next best card on the list (that didn't make it into the original deck) is a lot worse.
The played statistic is useless. You want the drawn statistic. By that number, for your one deck, aya blackpaw is the best card in the jade druid deck. As other have said, spreading plague sits in your hand a lot vs. slow decks, but it turns what would be a 40% matchup into close to a 50% matchup against aggro decks. Without it, jade druid actually loses to something, and thus can be pushed out of the meta.
You are also looking at data for one specific list, and likely not sorting based on anything like matchup/rank. Blizzard is going to be using much more sophisticated methods of data analysis. And honestly, I don't understand why people don't think spreading plague should be nerfed. Before the nerfs were announced, I saw piles of threads complaining about how spreading plague is too good at covering druid's weaknesses.
I pointed out earlier in the thread a specific metric that Blizzard can probably measure with their dataset that hsreplay doesn't have enough games to support: Comparing performance of a popular deck to performance of a nearly identical deck with a single card swapped out. Do the searches yourself, you'll see they don't have enough data on those variations. Blizzard may not either, but with 300x more games, they'll come a lot closer.
How's about we look at this Spreading Plague card mentioned in the initial post.
"...Spreading Plague sits much further outside of the normal range. It's in a league of its own when it comes to absorbing damage."
"Given how good it is at defense, you could argue the card would see play even if it had zero capacity to deal damage."
"The data suggest that for Spreading Plague to be more in line with the rest of the card pool, the scarabs it generates would need to be closer to 1/2’s (60% lower base stats)."
You know who said all that? Hsreplay themselves, yesterday.
https://hsreplay.net/articles/28/deep-dive-into-spreading-plague
Ibn Fahd.
UI is definitely a problem.
As people pointed out, Spreading Plague is generally played to try and salvage lost games (which it does extremely well). It's generally heal for 15-35 (and sometimes even fight back for board). It's incredibly strong and needed a nerf as well.
That said, in the mirror, whoever casts UI first is at a HUGE advantage. It really shows how powerful UI is. That Blizzard didn't nerf UI says a lot about their attitude towards nerfing expansion cards.
P.S. Spreading Plague needed a stat nerf more than a mana nerf.
You're forgetting that blizzards statistics also factor in the dust value of the card in determining it's statistical performance, so UI isn't nearly as op when considering it costs 400 dust. It's also not a part of basic and classic which is another heavily weighted factor in their really great stats. The best stats...
Leeroy was nerfed ages ago. Pyroblast is pretty good though. That was nerfed too!