Gotte
Joined: Dec 16, 2010
|
 Posted:
Dec 08, 2011 - 14:14 |
 
|
| PurpleChest wrote: | | Oh, and the interesting surprises for me were that Dorfs come in at tier 1, and that woodies do not. |
Woodies and dwarfs are both good in the right hand. Difference is woodies stink in the wrong hands and dwarfs are still decent. |
|
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Dec 08, 2011 - 14:36 |
|
| PurpleChest wrote: | | I cede that the study and anlysis of statistical data is a science. the wild innacuracy of the understanding of the question, the answers shown and the factors unaccounted for give grounds for the art of interpretting said data into RL useable conclusions. |
Art of interpretation seems to be a very generous phrase. More appropriately it should be "Wikipedia and intuition are not substitutes for an education on any given topic". Being accurately self-aware isn't really an art, per se.
| PurpleChest wrote: | | In its simplest form, any couple marrying in the Uk can expect to have 1.8 children. Yet you will struggle to find anyone with 1.8 children. this lies at the heart of my contention that the actual experience of running a team in b may differ from the average profile gleaned from matches played by all teams of that race. While of course most teams, and to some extent all teams, will trend towards that average. |
That is a shockingly bad analogy - you can't have fractional children, but you can have win records that match the ones given. With few exceptions you choose how many children you have - national averages are showing trends in the choices couples make. Win percentages are not a measure of how often people choose to win, they are simply descriptive measures of how often they do, and in which set of measured circumstances.
That said, I don't think anyone capable of spelling their own name is under the impression that any average means they are guaranted to see exactly that number. However, if you need to guess what your win percentage will be against a given race then those exact numbers are your best estimation in the long term.
Remember that the data included all 55,000+ games and that it includes the games of really good coaches and really bad coaches.. of chaos teams that were chock full of CLPLOMB players, and amazon teams that made sure to keep their TVs too low to face them, and so on. All the different strategies people have employed are factored into that, and across all the games, those were the hard numbers.
FUMBBL represents far more data on Blood Bowl than the BBRC can honestly say they ever had access to - similarly, their data is much less likely to be useful for these purposes, as they came from leagues where it is quite often unlikely you'd see all races being played against all other races, nor with all the same rules being used. The real point of using FUMBBL data is to move beyond the BBRC's gut feeling on numbers and into actual measured numbers, and then to work from there. |
|
|
koadah
Joined: Mar 30, 2005
|
| Posted:
Dec 08, 2011 - 15:07 |
|
|
uuni
Joined: Mar 12, 2010
|
| Posted:
Dec 08, 2011 - 15:23 |
|
I gladly use this time to give a bit off topic example to hopefully enlighten people! 
Purplechest has a point in the interpretation and VoodooMike's reply is on topic. Also many others have brought up interesting and enlightening points on the matter.
For many of the rest of us, it is useful to remember for example that typical and average are different subjects.
Consider the following drivers:
| Code: | Name Accidents
Albert 0
Benjamin 8
Charles 1
Daniel 0
Eric 0 |
Here the average amount of accidents/driver is (8+1+0+0+0)/5 = 9/5 = 1,8. Four drivers, Albert, Charles, Daniel and Eric have less accidents than 1,8 and so are better drivers than average.
On the other hand, the median (middle value, typical) of the set (8,1,0,0,0) is 0. We can also say that the drivers that have 0 accidents in the set, that is Albert, Daniel and Eric, are typical drivers.
In this set, typical driver can be said to be better driver than average.
It is quite easy and normal to mix for example these subjects up. It is ok, but we will benefit for trying to use appropriate subjects, we can then learn better from each other. |
|
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Dec 08, 2011 - 22:51 |
|
| koadah wrote: | | But then some might say that the Box is not 'real' blood bowl. |
Usually as the sole support for their contrary position. All blood bowl that follows the CRP is real blood bowl, by definition.
| koadah wrote: | | What with all that TV matching, sweet spotting and 'ultra tight team management'. |
All of which are completely within the rules of the game. You can certainly take the results of a large number of games and use those to support arguments against things like TV matching, but dismissal on the grounds that you don't feel such-and-such is "real enough" is just hand-waving.
| koadah wrote: | | Being a flat earther I've more interest in who wins the tournaments, leagues and of course Sprints |
Which would be spiffy if there we a large enough number of such things to run decent stats on... without that, its just eyeballing, which is again only one step up from gut feelings.
| uuni wrote: | | In this set, typical driver can be said to be better driver than average. |
The median is not the "typical" case. In a normal distribution it is the same as the mean, in a unimodal skewed distribution comparing the median and mean will tell you the direction in which the data is skewed. In those cases, the median tends to be less affected by outliers than the mean.
Nobody would ever use BOTH the median and mean to say something moronic like "the typical driver is better than average".. or rather, nobody who knew what they were talking about. You would use the median in place of the mean if you felt the distribution was too skewed for the actual average to be appropriate for the task. You'd also say, outright, that it was the median, not the mean... you wouldn't try using a variety of synonyms.
You are, once again, everything I meant when I referred to wikipedia not being an education. |
Last edited by VoodooMike on %b %08, %2011 - %23:%Dec; edited 1 time in total |
|
JimmyFantastic
Joined: Feb 06, 2007
|
| Posted:
Dec 08, 2011 - 23:03 |
|
VoodooMike, did you look on wikipedia for your social skills? |
_________________
Pull down the veil - actively bad for the hobby! |
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Dec 08, 2011 - 23:10 |
|
| JimmyFantastic wrote: | | VoodooMike, did you look on wikipedia for your social skills? |
When I'm looking to be voted into public office I'll worry more about boosting people's self-esteem than I do about being correct. Since the truth isn't a democracy, this topic doesn't require that I say anything other than "you're wrong" when you're wrong.
If that hurts your feelings then hey, that's a shame, but it's really more your problem than mine, isn't it? |
|
|
Were_M_Eye
Joined: Sep 24, 2007
|
| Posted:
Dec 08, 2011 - 23:14 |
|
|
koadah
Joined: Mar 30, 2005
|
| Posted:
Dec 09, 2011 - 00:02 |
|
| VoodooMike wrote: | | koadah wrote: | | But then some might say that the Box is not 'real' blood bowl. |
Usually as the sole support for their contrary position. All blood bowl that follows the CRP is real blood bowl, by definition. |
That's one way of looking at it.
But our goblins are missing key skills so I'm not sure it's fair to call them 'real' goblin teams. Our ogre teams may have 10 ogres. Is that a 'real' ogre team?
I've never bothered looking at [R] division data. The reason is that I (and I guess at least a few others) often play those games like pre-season/friendly games. i.e. returning MNGers, not getting injured, gaining skills etc have a higher priority than winning.
| VoodooMike wrote: |
| koadah wrote: | | What with all that TV matching, sweet spotting and 'ultra tight team management'. |
All of which are completely within the rules of the game. You can certainly take the results of a large number of games and use those to support arguments against things like TV matching, but dismissal on the grounds that you don't feel such-and-such is "real enough" is just hand-waving. |
I don't think that we're dismissing the data. Just saying there is probably more to it than that.
e.g. Do orcs look bad because they are often recommended to rookie coaches? Do woods look good because only good coaches can keep them alive long enough to play a lot of games?
You'll have to wait until the new CR formula has stabilised to look into that. It hasn't been worth doing before.
| VoodooMike wrote: |
| koadah wrote: | | Being a flat earther I've more interest in who wins the tournaments, leagues and of course Sprints |
Which would be spiffy if there we a large enough number of such things to run decent stats on... without that, its just eyeballing, which is again only one step up from gut feelings.
|
That's the point really. Not that this data is wrong but that it doesn't tell us what we want to know. i.e. how would the teams do in a non TV matched environment.
We already saw this when you did your inducement analysis.
So, you have few takers for dwarves in (their) tier 1 as the gut feel is dwarves (and zons too) wouldn't cut it vs the 2000+ claw boys. That's what I want to know and there isn't enough data to prove the gut feel right or wrong. |
_________________
O[L]C 2016 Swiss! - April ---- All Star Bowl - Teams of Stars - 2 more teams needed |
|
dode74
Joined: Aug 14, 2009
|
| Posted:
Dec 09, 2011 - 00:26 |
|
The data largely includes games which are matched by TV (there are a few big differences, but over 90% take place within 120TV of each other). Because of this it's not really possible to say anything about matchups outside a TV-matched environment. 1200TV zons might do very well when everyone else is 1100 to 1320TV, but throw in a few teams at 1500-1600 and there may be a change.
Is there enough data from [L ] to run a similar study? That would remove the TV-based MM, which is itself a house rule. |
|
|
koadah
Joined: Mar 30, 2005
|
| Posted:
Dec 09, 2011 - 00:40 |
|
|
licker
Joined: Jul 10, 2009
|
| Posted:
Dec 09, 2011 - 00:58 |
|
| koadah wrote: | | That's the point really. Not that this data is wrong but that it doesn't tell us what we want to know. i.e. how would the teams do in a non TV matched environment. |
Who is 'we'? The data clearly isn't wrong, and it was clearly stated that it is applicable only to B. Anyone else trying to draw any other inferences from that is making what is likely a rather poor assumption.
| koadah wrote: | | So, you have few takers for dwarves in (their) tier 1 as the gut feel is dwarves (and zons too) wouldn't cut it vs the 2000+ claw boys. That's what I want to know and there isn't enough data to prove the gut feel right or wrong. |
And they are free to define their 'tier' however they damn well please, and then to back up their definition with whatever statistics they can dig up.
Is that what this is all about? Dwarves shouldn't be considered 'tier 1' in the box? Why should anyone find that to be offensive? Dwarves min/maxing to stick sub 1500? Just like other teams to stick in whatever sweet spot they find?
Tiers are rather pointless anyway if you ask me, but at least VM defined his tiers based off of the results of his statistical analysis. It's impossible to argue with it, unless you wish to appear dense.
Oh, it doesn't answer the question 'we' want answered. Funny, I don't remember anyone asking that question until after VM provided his analysis. And even then, when that question gets answered (whatever that question is, even though I apparently asked it too) someone will be there to tell us why the answer doesn't fit with their own perception of reality. |
|
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Dec 09, 2011 - 08:10 |
|
| koadah wrote: | | But our goblins are missing key skills so I'm not sure it's fair to call them 'real' goblin teams. Our ogre teams may have 10 ogres. Is that a 'real' ogre team? |
That is a valid point, yes, but the numbers agree with everyone's intuition on this matter: goblins and ogres are both crappy teams, regardless. Goblins have the missing skills in the Cyanide version, and they still perform poorly. Certainly in pretty much everyone's definition, both are tier 3 teams.
| koadah wrote: | | e.g. Do orcs look bad because they are often recommended to rookie coaches? Do woods look good because only good coaches can keep them alive long enough to play a lot of games? |
It is highly unlikely that either of those things is significantly skewing the data, given the large number of cases being examined. We specifically don't try to control for everything, which is something people often don't get - in order to have numbers that are applicable to the topic in general you have to avoid doing so.
What your objection is, in essence, is that these numbers might not apply in a perfect world, which is fine, because this is the real world, and those are the actual numbers from the actual results.
| koadah wrote: | | You'll have to wait until the new CR formula has stabilised to look into that. It hasn't been worth doing before. |
From what I hear, the CR formula has changed a dozen times, which, no offense to the people creating it, sounds like they're just throwing things at a problem and seeing what will stick. Unless there's a very well reasoned system in place then what is really being developed is a system that will create a number that is in tune with people's intuition, and that's not really a helpful measure - it won't be about the truth, it'll be about falling into line with opinion.
Maybe I'm wrong about it, but I don't see why you'd need to continually change a system if it had a sound statistical and mathematical basis.
| koadah wrote: | | That's the point really. Not that this data is wrong but that it doesn't tell us what we want to know. i.e. how would the teams do in a non TV matched environment. |
FUMBBL is not that environment, so I'm not sure why you'd be focused on that question. The vast, vast majority of games played on FUMBBL are, and always have been, based on TV matching in one form or another. B sets them up for you, R restricts TV difference. L has less than half the games that R does, R has less than half the games B does.
| dode74 wrote: | | The data largely includes games which are matched by TV (there are a few big differences, but over 90% take place within 120TV of each other). Because of this it's not really possible to say anything about matchups outside a TV-matched environment. 1200TV zons might do very well when everyone else is 1100 to 1320TV, but throw in a few teams at 1500-1600 and there may be a change. |
My question is - where does that become a relevant question? Maybe in a challenge-based system with no TV range restrictions it would help you pick opponents you're most likely to beat? Is that really what you want to know about the game?
Looking into inducements on both FUMBBL and Cyanide showed that they do not make up the gap in TVs, so we can already say that teams that do well at high TVs will tend to win if teams that are most successful at lower TVs are forced into playing them. If anything, all the data suggests that things like tournaments are a poor method for determining a real "winner" or "best" in an open division like B or R. Controlling TV lets most teams excel, but if they then have to take on other teams with a good record, the game's dynamic suddenly changes... is that really a legitimate "play off" for coaches?
| dode74 wrote: | | Is there enough data from [L ] to run a similar study? That would remove the TV-based MM, which is itself a house rule. |
Not enough to get as reliable information, but enough to get some numbers on. I don't know much about L's mechanics, but there are 8126 L games in the data (compared to 55,446 B games and 21436 R games), and those will be broken down into 276 race matchups even if we don't want to take TV into account... |
|
|
Gotte
Joined: Dec 16, 2010
|
| Posted:
Dec 09, 2011 - 09:27 |
|
The CR formula has been changed becuase it only was for ranked before.
The importance of non tv matched results is because what everyone in the end cares about is closed-leagues and tournaments. Seeing as this is what this game was actually created for. Not the huge blackbox thingy we got going on.
And this means that any actually rebalancing that might be done down the line. Should be made with that in mind. We can't balance teams for blackbox and then see that the game is fucked for the rest of the world.
ps. I dont actually belive that a new version will come out in the five years. But one can dream. |
|
|
dode74
Joined: Aug 14, 2009
|
| Posted:
Dec 09, 2011 - 09:28 |
|
| Quote: | | My question is - where does that become a relevant question? |
Leagues, tournaments, anywhere where matching is not done by TV.
| Quote: | | Looking into inducements on both FUMBBL and Cyanide showed that they do not make up the gap in TVs, so we can already say that teams that do well at high TVs will tend to win if teams that are most successful at lower TVs are forced into playing them. |
That was only true for small TV gaps with any confidence. Similarly, it didn't take racial factors into account - we can only say that TV was a factor in that analysis, not anything else.
The design of the game was such that certain races were designed to do better in certain matchups if they carried inducements than if they did not (I can find you a quote from Galak somewhere if needs be). An example would be against high-TV Chaos, where many teams would likely benefit more from a couple of apoths and babes than they would from some nugatory skills.
| Quote: | | Controlling TV lets most teams excel, but if they then have to take on other teams with a good record, the game's dynamic suddenly changes... is that really a legitimate "play off" for coaches? |
TV matching is a house rule, so it is legitimate, yes. |
|
|
|
|
| |
Gnomes are trash
Search
