Most bashy to least bashy

[gjnoronh]

Okay my seven year old son has developed a love of Blood Bowl and is FREQUENTLY asking me what teams are 'the most bashy' and 'most dashy.'
We've explained team types to the kids as Bashy/Hybrid/"Dashy"

I've told him that any one of the elf teams is a good guess for 'most dashy' (though I recognize skaven can get there.) I recognize amongst the elves some are a little more likely to use an agility game then others.

But the bashy question I've tried to explain to him is hard to say at best and depends on team value etc.

He has an iPad team of Orcs and a tabletop team of Chaos.

He's 7 and he's not going to want the stats and nuance

So how would you answer 'most dashy' and 'most bashy' for a seven year old

And yes I am totally proud of him and super happy to be playing blood bowl with my kids!

[rolo]

I'd divide them up into categories:

Super Dashy:
Wood Elves, Skaven, Slann, Pro Elves

Pretty Dashy:
Dark Elves, High Elves, Humans

Mixed:
Necros, Lizards, Underworld, Amazon

Pretty Bashy:
Pact, Undead, Norse, Chaos Dwarves

Super Bashy:
Ogres, Orcs, Dwarves, Chaos, Khemri, Nurgle

Just for Lulz:
Vampires, Goblins, Halflings

I'd put Khorne in "Pretty Bashy". I don't know enough about Bretonnians to pigeonhole them.
Keep it simple, no need to start parsing whether Nurgle or Khemri are bashier - both play smashmouth as plan A and there is no plan B.

[CyberedElf]

I did try to do some statistical analysis to determine the answer to this question. The results were not as expected. I tried to compare CAS/net CAS to winning/TDs. It was an attempt to say which teams need to cause casualties to be successful.

The stats that coincide with MUCH more commonly accepted/expected spectrum of bash to dash is actually very simple. "Average casualties caused" gives closest to the expected list.

Elf 1.16
Wood Elf 1.17
High Elf 1.24
Vampire 1.34
Dark Elf 1.36
Slann 1.42
Skaven 1.61
Amazon 1.94
Goblin 1.95
Lizardmen 2.05
Human 2.07
Halfling 2.08
Orc 2.26
Norse 2.36
Necromantic 2.38
Underworld 2.39
Dwarf 2.56
Chaos Pact 2.57
Undead 2.59
Ogre 2.61
Khemri 2.67
Chaos Dwarf 2.82
Nurgle 2.98
Chaos 3.08

Website I copied this from did not have variance.

[dode74]

http://www.nufflytics.com/post/bash-das ... e-numbers/

[CyberedElf]

http://www.nufflytics.com/post/bash-das ... e-numbers/

All they did was view 17 stats from the two dimensions that separate the team average the most, then place the author's own interpretation on what they saw. Nothing in the actual statistics done on that page actually say anything about bash/dash. It showed that some teams perform differently along a spectrum already recognized. The stats make no statement on why or how they perform differently. Everything written there about why and how the teams perform differently is purely the authors opinion.

[Digger Goreman]

To try and (re?)answer the question: consider the skills....

To "bash" you must down your opponent, usually through blocking (one might consider tackle as bash). If you consider breaking armor and injury as part of the "bash", then other skills come into play. So... certainly Norse, dwarves, cds and any other teams with access to initial block can be considered "bash".... Teams with Mighty Blow "bash" harder.... Tacklers, Diving Tacklers, players with prehensile tail... whatever gets you to the armor breaking roll and gets the opponent off the field....

One of the bashiest teams I ever saw was a fumbbl monstrosity elf team with well over 200 games and all players had Mighty Blow and Piling On!

I think one can somewhat qualify starting teams... then it becomes "what if" and probable access to skills to take the other off the field. Another consideration could possibly be length of season (if starting from scratch) as my Necros have a hard time bashing at first (only two blockers and a pair of iffy frenziers), but from 6-12 games into a season they usually have block and (with a double) mighty blow to go with claw... then they are certainly bashier... as well as finally getting those strength skills on the golems.... Fumbbl 200+ game, chaos and nurgle can be frightening... tabletop equivalents rarely to that extent....

Dash (what I call avoidance?) teams use skills and agility to (generally) avoid contact, mitigate bash, and preserve themselves while scoring....

So, can anyone reach a consensus on what the qualifications are before labeling begins?

Maybe?

[CyberedElf]

I just realized I never came back with the Cas caused vs TD/win analysis. Same link:
https://docs.google.com/spreadsheets/d/ ... sp=sharing
Still not really as expected.
cas vs td
Norse
Chaos Dwarf
Lizardmen
Underworld
Chaos Pact
Undead
Dwarf
Chaos
Amazon
Ogre
Goblin
Khemri
Halfling
Orc
Elf
Necromantic
Human
Nurgle
Slann
Dark Elf
Wood Elf
High Elf
Vampire
Skaven

cas vs win
Norse
Lizardmen
Underworld
Chaos Dwarf
Chaos Pact
Amazon
Halfling
Dwarf
Undead
Goblin
Human
Khemri
Chaos
Orc
Ogre
Necromantic
Elf
Dark Elf
Slann
Nurgle
Wood Elf
Skaven
High Elf
Vampire

The real shockers for me were Khemri and Nurgle. Nurgle was one of the worst for number of casualties predicting if they scored or won. Khemri was only in the middle.
For Humans and Halflings casualties were a better predictor of winning than scoring in relation to other teams. Chaos and Ogre were the opposite. For all the rest the order of CAS vs TD and CAS vs win were fairly similar.

It really depends on how you define Bash/Dash. There is some consensus as to where things lie on the spectrum or get grouped, but there is not consensus on how it is defined.

Some might look at my list and assume it is faulty because of Khemri and Nurgle, but I clearly defined my definition before looking at the stats. Don't argue with the results, argue with the definition or data sample.
I presented 5 possible definitions of bashiness:
Causing casualties in a game predicts scoring
Causing casualties in a game predicts winning
Net casualties in game predicts scoring
Net casualties in game predicts winning
Causes the most casualties

Data:
Three months of Black Box matches on FUMBBL (for the first four definitions)
All CRP FUMBBL Black Box games for the last definition (not my stats)

[dode74]

http://www.nufflytics.com/post/bash-das ... e-numbers/

All they did was view 17 stats from the two dimensions that separate the team average the most, then place the author's own interpretation on what they saw. Nothing in the actual statistics done on that page actually say anything about bash/dash. It showed that some teams perform differently along a spectrum already recognized. The stats make no statement on why or how they perform differently. Everything written there about why and how the teams perform differently is purely the authors opinion.

The two extremes:

THE BEAUTIFUL GAME
With lots of passing and running, these are the teams to look for when composing highlight reels.

TAKE NO PRISONERS
For the teams in this group, the only safe touchdown is one where there are no opposition players left on the pitch. Lots of injuries caused and very few sustained will cause a team to end up in this category

Seems pretty much like the definitions of what bash vs dash would look like, to me. What more, exactly, would you want?

[rolo]

Some might look at my list and assume it is faulty because of Khemri and Nurgle, but I clearly defined my definition before looking at the stats. Don't argue with the results, argue with the definition or data sample.

Having played Khemri in a league, this isn't really all that shocking. I have two possible explanations:
- First of all, Khemri at low-to-medium Team Value just don't hit that hard. Blocking is for position most of the time (getting opposing players out of the way, pinning them, keeping them away from the ball). Sometimes this causes casualties, but that's a bonus. Nobody starts with Mighty Blow, only two players start with Block. Players skill up VERY slowly and are more vulnerable to attrition than you'd think. Rerolls are expensive. When Tomb Guards are in a good position (ideally, tagging 1-2 important ST3 players), it's better to just stand there.
My prediction, if this theory is true, is that Khemri throw fewer blocks, and cause fewer casualties per block thrown, than teams like Norse, Dwarves, Chaos Dwarves.
Counterexample: I'm kind of surprised that Lizardmen are as high as they are on your list, as at low team value (where the vast majority of games are played), Lizards have a lot of the same skill issues as Khemri.

- Second of all, when Khemri DO hit, it's often by fouling. Foul CAS often don't show up in statistics (I haven't looked into your numbers to see if that is the case).
My prediction, if this theory is true, is that Khemri will end up much higher on that ranking if Foul casualties are included.

[CyberedElf]

The two extremes:

THE BEAUTIFUL GAME
With lots of passing and running, these are the teams to look for when composing highlight reels.

TAKE NO PRISONERS
For the teams in this group, the only safe touchdown is one where there are no opposition players left on the pitch. Lots of injuries caused and very few sustained will cause a team to end up in this category

Seems pretty much like the definitions of what bash vs dash would look like, to me. What more, exactly, would you want?

Nothing in that analysis shows "The beautiful game" teams pass or run the ball more.
Nothing in that analysis shows "Take No Prisoners" teams cause more injuries and certainly nothing there shows they score when there are fewer opponents left.
They might be fine definitions, but they do not relate to the stats that were presented. The groups were defined by opinion, both in definition and selection of teams.
Also, as two definitions it makes two scales that were not proven to be directly opposed. All the author did was say, "When I simplify the data, some teams appear to perform (sometimes) more similarly to each other than other groupings I can define. I will define labels for those sets based on my personal experience." Except the author even acknowledged the sets are not clearly defined and some teams in a set are closer to teams in other sets than they are to some teams in their own set.

What would I want? Define a scale (either of the definitions you listed would work) then see which teams match that definition the most. Don't simplify the data, then define groups based on preconceived notions.
In fact, "Take No Prisoners" is very similar to one of the definitions I used. Net casualties compared to TDs. Which teams are more likely to score more TDs when they have a larger net casualty difference? By that definition Nurgle and Khemri still don't make the top third.

The author never justified his definitions for each group. They grouped based on visual inspection of simplified data, then applied their own personal descriptions to the group. How is that a sound statistical analysis?

[dode74]

I think you're dismissing it out-of-hand. The write-up doesn't have as specific much info as I'd like, either, and it takes some reading to even conclude that the pass/run and cas caused/sustained parameters are likely the principal components in each case, but the dataset used is defined, as is the methodology. It's the interpretation which you appear to have issues with. The author states pretty well that it's an interpretation of the data, and asks others for their own insights:

Despite this spectrum making an hard classification essentially arbitrary, how would you group races based on this analysis?
...
If anyone has any insights from within these leagues as to why specific races might look out of the ordinary I’d be glad to hear them. Or if you think there’s a better grouping of teams based on the above graphs let me know below.

Personally I'd agree with the statement that any metrics from the teams will be on a spectrum, making it essentially arbitrary as to where you draw the line (similar to how people try to put the T1 teams into tiers within T1 itself). Perhaps you're asking for something which isn't there in quite as well-defined a manner as you'd like it to be?

[CyberedElf]

Dode: Mostly I agree with what you just said.

Article Title: BASH/DASH/HYBRID BY THE NUMBERS
If the author didn't know which teams were bash or dash before the analysis, no numbers were presented that would inform. I don't see how the title claim was actually addressed. That is what I have an issue with.
I think it is wrong to say something is "????? by the numbers," then address "?????" in an admittedly arbitrary manner.

we have been able to create a data-driven classification of Blood Bowl races based on team performance

I disagree that changing the viewing angle and removing 88% of the data before a visual inspection qualifies as a "data-driven classification." The author stated that the classifications were "essentially arbitrary," but it was also stated that a "data-driven classification" was created. My standard for labeling as "data-driven" is higher than the authors.

As for methodology, the only stats they did was PCA and they did not use it like it's supposed to be used. The purpose of PCA is to see the relationship between original axis and the principle component axis. Pass/run and casualties might be highly associated with the primary principal components, determining that is what PCA is actually for, but that is not what the author did.

it takes some reading to even conclude that the pass/run and cas caused/sustained parameters are likely the principal components in each case

That is exactly what a good use of PCA would answer. That would be the analysis, not something the reader has to guess because they know the system being looked at. The loading output of PCA gives you that information, but the author chose not to discuss that at all. The author was clear about the process but did not follow through. Discussion of the loadings would have helped define the groups in an actual data-driven manner. The cutoff between groupings would still be arbitrary, but the definition of the group did not need to be.

Honestly I enjoyed reading and thinking about the article. With a different title and removal of the two "data-driven" sentences, I would not have a problem with it. I would have just wished they had done a little more by discussing the loadings of the observations.

[Lyracian]

http://www.nufflytics.com/post/bash-das ... e-numbers/

While very nice graphs I can not actually tell what PC1 and PC2 are. My Guess is net passes and net Cas from others comments.

[CyberedElf]

http://www.nufflytics.com/post/bash-das ... e-numbers/

While very nice graphs I can not actually tell what PC1 and PC2 are. My Guess is net passes and net Cas from others comments.

Quick explanation of PCA used there:
He had 16 variables for each coach each game plus race. Some overlap from one coach to the other, sustained casualties and inflicted casualties, for example.
Plot the data in 16 dimensional space.
Pick the best fit line in that space, that is PC1. It does not have to be one of the axis.
Pick the best fit line that is orthogonal to the previous N lines, that is PC(N+1).
Where each PC line is in 16D-space and how much it follows each of the original axis is called the "loadings."
Each principal component (PC) is a linear function of all 16 variable.
Since 2D pictures are the easiest to show, people often stop at PC1 and PC2.
Plot the original data or plot the race average of the original data. This is the 2D view of the 16-space that spreads the data out the most.
It gives you the best 2D view of how different the data points are (and ignores the 14 other principal components because they are less meaningful).

The loadings would tell you how the passing and casualty axis related to the lines of PC1 and PC2. Without the loadings, everyone is just guessing.
Looking at the data (and guessing), PC1 is actually probably primarily a mix of running and passing yardage. PCA can easily be biased by different scales on the axis. A bigger scale spreads the data out more and therefor appears more significant. 14 of the axis measurements often range 0-5 (TD, casualties, passes, etc.) with caused and sustained KOs occasionally slightly exceeding that range. Yardage will vary over a much larger range.
(I don't know the range of the occupation variables, I think it is 0-1, and I don't know the definition of "tackles." I assumed ball carrier knockdowns, but if it's total knockdowns then I would expect it to be viewed as more significant by PCA, again because of bigger range.)

Summary: PCA can give you a summery view to show you the spread of the data. Only by knowing the loadings, which we don't, do you know what "angle" you are looking at the 16D-space. The loadings tell you the function of the original axis in the data was found to be most significant. Without them, we don't know what PC1 and PC2 are.

[dode74]

I would have just wished they had done a little more by discussing the loadings of the observations.

There's a follow-up: http://www.nufflytics.com/post/bash-das ... ng-deeper/