NTBB2.0/2014

[VoodooMike]

Looking at Box+FOL and NAF stats, this is what I found

If you have multiple sources of data which, individually, are large enough to work with then you shouldn't combine them, you should look at them independently and use them to verify conclusions. If they're not large enough to work with on their own, and the environments are different... the results you get will be nonsensical.

An example from the thread you added your gut feelings to last week or so... using Box data we found a significant correlation between match outcome and games played. Using OCC data we found a *negative* relationship. Had we just combined the datasets we'd still have found a significant correlation. Could we have really concluded that games played is a significant predictor of match outcome then, just because we'd combined two datasets and then found one? By looking at them separately we found that the correlation IS environment specific, and thus, we know not to use it outside of that environment.

You'll agree that it's no good applying a straight short term nerf to the latter if they collapse in CRP after 20 games, for instance.

He has stated that he only cares about short term play. If you object to the idea of changes that only affect one small subset of the play that might happen in any environment then I wholeheartedly agree, but plasmoid won't since he doesn't care about what happens after 20 games <shrug>.

B is a pretty outlandish environment, part of the reason zons are so successful there is the vast amount of bash that love claw and lack tackle.

I'd say pretty much 100% of what makes amazons successful in TV-matched MM is that they can have a full roster of blodge very quickly and for almost no cost... and that, for them, represents "low development". Against other low development teams they are prison rape.

I first looked at all teams in 10(0)TV increments, making note of any team with a mean win% below 45 or above 55.

Ow ow ow... stop. Did you note, by any chance, the number of datapoints you had in each of these 10 TV increments? I'd bet my left testicle that if you ran an ANOVA on these little groups for any race you'd find next to no significant difference between them past the first handful of TVs. You're eyeballing again, under the guise of doing stats.

Apart from the non-mirror-match win-percentages, tier 1 is still tier 1 (45-55). I'd hope to put any nerfed team in the top half of that.

You're not talking about the BBRC's tiers anymore when you exclude mirror matches, if they included them. While I agree that they're just wasteful, comparing two figures that aren't created the same way, and saying "see, look... they're different" is sort'v a "no shit, sherlock" thing. Your Tier 1 is automatically different from the original Tier 1 because the ranges are calculated in a different manner... and if they're calculated in a different way, why are you sticking to the same ranges?

They are winning with ease at the moment because they have min maxed to the max. In a league setting this team would really struggle but in match maker environment they are slaughtering teams that are not min maxing.

Minmaxing close to 1500? Minmaxing, more and more, is defined as "doing better than I think they should" with little else as a qualifier. Why do you think they'd struggle in a league setting? That sort of team is hard to *create* in league settings, but I've yet to see someone using such a setup to poor effect in a league setting to support this often tossed around idea.

Let's be fair here Spub, people don't min-max to win in Box, they min-max to build and destroy teams/legends.

Man, I wish that when they were handing out superpowers I'd picked up long-distance mind reading, too. I can't help but note that the numbers seem to show the win% of minmaxed teams (or, at least, highly developed teams facing fairly undeveloped teams) being higher than usual for the roster in the given TV range, which sort'v suggests they're trying to win games (in that they ARE winning games).

To be fair I hated everyone of those teams you mentioned. I couldnt care less if they are a killer min max team or an elf one, bottom line is that is not how the game should be played and I personally think those coaches who are already good at the game should lead a better example. Its really a sorry state of affairs.

Thus spake Garion Starscraper on how Blood Bowl was intended when handed down on stone tablets from on high! Lets hear your views on gay marriage next!

All of this discussion about min maxing being an optimal strategy for this or that team, about such-and-such team needing a nerf or buff; it's all speculation. If you read the OP, the NTBB 2 is purportedly going to take, as a basis for its changes, a slice of the vast amounts of data that exist out there; not you waving your arms about which incarnations of particular teams you hate, or why CRP is a pile of crap.

You probably didn't get the memo, but people think that their data-unsupported opinions are on par with analysis, and they throw them into the ring whenever people discuss data to hope legitimacy rubs off on them. It's all the rage.

[Shteve0]

Mike, re: NTBB2 only being interested in short term leagues - not so.

IMO, there are 3 popular and distinct environments in BB. Tournament (low TV, a few skills, resurrection); TableTop (short term, season/trophy) and Survival (Long term/perpetual). I think it would be great if NTBB could take all 3 environments into account, making roster changes to teams that are overly strong/weak in one environment, as long as it would not push the same team outside of tier 1 in another environment.

This suggests equal weighting to short term, resurrection and perpetual, no?

[Hitonagashi]

Let's be fair here Spub, people don't min-max to win in Box, they min-max to build and destroy teams/legends.

Man, I wish that when they were handing out superpowers I'd picked up long-distance mind reading, too. I can't help but note that the numbers seem to show the win% of minmaxed teams (or, at least, highly developed teams facing fairly undeveloped teams) being higher than usual for the roster in the given TV range, which sort'v suggests they're trying to win games (in that they ARE winning games).

Of course they are trying to some extent to try and win. If there's proof that you are completely ignoring the game to try and kill/survive, the FUMBBL admins will ban-hammer your entire account. Therefore, the fact there are no min-maxers who ignore the ball entirely is an example of survival of the fittest.

Because they a) have the better teams, and b) have to try and win, then of course they will exhibit better than normal performance. Also, if you actually played Bloodbowl, you'd know the best way to keep players alive is to pressure and force the opposing coach to concentrate on defense as opposed to killing/fouling.

You can only observe the trivially obvious (that min-maxed teams win more/skill players higher than you would normally find).

I'm making an observation drawn out of games, that min-max coaches I have played are willing to sacrifice winning in order to keep key players alive. The fact that they don't tend to expose key players that often (usually after a quad push meaning an exposed blitz) means that they can marry a "keep alive" with a "win" objective the majority of times.

It's not 'mind reading' as much as it is 'playing'. You should try it sometime, it's a fun game .

[Shteve0]

The point is: What exactly is the problem with minmaxing?

If it's because it's beardy, then it will always be thus.

If it's because it's it skews matching by TV, that's an inefficiency of the format, not the roster. Teams that rely on inducements will never be effective in this format, but that, too, is a format limitation - and TV matching environments are outside the scope of this project.

If it's because the teams that habitually min max are too competitive - ie enjoy a disproportionately high win % in comparison to their tier - that should emerge in the overall win % for the format in the same way that an exceptionally good (efficient) and sufficiently proliferated strategy with any team would do so.

One would expect that the onus is on the cpmpetitors within that format to generate specialist answers to issues of its own creation (much in the same way that MTG decks do). If teams are naturally equipped to perform better, it's an issue that one expects to show in the stats; if it's becuase someone or some group is playing the meta, the meta will in time generate an answer. If the amazon roster garion described is so powerful (likely an innovation in counter to the CPOMB meta) then a new solution will arise in balance. Perhaps a Khorne team - 2 heralds (PO,MB), 10 Linemen (Tackle, with smatterings of PO and DP), or a Norse package with a similar skill array. With a degree of balance in the rosters, correctional solutions are always just around the corner, and the format will find them (or eliminate the problem roster through reform only if the win% is disproportionate and sustained).

So, if a team is enjoying a disproportionate win% in the format, it will receive attention through a stats basis either way; whether that team is often minmaxed or not is pretty much irrelevant.

Make sense?

[VoodooMike]

This suggests equal weighting to short term, resurrection and perpetual, no?

It does, yes. I missed that the goals of NTBB2 are different from NTBB. Plasmoid was quite clear that he only cared about very short term play in previous discussions of NTBB, but what you're quoting absolutely does suggest the aims have expanded.

Of course they are trying to some extent to try and win. If there's proof that you are completely ignoring the game to try and kill/survive, the FUMBBL admins will ban-hammer your entire account. Therefore, the fact there are no min-maxers who ignore the ball entirely is an example of survival of the fittest.

So we have evidence that they have a statistically significant higher win rate than teams with lower development at that TV range, but that's not why they do it, based on your anecdotal experience in the matter. Again, seems like you'd need to either be part of some secret minmaxing conspiracy, or have mind-reading abilities, to know the intentions of minmaxers, given the fact that the data only supports the increased win% aspect.

I'll point out something that is obvious to non-FUMBBL people, but apparently not so obvious to FUMBBL coaches.. not all Blood Bowl is played on FUMBBL. Unless you're suggesting that we should ONLY see a higher win% in Box, and not in Cyanide's matchmaking leagues, what FUMBBL admins will do to FUMMBL coaches is irrelevant as far as trends outside of FUMBBL go. And again, the game and environment exist outside of FUMBBL's B division.

Because they a) have the better teams, and b) have to try and win, then of course they will exhibit better than normal performance. Also, if you actually played Bloodbowl, you'd know the best way to keep players alive is to pressure and force the opposing coach to concentrate on defense as opposed to killing/fouling.

They don't HAVE to.. again, that's a FUMBBL thing. There is no punishment for minmaxing in Cyanide's matchmaking leagues, yet we still see the same trend. A team that is designed to supposedly be able to massacre lower teams... doesn't much need to "keep up the pressure" to avoid losing players they care about.

You can only observe the trivially obvious (that min-maxed teams win more/skill players higher than you would normally find).

Yes, you can only observe things that actually happen (or, as you call them "trivially obvious").. everything else would, I'd imagine, require ouija boards and mind reading. Luckily you can hook us up with that information!

I'm making an observation drawn out of games, that min-max coaches I have played are willing to sacrifice winning in order to keep key players alive. The fact that they don't tend to expose key players that often (usually after a quad push meaning an exposed blitz) means that they can marry a "keep alive" with a "win" objective the majority of times.

So, a long-winded way of saying "anecdote stated as fact" then? Got it. That said, MM data in general (Box and FOL) demonstrate that a huge proportion of coaches, minmaxing or not, focus on keeping players alive over winning the match. I'm not sure what kind of half-cocked logic you're using to marry the two concepts in such a way as to support your anecdotes.

It's not 'mind reading' as much as it is 'playing'. You should try it sometime, it's a fun game

I always find it amazing that FUMBBL people so often can't wrap their heads around the fact that people need not play on FUMBBL to play Blood Bowl. I play table top and online using Cyanide's LE. I'm not interested in playing on FUMBBL because I don't much enjoy the client, and I can't guarantee 90-120 minutes of my time without having to leave the computer for extended periods of time (which is required, by the site rules, to engage in matches there). I also find a lot of the FUMBBL community to be pretensious, reactionary douchebags that I don't much want to associate or interact with.

No offense, of course, to you or Garion...

[Shteve0]

Play nice, guys.

My core point, digested: If a minmax team is hitting a high win% after 50+ games, that will show in stat analysis and will (purportedly) be adressed by the OP goals. If it's not seeing an excessive win% in the format's defining range, I genuinely don't see what the issue is (ie the meta will readjust itself in time).

[garion]

unlucky plasmoid. Another thread ruined by the troll king.

[garion]

Garion mate

All of this discussion about min maxing being an optimal strategy for this or that team, about such-and-such team needing a nerf or buff; it's all speculation. If you read the OP, the NTBB 2 is purportedly going to take, as a basis for its changes, a slice of the vast amounts of data that exist out there; not you waving your arms about which incarnations of particular teams you hate, or why CRP is a pile of crap.

I agree with you, that's exactly my point. Plasmoid said he is looking at data from Box and match maker to make changes to teams. All I am saying is those environments are not a good basis for making changes as they are exploiting flaws in the rules to gain better win percentages. In league keeping your TV down will not achieve any good in terms of winning ability, i'm sure I have seen data numerous times from dode and voodoo mike that says the higher TV teams have a better chance of winning a match than a lower TV team, i cant remember the figures but iirc it was roughly 70%. But in box it keeping your TV down and min maxing will help.

Saying zons and lizardmen need a nerf based on that data is what all the above post from my self hito and Spubbba were about. Bottom line is I have never seen a zon team win any majors or leagues. Same for Lizardmen in a perpetual environment. I'm sure they probably have somewhere but making nerfs based on Box data is going in the wrong direction. It is just far too open to exploitation of the rules. And yes Voodoo mike you might not like me saying it, but Galak has said many many times Match Maker environments are not what the rules were written for.

I hope Plasmoid can get the data he needs from league. Personally I don't think it exists. But good luck to him.

[spubbbba]

I also find a lot of the FUMBBL community to be pretensious, reactionary douchebags that I don't much want to associate or interact with.

No offense, of course, to you or Garion...

Hey! I object to that, you forgot to include me.


Garion, I believe Zons won a very early FUMBBL major and I’m pretty sure they won the WIL once. Brinoth’s Lizardmen are probably still WIL’s most successful team, but those were all under lrb4.

Large, high TV tournaments are unique to FUMBBL as far as I’m aware, so not sure how much use they are. Being able to spend as long as you want pimping a team until they are just right gives some coaches quite an advantage in the R majors, the same with B except you don’t get to choose your opponent but won’t have to spend as long hunting for opponents so can play more.
There’s the added problem that a lot of the unsuccessful races tend to not have many entrants in the 1st place compared to the likes of Orcs, Chaos and the elves.

I’m sure some of our more statistically minded members could factor that in and compile stats though.

[dode74]

i'm sure I have seen data numerous times from dode and voodoo mike that says the higher TV teams have a better chance of winning a match than a lower TV team, i cant remember the figures but iirc it was roughly 70%.

It really isn't that simple. There is a correlation between having a higher TV team and winning, but there are also all sorts of other correlations there too. What was actually seen was that as TV difference increased the higher TV team had a higher win% than the lower TV team. There are better indicators though: perhaps Mike will post a link to his blogs about it.

But in box it keeping your TV down and min maxing will help.

Depends on what you mean by minmaxing. Certainly there is a correlation between having a higher number of games played and winning, but that correlation is weaker than for TV. What noobhunters (those who stay at very low TV while developing teams: Pact in particular are good at this while Zons could be said to be designed for it) do is increase development while reducing TV, meaning that they do tend to win more. It just doesn't happen that often though. Of the FUMBBL data I looked at <=1250TV, <6% of all Pact (the worst offenders) games were played where the Pact team had 25+ games and the opponent had 5 or less (i.e. noob-hunting), and for Amazons it was about 3%, which is actually about the same as the mean for all teams. In short, it's not massively prevalent and makes little difference to the metagame.
Minmaxing itself (as opposed to noobhunting) is just a strategy: you get the most for the cost. It happens in every competitive game, so why you'd complain about sound strategy I'm unsure.

Bottom line is I have never seen a zon team win any majors or leagues. Same for Lizardmen in a perpetual environment.

I can't offer anything on Zons (and would agree they are a dull roster and need a rewrite) but Lizards are the second most successful team in terms of winning the championship in my league after Wood Elves. WEs won 7 times (5 of those by Flix, who is an exceptional player) with 2 different coaches; Lizards won 4 times with 3 different coaches (iirc). No other race has won more than twice.

It is just far too open to exploitation of the rules.

It's not the rules being exploited, it's the matching system.

[garion]

It's not the rules being exploited, it's the matching system.

Okay fair point, and I would love to see blackbox or Match Maker work in a different way, but the point remains the same. Using data from such house ruled environments, places where teams are not built in the typical way, and making changes based on that data is not a good idea imo. How teams perform in leagues and how they perform in BB/MM is a very different kettle of fish, and if plasmoid make a change based on MM data who is to say what kind of influence it will have on the same race in a league, both short and/or long term?

The other problem is there are already a number of rule changes in place in his NTTB rules, so who knows how they will effect how well certain teams perform. For instance is there actually any data that shows CPOMB teams are the best at winning? Yet plasmoid has made changes that he 'feels' will make the game more balanced. He has also made a change to sneaky git which makes Elves good at fouling... Do elves need this boost? Most data I have looked at puts elves amongst the best teams at winning.

Like it or not making wholesale changes to team rosters and the rules is ultimately suck it and see. So you can look at all the data in the world. As soon as a couple of big rule changes are made you just don't know how it will affect any of the races win percentages, their bash ability and so on. Any changes made to rosters there on are just speculative...

[dode74]

I've got absolutely no problem with any rules changes which anyone wants to apply as house rules. The issue occurs when claiming that they do something to the tiers, which are statistical entities. This goes back to the "stats" thread in this sub-forum, so I'd be happy to continue this there rather than here.

[VoodooMike]

My core point, digested: If a minmax team is hitting a high win% after 50+ games, that will show in stat analysis and will (purportedly) be adressed by the OP goals. If it's not seeing an excessive win% in the format's defining range, I genuinely don't see what the issue is (ie the meta will readjust itself in time).

The idea has already been examined in a thread over on the Cyanide forums... starting at about this post, meaning that's when numbers actually start getting used. The summary is that in lower TV ranges teams with a higher number of games played have a higher win% (statistically significant) against teams with lower games played.... as redefined many times during that thread, but an example would be... < 1200 TV with 10 or fewer games played versus same TV range with 20 or more games played.

All I am saying is those environments are not a good basis for making changes as they are exploiting flaws in the rules to gain better win percentages.

Using your almighty gut, why don't you give us an estimate of the % of matches you think involved minmaxing in, say, the 1200 or lower range, in box? Or, really, pick a range and give a percentage. One would expect it to be a very HIGH percentage if we're going to declare that the environment's data is somehow unusable in terms of looking at the game in general, right? Would it surprise you to hear that the % of games that *potentially* involve minmaxing, as defined by the above criteria, is somewhere between 0.8% and 2.8%? That % will grow as you raise the TV limit toward the mean TV for box as a whole, but that's because it becomes harder and harder to isolate high development in a supposedly inappropriate TV range, when you're close to the TV that most teams of ANY development level tend to hang out at. Minmaxing appears to be a primarily perceptual issue, not one that is actually ruining the game.

And yes Voodoo mike you might not like me saying it, but Galak has said many many times Match Maker environments are not what the rules were written for.

Yeah and I don't see why anything he says particularly matters to any of this. His statements don't change the way the game IS being played, nor should they... and the rules he helped create contain plenty of issues that even his biggest fans can point out. The general community has access to far, far more data than the BBRC ever did, and we have people who are better at working with the numbers than they were. So, while it's swell that he was involved in the creation of the ruleset that doesn't mean things need, or should, stop there.

Hey! I object to that, you forgot to include me.

I've never had any issues with you. I don't find everybody on FUMBBL to be that way, just a vocal majority that reminds me of the republican party, with all their progressive ideas about science, change, and doctrine. I see that half of my "no offense" statement was removed, such that it looks like I was actually telling them not to take offense. Funny!

Certainly there is a correlation between having a higher number of games played and winning, but that correlation is weaker than for TV.

Remember that this is not actually true outside of Box data. OCC showed a negative correlation between games played and match outcome. Using Box data we're already controlling for TV difference to a great extent, which is why TV difference shows a significant, but fairly weak correlation with match outcome, and games played shows up as a positive predictor. When that isn't controlled for, as in a league setting like OCC, we're not seeing it as a helpful predictor at all.

There's a good chance that the issue is this... games played IS a predictor, but it faces a massive ceiling effect. There's only so developed a team can get. The more likely coaches are to stick to a single team, the worse games played becomes as a predictor. Box has the "benefit" of being chock full 'o throw-away teams, with the median games played being all of 5. Even then games played isn't a great predictor... so we can sort'v assume games played is, as a whole, not worthwhile.

Using data from such house ruled environments, places where teams are not built in the typical way, and making changes based on that data is not a good idea imo.

I think you mean "exclusively". If the intention is to adjust the game such that it works in all the major environments where it is being played then you'd want to use data from all those environments in order to decide on what changes should be made to deal with the issues that crop up in each of the environments.

How teams perform in leagues and how they perform in BB/MM is a very different kettle of fish, and if plasmoid make a change based on MM data who is to say what kind of influence it will have on the same race in a league, both short and/or long term?

Making sure that changes meant to address and issue in one environment don't break play in another is tricky stuff, which is why such a project will be incredibly difficult. Not impossible, just difficult. People declare difficult things to be impossible because they're too vain to admit they're not smart enough to accomplish it. Whether or not plasmoid can, remains to be seen!

For instance is there actually any data that shows CPOMB teams are the best at winning? Yet plasmoid has made changes that he 'feels' will make the game more balanced. He has also made a change to sneaky git which makes Elves good at fouling... Do elves need this boost? Most data I have looked at puts elves amongst the best teams at winning.

This is an aspect that also causes me concern with NTBB and its previous claims of being data-driven. I've never seen any numbers that suggest CPOMB is giving teams a significant win advantage but lots of people declaring that it is armageddon, in the same way minmaxing represents a trivial amount of play, but people declare IT to be armageddon too. I'm quite opposed to making changes based on people's misperceptions and mistaken beliefs.

Like it or not making wholesale changes to team rosters and the rules is ultimately suck it and see. So you can look at all the data in the world. As soon as a couple of big rule changes are made you just don't know how it will affect any of the races win percentages, their bash ability and so on. Any changes made to rosters there on are just speculative...

It's not a boolean thing. There's a pretty major difference between making a change that can be justified by actual data, and making a random change based on using the force. Certainly it is impossible to predict outcomes perfectly, but that doesn't mean that all methods for deciding on changes or courses of action, are equal... or even close to it.

[dode74]

Remember that this is not actually true outside of Box data.

Yep, but we were talking specifically about minmaxing in box, as per the quote from garion I posted. I do agree with your assessment on the reasons why, particularly with regards to the "ceiling effect" being the reason why this doesn't happen in OCC. After all, while the median GP in Box is 5, the minimum for a season in OCC is 9.

[Hitonagashi]

I see that half of my "no offense" statement was removed, such that it looks like I was actually telling them not to take offense. Funny!

Your subtlety is only matched by your netiquette .

Regardless, we are thread-jacking again. I would like to hammer out our differences, because you are clearly intelligent (and most of the time, correct), but this thread is not the time or place for it. I'm also pretty sure that even offering to discuss will be met out of hand with a scornful insult, probably comparing talking to me to negotiating with a terrorist, but hey, I offered.

Looking over my original post, I was wrong *shrug*. It happens, I'm not called VoodooMike, I occasionally/frequently make mistakes . I (and I imagine Garion) don't spend hours carefully crafting the ideal retort/response. The line I missed was:

The CRP+ rules adress the long term anyway, in an attempt to change the metagame, while the NTBB roster changes are focused on the shorter term.

If you are only looking at short term, cool! You could be right about CRP+ being a long term boost which negates the short term nerf. The games I mostly care about are high (1600+) TV, so the team charts look different at that size. I'd like to see NTBB attempt to narrow the tiers at all ranges, rather than just short term, but that's a much harder proposition.