Talk Fantasy Football

dode74 wrote:So, FUMBBL data:

Hey dode, could you make a graf like that, using the stunty leeg data?

That would be really cool.

Sure. Be very, very wary of the 3 races on the left (notwithstanding one of them is unknown!). So few games played isn't enough to say anything about them, really. I've therefore ignored them for assessment of correlation, and r=-0.09.

Yeah. "halfings" and "goblins" are in non-use and considered Grandfather teams

Thanks for the extra graphs and insight on FUMBBL and Stunty Leeg dode.

In terms of the the graphs that you linked to that were posted in 2015, the graphs are quite convincing, especially with reference to confidence intervals (nice!).

Looking at the table linked to the first graph though, the win percentages are based on numbers of matches that vary quite wildly between races. It would be nice to see a similar analysis where the number of matches was controlled i.e. each race had played the same amount of games when the win percentage was calculated. This is what I was trying to get at in my previous post talking about an idea for a simulation/experiment.

Do you have access to match level data that would allow for this analysis?

Best Wishes,

AndeeT

Actually, now that I think about it, the graphs posted in 2015 should have already taken into account the number of matches, as the confidence intervals will have been calculated by dividing by square root of the sample size (i.e. no. of matches).....hmmm, maybe what I am suggesting would just be another way of showing your original analysis in 2015 dode? (Although based around randomised sampling and ANOVA, rather than confidence intervals and total data; nonetheless, I guess it would show similar statistically significant differences?)

Best Wishes,

AndeeT

I was going to ask why you'd deliberately reduce your sample sizes

Well as long the confidence intervals are there and are taking into account number of matches (which I'm sure they do, looking at the length of the bars) reducing sample isn't a great thing definitely!

A slightly different angle on it would be the fact that it is randomly sampled and might control a bit more some of the bias' we have discussed in this thread. It's a bit more laboratory style, which depending on how you look at things, may or may not necessarily be any better .

Best Wishes,

AndeeT

Have been thinking a bit more around my suggestion of an 'experiment' and I guess there isn't really much worth in it, and it has, as you say, already been done; if it were truly random sampling (given a big enough sample), the sample should still technically mirror the total population, and thus mirror the graphs presented in 2015. I should think about these things a bit more before I dive in! Still, at least no one has gone to the pain of actually doing it.


Trying to think of ways to analyse data we have available to better dissect the trend we are seeing and I am struggling to find other ways of doing it...

I guess one problem that might be clouding things is the fact that we are looking at data from many different coaches, so coach experience/skill is always going to have a big effect. Though, even if we could look at individual coaches, I doubt anyone has accumulated enough data just by themselves with multiple teams to allow a fair comparison.

Thanks everyone for all the replies so far, I've enjoyed taking about BB data and learning a few things along the way.

Best Wishes,

AndeeT

I am happy with all this data showing that a lot of the work we did for CRP did what we tried to make it do. A few outliers yes ... but overall ... I'm pretty happy with those graphs.

Tom

Hi Tom,

Do you mind expanding on "work we did for CRP did what we tried to make it do" please?

Looking for the first time at the FUMBBl database that dode linked to - something that surprised me is the difference in home/away CAS. I didn't realise that home/away had any functional effect in the game, and personally, I don't pay enough attention to it for it to have a psychological effect on my blocking game...am I missing something here? Saying that, there doesnt seem to be a particular trend in one direction...maybe this is just regression to the mean CAS?

Best Wishes,

AndeeT

AndeeT wrote:Hi Tom,

Do you mind expanding on "work we did for CRP did what we tried to make it do" please?

Was talking about win % ... not CAS.

Tier 1 was 55% to 45%
Tier 1.5 was 50% to 40%
Tier 2 was 45% to 35%
Tier 3 was 35% to 30%

And while there are some minor outliers (and not necessarily the same ones from each data source) ... the vast majority of the 24 teams are performing in the bands we worked to modify the rules to get them to.

Hi Tom,

Sorry, my point about CAS was in reference to the FUMBBL data that dode had posted earlier in the thread http://fumbbldata.azurewebsites.net/stats2.html. Specifically the columns on home and away CAS.

I still can't see why home/away should influence CAS. I know that some things like 'throw a rock' on the kick off table are affected by FAME but I'm not sure if home/away is factored in to FAME, or any other part of the game for that matter.

In terms of CRP rules to get teams into the specified bands - I didn't realise that there was a conscious decision to stratify them in that way, I thought they just sort of 'ended up there'. Shows my naivety when it comes to games design!

Best Wishes

AndeeT

Home/away has virtually no effect. The only reason you might see bias in the NAF data is that when you use Score (and possibly other software), the higher-ranked team is usually Home. So while there is no in-game effect, there might be a knock-on one.

"Home cas/game" on FUMBBL simply means cas for that race. Away means cas against.

Tom - bear in mind the data I posted in this thread does not include CI ranges.

AndeeT wrote:In terms of CRP rules to get teams into the specified bands - I didn't realise that there was a conscious decision to stratify them in that way, I thought they just sort of 'ended up there'. Shows my naivety when it comes to games design!

Best Wishes

AndeeT

No worries ... but yes .. it was a very conscious design. Much discussed on this forum back in 2007 and 2008 when we were working to create the tiers (or bands as you called them).