VoodooMike
Joined: Nov 07, 2010
|
 Posted:
Sep 26, 2011 - 13:35 |
 
|
In order to crunch some numbers for the Analysing Data thread I pulled out the most recent 25,000 or so matches from the FUMBBL database. Since I still had all that data I thought I'd take a look at some other relationships, and I found something that (I think, anyway) is strange, or at least surprising.
GalakStarscraper has stated that inducements were designed to allow the underdog to win 35% of the time. Specifically, he said that if inducements bridged the gap between teams of different TVs, it would discourage team building. Based on that, I decided to look at how often the TV underdog wins.
The flat number of underdog victories in the data was 40.3% which, at first glance, seems about right according to the stated design. However, if we remove tied games and then check how often the underdog wins a game that is won at all, this figure goes up to 49.9% meaning that when the games are won, the TV underdog has the approximately the same chance of winning as the team with the higher TV.
(Total Games: 25610, Higher TV Wins: 10391, Underdog Wins: 10339, Draws: 4880)
Now, this is already a strange result, but to examine the effect that inducements might be having on this we need to look at the relationship between the difference in TVs between the teams, and the likelihood that the underdog will win. If inducements are weak compared to their cost, we should see a decently strong negative relationship between this (meaning that as the gap between TVs goes up, the chance of the underdog being the victor should go down)...
...but we don't. Instead we see a weak (but statistically significant) POSITIVE correlation between the difference in TV and the likelihood that the underdog will win (r = 0.02, p < 0.05, N = 20730) which suggests that inducements are, contrary to Galak's stated design intention, worth slightly more than their TV equivalent. If we control for the difference in ratings of the two teams, we end up improving the strength of the correlation between TV difference and victory likelihood (r = 0.05, p < 0.01, N = 20730).
Now, ordinarily we might say that the correlation is so small as to dismiss its practical significance regardless of its statistical significance (if you use a large enough sample size, you're far more likely to find SOME effect, regardless of how small) but given that we know what the correlation *should* be based on the designer's statements (and thus, what everything has previously been based around) it is a pretty significant result.
So, my question is this - is FUMBBL's implementation of inducements significantly different than the ones listed in CRP? If not, then inducements might not have their intended effect on games in an open setting, as most games played here are. Alternately, I may be missing a possible explanation for the data that someone else can suggest - inducements are simply the only thing I can think of that explain the difference in likelihood of the TV underdog winning a game as the TV difference increases. |
|
|
Garion
Joined: Aug 19, 2009
|
| Posted:
Sep 26, 2011 - 13:39 |
|
If you can look at all the games wizards were taken by the underdog and see how that has effected the win percentage that will be quite interesting, also Babes if possible, they imo are far and away the best inducements.
Edit: and no fumbbls implementation of inducments is no different really, just mecs are missing now I think, oh and also some of the secret weapons.
edit edit: you will also have to remove the games where the TV difference is less than 50 because that means no inducements were taken. |
_________________
 |
|
Reisender
Joined: Sep 29, 2007
|
| Posted:
Sep 26, 2011 - 13:46 |
|
awesome voodoo....
now the weird feeling of mine that inducements were to powerful has some statistical backup.
i think there maybe tresholds of winning probability at 50k (1 babe, 40k is just nothing) 150k (wizzie) and maybe some other levels of good inducements (skitter, eldril, etc.) |
|
|
freak_in_a_frock
Joined: Aug 02, 2003
|
| Posted:
Sep 26, 2011 - 14:02 |
|
It is going to be very hard to prove one way or another whether the statistics mean anything. For example, how many of these games would have been won if the inducements hadn't been taken? How many of the losses were due to the failure of an inducement (bribe, wizard or chefs for example). It is interesting, but to be honest not that surprising. |
|
|
Hitonagashi
Joined: Apr 09, 2006
|
| Posted:
Sep 26, 2011 - 14:05 |
|
You beat me to it!
I was going to write a blog post on this effect tonight, as a counterpoint to Purplegoo's blog.
To me, it's because an inducement is customisable TV. If I play 3 games at 100 TV down with elves, against a killer chaos team, a norse team and an elf team, I can pick between an apo for the chaos, a bribe for the norse, or an extra rr/babes against other elves. You can match it to what your team needs.
In addition, in some cases they are even more devastating, as they allow your team to be built in such a manner that you couldn't normally do it.
The obvious example is Eldril, giving an elf team a gaze they *cannot* get otherwise, but an even more powerful example is Silibili/Morg for a lizardman team. The balance of the lizardmen is inherit in the fact they are limited to 6 saurii and a kroxigor. If you allow them to go to 7 saurii (using sili), or even 7 saurii and 2 kroxigors (using sili and morg), they are just so strong that only a dedicated killer chaos team can match them for strength wars.
In my experience, 50-100k is a slight advantage (a babe over 2 skills is usually a good trade off), 100k-150k is a pain (no wizard, still no real improvement over 2 babes), and 200k + gets really fun as you can customise what you take to match your opponent. |
_________________
http://www.calculateyour.tv - an easy way to work out specific team builds.
 |
|
Juff
Joined: May 24, 2006
|
| Posted:
Sep 26, 2011 - 14:10 |
|
Problem with this, is that it's assuming that all equal- TV teams are equal, if you get what I mean.
Let's say a minmaxed TV1250 team plays a poorly-built TV1320 team. The "underdog" is already a favourite here, *before* it adds a babe or whatever.
I would suggest that such matchups are common enough to skew the results as you're seeing. |
|
|
Garion
Joined: Aug 19, 2009
|
| Posted:
Sep 26, 2011 - 14:14 |
|
| Juff wrote: | Problem with this, is that it's assuming that all equal- TV teams are equal, if you get what I mean.
Let's say a minmaxed TV1250 team plays a poorly-built TV1320 team. The "underdog" is already a favourite here, *before* it adds a babe or whatever.
I would suggest that such matchups are common enough to skew the results as you're seeing. |
Yes but it is still a huge sample of games, so while some teams may have min maxed and the other was badly designed etc.. the size of the sample should be enough to make the anomlies have less of an impact on the stats. No stats are ever perfect but it is still interesting and confirms what many already believed to be true. There is also a racial factor to take into a account as well like dwarves vs zons etc.. but again that should be swalloed up by the size of the sample. |
_________________
|
|
koadah
Joined: Mar 30, 2005
|
| Posted:
Sep 26, 2011 - 14:30 |
|
Where did you get all those games and what do you consider and under dog?
I'm only looking at the Box so due to TV matching I couldn't find many games at all.
I didn't bother looking at games with a gap less than 200.
I don't have the stats here but I'm pretty sure that at
a 200+ gap the win% was around 40% but with 300+ gap the win% dropped to 34% in line with Galaks expectations.
As the gap grew the underdogs win% fell. (Yes it did Goo!  ) |
_________________
O[L]C 2016 Swiss! - 19th June! ---- All Star Bowl XII - Teams of Stars - Sign up NOW! |
|
uuni
Joined: Mar 12, 2010
|
| Posted:
Sep 26, 2011 - 14:30 |
|
| VoodooMike wrote: | | So, my question is this - is FUMBBL's implementation of inducements significantly different than the ones listed in CRP? |
I think your question is valid and on point.
My feeling is that FUMBBL's decision to limit the pairing of the teams to 15% TV difference may be much stricter than the game generally was designed to.
The inducements have a lower limit of 50k, and I suppose the sweet spot for the most effect gained by inducements would be somewhere around 200k-400k, where you many times can use your inducements to the penny, if you so wish. In FUMBBL, inducements of 200k would require the overdog to be of 1540k TV and 400k inducements would require the overdog to be of 3070k TV. Most games are played with lower overdog TV than this.
My understanding is that the inducements were designed to much wider team imparities than are allowed on FUMBBL.
Also worth of note in this calculation is that the 15% is a hard limit, not an average. In box the scheduler tries to find pairings near TV parity and I suppose that is also what the people in R do (I guess there are also some metarules in R regarding this). This means that the TV differences are normally much smaller than this - probably they are normal distributed around the parity.
***
What inducements are available in the rules and are they available currently on FUMBBL:
| Code: | A Bloodweiser Babes (0-2) - 50,000 gold pieces
A Bribes (0-3) - 100,000 gold pieces
A Extra Team Training (0-4) - 100,000 gold pieces
A Halfling Master Chef (0-1) - 300,000 gold pieces
A Igor (0-1) - 100,000 gold pieces
A Star Players (0-2) - Various prices
A Wandering Apothecaries (0-2) - 100,000 gold pieces
P Wizards (0-1) - 150,000 gold pieces
N Mercenaries (Unlimited) - Various prices
N Special Play Cards - Various prices |
A-Available, P-Partially, N-Not available |
Last edited by uuni on %b %26, %2011 - %14:%Sep; edited 4 times in total |
|
DukeTyrion
Joined: Feb 18, 2004
|
| Posted:
Sep 26, 2011 - 14:32 |
|
I would like to see the statistical differences in brackets.
10TV to 40TV difference (No Ind)
50TV to 90TV difference (Babe)
100TV to 140TV difference (Babes/Bribe/Apo/Star)
150TV to 240TV difference (Wizard Territory)
250TV plus difference (Big Star Players kick in) |
|
|
Irgy
Joined: Feb 21, 2007
|
| Posted:
Sep 26, 2011 - 15:18 |
|
The way I see it, inducements, with some exceptions, are generally worth less than their cost. However, what makes up for it is choice. The higher TV team is stuck with whatever team make-up they've got. The lower valued team on the other hand can choose something which, while maybe slightly overpriced in theory, is well suited to the match-up.
I find with the new rules I don't care so much for keeping my players alive for just that reason. I'm much happier to use a strategic apo on a badly hurt player rather than save it purely for (sometimes) keeping players alive.
On the whole your statistics suggest that inducements are better than they're intended to be, but at least they haven't overshot completely by making it better for the underdog. I don't think it's such a bad thing that both teams have a fair shot. |
|
|
nin
Joined: May 27, 2005
|
| Posted:
Sep 26, 2011 - 15:25 |
|
-Match reports include the actual inducements, if you have access to that, some extra analysys could be done.
-Nowadays I don't have any problems with R, but I'm a bit suspicious of data from that div when talking about statistics...any chance of division specific analysys or that would make the samples too thin?
The brackets thing that Duke Tyrion proposes looks good too, that 35% of Galak was a very loose statement. Does it refer to "allways", "overal" or what? A game bellow 150TV diference feels diferent from one over that.
(Just in case, statistics and the way I experience competition and games are different subjects) |
|
|
Reisender
Joined: Sep 29, 2007
|
| Posted:
Sep 26, 2011 - 15:30 |
|
160k can be
7 4 4 8 Block Dodge Dark Elf
A Wizard
6 3 3 8 Block, Tackle, ClawPOMB Beast
3 Zombies and a Snotling
A Rat Ogre...
More Bloodweiser Babes than legally allowed for one team
---
This is just random nonsense but i should have made a poll...
----
80000 Pies 2 gk each |
|
|
WhatBall
Joined: Aug 21, 2008
|
| Posted:
Sep 26, 2011 - 15:34 |
|
I think in Ranked that there may be a trend to cherry pick upwardly against weaker coaches with an ideal match up, which may skew numbers. As mentioned, you can customize your inducements to the opponent's team, and if they are a weaker coach and/or the match up favours the underdog race, then those percentages are not surprising. |
_________________
 |
|
clarkin
Joined: Oct 15, 2007
|
| Posted:
Sep 26, 2011 - 15:45 |
|
| nin wrote: | | -Match reports include the actual inducements, if you have access to that, some extra analysys could be done. |
Wizard isn't listed in match reports (a bug, I guess) |
|
|
|
|
| |
Borg Invasion
Search
