Team ratings...onto OT football/soccer discussion

[Zombie]

Have you even cared to try it? It's actually easier to use than the current system (less calculations involved) and a lot more accurate. I agree that it doesn't say anything about the value of a goblin team compared to that of a dark elf team, but no system could.

[Ghost of Pariah]

Exactly how is it more accurate for predicting a winner?
I've tried it. I didn't find it any easier to use or any more accurate.

[Mestari]

Pariah:

To begin with, we need to define two things. Team rating and coach rating. The current coach rating system is a self-correcting one: when you win, it goes up, when you lose, it goes down, converging towards the rating which you should have in your current playing environment (i.e. the other coaches). The TR system is not good, as I've noted before, for this purpose, but it'll do.
These two make the calculations possible but are also sources of error: If someone has CR of 150, but should be something like 200 the calculations do not give the right percentages - but that's just an extreme case - statistically our CR150 coaches should win about as many games as the formula suggests. Those wins (as they are derived from statistical results) include the probability of a better-than-150-coach winning the game, as there are bound to be such cases in the original data to change the odds to take this into account.

As winning is defined clearly to have more Touchdowns than the other team:
A simple example:
Over a large amount of data, we've noticed that TR200 teams score on the average (regardless of the opponents TR) per game:
0 TD 18%
1 TD 22%
2 TD 28%
3 TD 22%
4 TD 8%
5 TD 2%
And TR 100 teams score on the average per game:
0 TD 30%
1 TD 40%
2 TD 25%
3 TD 4%
4 TD 1%

Now, assuming equal coaching skill, we can easily see that the TR100 team wins 20,82% of the games, loses 57,02% and there is a draw 22,16% of the games.
THIS WAS A SIMPLIFIED EXAMPLE ONLY - Don't send messages saying "that's not true - my TR200 team scores a lot more than that"...

Now we now that the TD percentages above depend to some extent on the coaches, so we need to fudge in the coach rating. And IMO NAF system does that pretty good.

Also I don't think Neo is claiming that the system NAF has is perfect. I rather think NAF will closely examine all the game data they get to improve the system and to make it more closely approximate the true probabilities of the game.

Personally I'd like to have the probability of a draw included (as in my example above) for the league matches where there can be a draw as an optional feature (two different formulas - other of which has three results coming out from it)

Anyhow the main point is:

With enough of data to back our statistical model up: we
-don't have to care about kick-offs and other lucky rolls - they're included
-don't have to care about inaccurate coach ratings: they are self-correcting and their effect is included in the statistics
-have to accept the fact that the numbers are probably not 100% accurate
-but still we can say that they model the real world rather accurately.


Zombie: I'll go and check your system out - I'll get back to you when I've thought it over.

[Zombie]

Well, first, the current TR system doesn't take nigglings and stat decreases into account, while mine does. This makes my system more accurate to express the strength of a given team.

Another important failure of the current TR system is that it puts as much value on a 7-skill player and ten 0-skill player than it does on eight 2-skill players and three 1-skill player, which is clearly inaccurate. My system corrects that as well.

The last point is that money in the treasury doesn't translate into how strong your team is on the field, so i simply didn't include it in my system.

If the system can better guess the strength of the team presented on the playing field, then clearly it can also make a better guess as to which team has the best chance of winning, and how big the gap is between two given teams.

[Ghost of Pariah]

Well that's great for a one off game but in a league setting it's just as inaccurate.
In a league setting money is important to a team's strength. In the example I used earlier with chaos and elves money could be an important factor in helping the elf team "recover" from it's "victory".
All yoursystem does is try to predict the winner of an individual game which is pointless and impossible. It doesn't give any indication of a team's overall strength in the league.

[Zombie]

Hey, i never said it's good at it. All i said was that it's better at it. You could at least acknowledge that.

[Ghost of Pariah]

I don't understand.
I think the current system is good. I don't think yours is good. How should I think that it's better?

[Zombie]

All right, then tell me how the current system is better. I've given you my arguments, but all you've given me is the finger. At least explain what you think is wrong with it.

[Mestari]

Pariah thinks that potential should be included in the TR. That's what money is, for example: potential that you can actuate earliest to the next game. Same goes for MNG-players: they're there, but not now - still they should be counted.

Whereas what Zombie and I myself aim for is a system for rating the teams current situation, its ability to compete in its current state.

These are opposite philosophies and choice is a matter of personal preference. Looking things from Pariah's side, I can easily see why he thinks the current system is better. Still, I have hoped that the majority would prefer TR to be an indicator of the teams current level.

The points that support our TR philosophy:
-the way NAF rating system is built would indicate a desire for TR's to be as accurate descriptions as possible about the teams ability to win a single game.
-"League-setting" that Pariah referred to is a rather blurry concept, whereas a single game is a well-defined environment where TR can be used as a rather accurate indicator should we develop the system with this aim in mind.
-Moreover, should we desire making TR system a better indicator of the teams strength in a "league-setting", we should be able to account for other potential apart from the obvious ones (gate, missing players) in our formula.

[Ghost of Pariah]

The reason I don't support a system that attempts to rate a team's current state is because it's impossible. There are so many variables involved in winning a game that any kind of Team Strength Indicator would need to be impossibly complicated to be accurate. Then if it somehow was accurate what would be the point in playing the game?

What is the point of predicting who is going to win?

[Dragoonkin]

Then you could just give it to Anthony and he could make his Championship Manager game-thing.

[Zombie]

What does the current TR system attempt to do? It's an indicator to the strength of the team, and who has the best chance to win (and then it tries to equalize that through handicap).

My system does exactly the same, only better. If you hate any system that attempts to do that, then you must hate the current TR system as well.

[Mestari]

Then if it somehow was accurate what would be the point in playing the game?

What is the point of predicting who is going to win?

Is there a point in rolling a 4+ on d6? Yes. The fact that we know there is a 50% chance of failing doesn't change that.

If you roll three 6+'s in a row, I predict that you have a 99,54% chance of failing. You can still try, and if you succeed, you could boast against how big odds you managed to succeed against.

It's the same with playing a game. Even if we know that statistically the other coach should win 99,99% of the time, there's a point in trying to beat him, because you still can! And if you do, you can say that you had a snowballs chance in Hell to win that game and still you did!

[Ghost of Pariah]

I'm talking about the TR, Zombie, not the handicap system. The Handicap system doesn't work but not only because it tries to use team rating.

The current Team Rating gives you a general idea of how well your team is doing in the league. The handicap system tries to determine who will win the match by using TR and then tries to even the odds. It isnt successful and neither is your system.
Like I said there is no system that can predict such a thing. Nor should there be.

Mestari,
If the system handicap system you come up with can predict a winner with 50% accuracy then I say flip a coin instead doing all your calculations.

If it can predict it with 99% accuracy then I say wow you must be omnipotent and then I would still rather just do the 5 minute calculation, write down my winnings and play the next calculation.

Neither one of you has told me what a WIN is. Do the Elfs who score 2 more TD's win over the Chaos team who killed or injured 6 players? If so, then do the Elfs still win if their next game is against a developed Undead squad?

[Zombie]

The current team rating has one and only one use: to determine the handicap. You can say otherwise, but you're only fooling yourself.

There's a difference between the chance of winning and the accuracy of the prediction. If a system could predict with 95% accuracy that a team has a 60% chance of winning, that does say something very interesting, but it still makes for an interesting game, because the team still has a 40% chance of winning. I don't think you understand the basic principles behind statistics.