I am a stats noob, and would be delighted to get enlighted.

If you don't know what to do, use T-test.

You can do very well, if you make an assumption, that the games are not based on luck, just on skill. Such model has a huge advantage of the ability to predict many matches; if player A beats player B and player B beats player C, you can already expect to know the result of A versus C. The other games can be assumed to be 50:50 (which can then be used to get predictions).

Edited 3/18/2022 13:58:10

MH makes subs


Template Strategic Greece 1vs1

out Swisster https://www.warzone.com/Profile?p=28867016

in LeQuébécois_Benoit https://www.warzone.com/Profile?p=2213363356


Template Great Lakes 1vs1

out knaan https://www.warzone.com/Profile?p=1338486456

in Corsaro93 https://www.warzone.com/Profile?p=49119746138

Linberson, to answer your questions:

This model is now already two weeks out of date, so it is not very relevant at the moment. I will compute new numbers soon.

Basically, i chose this method because
1) The method is quite easy to understand
2) Even so early in the season, patterns are already clearly visible
3) The model most likely overestimates the worst performing clans and underestimates the best performing clans. This has the advantage that at least the ranking is correct.

If you are going to play around with other percentages than 50, you will quickly conclude that this problem asks for ELO. Only, it is quite hard to implement:
1) Clans have not a uniform rating. In some tourneys, some clans will perform way better. This causes issues when giving a single clan a rating as tourneys with many games finished will influence the tourneys with many games to go a lot.
2) Giving an ELO rating to each tournament is the next option, but the limited amount of games, the inclusion of subs, and the influence a single decision can have on this model (some games are decided by one 50/50 predict), it would just be a mess.
3) Using ELO from previous seasons is an even bigger mess. Teams change, opponents change, motivation changes, some clans have an entire different roster, some teplates have never been played competively etc. It is just a mess.

> the outcome of the model are probabilities, in the case of discovery (who are leading right now) the answer this model gives is then the following: if they win 50% of the games from now on, then they will have probabilty to promote of ~54% after redoing this fo 1.000.000 times.
Not quite. Each game they get a 50/50 win chance. This is different than winning 50% of the games. For example, if they play 4 games, they now have a 0.5^4 = 0.0625 chance to win all four.

> the model should favour total points at this point, right? we see this with Blitz
Yes, Blitz is leading in this graph due to their points. A 4-1 record is way better than 1-0 in such a probabilistic model.

> are there any other models that could be used, and if so why did you chose this way?
This is already quite an obscure problem, and I think that sport's betting is about the closest equivalent to doing this. For most team sports, ELO is used (together with obscure stats as home advantage etc.) Only, there the bookees have a few more advantages:
1) Much more results are known and teams don't change as often.
2) Most of these sports are about one result while here, we have to combine many of them while the competition is ongoing. I know that for example in tennis, you can bet on single points, but there a lot of data is known.
I am just an amateur with a computer and maybe by sieving through old CL data I can match model and outcome much better, but it is barely worth the effort.

Edited 3/19/2022 13:38:04

Thank you for satisfying my curiosity!

Not quite. Each game they get a 50/50 win chance. This is different than winning 50% of the games. For example, if they play 4 games, they now have a 0.5^4 = 0.0625 chance to win all four.

Ah, yes. I understood this, but my summary was wrong.

I am just an amateur with a computer and maybe by sieving through old CL data I can match model and outcome much better, but it is barely worth the effort.

No pressure from me, I was just curious.


Tbh my takewaway is that due to individual tournaments and different players and subs this is hard to predict. Out of my head I can only think of some boxing or wrestling leagues that are structured similarly, where you have the different weight divisions but count wins and losses together in order for some club to win the championship.

The best ways to do this with so limited data would be

* using frequentist statistics: fit a logistic model with as outcome win (1) or loss (0) for "the home team", code the home team as "1" and the away team as "-1" and all other clans as 0 for each game. You can predict the probability of a win for any other game now by coding the two opponents as a home team and away team. The predicted probability of the win would be for the home team.

* using Bayesian statistics: assign a binomial prior and a beta posterior to each team based on the number of wins they already have. Calculate the probability of a win in the remaining games based on difference of the posterior distributions (probability that a draw from distribution A is larger than a draw from distribution B).
This does ignore the strength of the opponent against whom the game was won or lost already.

* use a rating system to estimate win probabilities.

For each method, you could either use the results of single tournaments (per tournament), the results of a clan overall (1 per clan) or by giving partial wins and losses of the same clan in other tournaments (per tournament but with adjustments for performance in other tournaments). The latter approach is easy for the Bayesian method and hard for the other methods.
Given the limited amount of data, you likely want to do this overall, not per tournament, although the assumption that the clan has an equal underlying strength for each tournament is likely violated.

You can estimate the final results in 2 different ways as well:

* Simulate repeatedly. You can update model, distribution or rating system after each game is simulated as well.

* Directly calculate the mean number of points based on the win %. This does not allow for updating.

Edited 3/23/2022 11:16:24

I am a stats noob, and would be delighted to get enlighted.

If you don't know what to do, use T-test.

Which of the (at least 8) different t-tests?

Also: if you don't know what to do, it's probably best to consult a specialist, in this case, a statistician or data-analyst. They have a clan on here, I heard.

You shouldn't operate if you're not a medical doctor.
You probably shouldn't try to fix your plumbing or electricity, if you are not a plumber, electrician or specialised in building.
You may not want to do your own data-analysis without following some classes first or consulting a statistician.

Ah, yes, that sounds like proper statitistics! Thanks for the enlightening advice. Even as I am now (weirdly) a professional mathematician, statistics was never my cup of tea. I indeed could have asked someone with more knowledge, but for such a simple model, I am quite convinced I can handle it.

Your suggestions are indeed decent options and I only considered the rating method, but still at that stage of Clan League, I was in no way able to justify any particular win percenatge for each game. The number of substitutions and lack of early results would just give unbelievably high win percentages. I indeed should have thought of the Bayesian approach.

It's that time again, 101st would like to make another sub please,

Template: Aseridith Islands 1v1
Player out: Bill Green
Player in: MoriOri

Subs Remaining: 6

Speed-run using all of 101st's subs, should be the last of it for a bit though hopefully...

Template: Biomes 2v2
Player out: Derfellios
Player in: TweeKeerJelle

Subs Remaining: 5



In addition 101st also needs to change out our secondary clan contact:

Old Secondary Contact: Derfellios
New Secondary Contact: Kant

Posting here for those that aren't in the discord server and missed the announcement:


Just a heads-up, the clot (bot & site) will be down for the next day or so

Also @Clan Contacts, please include me in mail (in addition to meld still) for any subs or changes

Times for another update, ladies and gentlemen of division B. The bot may be dead at the moment, but now I had some time to pen down a few remarks.

French Community is now leading the division with 61.8% PC, which is impressive, but also shows how close the division is this season. Seldom, the leader of division B had such a low PC %. FC is now a favorite to advance to division A and fight for their second title.

[Blitz] is not far behind, but has many more games completed already, making them the actual leaders in terms of points. If the small penalty of FC at the start of the season is taken into account in the PC % computation, Blitz is actually in the lead.

Discovery had an impressive start of the season but has now fallen back to third place. Normally, around halfway through the season, this would already imply that getting first place is difficult, but this race is still wide open and Discovery has every chance to make the cut.

The next three clans M’Hunters , |GG| and {101st} are incredibly close, and the battle for relegation is still on, yet all these clans also have decent chances to make it to A. Did I already say that B was close?

Only CORP seems to be out of the running. With just over 20% PC, the line-up just does not seem strong enough. Come on guys!

3v3 Middle Earth in the Third Age

Discovery 		1 - 0
M'Hunters 		1 - 0
[Blitz] 		1 - 0
|GG| 			1 - 1
{101st} 		1 - 1
French Community 	0 - 1
CORP 			0 - 2

The first tournament seems to be quite slow. All clans have one game completed, and 3 clans have two games completed while already 6 rounds of games have been created. 3 clans are leading, but everything is possible. The only thing we know for certain, is that CORP has had a rough start.
3v3 Deadman's Rise of Rome

[Blitz] 		2 - 0
M'Hunters 		1 - 0
CORP 			1 - 0
|GG| 			1 - 1
French Community 	1 - 1
{101st} 		1 -2
Discovery 		0 - 3

So far, Blitz seems to be running away with this tournament (but, with that team, anything less would be disappointing) while Disco is in problems. Again, already 6 rounds have been created so some clans are playing 4 or 5 games at the moment. Yikes. The schedule seems a bit too tight for the 3v3s. Still, this tournament is wide open.
2v2 Szeurope

French Community 	2 - 0
{101st} 		2 - 1
Discovery 		1 - 1
|GG| 			1 - 1
[Blitz] 		1 - 2
M'Hunters 		0 - 1
CORP 			0 - 1

Two old-timers, Wazz and Phaeril show that Napolean is not dead and the French still can conquer Europe in a stacked tournament. Unexpectedly, 101st with two new players is able to compete with these players.
2v2 Strategic MME

Discovery 		2 - 0
{101st} 		2 - 0
French Community 	1 - 0
[Blitz] 		1 - 1
M'Hunters 		0 - 1
|GG| 			0 - 2
CORP 			0 - 2

Disco and 101st lead this tournament comfortably while CORP and GG are in trouble. Again, little progress has been made here, so a lot is still possible.
2v2 Biomes of America

|GG| 			1 - 0
M'Hunters 		1 - 0
[Blitz] 		1 - 0
Discovery 		0 -0
{101st} 		0 - 1
French Community 	0 - 1
CORP 			0 - 1

And here is the slowest of all tournaments. Only 3 games completed, so I have to fill this box somehow with a good observation: All 2v2’s are based on real world maps, something which seems quite a bit more common for team games (does anyone know why?). The only team templates ever played on fictional maps are Cookies 2v2, Volcano Islands 2v2 and Middle Earth 3v3.
1v1 Timid Lands

M'Hunters 		4 - 0
French Community 	3 - 0
Discovery 		2 - 1
CORP 			1 - 1
|GG| 			1 - 3
{101st} 		0 - 2
[Blitz] 		0 - 4

MH in this tournament has already twice as many wins as any team has in the team tourneys. Only two more games left for Jacob the Restless before he sweeps the tournament. Jo Jo from FC is closeby, and it is not unlikely that their game will decide the tourney. 101st and especially blitz are in trouble.
1v1 Aseridith Islands

|GG| 			2 - 0
French Community 	2 - 0
[Blitz] 		1 - 1
M'Hunters 		1 - 3
{101st} 		0 - 0
Discovery 		0 - 1
CORP 			0 - 1

Rynofury (GG) and Fan (FC) are showing good results on the new INSS template both beating the strong Emu Pub, who has completed 4 games already while 101st has zero games completed at the moment of writing.
1v1 Battle Islands V

|GG| 			3 - 0
[Blitz] 		3 - 1
{101st} 		2 - 1
French Community 	2 - 2
Discovery 		1 - 2
M'Hunters 		1 - 2
CORP 			0 - 4

MIFRAN (GG) shows that he stills masters the game and shows how this template is played. Jefferspin (Blitz) is trailing him closely, but a lost head-to-head makes winning the tournament a lot more complicated for him. CORP seems to be in big problems in this tournament.
1v1 Strategic Greece

[Blitz] 		3 - 0
{101st} 		2 - 1
CORP 			1 - 0
Discovery 		1 - 1
French Community 	0 - 1
|GG| 			0 - 2
M'Hunters 		0 - 2

Again, Blitz is showing good results here, showing that experience helps on the classical templates. Only, a newcomer form 101st might disagree with that and show that that playing the same stuff for ten years does not make you a better player. GG and MH have difficulties.
1v1 Numenor

Discovery 		2 - 0
French Community 	2 - 0
M'Hunters 		1 - 0
{101st} 		1 - 2
|GG| 			0 - 1
[Blitz] 		0 - 2
CORP 			0 - 2

In the island realm of Numenor, little is happening as of yet, compared with the other 1v1s. Disco and French Community show good results while this is one of the few tournaments where Blitz has trouble gaining points.
1v1 Great Lakes

Discovery 		3 - 0
[Blitz] 		3 - 1
CORP 			1 - 0
|GG| 			1 - 1
French Community 	1 - 2
{101st} 		0 - 2
M'Hunters 		0 - 3

SkekUng of Discovery shows that a new template is not a problem for him and already has three wins. Bechaa of Blitz trails closely, but lost the head-to-head. Despite being the last tournament to start and neutrals of 3, Great Lakes progresses quickly.

Edited 4/4/2022 17:18:14

Thanks for the update. And, yes, the 3v3 DROR games are piling up.

I am of Discorvery :P Thanks Derf!

I am of Discorvery :P Thanks Derf!

Sorry about that. Fixed

Could CORP please substitute the following player on aseridith islands:

Player in: machine-washed cloth
Player out: Jurai Gumo

Will send a PM to meld shortly also :)

I don't want this thread to lock up so I'll mention that my game against The Gunslinger will decide the winner(s) of the Great Lakes Tournament.