ELO adjustments

[quietmint]

I want your thoughts about how ELO points should work. Currently Now Boarding adds/subtracts a flat number of points for each game. A win adds +1 per delivered passenger (so around +40 points for a 2-player game or around +66 in a 5-player game) while a loss subtracts -1 per complaint (so around -4 points for all player counts).

I am considering resetting all player's ELO and rebalancing the points in some way. I want to add or subtract the same number of points for each game (based on player count and the VIP Variant option). I am thinking of adjusting the values to something like this:

elo.png (20.79 KiB) Viewed 1675 times

For example, it would mean winning a 2-player game would add +16 to your ELO and losing a 2-player game would subtract -16 from your ELO. To move from "average player" to "good player" would require you to win or lose about 6 games in a row as 2-player or about 3 games in a row as 5-player.

Some co-op games never subtract points for losing. What are you thought on this? I feel like it's necessary to subtract something, otherwise ELO will just indicate the number of times you've played the game instead of indicating your skill. Should the subtraction amount for losing be equal to the addition amount for winning, or should it be something smaller like only 50% of the amount?

Some co-operative games on BGA use a more complex approach for calculating ELO (for example, Hanabi -- you can read more at https://en.doc.boardgamearena.com/Gameh ... ELO_rating and viewtopic.php?f=3&t=24363), but I think this is overly complicated. It also doesn't seem to solve any problems because every game that implements this still receives player complaints.

[tah2789]

I would assume not subtracting points is because getting grouped with bad players over and over could kill your ELO? I can see the argument there...

Slightly off topic but is there a way to see how ELO is calculated for any other game?

[quietmint]

For competitive games and the normal ELO calculation, see https://boardgamearena.com/faq?anchor=faq_account_elo

For co-op games, I'm not aware of any central place listing how they work/how many points they award (or subtract). Each game developer largely does their own thing and there's no standards around this. The linked forum post mentions The Crew using the simple additive method and Hanabi using the more complex pretend player method.

[ChiefPointThief]

Don’t reset elo. I played too much already
I did think the elo given out was rather high but don’t stress elo for this game. Either way elo will not show your true skill but only how much you play. The game isn’t that difficult. I’ve been picking the same alliance airline as my p to make it harder.
To your point about other co ops elo you are correct. Everyone will not always be happy. I’ve been going back and forth about hanabi elo w/ ppl for the last month. Other games like crew some ppl think their elo makes them great but it just means they play a lot.

I never read the rules (don’t know if timer is part of the game) but possibly making a faster timer w/ higher elo may be an idea. Taking the timer into consideration you would have to rethink real time w/o it and turn based games. Maybe a perfect score bonus would be good or more negative points per complaint. Still looking at one’s score based on some of these suggestions wouldn’t show you how skilled they are but how much they played.

Im basing this on the assumption that people will learn how to win after a few tries. Once you do it is victories moving forward.

[ufm]

Currently there are three rating systems in BoardGameArena co-op games:

1. Compare to an artificial rating value against the game according to the difficulty chosen
- Hanabi, Pandemic

EDIT: Forbidden Island, Tranquility, Yokai have been switched to option 3.

2. Award fixed amount regardless of the difficulty setting/game result and never decrease rating
- Bandido, Similo, The Crews, My First Castle Panic, Burgle Bros., Tiny Epic Defenders

3. Award fixed amount based on the game result and add bonus rating for the game difficulty
Rating loss on defeat: Now Boarding (now here)
No rating loss on defeat: Narabi, Regicide, Regidice, Just One, Paint the Roses, Tranquility: Ascent, Baby Dinosaur Rescue, La Marche du Crabe, Coupell, Ice and the Sky, Robots Ate Our Pizza, Crime Zoom, None Shall Pass!, Solar Storm, Zefiria, Art Thief

Hardback, Room 25 and Timeline Twist adjust rating in competitive mode only.


Unless you plan to switch the calculation algorithm into artificial ELO simulation, resetting the rating is not recommended as it makes all players lose their progress and it causes unnecessary confusion.
I've seen it once in Tranquility, but considering bug reports regarding ratings I can't say the change was well-received.


From my perspective, trying to simulate ELO artificially is detrimental.
You'll understand what I'm saying when you read disputes over 'ELO' in BGA Hanabi forum...
Honestly most players just don't want to mess with ELO-like things in co-ops.
In all co-op games with an artificial ELO simulation, bug reports complaining about that exact system can be found easily in the bug/suggestion list.
I don't even call ratings in co-ops 'ELO' as they are not using a proper ELO system.
In fact, 0 co-op games adopted artificial ELO simulation since 2021 Summer.


Co-op games I implemented never subtract rating for two reasons:
- Subtracting points encourages experienced players to actively avoid beginners as they increase the risk to rating, as tah2789 already said.
- With rating subtraction, conceding (previously abandoning) the game can be abused as it can be exploited to negate rating loss.
FYI, conceding co-op games forces the final score to be 0 to avoid possible bugs (for example, some Hanabi games after co-op abandon ban update crashed at the end of the game as the game's coop_elo_mode failed to address negative scores).


I'd rather opt to apply multipliers to the final winning score according to the difficulty instead, and remove rating loss.
Like, 1/10 (minimum 1) for no timer, 1/4 for double timer, full rating for the standard mode.

I never read the rules (don’t know if timer is part of the game) but possibly making a faster timer w/ higher elo may be an idea.

viewtopic.php?t=31196

[Oloof22]

What your are trying to do here is to outsmart a human mind with a mathematical system, I don't think it can be done. Whatever system you will do, as long as you have freedom to choose your opponents (partners), you have the ability to pick ones that give you an advantage. The only way to fix that is to remove the ability to choose other players. Picking your opponents is a skill based activity and it is reflected in your rating (although it shouldn't). And again it will be there as long as you have the freedom to choose opponents. But I don't want to have fully randomly chosen opponents and I don't think you want that either.

On the other hand you can remove any rating loss leaving only ranking gain. This approach means just removing any rating entirely. Whatever you call it, it is not a skill rating anymore. Rating will not be rating at all, but just a number representing something else, probably your experience at the game, but not skill. I don't like this approach too.

For me the best way to do it (and still call it a skill rating) is already mentioned system:

1. Compare to an artificial rating value against the game according to the difficulty chosen
- Hanabi, Forbidden Island, Pandemic, Tranquility, Yokai

But you mentioned 2 issues: people avoiding begginers, and people conceiding the game.

If people avoid begginers it means that they are overrated, maybe they real skill rating should start at -100, not 0.
But I would say another approach to fix it. Disable the rating change with begginers at table. If you have 100+ rating whether you win or lose a game your ranking will not change at all. And if you have less than 100 you could gain a fixed amount per every won game.
And when all players have 100+ it should work normally.

And about conceiding games, it should just count as a loss.

Either way we go, it will not be perfect as long as there are "Players who pick their opponents carefully" but in my opinion a skill rating should be a skill rating and we should not change it into an "experience rating".

[ufm]

Either way we go, it will not be perfect as long as there are "Players who pick their opponents carefully" but in my opinion a skill rating should be a skill rating and we should not change it into an "experience rating".

Co-op ratings are not always meant to be skill rating.
That's why none of co-ops have adopted your approach for 2 years as 'experience rating' worked better than artificial ELO in general.
Ultimately, the developer and the publisher decide it, not some vocal users.

[quietmint]

Okay, if no other co-op games subtract points for a loss then I suppose I shouldn't do so.

In that case, I'll change to something like this.

• With normal timer: playerCount * playerCount + 10
• With double/unlimited timer: half of the above

elo3.png (10.95 KiB) Viewed 1561 times

[raspberryoolong]

I like the adjustment based on player count. Would it be feasible to add part of what you originally implemented in terms of different scores based on number of passengers delivered? I think there is a significant difference between winning the game barely and having all passengers delivered and both count equally under the new system. So perhaps a base score and add points based on number of passengers delivered past the base (e.g. 37 for 2p) or subtract points based on number of undelivered passengers or complaints at the end of the game.

[ChiefPointThief]

The added cities variant needs to be considered as well. I've only tried the jfk addition for 2p so far.