BGA has a serious bully problem

[FrankJones]

There's another problem with the randomness/volatility of your rating for luck-based games. If you want to be the number one arena player, the "correct" strategy is to keep playing until you get a winning streak and are number one. Then, quit playing and hope nobody else can catch you.

Getting to number 1 arena for a luck-based game is about grinding for a good winning streak, and not being the actual best player.

Lowering the K-factor wouldn't help, because you still would lose points on average when you play a lower-rated player who can win due to luck more than what the elo formula expects.

The only solution is to use a different probability distribution for luck-based games. Elo has an implicit assumption, 400 rating point difference equals 90% chance of winning and it scales multiplicatively/exponentially. The assumptions of the elo system are not valid for luck-based games.

The goal of the rating system should be "figure out who actually is the best player" and not "who went on a winning streak and got to number one".

Or we could just accept the fact that the highest rated player at any arbitrary point in time (in the case of Arena, the 13th of January, April, July, October)) isn't necessarily the best player.

No system is going to guarantee that the best player at some arbitrary point in time is going to be the highest rated player, or vice versa.

[tbhp]

There's another problem with the randomness/volatility of your rating for luck-based games. If you want to be the number one arena player, the "correct" strategy is to keep playing until you get a winning streak and are number one. Then, quit playing and hope nobody else can catch you.

Getting to number 1 arena for a luck-based game is about grinding for a good winning streak, and not being the actual best player.

Lowering the K-factor wouldn't help, because you still would lose points on average when you play a lower-rated player who can win due to luck more than what the elo formula expects.

The only solution is to use a different probability distribution for luck-based games. Elo has an implicit assumption, 400 rating point difference equals 90% chance of winning and it scales multiplicatively/exponentially. The assumptions of the elo system are not valid for luck-based games.

The goal of the rating system should be "figure out who actually is the best player" and not "who went on a winning streak and got to number one".

Or we could just accept the fact that the highest rated player at any arbitrary point in time (in the case of Arena, the 13th of January, April, July, October)) isn't necessarily the best player.

No system is going to guarantee that the best player at some arbitrary point in time is going to be the highest rated player, or vice versa.

Or we can accept that the player who wins a competition was the best of that competition.

In any sports, it would be considered terrible sportsmanship to suggest that the winner is not also the best. You wouldn’t say « if we were to replay the final, statistically the team who lost would have won it 90% of the time, so they were actually the best».

Competition isn’t about « what ifs », it’s about reality and actual results. The winner of any game or competition is definitionally the one that played the best, because they managed to win while others didn’t.

[FrankJones]

There's another problem with the randomness/volatility of your rating for luck-based games. If you want to be the number one arena player, the "correct" strategy is to keep playing until you get a winning streak and are number one. Then, quit playing and hope nobody else can catch you.

Getting to number 1 arena for a luck-based game is about grinding for a good winning streak, and not being the actual best player.

Lowering the K-factor wouldn't help, because you still would lose points on average when you play a lower-rated player who can win due to luck more than what the elo formula expects.

The only solution is to use a different probability distribution for luck-based games. Elo has an implicit assumption, 400 rating point difference equals 90% chance of winning and it scales multiplicatively/exponentially. The assumptions of the elo system are not valid for luck-based games.

The goal of the rating system should be "figure out who actually is the best player" and not "who went on a winning streak and got to number one".

Or we could just accept the fact that the highest rated player at any arbitrary point in time (in the case of Arena, the 13th of January, April, July, October)) isn't necessarily the best player.

No system is going to guarantee that the best player at some arbitrary point in time is going to be the highest rated player, or vice versa.

Or we can accept that the player who wins a competition was the best of that competition.

In any sports, it would be considered terrible sportsmanship to suggest that the winner is not also the best. You wouldn’t say « if we were to replay the final, statistically the team who lost would have won 90% of the time ».

Competition isn’t about « what ifs », it’s about reality and actual results. The winner of any game or competition is definitionally the one that played the best, because they managed to win while others didn’t.

Agreed.

I think the issue (not that I think there is an issue) is that some in this thread feel as though the system is set up in a way such that the actual winner is not the best player because they feel the system makes it difficult for the actual best player to be at the top of the leaderboard when the season ends.

Whereas, I say, as you do, "Whoever ends the Arena season atop the leaderboard was the best player for that Arena season."

That doesn't mean that player is the best player in the world, or even the best player on BGA for that game. It just means that player accomplished something nobody else did in that game - specifically ,finishing the Arena season with the most Arena points.

There are numerous sports analogies.

In the early '90s, The Minnesota Twins won a world series in a season in which they won 87 games (I think). Were they the "best" team? Quite a few teams won more than 87 games in the regular season. So what? The Twins won the World Series.

In 2008, Famously, The NY Giants beat the Patriots in the Super bowl. The Patriots has been 16-0 in the regular season compared to 10-6 for the NY Giants. Oh Well. In a short-term small-ish sample, the team that is most talented and most likely to win in the long run will not always win.

Whether this is a "flaw" or "problem" is a subjective matter of opinion and personal preference.

Also, for Arena seasons here, yeah, you can get a high score and "park" there. But then others can pass you. For the games I play, the Arena winner has kept playing up until the end, trying to gain further points rather than passively sitting there and risking being overtaken while sitting there doing nothing.

Furthermore, if someone can just go on a "lucky streak" and "park atop the rankings" to win the Arena season, then, who cares anyway? That would be acknowledging that it's just an arbitrary time cutoff luck-fest.

I still think this is much fuss about nothing. There is skill inherently embedded into this strategy anyway.

"Okay, I'm in the number one spot with this many Arena points. Do I stay here? Do I keep playing?" These questions require knowledge of the variance of the game, knowledge of one's own ability to continue winning, while weighing the probabilities of Arena points gained and lost by playing further.

If the game has so much luck (as has been indicated by CHiefPointThief) that such determinations cannot be made, then by definition, it is a high-luck game in which there will be a lot of volatility in the rankings at any given time, including when the Arena season ends.

The notion that the same player should win Arena season every time, in a luck based game, seems absurd to me. The more Luck involved in a game, the more likely it is that the "best" player will not finish atop the standings at any arbitrary point in time.

[Ceaseless]

There's another problem with the randomness/volatility of your rating for luck-based games. If you want to be the number one arena player, the "correct" strategy is to keep playing until you get a winning streak and are number one. Then, quit playing and hope nobody else can catch you.

Getting to number 1 arena for a luck-based game is about grinding for a good winning streak, and not being the actual best player.

Lowering the K-factor wouldn't help, because you still would lose points on average when you play a lower-rated player who can win due to luck more than what the elo formula expects.

The only solution is to use a different probability distribution for luck-based games. Elo has an implicit assumption, 400 rating point difference equals 90% chance of winning and it scales multiplicatively/exponentially. The assumptions of the elo system are not valid for luck-based games.

The goal of the rating system should be "figure out who actually is the best player" and not "who went on a winning streak and got to number one".

And what would this alternative be? The nature of trying to scale the probabilities with large gaps is probably a big part of where things go haywire. There's a good chance you simply have to not have rated games at a certain gap, which may have some variation per game, but probably exists somewhere for every game. Extreme gaps always offer abuse, just in different directions when it comes to high variance vs low variance.

Or we can accept that the player who wins a competition was the best of that competition.

In any sports, it would be considered terrible sportsmanship to suggest that the winner is not also the best. You wouldn’t say « if we were to replay the final, statistically the team who lost would have won it 90% of the time, so they were actually the best».

Competition isn’t about « what ifs », it’s about reality and actual results. The winner of any game or competition is definitionally the one that played the best, because they managed to win while others didn’t.

Or we can accept that whoever won the competition... won the competition. That is the fact of the matter. That they are the best, or that they even played the best, is an interpretation, not fact. You can make a judgment call and use data to support your claims, but there's nothing guaranteeing you'll be right.

Agreed.

I think the issue (not that I think there is an issue) is that some in this thread feel as though the system is set up in a way such that the actual winner is not the best player because they feel the system makes it difficult for the actual best player to be at the top of the leaderboard when the season ends.

Whereas, I say, as you do, "Whoever ends the Arena season atop the leaderboard was the best player for that Arena season."

That doesn't mean that player is the best player in the world, or even the best player on BGA for that game. It just means that player accomplished something nobody else did in that game - specifically ,finishing the Arena season with the most Arena points.

There are numerous sports analogies.

In the early '90s, The Minnesota Twins won a world series in a season in which they won 87 games (I think). Were they the "best" team? Quite a few teams won more than 87 games in the regular season. So what? The Twins won the World Series.

In 2008, Famously, The NY Giants beat the Patriots in the Super bowl. The Patriots has been 16-0 in the regular season compared to 10-6 for the NY Giants. Oh Well. In a short-term small-ish sample, the team that is most talented and most likely to win in the long run will not always win.

Whether this is a "flaw" or "problem" is a subjective matter of opinion and personal preference.

These big events are more for show than substance, which may be more praise for the show than criticism of the substance. Winning one event does not mean as much as many like to believe, at least with respect to the things they often try to tie it to. To win may be a very remarkable achievement and can be very important for all sorts of other reasons, but the idea that whoever wins "this one big thing" is the ultimate champion is for entertainment, not reality.

[tbhp]

Or we can accept that whoever won the competition... won the competition. That is the fact of the matter. That they are the best, or that they even played the best, is an interpretation, not fact. You can make a judgment call and use data to support your claims, but there's nothing guaranteeing you'll be right.

There's no judgment call to make. The outcome of the game already tells you who played best. The player who manages to win according to the rules is by definition the one who was the best. That's how games work. How much luck or non-luck there is in a game doesn't change this basic concept.

These big events are more for show than substance, which may be more praise for the show than criticism of the substance. Winning one event does not mean as much as many like to believe, at least with respect to the things they often try to tie it to. To win may be a very remarkable achievement and can be very important for all sorts of other reasons, but the idea that whoever wins "this one big thing" is the ultimate champion is for entertainment, not reality.

Imagine calling a world champion title or an Olympic gold just "entertainment".

Titles are what games are played for, and are the final arbiter of who did best in that discipline. It doesn't matter how good you think a player is, if they never win anything big, then they failed their career.

[Ceaseless]

Or we can accept that whoever won the competition... won the competition. That is the fact of the matter. That they are the best, or that they even played the best, is an interpretation, not fact. You can make a judgment call and use data to support your claims, but there's nothing guaranteeing you'll be right.

There's no judgment call to make. The outcome of the game already tells you who played best. The player who manages to win according to the rules is by definition the one who was the best. That's how games work. How much luck or non-luck there is in a game doesn't change this basic concept.

These big events are more for show than substance, which may be more praise for the show than criticism of the substance. Winning one event does not mean as much as many like to believe, at least with respect to the things they often try to tie it to. To win may be a very remarkable achievement and can be very important for all sorts of other reasons, but the idea that whoever wins "this one big thing" is the ultimate champion is for entertainment, not reality.

Imagine calling a world champion title or an Olympic gold just "entertainment".

Titles are what games are played for, and are the final arbiter of who did best in that discipline. It doesn't matter how good you think a player is, if they never win anything big, then they failed their career.

If I play a game where a player tells me if my 100-sided die will land on a 1 or not and they just keep saying it will, we don't say they won the 1/100 games they take because they played the best. Winning has a correlation to better play, but they're not synonymous. We are capable of using our brains and going "oh, yeah, that happens 1% of the time, they just got lucky there" and move on.

No need to imagine, I just called them entertainment. The players in these events are entertainers first, competitors second. It's a spectacle for the benefit of the spectators. To give credit where credit is due, it's quite an impressive scale for entertainment. I hear there's quite an event in the Superbowl, the commercials.

To say a player that hasn't achieved any titles is a failure in their career is ludicrous, especially if we're talking team sports, where a player's impact on the outcome is even more limited than in standalone events. Results are a useful tool for analyzing things, but trying to substitute your brain with the outcome of the game is nothing but delusion. The outcome of the game is just another fact, which you can analyze with other data to come to a conclusion. Without proper judgment, that conclusion is going to end up being rubbish.

[Fuchur]

Sorry, this is (as the vast majority of posts in this thread) completely off-topic.

It seems to me that most of posts keep messing up two sides of ELO (in any of the infinite number of ways you can implement it) :

• Describe the past performance of someone.

• Be an good approximation of a "skill-meter" (see below), enabling to predict the chances of winning when playing against somebody else.

While any implementation of ELO will do for the first point, the second is much harder to meet -- because there might be nothing to approximate :

Definition skill-meter : Assign à real number S = S(A,t) to measure the "skill" of player A at a given time t with the following property :
If player B has, at a given moment, a skill-meter value T = S(B,t) below the value of player A : S > T, than the chance of A to win against B is higher than 50 %.

Consequence : If A has a "higher skill" than B (i.e. S(A,t) > S(B,t)) and B is better than C (at the same moment) than A must be better than C.

• It is quite easy to come up with exemples where it is impossible to have a consistent skill-meter
  • (think of rock-paper-scissors where player A always plays rock, B always paper and C always scissors,
  • or, even more striking, something like the intransitives dices https://en.wikipedia.org/wiki/Intransitive_dice : In every single game players A, B et C are perfectly ordered : Someone will bee the best, someone the last et someone in between, but in the long run (i.e. on average) A will win against B, B against C and C against A).

• In games which are 100 % luck based the skill-meter is the same (arbitrary) value for anybody.

In both cases ELO has no predictive value.

P.S. I've never seen any discussion whether a given game admits a skill-meter or not; the common use of ELO in chess seems to indicate that it comes (at least reasonably) close to a skill-meter. In "high luck games" where it exists its predictive value is quite limited.

[Ceaseless]

Definition skill-meter : Assign à real number S = S(A,t) to measure the "skill" of player A at a given time t with the following property :
If player B has, at a given moment, a skill-meter value T = S(B,t) below the value of player A : S > T, than the chance of A to win against B is higher than 50 %.

Consequence : If A has a "higher skill" than B (i.e. S(A,t) > S(B,t)) and B is better than C (at the same moment) than A must be better than C.

If you're going to use t for time, even if you're capitalizing to T, I would suggest against using that letter for something else. Use a different letter, or even just use S' or something, lol.

Definition skill-meter : Assign à real number S = S(A,t) to measure the "skill" of player A at a given time t with the following property :

• It is quite easy to come up with exemples where it is impossible to have a consistent skill-meter
  • (think of rock-paper-scissors where player A always plays rock, B always paper and C always scissors,
  • or, even more striking, something like the intransitives dices https://en.wikipedia.org/wiki/Intransitive_dice : In every single game players A, B et C are perfectly ordered : Someone will bee the best, someone the last et someone in between, but in the long run (i.e. on average) A will win against B, B against C and C against A).

• In games which are 100 % luck based the skill-meter is the same (arbitrary) value for anybody.

In both cases ELO has no predictive value.

P.S. I've never seen any discussion whether a given game admits a skill-meter or not; the common use of ELO in chess seems to indicate that it comes (at least reasonably) close to a skill-meter. In "high luck games" where it exists its predictive value is quite limited.

Even Chess won't have absolute transitivity, though it's not so severe to cause notable issues. Generally the application of transitivity isn't the problem for games of luck, it's the degree of transitivity that messes things up for a lot of players.

[Fuchur]

P.S. I've never seen any discussion whether a given game admits a skill-meter or not; the common use of ELO in chess seems to indicate that it comes (at least reasonably) close to a skill-meter. In "high luck games" where it exists its predictive value is quite limited.

Just found this discussion of the predictive value of chess-ELO : https://blog.russelldmatt.com/2018/11/1 ... s-elo.html

[tbhp]

If I play a game where a player tells me if my 100-sided die will land on a 1 or not and they just keep saying it will, we don't say they won the 1/100 games they take because they played the best.

It doesn't matter what the odds are. They guessed correctly where it would land, so they were the best at the games they won. The results are what counts.

Your attitude is very unsportsmanlike and stands against everything that games are about.

By your logic, if someone has a 99% chance of winning when they play their last turn, but then on their opponent's last turn, they get the winning roll they had a 1% chance of getting, then according to you the first player played better because they were much more likely to win before the last moment.

That's not how games work. You don't rewrite history by invoking probabilities. Probabilities help you make your choices and determine the most likely path to victory, but the moment the curtains fall and the game is over, probabilities don't matter anymore and there is only one reality. The player who actually wins is always the deserving victor and the one who played best.

No need to imagine, I just called them entertainment. The players in these events are entertainers first, competitors second. It's a spectacle for the benefit of the spectators. To give credit where credit is due, it's quite an impressive scale for entertainment. I hear there's quite an event in the Superbowl, the commercials.

Are you aware there are sports that barely anyone watches?

Imagine telling the best athletes in the world that they are entertainers before being competitors.

The fact that many people watch certain sports is a bonus. It's not why athletes play sports. They play because they want to surpass themselves or be the best in a discipline. You don't need a crowd for that (just as when we compete for titles here, we generally don't have a crowd watching us).

On your last line, if this is your outlook on sports and, well, life, why are you even on this website?

To say a player that hasn't achieved any titles is a failure in their career is ludicrous, especially if we're talking team sports, where a player's impact on the outcome is even more limited than in standalone events. Results are a useful tool for analyzing things, but trying to substitute your brain with the outcome of the game is nothing but delusion. The outcome of the game is just another fact, which you can analyze with other data to come to a conclusion. Without proper judgment, that conclusion is going to end up being rubbish.

You are talking about victory as if it was a detail, just another metric.

Winning is the point of games, and the way you play is the tool to get there. Analyzing things is good, but that should never lead us to rewrite history and act like the person who won something wasn't actually the best at that event. You have to understand that those are two separate discussions.