Rating Inflation in FargoRate

[LHP5]

So another question/thread about Fargo. I know this is sort of based off of the Elo rating system in chess or what not, but my understanding is that the average chess ratings have been climbing the past few decades. Now this can be the result of an actual better population of players due to better instructors/training or better computer training programs, or etc. However, some people tend to argue though that the ratings are actually artificially high and that may be due to flaws in the system itself. Rating inflation I think is what it's called.

I have noticed that the number of 800 level players has significantly increased within the past few years. I remember there being only 10 or so players at 800+. Today I think that number is hovering around 20 players. Even the lowest rating at the top 100 players list is at a minimum 760 when I believe it was around 740/750 a few years ago. More glaring would be the US top 100 players list that just 3 years ago had the player listed at #100 at around 700. Today that number is at a minimum 710. This change might not seem like much, but if this type of trend continues the lowest rating maybe at 730 in a few years. This is not necessarily a bad thing and it's possible that the skill level has risen, but is it real?

Looking for Mike Page or anyone with this experience to chime in. I'm actually interested to see if rating inflation has been addressed in the FargoRate system and not trying to start any fires. Thanks

 

[mikepage]

So another question/thread about Fargo. I know this is sort of based off of the Elo rating system in chess or what not, but my understanding is that the average chess ratings have been climbing the past few decades. Now this can be the result of an actual better population of players due to better instructors/training or better computer training programs, or etc. However, some people tend to argue though that the ratings are actually artificially high and that may be due to flaws in the system itself. Rating inflation I think is what it's called.

I have noticed that the number of 800 level players has significantly increased within the past few years. I remember there being only 10 or so players at 800+. Today I think that number is hovering around 20 players. Even the lowest rating at the top 100 players list is at a minimum 760 when I believe it was around 740/750 a few years ago. More glaring would be the US top 100 players list that just 3 years ago had the player listed at #100 at around 700. Today that number is at a minimum 710. This change might not seem like much, but if this type of trend continues the lowest rating maybe at 730 in a few years. This is not necessarily a bad thing and it's possible that the skill level has risen, but is it real?

Looking for Mike Page or anyone with this experience to chime in. I'm actually interested to see if rating inflation has been addressed in the FargoRate system and not trying to start any fires. Thanks

I think the US-100 player going from 700 a few years ago to 710 now is not rating inflation but rather more players being on our radar.

We don't suffer from the same inflation problem the ELO schemes do, but then again if there is an actual overall tide rise of a point or two over several years we would not detect it. The ratings can be arbitrarily shifted up or down without affecting anything (add 50 points to everybody and nothing changes). We actually anchor the ratings by assuming the average level amongst a number of active top players doesn't change.

 

[LowRight]

With FargoRate, the only thing that really matters is the difference between players to determine a fair match. If player 1 is twice as good as player 2, he or she will have a rating that is 100 points higher and would have to give player 2 half of the games on the wire to make an equal match. If the top ratings are in the 800s or the 8,000s, it doesn't make a difference for all intents and purposes.

Some have trouble with the fact that an 800 is twice as good as a 700. Why wouldn't it be 800 vs. 400? The answer is that the ratings are not linear; they are logarithmic.

 

[hang-the-9]

So another question/thread about Fargo. I know this is sort of based off of the Elo rating system in chess or what not, but my understanding is that the average chess ratings have been climbing the past few decades. Now this can be the result of an actual better population of players due to better instructors/training or better computer training programs, or etc. However, some people tend to argue though that the ratings are actually artificially high and that may be due to flaws in the system itself. Rating inflation I think is what it's called.

I have noticed that the number of 800 level players has significantly increased within the past few years. I remember there being only 10 or so players at 800+. Today I think that number is hovering around 20 players. Even the lowest rating at the top 100 players list is at a minimum 760 when I believe it was around 740/750 a few years ago. More glaring would be the US top 100 players list that just 3 years ago had the player listed at #100 at around 700. Today that number is at a minimum 710. This change might not seem like much, but if this type of trend continues the lowest rating maybe at 730 in a few years. This is not necessarily a bad thing and it's possible that the skill level has risen, but is it real?

Looking for Mike Page or anyone with this experience to chime in. I'm actually interested to see if rating inflation has been addressed in the FargoRate system and not trying to start any fires. Thanks

I made this same post a while ago but I think in another thread about Fargo. If my memory is good, there was only ONE 800 level player at one point. Seems like there is a bit of a ratings creep where the top players keep bumping into 800s and everyone at the top is going up 10-20 points over a few years.

 

[iusedtoberich]

Here is a 3 year old list from the players entered in 2015's US Open for comparison.

https://forums.azbilliards.com/showpost.php?p=5335253&postcount=17

 

[Poolplaya9]

I was curious about a few things as well. Some of it was already touched on and in those cases I am seeking a bit more clarification.

What exactly anchors FargoRate? What is or what defines the rating baseline that everything else is based/gauged off of?

Can what any individual rating number means change over time? For example, will an 800 level player 30 years from now play at the exact same speed as an 800 level player from today, or can the actual skill level that an 800 rating represents (or a 500 rating or any other) change over time? Or to clarify for some, if an 800 rated player 30 years from now were able to time travel back to present day and match up with an 800 rated player from today, is there any chance that it would not be a statistically dead even match up (presuming both scores were perfectly accurate at their respective times)?

I have come across a player with a negative rating (it is not an established rating yet though), and not by just a couple of points. Can you confirm that this is correct? I would presume that it is correct given that the algorithm is set to distribute rankings in a way such that every player is twice as good as those exactly 100 points beneath them, and only half as good as those exactly 100 points above them, and there could in fact be enough separation between the very best and very worst that negative numbers are having to be utilized particularly when taking into account that pretty extreme outlier ratings at the ends of the scale can more easily occur when players only have a few games in the system.

Is it correct to assume that you have shifted the scale upwards enough that negative ratings only occur on rare occasion, and can only really happen with people that have just have a handful of games in the system, or is this something that isn’t uncommon and also occurs at times even with players with more than a handful of games in the system (but not yet established), say more than 50 games? In the worst case scenario, do you think a negative rating for an established player (200+ games in the system) could ever actually happen? And finally, can you confirm what the theoretical lowest and highest possible ratings could be (with no robustness levels excluded) based on the current algorithm, if any?

Just curious about these and some other questions that have crossed my mind in the past (can’t remember the others at the moment), and if any of them would cross into revealing parts of the secret sauce that you would prefer not to share I understand and respect that.

 

[Bob Jewett]

... What exactly anchors FargoRate? What is or what defines the rating baseline that everything else is based/gauged off of?
...
I have come across a player with a negative rating (it is not an established rating yet though), and not by just a couple of points. Can you confirm that this is correct? ...

Mike answered the first question above. A number of top players as a group set the upper end.

As for a negative rating, that's entirely possible. Many years ago the NPL (which also uses a logarithmic scale but with 30 points per doubling) had one player who played so badly that his established rating should have been negative. Instead he was pegged at a small positive number. I think a better solution would have been to move everyone up so even the worst player in the system would have a positive rating.

I can imagine resetting Fargo so that there are one or two players over 1000 which would probably move the negative players into positive territory. Also, psychologically, if you change a 25 player to 225 he might feel a lot better compared to the top players at 225/1000 versus 25/800. But mathematically it makes no difference.

It is interesting that there are ability ratios at pool of up to 100 or so among people who play regularly. That is, SVB could expect to win a race to 100 with a shutout against some players. (Maybe at that large a difference the randomness of nine ball would change the theoretical odds.) I suppose you see the same ratio or higher in other sports. Some people who play tennis regularly probably have a chance in a million of winning a set against Federer.

 

[misterpoole]

My rating has drifted higher even though i have not been playing in many bca rated matches in the last few years. When i have played, I would have expected my mostly losses to lower my high rating (over 600) but that hasnt happened. this means that i can no longer play in some fargo restricted tournaments. Those restrictions have stayed the same.
Seems like my better wins years ago dont fade over time with this rating system. Just my own personal observation for this year now that i want to play more.

 

[mikepage]

I made this same post a while ago but I think in another thread about Fargo. If my memory is good, there was only ONE 800 level player at one point. Seems like there is a bit of a ratings creep where the top players keep bumping into 800s and everyone at the top is going up 10-20 points over a few years.

You are probably remembering a USA list, which may have had only 1 800+ player. I think there has always been a bunch (10ish or more) at the world level

 

[mikepage]

Mike answered the first question above. A number of top players as a group set the upper end.

As for a negative rating, that's entirely possible. Many years ago the NPL (which also uses a logarithmic scale but with 30 points per doubling) had one player who played so badly that his established rating should have been negative. Instead he was pegged at a small positive number. I think a better solution would have been to move everyone up so even the worst player in the system would have a positive rating.

I can imagine resetting Fargo so that there are one or two players over 1000 which would probably move the negative players into positive territory. Also, psychologically, if you change a 25 player to 225 he might feel a lot better compared to the top players at 225/1000 versus 25/800. But mathematically it makes no difference.

It is interesting that there are ability ratios at pool of up to 100 or so among people who play regularly. That is, SVB could expect to win a race to 100 with a shutout against some players. (Maybe at that large a difference the randomness of nine ball would change the theoretical odds.) I suppose you see the same ratio or higher in other sports. Some people who play tennis regularly probably have a chance in a million of winning a set against Federer.

There are about 22,700 established players now

Top 1% 748
Top 0.1% 797
bottom 1% 225
bottom 0.1% 121

The average is 502.4, median is 506.6
standard deviation is 111

It is tempting to view this as some sort of fundamental distribution. But if we look at the 14000 players with 300 or more games (as opposed to all players with 200 or more) the average goes up to 514. This trend continues and goes the other way too.

There are only a couple (out of 20,700) negative established ratings. I looked at one. It is a female who played in an all-female league for which the average opponent is rated 258. This player won 25 games and lost 192.

Amongst opponents above 300 (averaging 398) her record is 2 wins and 60 losses.

 

[Poolplaya9]

Mike answered the first question above. A number of top players as a group set the upper end.

As stated in my post I saw that he had addressed it, but without specifics. I was curious about the specifics. How are the players chosen, how many of them are there, what method is used (their average or?), etc?

I think a better solution would have been to move everyone up so even the worst player in the system would have a positive rating.

Wondering whether FargoRate would be better served by shifting everyone up is one of the reasons for some of my questions (simple curiosity and possibility to learn something is probably more of the reasons for them though). I agree that within reason, shifting the scale to keep most players out of the negatives is best, and all would be ideal if feasible, but I think that keeping them all out of the negatives (particularly some of those that aren't fully established) is probably impossible unless he has set some kind of cutoff limits for how low/high the ratings can go (and how he is handling the extreme ends of the scale was essentially another of my questions). I suspect that where he has chosen for the distribution of people to fall on the scale is probably already pretty close to ideal all things considered.

I can imagine resetting Fargo so that there are one or two players over 1000 which would probably move the negative players into positive territory.

My feeling that it is probably essentially impossible to prevent any possibility for a negative rating (without imposing scale end limits) is for the exact reason you hit on with your SVB race to 100 example where some people would never be able to win a game. With the most extreme examples at both ends of the scale there is so much skill level difference that to keep everybody in the positives you might have to shift the scale to where SVB would have to be rated some ridiculous number like say a 20,000, and with the vast majority of players still falling within an 800 point range (from say 19,200-20,000) as it is now, but that would obviously look pretty silly to most people, not to mention you still have the psychological issues with people seeing that they are thousands of points beneath him (even though they are still in the positives) while most others are within hundreds of points.

Also, psychologically, if you change a 25 player to 225 he might feel a lot better compared to the top players at 225/1000 versus 25/800. But mathematically it makes no difference.

Fully agree on both points.

Hopefully Mike will answer my still as of yet unanswered questions or let us know that he would prefer to keep the answer proprietary when/if that is the case.

 

[skip100]

Inflation has continued with 32 players now rated 800 or above. The minimum rating in the top 100 has also increased to 767 and the minimum rating in the US top 100 has increased to 721.

 

[Texas Carom Club]

Mine just goes down

 

[skip100]

I wonder if the addition of so many league players in recent years has caused the apparent "inflation" of pro ratings? If there is a centering mechanism for the entire data set that hasn't been adjusted then this could be a result.

The blurb on the front page FAQ does need to be changed to match up with the most recent data: "Fewer than 20 players worldwide have ratings that exceed 800."

 

[Fatboy]

There are about 22,700 established players now

Top 1% 748
Top 0.1% 797
bottom 1% 225
bottom 0.1% 121

The average is 502.4, median is 506.6
standard deviation is 111

It is tempting to view this as some sort of fundamental distribution. But if we look at the 14000 players with 300 or more games (as opposed to all players with 200 or more) the average goes up to 514. This trend continues and goes the other way too.

There are only a couple (out of 20,700) negative established ratings. I looked at one. It is a female who played in an all-female league for which the average opponent is rated 258. This player won 25 games and lost 192.

Amongst opponents above 300 (averaging 398) her record is 2 wins and 60 losses.

This distribution of rating could be applied to marketing of goods and services for active players in this group of data. Good stuff.

Best
Fatboy

 

[mikepage]

I wonder if the addition of so many league players in recent years has caused the apparent "inflation" of pro ratings? If there is a centering mechanism for the entire data set that hasn't been adjusted then this could be a result.
[...]

Here is an update of the numbers


FOUR YEARS AGO
There are about 22,700 established players now

Top 1% 748
Top 0.1% 797
bottom 1% 225
bottom 0.1% 121

The average is 502.4, median is 506.6
standard deviation is 111

NOW

There are about 62,150 established players

Top 1% 726
Top 0.1% 795
Bottom 1% 220
Bottom 0.1% 122

The average is 486, median is 490
standard deviation is 106

 

[skip100]

Interesting. The 15-point drop in the median rating mirrors the roughly 15-point inflation for the top players.

 

[fastone371]

Wow, I gotta remember to check dates so I don't keep reading these outdated posts.

 

[jtompilot]

Interesting. The 15-point drop in the median rating mirrors the roughly 15-point inflation for the top players.

Say what? Looks like deflation to me. The top dropped from four years ago

 

[skip100]

Say what? Looks like deflation to me. The top dropped from four years ago

Those percentiles don't mean the same thing. The top 0.1% was made up of 22 players four years ago while it's 62 players now.

Taken at face value, this would suggest that the 60th best player in the world now is as equally skilled as the 20th best player in the world four years ago. Unlikely. These are largely the same folks who are playing largely at the same level.

It's true that the ratings are only important in a relative sense. However, we have gotten used to the 600, 700, 800 levels "meaning something" and that meaning seems to be drifting slowly over time. I wonder when we'll see Fargo ratings 2.0 with the great recentering