2021 US Championship

[Pargat Perrer]

Okay, so experiment #1 did not give the result I expected.

https://www.chess.ca/en/ratings/t/?id=202110013

As you can see, parameters of experiment 1 are:

8 unrated new players.
5 round Swiss with a few byes included

Comments:

Since all players are unrated, the rating system is unable to proceed without some user input.
It stopped and asks for the average rating of unrated players.
I entered 800. Program continued with this results.

Performance ratings are correct.
The formula for performance rating is
win - add 400 points to your opponents rating.
draw - use your opponents rating
loss - subtract 400 points from your opponents rating,
to calculate your performance rating.

For example, for Brenda
1200 +1200 +400 +1200 +1200 = 5200 / 5 = 1040

For Evelyn
+1200 + 400 + 1200 + 400 = 3200 /4 = 800

Provisional ratings should simply be your cumulative average performance rating,
so I was expecting all the provisional ratings to be equal their performance ratings.

That didn't happen. I am disappointed but not all that surprised.
The program does give unexpected results occasionally in very unusual conditions.

I am going to try some further experiments and see what happens.
If any math wizards out there can decipher the differences between perf and ending ratings, I would love to hear it.

Bob I don't want to spook you, but I literally knew your next post would begin by saying in effect "things didn't go as I expected." I literally visualized it, not by willing myself to visualize it, but it popped into my head. It's not the first time this has happened. It has happened to me in card games, when I literally know what card in the deck someone has just been dealt. When it happens, which is quite rare and unpredictable, I am never wrong.

Also, I want to point out that you are grammatically incorrect to say for your performance rating,
"win - add 400 points to your opponents raging"
and
"loss - subtract 400 points from your opponents rating"
This makes it sound like you are to modify your opponents rating.

Ok, that is nitpicky (?).....

My point is proven. To begin an ELO rating system, to kick it off, some default number has to be provided. Even Arpad Elo MUST have assigned a default rating to the first beginning player or players. His formula requires an existing rating to be there.

Based on the answers I have seen on Quora, this default beginning number was likely 1200. No one seems to know for certain.

So from 1200, a number that was basically picked from a hat, men have developed in chess to have peak ratings around 2850, which might eventually get to 2900.

There is no reason why women, IF THEY ARE NOT INFERIOR TO MEN AT CHESS, cannot do the same by playing only amongst themselves. Men had no one rated 2800 to play at the beginning, women don't need it either.

Furthermore, if we started a new rating group of people who are left-handed, that group would obviously have lower membership than the mixed section, but if we assume that left-handed and right-handed people are equal in chess ability, the left-handed group would eventually reach par with the mixed group even if left-handers only played against left-handers.

I consider my point proven at this time.

[Pargat Perrer]

In the meantime, you will need to be patient young grasshopper.
It will take several lessons to learn the basics of the CFC rating system.
Only then, will I wade into the discussion between yourself and Bob A.

I do know the basics of Elo rating system, and i know enough to realize that at the very beginning, when no one is rated, a default rating must be provided. You, on the other hand, were surprised when you submitted your completely-unrated section and the program asked you to provide that default number.

It's ok, Bob, even the GMs still learn something new every time they play chess lol.

The key is to be ready at all times to learn something new. Don't worry, I won't be "going Belzberg" on you! (a modification of the American saying "going postal") LOL

[Bob Gillanders]

I do know the basics of Elo rating system, and i know enough to realize that at the very beginning, when no one is rated, a default rating must be provided. You, on the other hand, were surprised when you submitted your completely-unrated section and the program asked you to provide that default number.

It's ok, Bob, even the GMs still learn something new every time they play chess lol.

The key is to be ready at all times to learn something new. Don't worry, I won't be "going Belzberg" on you! (a modification of the American saying "going postal") LOL

Oh excuse me, you are so smug aren't you. I guess my grammar must be terrible as you so frequently misinterpret what I say.

I was NOT surprised that the rating program asked for an average rating estimate for the unrated, as that has happened many many times over the years. If you have a small percentage of rated players, it will do the same. What did surprise me was that the provisional ratings were not calculated as expected. The program gets it right most of the time, but not this time.

Many new players ask about how their first ratings are calculated. A common misconception is that there is a beginning default (same) rating for all new players. Some believe it is zero, some think it is 1200 some think 1500. These seem to be the most common beliefs. I wanted to demonstrate how it actually happens. I could have picked any estimated rating for the group other than 800. But I assure you, some thought does go into it. Often I consider how old the group is (as it is usually kids) or I ask the TD /Coach for an estimate. It isn't the same default number, it depends upon the group.

I am greatly insulted by your tone.

[Pargat Perrer]

Oh excuse me, you are so smug aren't you. I guess my grammar must be terrible as you so frequently misinterpret what I say.

I was NOT surprised that the rating program asked for an average rating estimate for the unrated, as that has happened many many times over the years. If you have a small percentage of rated players, it will do the same. What did surprise me was that the provisional ratings were not calculated as expected. The program gets it right most of the time, but not this time.

Many new players ask about how their first ratings are calculated. A common misconception is that there is a beginning default (same) rating for all new players. Some believe it is zero, some think it is 1200 some think 1500. These seem to be the most common beliefs. I wanted to demonstrate how it actually happens. I could have picked any estimated rating for the group other than 800. But I assure you, some thought does go into it. Often I consider how old the group is (as it is usually kids) or I ask the TD /Coach for an estimate. It isn't the same default number, it depends upon the group.

I am greatly insulted by your tone.

Relax, Bob, take a chill pill or have a drink.

If I was being (moderately) smug, it was only because you started that by saying to me:
"In the meantime, you will need to be patient young grasshopper.
It will take several lessons to learn the basics of the CFC rating system."

What, am I a child? If anyone should have been offended it is me, you started with the smug tone as if you were the Oracle of Wisdom on ratings and were going to teach us all some great lesson.

I admitted the grammar thing was nitpicky, and look, I even made a mistake. I thought you were surprised at needing to provide a default value, but instead you were surprised by what your CFC rating system did. So let's just dial back this "I am greatly insulted" nonsense, shall we? No insult was intended, I assure you, even though as I said you were the initlal one to be smug.

So let's get back to the matter at hand.

All I wanted to know was did you need to provide an initial value. As soon as you admitted that, my point was made. True, a new player today does not get a default value, but if a whole new rating group was begun, and it was a group which was NEVER going to play an established rated player, then yes, a default value is needed.

So this is how all ELO chess ratings began: with some initial default value. No one seems to know what value Arpad Elo used, maybe its in his book, which was republished in 2008 and is available on Amazon, but I don't want to buy it just to find out what initial value was used.

And if women had begun by refusing to play men at all, and thus became their own rating group playing only amongst themselves, they too would have required a default starting value.

Which means, assuming women are in no way inferior to men at chess and work as hard at chess as men do, they would as a separate rating group eventually reach rating parity with men. Meaning that an equal % of women would be over 2800 as is the case with the men, would be 2700-2799 as is the case with men, would be 2600-2699 as is the case with men, etc.

All without ever playing any men. Bob Armstrong says this is not possible. He is wrong.

So now Bob you have opened up another can of worms. You say "The program gets it right most of the time, but not this time." Exactly what does that mean? Is there a bug in the program? Are all CFC ratings suspect?

And what about this initial value you have the group of 800, what if you had given 1,000,000? Would the program accept that? What would the consequences be once all those millions of rating points got mixed into the rest of the CFC?

[Bob Gillanders]

assuming women are in no way inferior to men at chess and work as hard at chess as men do, they would as a separate rating group eventually reach rating parity with men. Meaning that an equal % of women would be over 2800 as is the case with the men, would be 2700-2799 as is the case with men, would be 2600-2699 as is the case with men, etc.

All without ever playing any men. Bob Armstrong says this is not possible. He is wrong.

Let me ask you a hypothetical question,

Start with 2 hypothetical groups of chess players:

Group 1 - 1,000 Men - average rating = 1700 with normal distribution.
Group 2 - 100 Women - average rating = 1700 with normal distribution.

Adequate tests have been done to verify ratings are accurate, and that 1700 in group 1 equal 1700 in group 2.

We go forward with everyone only playing those in the same group.
Men play men, Women play women.

A year later, what is the result?

[Pargat Perrer]

Let me ask you a hypothetical question,

Start with 2 hypothetical groups of chess players:

Group 1 - 1,000 Men - average rating = 1700 with normal distribution.
Group 2 - 100 Women - average rating = 1700 with normal distribution.

Adequate tests have been done to verify ratings are accurate, and that 1700 in group 1 equal 1700 in group 2.

We go forward with everyone only playing those in the same group.
Men play men, Women play women.

A year later, what is the result?

Well, I asked you a set of questions about your CFC rating experiment and you haven't answered them. You said you would answer all questions.

Also, your hypothetical question is loaded because you are limiting the time period to a single year. How about 100 years? Also, you are ignoring that each group of players will grow over time, new members joining.

But I would say that the distribution of ratings should be roughly the same in each group, accounting for margin of error and assuming men and women have equal abilities and work habits in chess. But in practical terms, if we say that the odds of any player in a normal distribution group of players reaching 2850 rating are 1 in 1000 (just to make it simple), the men could see a single member get over that rating while the women most likely won't. That doesn't demonstrate anything, because if the men began with 100,000 members and the women began with 10,000 members, then the men should have roughly 100 men over 2800 and the women should have about 10 members over 2800.

Now I hope you will address the questions I asked about the CFC rating system. I'm especially interested in what range of numbers the program will accept for the initial average rating of the group, and if it would accept some ridiculous value like 1,000,000, what are the consequences?

I'd like to add a few more questions: where did this provisional rating "formula" come from? Who decided wins should add 400 points and losses should subtract 400 points?

I wonder if you noticed that even if the provisional ratings had been just as you expected, the group as a whole lost rating points. When everyone began with 800 rating , total rating points were 6400. When the tournament ended, even with your expected results the total rating points only totaled 6253.

The tournament leader could have taken byes after her first win and still ended up with the same rating, 1200. That means her 2nd, 3rd, 4th and 5th wins added 0 to her rating. So the whole provisional rating "formula' goes against what the Elo rating system is all about. You would need to use a different number than 400 for adding and subtracting in order to not lose or gain rating points.

[Erik Malmsten]

where did this provisional rating "formula" come from? Who decided wins should add 400 points and losses should subtract 400 points?
.

When I first learned how to calculate ratings (by hand) in 1972 the 400 rating limit was already established. It assumes that the player rated 400 points above will win 100% of the time. A winner got 16 rating points for a win plus 1 point for every 25 rating points of his higher rated opponent up to a maximum of 16 (at 400). The system has had some adjustments since then including bonus points and once giving everyone 100 points. Always fighting deflation.

Masters CFC rating are often over 100 points higher than their Fide, perhaps due to many more CFC rated games being played and rolling over weaker opposition. But in our small pool of players there was a rating ceiling. To get a higher Fide rating they needed to play outside the country to meet higher rated Grandmasters.

[Peter McKillop]

..... You said you would answer all questions. .....

I haven't bothered to read the whole thread. It's my style to jump into the middle of a discussion with information that may be irrelevant or redundant. With that in mind, here is a link which may (or may not) be helpful:

https://en.m.wikipedia.org/wiki/Elo_rating_system

[Pargat Perrer]

When I first learned how to calculate ratings (by hand) in 1972 the 400 rating limit was already established. It assumes that the player rated 400 points above will win 100% of the time. A winner got 16 rating points for a win plus 1 point for every 25 rating points of his higher rated opponent up to a maximum of 16 (at 400). The system has had some adjustments since then including bonus points and once giving everyone 100 points. Always fighting deflation.

Masters CFC rating are often over 100 points higher than their Fide, perhaps due to many more CFC rated games being played and rolling over weaker opposition. But in our small pool of players there was a rating ceiling. To get a higher Fide rating they needed to play outside the country to meet higher rated Grandmasters.

I"m assuming Erik that when you wrote, "But in our small pool of players..." you really meant "But in our small pool of FIDE-rated players...."

In the times you are mentioning, FIDE was very restrictive about who could join FIDE and get a FIDE rating. I don't know the actual restrictions, but yes, the result was that a country like Canada would have had very few FIDE-rated players.

Also, back in those times learning was more limited because there were no computer chess engines to evaluate positions or to play against. Even endgame knowledge was more limited, since endgame tablebases didn't exist or were very limited in extent.

I think this may be why Bob Armstrong is so entrenched in believing that women players must play men in order to improve their ratings. He doesn't realize that women don't need men in order to learn and advance when tools like Stockfish and Fritz are available, and also the pool of rated women in FIDE is now quite large and growing all the time.

[Pargat Perrer]

I haven't bothered to read the whole thread. It's my style to jump into the middle of a discussion with information that may be irrelevant or redundant. With that in mind, here is a link which may (or may not) be helpful:

https://en.m.wikipedia.org/wiki/Elo_rating_system

I wasn't talking about basic questions on Elo rating system. But in going to your link, I did scroll way down and came across the section on Performance Rating.

And here is the interesting thing about this rating, that the CFC uses when an event is held between all-unrated players:

Example: If you beat a player with an Elo rating of 1000,

If you beat two players with Elo ratings of 1000,


So you beat the first 1000-rated player, you jump to 1400.
You beat the second 1000-rated player, you stay at 1400.
If you beat a million 1000-rated players, you stay at 1400.
What a ridiculous system.
This is not how FIDE does it. They use some kind of a look-up table.

[Bob Armstrong]

Hi Pargat:

In a rating pool, studying and playing a lot does NOT increase your RATING.

What happens is that a good woman player of 2300 is stuck with her rating at the top of the pool with other women of 2300, though her strength may well be, outside of the pool, higher.

So she goes to play in an open tournament, and lo and behold, though 2300, she is beating men with 2400 ratings! She will pick up points.

Then when she returns to the women's pool, she will contribute these new additional points to the pool, and the average top rating can then rise a very little bit.

For the top woman in the woman's pool to pull away from her peers in the group, she must be exceptionally more talented than the rest at the top. Then she can raise her rating by regularly beating her top peers, and taking points from them. This is why Magnus Carlsen was able to get such a gap in his high rating, as against the ratings at the top of his peers.

Bob A

[Bob Gillanders]

I wasn't talking about basic questions on Elo rating system. But in going to your link, I did scroll way down and came across the section on Performance Rating.

And here is the interesting thing about this rating, that the CFC uses when an event is held between all-unrated players:

Example: If you beat a player with an Elo rating of 1000,

If you beat two players with Elo ratings of 1000,


So you beat the first 1000-rated player, you jump to 1400.
You beat the second 1000-rated player, you stay at 1400.
If you beat a million 1000-rated players, you stay at 1400.
What a ridiculous system.
This is not how FIDE does it. They use some kind of a look-up table.

This calculation is only relevant for the first 25 games. Once a player has played 25 rated games, his/her rating now is considered established.
Then the more standard ELO system kicks in.

[Bob Gillanders]

I wonder if you noticed that even if the provisional ratings had been just as you expected, the group as a whole lost rating points. When everyone began with 800 rating , total rating points were 6400. When the tournament ended, even with your expected results the total rating points only totaled 6253.

I did not notice that, but your comment caught my attention.
Yes, I would expect the total rating points to be 6400.

So, I gave it some thought. The cause is an effect from the byes.
If you do a weighted average of the performance rating based on number of games played,
you will see the average rating of 800 is maintained.

[Bob Gillanders]

Also, your hypothetical question is loaded because you are limiting the time period to a single year. How about 100 years? Also, you are ignoring that each group of players will grow over time, new members joining.

But I would say that the distribution of ratings should be roughly the same in each group, accounting for margin of error and assuming men and women have equal abilities and work habits in chess. But in practical terms, if we say that the odds of any player in a normal distribution group of players reaching 2850 rating are 1 in 1000 (just to make it simple), the men could see a single member get over that rating while the women most likely won't. That doesn't demonstrate anything, because if the men began with 100,000 members and the women began with 10,000 members, then the men should have roughly 100 men over 2800 and the women should have about 10 members over 2800.

Honest, it was not intended as a gotcha question!
I like hypothetical questions because you can filter out variables and focus on a single key element.
Picking 1 year was just random, it doesn't matter.
The size of the pools, it doesn't matter.

The key element is the relative size of the 2 pools.

At whatever elite rating you pick, there will be more men achieving that rating than women.
It is simply a result of mathematics. Totally divorced from the skill level of Men vs. Women.

If we have Alice and George, equal skill levels.
You may end of with result:
Alice rating = 2700, George rating = 2800.

It is simply the math.......read Bob A. posts again.

[Pargat Perrer]

Honest, it was not intended as a gotcha question!
I like hypothetical questions because you can filter out variables and focus on a single key element.
Picking 1 year was just random, it doesn't matter.
The size of the pools, it doesn't matter.

The key element is the relative size of the 2 pools.

At whatever elite rating you pick, there will be more men achieving that rating than women.
It is simply a result of mathematics. Totally divorced from the skill level of Men vs. Women.

If we have Alice and George, equal skill levels.
You may end of with result:
Alice rating = 2700, George rating = 2800.

It is simply the math.......read Bob A. posts again.

Very good, Bob G., you have reached the correct key variable. And for the case where the women pool size < < men pool size, yes, the highest woman rating will almost certainly be below the highest man rating.

This does NOT make correct Bob A.'s assertion that to improve their Elo ratings women must play the higher-rated men.

Instead, the key is to grow the women's pool to be the size of the men's pool. Then you will have the women's equivalent of Carlsen, the women's equivalent of Caruana, the women's equivalent of everyone in the men's pool, if we assume that women's chess abilities and work habits are just the same as men's.

It is a sexist view to tell women that they must play men to get better at chess, even if you are just talking about Elo ratings. It would be more correct and not sexist at all to say they should encourage more women and girls to play competitive chess, and grow the women's pool. I know that Bob A. isn't sexist, he just doesn't realize that his view is sexist. He says he is being accused for years as sexist, and it is his view that is responsible for that. Women aren't dummies.

One of the ways to grow the women's pool is to offer women titles at lower rating levels than the men. They can still earn the men's titles under the same conditions as the men, but they have an additional enticement. It is only to get more girls and women into chess, nothing else. Bob A. feels that this additional enticement isn't needed any more, and i don't argue his view there, nor do I say that that view is sexist. He could very well be correct. I really don't know on that issue.

A lot of women are indeed playing against men in mixed tournaments, have been for many years. But here's the key thing: if women and men are both improving or getting worse at chess at the same rate (obviously some individuals in each group will improve or get worse randomly, but I am taking them each as a group), then women are not going to "steal" points from the men by playing against them. The only thing that would enable women to steal points from men as a group would be if women are improving at chess AT A GREATER RATE than men.

Bob, I believe you organize tournaments, is that correct? Would you have an opinion based on your observations that young girls improve at chess faster or slower than young boys? I would almost bet that you can't objectively say one way or the other.

I just think we men need to stop telling women they must play men to get better. We only turn them off with such statements, especially when the statements are entirely false.

Now Bob G., please tell me because I'm really wanting to know:

What is the range of numbers that the CFC program you just mentioned for the performance rating will accept for the average rating of the unrated group? I'm especially interested in the maximum number it will take.

Thanks!