# Proposed solution for ratings calc

RookRAK
Site Ideas 07 Apr '05 17:20
1. RookRAK
Out of drinks
07 Apr '05 17:202 edits
I've only been on RHP a few months. I've already seen a number of threads concerning what happens to our ratings when we lose/draw a game against someone whose rating is artificially low (usually due to a quick string of timeouts). I've already been hit by this a couple of times myself, not to mention watching jira, who had previously been rated over 2000, rip through my group in the beta bander 8x8 1400-1599.

I have a solution to propose. I'm borrowing, and expanding on, a concept that US Chess Federation uses - the "rating floor". They use this so highly skilled players can't intentionally drop their ratings then win money by entering lower sections of tournaments.

In this solution, everyone has their normal rating, but they additionally have a rating floor. The rating floor is calculated to be a fixed number of points lower than the highest rating you've ever achieved. For this discussion I'll use 200 points, as USCF does.

So, for me, my rating today is 1396. But two weeks ago before my recent losing streak I was at my highest rating to date of 1574. Therefore my rating floor is 1374, and it can never go lower. Today ironman31 is at 2459 which I think is his highest ever, so his rating floor is 2259. If the highest jira was ever rated was 2018, then his floor is 1818.

This value is used 2 ways:

1) When you complete a game, your opponents rating used is the *higher* of their rating, and their rating floor. So losing a game against a player who just dropped 1000 points doesn't hurt so much. Beating that same player rewards with points as if you've been a player rated at their rating floor - maybe not quite what you would nornally get, but it still hurts less.

2) Entry in banded tournaments again use the *higher* of their max rating in 30 days, and their rating floor. People who have been rated above 2000 would not be allowed to enter a 1400-1599 tournament.

I can anticipate the objection of "what you I really am far worse than I used to be?" For example, I get brain damage or alzheimer's, making me a worse player. While this would be the exception, we could make the ratings floor reflect the highest rating you've achieved in say, the past 2 years.

What do you think?

--rich
2. MIODude
me, not you
07 Apr '05 18:08
Now I see why your profile says you are a Babbling Psychopath ðŸ˜‰
I like the idea.. I'm sure someone will point out a flaw though.. but.. in every system, there are flaws.. current system included.
3. RookRAK
Out of drinks
07 Apr '05 18:17
Originally posted by MIODude
Now I see why your profile says you are a Babbling Psychopath ðŸ˜‰
Are you referring to the "Babbling" or the "Psychopath"? Or ... both?ðŸ˜€
4. Ragnorak
07 Apr '05 18:43
Originally posted by RookRAK
I've only been on RHP a few months. I've already seen a number of threads concerning what happens to our ratings when we lose/draw a game against someone whose rating is artificially low (usually due to a quick string of timeouts). I've already been hit by this a couple of times myself, not to mention watching jira, who had previously been rated over 2000 ...[text shortened]... t the highest rating you've achieved in say, the past 2 years.

What do you think?

--rich
Seems like a good idea. It would help prevent people who stall their losing games to get their rating as high as possible as well.

It was discussed here, ages back...

D
5. RookRAK
Out of drinks
07 Apr '05 19:01
Originally posted by Ragnorak

It was discussed here, ages back...
I did a search for the word "floor" figuring something like this must have been discussed before, but didn't get anything. Thanks for the reference to the previous thread.

My suggestion is nearly identical, with one small difference (and I think improvement).

The previous suggestion was to make use of the highest rating for 30 days. I think that is too short a period. For one thing, some players seem to be "seasonal", for a variety of reasons (loss of access to a PC, travel, alien abduction, etc). They play frequently for several months, build up a fairly high rating, then drop off the face of the each. When they return two or three or four months later, their rating is 1000 points lower. They start playing regularly again (at their true skill level), stomping inappropriate competition who have to suffer now only the loss, but an inaccurately large rating hit.

A lifetime, or multi-year ratings floor would prevent this. If ironman suddenly abandons all his games and spends 3 months mountain climbing in the Himalayas, who would want to play him when he returns and his rating is 950? You'd be ground into the chessboard *and* it would cost you 32 ratings points for the privilege.

--r
6. Ragnorak
07 Apr '05 19:35
Originally posted by RookRAK
I did a search for the word "floor" figuring something like this must have been discussed before, but didn't get anything. Thanks for the reference to the previous thread.

My suggestion is nearly identical, with one small difference (and I think improvement).

The previous suggestion was to make use of the highest rating for 30 days. I think that i ...[text shortened]... ground into the chessboard *and* it would cost you 32 ratings points for the privilege.

--r
That earlier thread was before there was even the "highest rating in 30 days" rule, so the situation wasn't quite the same as it is now.

D
7. flexmore
Quack Quack Quack !
07 Apr '05 20:53
Originally posted by RookRAK
I've only been on RHP a few months. I've already seen a number of threads concerning what happens to our ratings when we lose/draw a game against someone whose rating is artificially low (usually due to a quick string of timeouts). I've already been hit by this a couple of times myself, not to mention watching jira, who had previously been rated over 2000 ...[text shortened]... t the highest rating you've achieved in say, the past 2 years.

What do you think?

--rich
i like it, this is an improvement.

the concept could be extended:
rating for entry to banded tournies could be the highest of:
1 rating now,
2 highest rating in the last month - 50 points,
3 highest rating in last three months - 100 points,
4 highest rating in last six months - 150 points,
5 highest rating in the last year - 200 points,
6 highest rating in the last 300 games - 300 points.

i do not know how much work it is for rhp to calculate all the things though.
8. 07 Apr '05 22:17
9. 07 Apr '05 22:51
Originally posted by RookRAK
I've only been on RHP a few months. I've already seen a number of threads concerning what happens to our ratings when we lose/draw a game against someone whose rating is artificially low (usually due to a quick string of timeouts). I've already been hit by this a couple of times myself, not to mention watching jira, who had previously been rated over 2000 ...[text shortened]... t the highest rating you've achieved in say, the past 2 years.

What do you think?

--rich
Nice idea but how would you deal with the inflation in ratings this would cause?

Taking the example of a 2000 player droping to 1000 then rising back to 2000: their opponents would gain in the drop period but wouldn't hurt as much on the way back - thus putting a lot of new points into the system.

The current system aviods this by always giving to one player what it takes from the other.
10. RookRAK
Out of drinks
07 Apr '05 23:47
Originally posted by The Swine Down Hope
Nice idea but how would you deal with the inflation in ratings this would cause?

Taking the example of a 2000 player droping to 1000 then rising back to 2000: their opponents would gain in the drop period but wouldn't hurt as much on the way back - thus putting a lot of new points into the system.
That is an *excellent* point. I did not deal with ratings inflation, because I believe it's impact would be negligible.

The reason being that RHP is not a "closed" system.

When a new person shows up, we gift them 1200 points. They play some amount of games, between one or two, to thousands of games. The vast majority leave at some point (there were 24K people who made at least one move in the past 100 days, yet the player numbers are up to 148K). When someone leaves, there is a net gain or loss to the total number of points within RHP. If they came in at 1200 and played their way down to 200, the remaining RHP'ers have found 1000 points. If they played their way up to 2200, remaining RHP'ers have lost 1000 points.

I think the impact of players coming and going to RHP will outweight the impact of occassional rating floor inflation. (I wonder, are people more likely to stay active on RHP is they are above or below 1200). Whether there is significant inflation could be measured by taking the average ratings of "active" players at any one time. I can't calculate that number, but the Players Table shows the median rating of the 6391 active, non-provisional players is 1289. Is that inflation significant enough to be bad? Would it grow dramatically with a ratings floor? My intuitive guess is "no", but I really don't know.

--r
11. flexmore
Quack Quack Quack !
08 Apr '05 11:201 edit
Originally posted by The Swine Down Hope
Nice idea but how would you deal with the inflation in ratings this would cause?

Taking the example of a 2000 player droping to 1000 then rising back to 2000: their opponents would gain in the drop period but wouldn't hurt a ...[text shortened]... this by always giving to one player what it takes from the other.
no no no no no no no no no no no no no no no no no no no no

the floor would not affect their rating at all .... the obvious use of a floor is for a player's eligibilty to entrance to banded tournaments

for example: a 1962 player may drop their rating to 854 points, they could then enter the 1700-1800 banded tourney , but not enter the 700 - 1000 tournament.
12. Ragnorak
08 Apr '05 11:55
Originally posted by flexmore
no no no no no no no no no no no no no no no no no no no no

the floor would not affect their rating at all .... the obvious use of a floor is for a player's eligibilty to entrance to banded tournaments

for example: a 1962 player may drop their rating to 854 points, they could then enter the 1700-1800 banded tourney , but not enter the 700 - 1000 tournament.
I think the floor should be taken into account when working out the rating as well, as per the original suggestion.

Rating infation doesn't matter that much anyway, as its already happening.

D
13. flexmore
Quack Quack Quack !
08 Apr '05 12:01
Originally posted by Ragnorak
I think the floor should be taken into account when working out the rating as well, as per the original suggestion.

Rating infation doesn't matter that much anyway, as its already happening.

D
what do you think is driving the inflation that is already happening?
14. RookRAK
Out of drinks
08 Apr '05 13:57
Originally posted by flexmore

the floor would not affect their rating at all .... the obvious use of a floor is for a player's eligibilty to entrance to banded tournaments
No.

The *primary* use of the floor is for ratings calculations. Keeping banded tournaments reasonable is it's part-time second job.

15. Crowley
Not Aleister
13 Apr '05 15:39
The idea surely has merit, but I think it's too 'static'.

I've been here for ages and I still can't get my rating to stay above 1400.
I'm probably a about a 1350, but let's say I get one of my patented rating spikes with some fortuitious resignations etc.
I suddenly spike up to 1600 in a month or so (not impossible). My floor value will now be 1400, which would mean I'd be excluded from tournies up to a 1399 max banded rating.
If I can't get my floor value down this would mean I couldn't play in banded tournies where I belong.

What I'd like to see is a 'career chess rating' (or something like that).
This rating would have different weightings for games according to time spent in certain brackets.
This will mean that spikes, up and down, would get <1 and >1 weightings respectively in order to level out the spikes.
Use this weighted total and calculate the average - this should be a more accurate average rating for a player and will definately be better than the current 30 day highest rating value.

I'm no statistician and the formula probably needs some tweaking, but this is what I would like to see.