English Chess Forum

My understanding is that to convert an ECF rating to a USCF rating (USCF = United States Chess Federation), you multiply by 8 and add 700. Thus ECF 100 = (8*100)+700 = USCF 1500, and likewise ECF 150 = (8*150)+700 = USCF 1900, and ECF 200 = (8*200)+700 = USCF 2300.

But from personal experience it appears to me that a USCF 1900 is playing at least as strong as an ECF 160, and likewise that an ECF 150 is not playng stronger than a USCF 1800. And an ECF 190, which theoretically should be USCF 2220, is nowhere close, and is probably around 2100 strength. And so I was wondering whether I got the formula wrong, and one was supposed to add 600 rather than 700. Of course the question of conversion of ratings is always a tricky one, particularly as one is comparing two populations which mostly can't be tested against each other to empirically determine the best conversion formula.

You did get the formula wrong.

For ECF -> FIDE : x 8 + 650
For ECF -> Other National Elo: x 8 + 600

This is explained here: http://grading.bcfservices.org.uk/help.php#elo

Arshad Ali wrote:My understanding is that to convert an ECF rating to a USCF rating (USCF = United States Chess Federation), you multiply by 8 and add 700. Thus ECF 100 = (8*100)+700 = USCF 1500, and likewise ECF 150 = (8*150)+700 = USCF 1900, and ECF 200 = (8*200)+700 = USCF 2300.

But from personal experience it appears to me that a USCF 1900 is playing at least as strong as an ECF 160, and likewise that an ECF 150 is not playng stronger than a USCF 1800. And an ECF 190, which theoretically should be USCF 2220, is nowhere close, and is probably around 2100 strength. And so I was wondering whether I got the formula wrong, and one was supposed to add 600 rather than 700. Of course the question of conversion of ratings is always a tricky one, particularly as one is comparing two populations which mostly can't be tested against each other to empirically determine the best conversion formula.

Hi,

I don't think your formula is correct. According to the USCF and Prof Glickman the conversion USCF - FIDE is

USCF = ([720 + 0.625 × FIDE] if FIDE< 2000 and [−350 + 1.16 × FIDE] if FIDE> 2000]

So, given that FIDE = ECF * 8 +650 we have

ECF 100 = 1626 USCF

ECF 150 = 1876 USCF

ECF 200 = 2260 USCF

As to how accurate these are I have no idea. Stewart Reuben would be a good source of info.

Very roughly I find that USCF appear to be about 100 points higher than FIDE ratings.

Hmmm, well, what Alex posted seems plausible:

USCF = (ECF*8) + 600.

In the case of conversion between FIDE and ECF, there is some empirical corroboration at the higher levels (as I presume there are many players who have both an ECF and FIDE rating, and if something was seriously out of whack in the formula it would reveal itself in the disparity of ECF and FIDE ratings of those who have both). But the empirical relationship between ECF and USCF is more tenuous. Again, there is a large group of players -- also concentrated at the higher levels -- who have both a USCF and FIDE rating. Those players can be compared with those who have both an ECF and FIDE to determine the fit of the formula between ECF and USCF at those higher levels. But even there it's a weaker and more tenuous empirical fit. At lower levels -- as far as I know -- there may be no empirical fit at all and we're just playing a guessing game.

Alex Holowczak wrote:You did get the formula wrong.

For ECF -> FIDE : x 8 + 650
For ECF -> Other National Elo: x 8 + 600

That's only appropriate for USCF ratings if you presume that FIDE ratings are inflated relative to USCF ones. It used to be the case that USCF ratings were about 50 points above the international ones.

Here's the USCF top 10 national (April 2010)

1 Nakamura, Hikaru (12641216) NY USA 2817
2 Kamsky, Gata (12528459) NY USA 2771
3 Onischuk, Alexander (12625186) VA USA 2765
4 Seirawan, Yasser (10509459) WA USA 2718
5 Akobian, Varuzhan (12740522) CA USA 2671
6 Ehlvest, Jaan (12514557) NY USA 2669
7 Shulman, Yury (12741541) IL USA 2668
8 Christiansen, Larry M (10460921) MA USA 2657
Hess, Robert L (12749774) NY USA 2657
10 Shabalov, Alexander (12544264) PA USA 2643
Lenderman, Aleksandr (12787646) NY USA 2643

and the top 10 under international FIDE (1st May 2010)


# Name Title Fed Rating G B-Year
1 Nakamura, Hikaru g USA 2733 4 1987
2 Kamsky, Gata g USA 2702 9 1974
3 Onischuk, Alexander g USA 2699 9 1975
4 Seirawan, Yasser g USA 2644 3 1960
5 Shulman, Yuri g USA 2613 9 1975
6 Akobian, Varuzhan g USA 2599 7 1983
7 Lenderman, Aleksandr m USA 2598 18 1989
8 Ehlvest, Jaan g USA 2591 19 1962
9 Hess, Robert L g USA 2590 8 1991
10 Shabalov, Alexander g USA 2585 14 1967

At the top of the list, it looks like the USCF ratings are still above their international counterparts by 50-70 points.

Lower down though, you may have the effect that last year's recalculations have left English players with grades that are flattering both historically and by international comparison.

As another benchmark, I looked up a former club colleague who is currently working and living in the States. His last ECF grade was 196 a couple of years ago, but he's a "200-standard" player. His (inactive) international rating is 2205 but his USCF is 2274.

Both USCF and international work on the premise that a difference of 200 points is about 25% expressed as a result, so any conversion between Elo based systems that isn't rating A = rating B + constant indicates that either or both rating systems have been distorted over the rating range.

Arshad Ali wrote:My understanding is that to convert an ECF rating to a USCF rating (USCF = United States Chess Federation), you multiply by 8 and add 700.

This used to be the formula, before the ECF increased most players' grades a year ago. For other foreign national ratings, the formula used to add 600 points instead of 700. As Alex Holowcak points out, only one conversion formula is now given on the ECF website, instead of the two there used to be. Perhaps someone from the grading team can tell us whether USCF ratings are now treated the same as other foreign national ratings, or whether this is an omission from the ECF website.

Ian Thompson wrote:As Alex Holowcak points out

Thanks Ian Thomson.

Roger de Coverly wrote:It used to be the case that USCF ratings were about 50 points above the international ones.

Yes, quite, this is a rough rule-of-thumb -- empirically corroborated -- of several years' standing. As you indicate later, USCF is about 50-70 points above the corresponding FIDE rating.

Lower down though, you may have the effect that last year's recalculations have left English players with grades that are flattering both historically and by international comparison.

That's my impression.

Both USCF and international work on the premise that a difference of 200 points is about 25% expressed as a result, so any conversion between Elo based systems that isn't rating A = rating B + constant indicates that either or both rating systems have been distorted over the rating range.

Not sure what that means. That 25% of the difference is to be expected as the usual statistical fluctuation in a player's rating performance, and so can be ignored? Okay so far. The rest is unclear, particularly the distortion part.

Postscript: No, that's a wrong interpretation on my part. You mean that to a small extent the difference can be corrected by adding some constant. Again, the rest is unclear. One would need to read a lengthy exposition to get a fuller idea about this.

Arshad Ali wrote:Not sure what that means. That 25% of the difference is to be expected as the usual statistical fluctuation in a player's rating performance, and so can be ignored? Okay so far. The rest is unclear, particularly the distortion part.

In the Elo systems of rating, a 200 difference in rating is supposed to translate into a 75% score for a sufficiently large number of games. So 1800 rated players should score about 25% against 2000 rated players.

If you use a formula (as quoted above by Sean) which doesn't preserve these differences across rating systems, then you are in effect saying that according to rating system A, players will score 25% against players 200 points above them, but if the same players are converted to use rating system B, then the expected score changes.

Of course when you plot the ratings of players who have ratings in both systems and solve for a conversion formula, you usually end up with a non-linear best fit.

There's a long list of things which have the potential to distort rating lists, the estimation processes for new players, the speed of improvement of players against the frequency of rating lists, the rules for qualification for the rating list, the rules of swiss system pairings, rules imposing rating floors, rules not removing points for tournament winners etc. As national and international systems will have different policies on these you end up with different relative ratings even for the same set of games.

You end up with a rating system where the 2000 players score 75% against the 1800 players and the 1800 players 75% against the 1600 players, but the 2000 players don't score 95% against the 1650 players (or the other way round)

Roger de Coverly wrote: In the Elo systems of rating, a 200 difference in rating is supposed to translate into a 75% score for a sufficiently large number of games. So 1800 rated players should score about 25% against 2000 rated players.

If you use a formula (as quoted above by Sean) which doesn't preserve these differences across rating systems, then you are in effect saying that according to rating system A, players will score 25% against players 200 points above them, but if the same players are converted to use rating system B, then the expected score changes.

Bueno. Gracias, senhor.

ICC did a monster study on this, with thousands of submissions

Ben

Alex Holowczak wrote:

Ian Thompson wrote:As Alex Holowcak points out

Thanks Ian Thomson.

Alex, it's frustrating, isn't it? My name seems to cause almost as many problems.

Ilyumzhinov is a slightly different case. When he first became FIDE President, he didn't spell his name in the Latin alphabet the same way as he does now. (From memory, it was Iljumzhinov, with a "j" rather than a "y".)