TPR Query

[John Upham]

As usual, I probably ought to know the answer to this....

On April 26th I was arbiter for the Basingstoke Rapidplay Open.

Danny Gormally played and scored 6/6 and 5/6 opponents were rated.

DGs ECF RP rating is 2591K

Danny's TPR over those six games was 2386.

Does this TPR imply this his RP rating will be reduced weighted over those six games?

[Roger de Coverly]

Does this TPR imply this his RP rating will be reduced weighted over those six games?

Rating changes using Elo methods are based around Actual Score minus Expected Score. I doubt Danny's expected score was much below 6/6 but to the extent that it was, he will gain a handful of points.

[Dragoljub Sudar]

His rating will go up.

Under the ECF system a player gets +10 for a win, before subtracting the Offset if the opponent was lower rated.

The maximum Offset is 8.4 so Danny will get +1.6 x 3 for rounds 1 to 3.

[Reg Clucas]

How is TPR calculated? Is it possible in this case that the one unrated opponent has been included in the TPR calculation with a rating of zero?

[IM Jack Rudd]

TPR is {mean of opponents' ratings} + {a number that is an increasing function of percentage score}. For the purposes of the rating system, it is irrelevant, because games are calculated individually.

[NickFaulks]

Danny's TPR over those six games was 2386.

I don't understand what that means.

[Roger de Coverly]

I don't understand what that means.

It's the number calculated by the chess-results software.

https://chess-results.com/tnr1136046.as ... =YES&snr=1

So the question really is how chess-results calculates it.

Stephen Jones, who only lost to Gormally came out at 1980.

[NickFaulks]

So the question really is how chess-results calculates it.

I stopped bothering about that years ago. I just know to ignore it.

[Colin Purdon]

There is a reasonable, but not detailed, explanation of different types of TPR calculation methods here: https://www.hirmulintu.fi/juha/chess/en ... ator.shtml (the colour scheme is less reasonable, imho).

If you plug in Steven Jones' results into the site, this site agrees with chess-results' 1980 figure for what it calls "Performance rating (FIDE)". I can get agreement for Danny Gormally's results with the chess-results figure if I change the 0 rating to 1000. The site also mentions a more simply calculated "Performance rating (linear)" value, saying that it is popular in Finland.

The ECF website provides another TPR calculator here: https://www.englishchess.org.uk/wp-cont ... rPage.html which gives yet another figure. However, I believe it is calculating what the Finnish site calls simply "Performance rating" which is calculated to find "such a rating that the expected score equals to the actual score". The figures for this on the ECF and Finnish sites are quite close, I suspect that they are simply using different numbers of iterations to calculate the value, but I couldn't say for sure.

Anyway, the takeaway for me is that there are different TPR interpretations out there, which mean different things.

[Paul Habershon]

At my club we run a 'challenge anyone any time' Rapidplay (25m+5s) throughout the season to give members a meaningful game when they don't have a match or a club championship game. Results are sent to the ECF. A minimum of ten games is required for consideration in the final standings, which are based on the TPR.

My query is in the calculation of the TPR. The ECF LMS gives a formula for it in league matches:
TPR=(sum(opponents' ratings)+400(wins-losses))/games

What if the ratings are more than 400 apart, e.g. a 1388 losing to a 2035? Does the 1388 get 1788 in his list of opponents' ratings? I seem to remember under the old system that if ratings were more than 50 apart they were considered to be 40 apart, so perhaps that 1788 should be more like 1738.

I am using a calculator for this, so I don't want any abstruse arithmetic or computer programs. I hope there is a simple answer, even if it's rough and ready.

[Ian Thompson]

What if the ratings are more than 400 apart, e.g. a 1388 losing to a 2035? Does the 1388 get 1788 in his list of opponents' ratings? I seem to remember under the old system that if ratings were more than 50 apart they were considered to be 40 apart, so perhaps that 1788 should be more like 1738.

Not quite. If they were more than 40 apart, not 50, they were treated as being 40 apart.

[Paul Habershon]

What if the ratings are more than 400 apart, e.g. a 1388 losing to a 2035? Does the 1388 get 1788 in his list of opponents' ratings? I seem to remember under the old system that if ratings were more than 50 apart they were considered to be 40 apart, so perhaps that 1788 should be more like 1738.

Not quite. If they were more than 40 apart, not 50, they were treated as being 40 apart.

So if a 100 fluked a win v a 200 they would get a score of 190 (140+50) and not the full credit of 250? Or would it be 210 by first counting the 100 as 160 to be forty away from the 200? However, I can see the point of the 100 not gaining points for a loss. Presumably the 100 would then score 90 instead of 150?

Proportionately, in answer to my original query, I suppose that if ratings are more than 320 apart they should be treated as 320 apart. I don't think the LMS formula allows for that. (40 is four fifths of 50 and 320 is four fifths of 400).

[Ian Thompson]

So if a 100 fluked a win v a 200 they would get a score of 190 (140+50) and not the full credit of 250? Or would it be 210 by first counting the 100 as 160 to be forty away from the 200? However, I can see the point of the 100 not gaining points for a loss. Presumably the 100 would then score 90 instead of 150?

100 v 200 game - 100 is treated as if they'd played someone graded 140; 200 is treated as if they'd played someone graded 160.

For the 100, win, draw, loss would score 190, 140, 90 points.

For the 200, win, draw, loss would score 210, 160, 110 points.

[Roger de Coverly]

My query is in the calculation of the TPR. The ECF LMS gives a formula for it in league matches:
TPR=(sum(opponents' ratings)+400(wins-losses))/games

That formula is sometimes known as the "rule of 400". It's equivalent to the old Clarke system if you take the view that Clarke 8 = Elo 400. If you use this formula and don't want to penalise players for winning, then it's necessary to restrict rating differences. Hence the BCF's 40 point rule which would translate to 320 Elo. You could use this, or narrow or widen the difference.