Starting Grade

[Kevin D Plant]

I can not find anywhere how the ECF work out a starting grade for a ungraded player. The only info I can find is that they go by the games they have played in a season or games they have played in previous years.

Does anyone know what formula they use

[Gavin Strachan]

Basically a new player has a gestimated grade (as they have no games under their belt a figure cannot be ascertained). This gestimated grade is what they or other people think they are in strength from casual play having played a range of strength opponents. If they know the Najdorf Sicilian 30 moves deep usually means they are pretty good, if they say "how do these bishop thingys move again" usually means they are a beginner!

By the end of the season if the player has played 30+ games these are calculated in the normal manner (+50 for a win + opponents grade, -50 for a loss+ opponents grade, nothing for a draw except the opponents grade, +-40 points from the gestimated grade is the grading boundary).

If the person plays less than 30 games but more than 7 in a season then they will have an estimated grade.

The problem is that there tends to be a tremendous flux in the person's grade for a few seasons before it settles (they get better or are a lot better than the gestimated grade or the gestimated grade is just too high).

Without looking I can't remember off the top of my head that the opponent's the new player plays have to wait until the end of the season and use the new grade or just use the gestimated, i think the latter.

[Greg Breed]

Without looking I can't remember off the top of my head that the opponent's the new player plays have to wait until the end of the season and use the new grade or just use the gestimated, i think the latter.

It's the former.
A new player does have an estimated grade, but if they play enough rated games in a season then their grade is calculated at the end and the opponents get theirs retrospectively.

[Richard Haddrell]

Gavin hasn't got this quite right. Nobody guesstimates - it's not as crude as that. The system calculates a "starting grade" which is consistent with the player's results in the current season (together with any results he may have had in the two previous seasons). All ungraded players are done in the same operation. To oversimplify a bit -

Stage 1 is to calculate a "grade" for each ungraded player on his games against graded opponents. The 40-point rule is not used. If all his opponents are graded, it stops there and the result will be used as his starting grade.

Stage 2 brings in games between the ungraded players. Once again the 40-point rule is not used. The players are "graded" on all their games, using as starting grades the figures obtained from Stage 1.

The resulting "grades" will not be very accurate. So they are fed in again as new starting grades, and Stage 2 is repeated. This continues, with increasing accuracy each time, until the figures (more or less) stop changing. What comes out is the same as the starting grade that went in, and the starting grades can be considered accurate.

These starting grades are then used in the grading proper.
__________

There is one awkward case where nothing works. A player with no graded opponents is unlikely to be a problem. He comes in at Stage 2, as you'd guessed, and picks up a starting grade from there. But he becomes a problem if none of his opponents have any graded opponents either. The system cannot handle a set of ungraded players who have played only amongst themselves. They will not be graded. Their games remain on record and are available for use next year.
__________

I said I would oversimplify, and I have. But I hope that gives a fair idea of the method.

Richard Haddrell
ECF Grading Administrator

[Kevin D Plant]

Thank you all for replying, I think i now understand how it is worked out

[Gavin Strachan]

I was sort of taking the angle of when you get a new player turning up to your club with no grade or competitive games behind his belt and you want to play him in your local league games, you can only guess his grade until he has played a sufficient amount of competitive games and then the season can be looked at retrospectively. As Richard states it is difficult to over simplify and he is the man who knows better than anyone!

[E Michael White]

Richard

If an ungraded player does not have a current LP grade but has an RP grade, isn't the RP grade used instead as an estimate for the main LP grading exercise, stages 1 and 2 being omitted ? For established players who take a season out of LP and play one or two RPs, this can cause big anomalies.

Are there any other occasions when estimates are calculated differently ?

[Gavin Strachan]

RP grades are a reflection of a persons performance in RP games only. Some people are very good at LP and not RP (and vice versa) or similar standard with both. So to use RP grades to justify a persons LP grade would not be correct or accurate as the two forms of chess are different even though it may give you a rough idea of their playing strength.

[Richard Haddrell]

If an ungraded player does not have a current LP grade but has an RP grade, isn't the RP grade used instead as an estimate for the main LP grading exercise?

Yes. I did say I was oversimplifying.

A player's grade will be used where available. The "right" sort of grade if he has one, and the "wrong" sort if he hasn't. If he has no grade, a starting grade is calculated as already described. There are no other cases in practice.

Richard Haddrell
ECF Grading Administrator

[E Michael White]

Richard

Not sure why this has caused me to sit up and take interest but I think there is probably an overall downward bias in the method used for allocating start grades which could result in low start grade for some players. This is associated with using an iterative process in combination with the ECF averaging formula and varying player activity rates.

I know you are not responsible for the theory behind the sums but do you know whether the grades of any players are fixed at any stage for the rest of the new starter process ( I know all grades are recalculated once all estimates are made )

ie :-

a) does an already graded player keep that grade throughout step 1, step 2 and each iteration or is his grade allowed to float depending on the actual results against ungraded players or the results derived during the iterative process

b) does an ungraded player who has an estimate from step 1, keep that grade after step 2 going into the iteration process and during the iteration process or is his grade allowed to float depending on the actual results against other players or the results derived during the iterative process

c) if grades of any players are allowed to float are grades of players whose grades were estimated in step1 but not included in step 2 recalculated in the iteration process

Also are there any crude or other adjustments made to the overall level of grades in each step of the iteration process or elsewhere, eg keeping the mean or a weighted mean of grades constant

[Richard Haddrell]

Richard

...do you know whether the grades of any players are fixed at any stage for the rest of the new starter process...

Michael,

(a) Published grades are not changed.
(b) Estimates created in the course of calculation are fluid till the iterations are complete.
(c) These are players who have no ungraded opponents. Their initial estimates don't change.

No adjustments are made beyond the calculations already described.

Richard Haddrell
ECF Grading Administrator

[E Michael White]

Richard + anyone interested

Thanks for your reply. Could you clear up 4 questions as I would like to know the basis for certain. Others may be interested as grading items seem to be viewed more than most other forum items.

a) When you say no 40 point rule applies for new starter grades, is a 50 point rule or similar used instead, either in stage 1 or stage 2 ? I expect there is not but if there is no similar rule, players could gain/lose points by losing/winning against much stronger/weaker players.

b) For example an ungraded player plays a player graded 100. If the two players move to more than 50 points apart during one stage of the iterative process does a 50 point rule or any other apply ?

c) Another ungraded player only plays 10 players graded 100, scoring 100% and does not go into the iterative process. He would get a starting grade of 150. You suggest in one posting that this starting grade is fed back into the grading proper; his graded opponents would then get 90 for those games as the 40 point rules applies in the grading proper. But what about his own grade. Is this recalculated in the grading proper using the start grade of 150 giving him a next years grade of 160 as the 40 point rule applies ?

d) Do you know when an iterative process was first used in the calculation of grades for ungraded players ?

Using an iterative process with the ECF grading formula will not usually give totally meaningful results. This is due to the ECF formula not being zero sum (contrary to what is written on Wikipedia). In any one result the total changes to grades of the players is not usually zero as shown in this example:-

Player A who plays 30 games and is currently graded 150, plays against player B who is currently graded 130 and plays only 10 games. If A wins, the change in his grade in respect of this result is (180-150)/30 = +1.0 and B's change is (100-130)/10 = - 3.0. The total loss to the grading system is 2 points. Similarly a draw/Bwin gives +1.33/+4.66. The only occasions when the results are zero sum, are when either the number of games played are the same or the players perform exactly as their grade.

So when a more active player performs above their grade some deflation arises. They do not have to be a stronger player as long as they are more active and overperform. This deflation is internal to the formula and is in addition to deflation in grades due to a player having to play an undergraded player. Players seem to agree that since about 1970 some grades appear to be deflated and what has changed most since 1970 is the number of weekend tournaments causing much wider variation in player activity rates. When players are having a good season ie playing above their grade, the structure of weekend tournaments encourages many to enter more events making this deflationary scenario more likely.

[Richard Haddrell]

Michael,

(a) No. No 50-point rule, no 75-point rule, no anything-point rule.

(b) See (a).

(c) I did indeed suggest, in fact I said, that the calculated starting grade is used in the grading proper. It is. Yes, this player will come out at 160. It isn't very satisfactory, but what method is satisfactory with 100% scores?

(d) No. I'm not an expert on these things. I was using one myself in the 1980s, but I won't have been the first.

I'm not a statistician either. But it is known that the mean grade can fluctuate for the reasons you give, with a downward trend if improving players play most games. Perhaps there are other factors? We've got the experts looking at things like deflation, as you know.

Richard Haddrell