Tuesday, May 03, 2011

On grades and grading II

[Due yesterday, but postponed for Ray Tweets the Wedding]

Another quiet day at work, a bit more time on my hands and, hey presto, another post based on a session with the calculator emerges. This one starts with a story. Are you sitting comfortably? Then we'll begin.

A month or two back I was talking with T.C. about how my season was progressing and I mentioned that while I was doing very badly everywhere else, I was at least scoring pretty well in the London League. Tom suggested that this could be something to do with Golden Lane - I feel comfortable there, so play better, so score more points - to which I replied that while it was true that I like playing at the London League's usual venue, I was also doing well against clubs like Athenaeum, Albany and Drunken Knights who play their home matches elsewhere.

That little exchange encapsulates the human condition pretty well, I think. We are built to create stories and to (try to) explain the world around us, but simplifying a complex set of circumstances into something more understandable is an error-prone activity because our perception processes are erratic - we pay attention to some things more than others - and we remember some of the things that we take in better than others.

Perhaps for that reason, the narrative that I'd constructed to explain my chess season was no more accurate than T.C.'s. I thought I was scoring well at Golden Lane, but when I crunched the numbers for last week's post I saw that my performance there was actually no better than par for the season. I only really did well in the London League at clubs that don't play at the central venue.

The +5 =1 -0 I scored at Albany, Drunken Knights, Athenaeum, Morley College, London Deaf and Civil Service highlights another major fault we have as storytellers: we tend to grossly underestimate chance as a determining factor. We like to believe that things happen for a reason so when events come together in a certain way we tend to think that 'x causes y' and totally overlook the fact that randomness is lumpy and not at all evenly spaced as we feel it should be.

After all, what possible reason could there be for my 90% score in non Golden Lane London League games this year? Two of the venues I actively dislike so it's certainly not anything to do with the playing conditions. No, much more probable is that it was just a slice of luck here, the absence of misfortune there and suddenly there I am with a 200+ grading performance over six games.

As the Gallagher brothers once observed, nobody ever mentions the weather can make or break your day. Games of chess can turn on similarly small points and sometimes, we would do well to remember, half-a-dozen good results just happen to come along all at once.

So with all that in mind, it's worth taking another look at the figures that came out of last Monday's On grades and grading and the stories that they seem to tell. In terms of grading outcome it appears that I,

  • perform much better when playing White than when playing Black;
  • do hugely better playing Black against queen's pawn openings than when facing king's pawn openings;
  • score well in the London League, but under-perform elsewhere (Surrey, Croydon, Thames Valley and Banks' Leagues)
  • enjoy a small but appreciable increase in performance when facing higher-graded opposition compared to games against lower-graded opponents.
  • do exceptionally badly in games that are concluded with a quickplay finish.

Is any of that really true though? The figures are fairly clear, but the sample sizes are generally pretty small - often only four or five games - and therefore vulnerable to chance factors skewing the outcomes.

I decided to find out if these stories were as true for the 2009/10 season as they were for 2010/11. To a fair extent they were, although with a bit more data it was possible to refine my chessplayer story somewhat:-

  • Yes, I do play much better with White than Black. Presumably, though, everybody does. Unanswered question: What would be a 'normal' differential?
  • Yes, I do better with Black against 1.d4 than 1.e4, although not as much as this season's games would suggest.
  • Yes, I do under-perform in the Surrey and Croydon Leagues although last year my London League games were not exceptionally successful for me (I did much better in County games and a couple of tournaments)
  • Yes, I do better grading-wise when I play higher strength opposition, but only when I'm White. When I'm Black, strength of opposition doesn't seem to be especially relevant.
  • Yes, I do worse in games played to a finish compared to those that are adjourned, but this year's abysmal results are much much worse than 'normal'.

And the moral of today's story?  Well, aside from the reservation that even two years' results are still not totally immune from a chance swing here or there (in a 20-game sample a single loss turned into a win or vice versa leads to a 5 point difference in grading performance), it does seem that I have a few areas that are worth investigating as potential causes of lost grading points.

Playing Black against 1.e4, playing outside of the London League, quickplay finishes - improvement in any of these areas could lead to a healthy gain in playing strength overall.  All I need to do now is work out why they might be problem areas for me and then come up with a some solutions.  Not exactly a piece of cake, but, now I've looked at the numbers, at least I can be more confident that the story I'm telling myself is a reasonably accurate reflection of reality.

This, apparently

JMGB 2009/2010 and 2010/2011

2009/10: 163.1 (35 games)*
2010/11: 166.8 (27 games)**
09 - 11: 164.7 (62 games)

Ron Harman Method (top and bottom 10% excluded):
2009/10: 165.6 (29)
2010/11: 166.1 (21)
09 - 11: 165.8 (50)


Playing White:
2009/10: 174.8 (20)
2010/11: 182.3 (12)
09 - 11: 177.6 (32)

Playing Black:
2009/10: 147.2 (15)
2010/11: 154.5 (15)
09 - 11: 150.9 (30)

Playing Black, King's pawn:
2009/10: 136.5 (8)
2010/11: 147.1 (9)
09 - 11: 142.1 (17)

Playing Black, Queen's pawn:
2009/10: 155.5 (6)
2010/11: 190.3 (4)
09 - 11: 169.4 (10)


London League:
2009/10: 162.1 (10)
2010/11: 181.3 (16)
09 - 11: 173.9 (26)

non London League:
2009/10: 144.9 (9)
2010/11: 146.1 (11)
09 - 11: 145.6 (19)


London League, White:
2009/10: 170.0 (7)
2010/11: 184.0 (7)
09 - 11: 177.0 (14)

London League, Black:
2009/10: 130.5 (2)
2010/11: 179.0 (9)
09 - 11: 170.2 (10)

non London League, White:
2009/10: 159.8 (4)
2010/11: 179.8 (5)
09 - 11: 170.9 (9)

non London League, Black:
2009/10: 133.8 (5)
2010/11: 117.8 (6)
09 - 11: 125.1 (11)


Higher-graded opposition:
2009/10: 170.1 (10)
2010/11: 168.9 (12)
09 - 11: 169.5 (22)

Opponents graded 0-9 points more:
2009/10: 170.4 (5)
2010/11: 168.8 (4)
09 - 11: 169.7 (9)

Opponents graded 10-19 points more:
2009/10: 169.8 (5)
2010/11: 168.0 (5)
09 - 11: 168.9 (10)

Opponents graded >= 20 points more:
2009/10: -
2010/11: 170.7 (3)
09 - 11: 170.7 (3)

Lower-graded opposition:
2009/10: 158.7 (15)
2010/11: 165.4 (15)
09 - 11: 162.0 (30)

Opponents graded 0-9 points less:
2009/10: 161.6 (11)
2010/11: 187.4 (5)
09 - 11: 169.7 (16)

Opponents graded 10-19 points less:
2009/10: 163.4 (10)
2010/11: 138.0 (4)
09 - 11: 156.1 (14)

Opponents graded >= 20 points less:
2009/10: 150.5 (4)
2010/11: 154.1 (6)
09 - 11: 152.7 (10)

Higher-graded opposition, White:
2009/10: 182.3 (6)
2010/11: 196.2 (5)
09 - 11: 188.9 (11)

Higher-graded opposition, Black:
2009/10: 151.0 (4)
2010/11: 149.4 (7)
09 - 11: 150.0 (11)

Lower-graded opposition, White:
2009/10: 171.4 (14)
2010/11: 172.3 (7)
09 - 11: 171.7 (21)

Lower-graded opposition, Black:
2009/10: 145.9 (11)
2010/11: 160.0 (8)
09 - 11: 151.8 (19)


2009/10: 154.3 (9)
2010/11: 118.2 (6)
09 - 11: 139.9 (15)

non Quickplay:
2009/10: 166.1 (26)
2010/11: 180.9 (21)
09 - 11: 172.7 (47)

* The ECF say I played 34 games last year.  I have included a win against a person who remains ungraded - presumably because he played too few games - estimating his grade based on those of the people who played on the boards around him.

**Today's figures for 2010/11 slightly different to last week's because I've had one more game since then.


kingscrusher said...

Hi Guys

I know Tom Chivers from Chessworld - and I was just wondering if I could do a video on your positional sacrifice series - using much of your content (with credit to your article), but adding my own thoughts and commentary- and for the video title to clearly give the link to this blog - and a link in the description?! I am very interested in positional sacrifices, and the recent article was great.

Best wishes

Jonathan B said...

Hi Tryfon,

Thanks for your comments. Feel free to use the material as a basis for a video.

Let us know when it's done and we'll link to it here.


ed g. said...

"Unanswered question: What would be a 'normal' differential?"

Well, figure that normally White scores something like 55% against even opposition. That's around 80 Elo, which is, what, 10 ECF points or so? Very, very roughly.

Jonathan B said...

Thanks Ed. I wonder if the White win percentage varies across playing strengths. Presumably it does, and that the higher-graded you are the more it's in White's advantage?

Anyhoo, I suppose I'm exceeding the 'norm' however you define it. Whether that means I'm particularly good at playing White - or particularly poor at playing Black, I don't know though.

ed g. said...

I seem to recall that White's winning percentage doesn't change radically once you're talking about players of a reasonable standard. Anyway, if it's 53% for 150 players and 58% for GMs, I think that difference is kind of lost in the noise of the ratings system and the small data sets you're dealing with,

Basically, if teh difference between your white and black scores is 4 points, maybe you're not getting enough out of your White opening; if it's 16, maybe sonething's wrong with your defences. (I think in Elo, so I drafted that as 30 and 130, heh.)

And of course, look for alternate explanations: if your Black games against e4 (which look like a problem) are mostly quickplay finishes and theresults are similar to your other quickplays, then maybe that's the actual problem.

These stats point out places to look, to focus your attention. Now look at the games: do you fall apart towards the end of quickplay finishes? Do you get positions you hate when you face e4? Both?