Stats

January 14th 2008

Gwent versus the rest

I have 1287 results, excluding games involving unrated players.
681 are against outsiders, 606 against other Gwent players.

Results have been around 50 Elo points better than expected,
except against the English. These figures don't seem to change very much.

Against the rest of Wales - 301 games and plus 50 Elo.
including East Glamorgan 245 games, plus 48

The rest of the world, (excluding England) - 180 games, plus 55 Elo points.

However, they have performed 38 points worse than expected against
ECF opponents. That is, using the ancient formula to convert the grades.
It worked fine thirty years ago, but nobody has checked it since then.
Perhaps we should multiply by 8 and add 700, instead of 600.

Overall, performance is plus 26 for 681 games.
The net result is a flow of points into the Gwent grading pool.
This is the main reason why the average Gwent rating has increased by 12 points
since the start of the season. The system is correcting itself, but slowly.

It is interesting to work out who is gaining the points.
The following graph shows cumulative points, with the games in grade order.
The hillside indicates it is mainly players in the 1650-1950 range.
On the right there is a little castle showing good work by the under-1240s.

How many games have people played?

Here is a graph of the number of games people have played:

There is a long tail of players with very few results.
It seems possible the WCU could lose them altogether.
I hope they are not just victims of best-available-team policies.

Two rating lists a year?

I would like to see this. One period could end at Christmas, the other
after the KO Cup Finals.

It might even save some time, as the results files would be less cumbersome.

A new system?

It is not clear whether your Welsh rating is supposed to assess your
play during the season; or your strength at the end of it.

Certainly the final batch tends to be the most important.
I would prefer a system in which all games are given equal weight.

There are many different versions of the Elo System.
The one used in Scotland is, I think, more accurate than the Welsh.
It is also simpler, and games don't have to be in date order.
I could automate the League.

Megan Owens has completed 3 batches.
She has performed overall about 25 points better than her grade.
So her "Scotch" rating has gone from 1591 to 1616.
Notice that her grade has tended to stabilise.

In the Welsh System she had a good first batch so went up from 1591 to 1649.
This counted against her during a poor second batch (G3=1536)
Now her low rating was an asset, and it climbed to 1662.
Are these changes really justified?

The Scottish system seems to make little difference in other cases.
Research is continuing