mchristie

Thanks for the positive feedback, to those that have given it, thanks to Gnasher for reading the Guide so fully and explaining it here. Some more thoughts from the developers... Improved data on web pages? Yes it'll come sometime, but is dependent on other unrelated EBU software changes. More emphasis on partnership grades? Yes, this is again a display (and search) problem. I'd like to show all partnerships where at least one player belongs to county X, and maybe all partnerships (with over 300 graded boards in the last 3 years) for player P. There are boring software constraints that make these more work to implement than they should be. Non-linearity of partnership performance... How well should we expect a 64/48 partnership to perform, if not half way between the two grades? I doubt anyone knows, and doubt we could ever get enough data to test any theory. I wouldn't worry about pairs who are so strong/weak their score in a session should be over 70% or under 30%; they are rare and will probably get a realistic grade when they play in a session more suited to their strength. Pick-up partnerships play worse than regular ones. There is now enough data to quantify this, if we had time to analyse it. An ad-hoc rule that reduced the expected score by x% for a partnership that had played at most 1 prevoius session in the last 3 years would work and not be hard to implement. The real issue with the A,B,C problem mentioned recently is when A stops playing with C, and FOREVER AFTER will have a lower grade than B. This feels wrong for a system that is supposed to reflect current playing strength, and would need another ad-hoc rule to fix it. The rule would be "if you nearly always play with B your grade is pulled towards B's grade". All ad-hoc solutions have downsides, and currently our principle is to keep the scheme as mathematically clean as practically possible, so no ad-hoc fixes in the foreseeable future. That's all for now folks.

That's what I thought originally, but in a head-to-head match the four players at THE ONLY OTHER table have an equal impact on the result, so such a result affects eight players, thereby contributing a second 2:1 factor. Once a board is played at several other tables the effect of the other tables' players rapidly diminishes. I think it's (n/n-1) if the board is played n times.

Yes, we are slowly working towards using teams events, for events such as SwissTeams where the boards are played multiple times. They will be graded by looking at the performance of individual pairs. For individual team of four matches, yes there are problems about data collection which could probably be overcome with a lot of effort, but the biggest issue is that there is only one result and it affects equally all eight players. That makes each board statistically worth only 1/4 of a dupliacte pairs board, so a 48 board match would be given as much weight as a 12 board pairs session. But..., the biggest problem is "systematic errors". The NGS suffers a bit from this, if a player is over or undergraded, this affects their partner's grade, and leads to people complaining about the system. For head to head team matches, you grade is susceptible to errors in any of the other seven players. I suspect that the inclusion of such events would simply make individual grades less reliable.

Actually, currently newcomers enter on a grade of 46.00. We monitor the average performance of the approximately 100 new players joining the system per week to see what their average strangth is. Over the last few months this has been around 45-46%. If it drops any lower, we'll reduce the newcomers' ranking to match. Zia has not yet played enough rated games to get a published grade, and you may feel it was lucky for his partners to be playing with such a skillful newcomer with an initial rating of 46, but it didn't seem right to make any exceptions. Should we?

At both MPs and IMPs some boards are more variable (wilder) than others which are flatter, and there is less opportunity to score well/badly on the flatter boards. We know that there are comparatively few extremely significant boards at IMPs whereas at MPs it is the extremely flat boards that are less common. What we compare when converting IMP scores to MP% equivalents is the average variability of IMP boards and the average variability of MP boards, based on actual data from a large set of boards, so that we CAN say the average variability of a bunch of MP boards is equal to the average variability of a bunch of IMP boards. Stephanie, I think this means we treat IMP pairs events fairly. It certainly isn't meant to be in the group of small effects that we ignore.

Just to say that, due to a display quirk, (well, ok, a bug) it is somewhat random whether you will see your NGS result for any multi-section pairs event, but it will have been scored, and if so it will affect your grade. Hopefully the display will be corrected later this month. Various IMP scored results, especially in Oxfordshire, have not been graded due to a bug in the reporting of these results by the scoring system. IMP results are compared to MP results and the conversion factor is essentially the ratio of the observed variances of IMP and MP scores for fields of equal players. (That doesn't read very clearly, sorry.) I'd like to use our large database of both MP and IMP events to refine these observations. Some day, year...

I'm Mike, the team leader for the NGS and have just found this thread. Interesting comments. I'll try to give some answers. 1) Andy Bowles now restored to grade of 60+. 2) Grade accuracy standard error of at best 2%, but you need a good mix of partners, and see below for caveats. 3) Biggest limitation: Ann usually plays with Bob and they score 50% (v average opps), but occasionally Bob plays with Dave and they score 45%? Who's the strong player? No grading system can know. NGS will (remarkably quickly) decide that Ann is the strongest getting a grade probably near 55 (but depends on starting conditions). From our unpublished analysis of playing patterns, we think this and similar problems significantly affect the grade of between 2 and 5% of the EBU 50,000 members. (based on Yorkshire test bleow) 4) We have lots of unpublished maths behind the NGS, for example: test of more complex estimation(Kalman filters/Least Squares estimatation) for the 4000 Yorkshire members over six months, which would give each player a personal grade accuracy as well as grade, but we rejected as it means your grade would change when you don't play, and takes very much more computation. measures of degree of mixing between clubs (around 50 clubs are sufficiently isolated that it will take over 5 years for their grades to be nationally standardised.) mathematical behaviour of the discrete filter that the NGS system uses. etc, etc. 5) NGS does use IMP scored pairs events not just MP. 6) Want to use pair-wise scores of teams events sometime in the future, but we need 100% reliable data on which players were which pair for each match of an event. 7) National/Regional events are weighted threefold. Thsi is primarily to assist in standardising grade averages between clubs ("diffusion"), but has the side effect that for strong tournament players, their grade is more dependent on these events than local club duplicate sessions. So I hope that in a couple of years, many strong players will get a more accurate grade, but I agree it would help if we could use at least some team events for such players. Some professional players suffer from the Dave effect (above) more than others. 8) Partnership grades are more meaningful, but players want individual grades. We're trying to improve the way P'ship grades can be displayed and searched. If anyone is interested, and has maths or software backgroud, we'd welcome some volunteer assistance with improving and monitoring the system. Contact ngsqueries at ebu.co.uk