I’ve renamed this topic from “Problems with the design of the 2022 Washington Open” to “Tournament format discussion”.
Most flyers aren’t perfect.
By “A player” and “B players” I’m assuming you are referring to “Class A” and “Class B” (just to be clear, I was not in my post when I wrote “200 prize A” and “200 prize B”).
In your scenario, under the rules the B player with 5 would get the $200 U2000 prize, the B player with 4.5 would get the $200 U1800 prize, and the A player with 4.5 would get nothing. However, the same thing would occur if the U2000 prize was $250 so saying having both under prizes the same amount is bad due to the above example doesn’t make sense.
I understand the rule very well. However try explaining your prize distribution to players who don’t when it’s not clear on the face as to which prize someone won.
You have pinpointed the problem; a performance rating for a game in isolation (and the crude TPR is basically doing a calculation a game at a time) really depends upon the context. The standard rating calculation (with non-linear winning expectancies) is basically doing a weighted average of your prior rating with a more accurate “performance rating” using a linearization around your prior rating. The exact same results against the exact same opponents would produce a different PR for a player with a different rating.
Forgetting /d5 on the time control in a flyer is “imperfect” (though in your view that seems to be a capital offense). Failing to describe who is eligible for which sections (both the unrateds in the main event and for the adults only section) is WAY beyond imperfect.
First, based on your previous comment, you don’t understand the rule well. You also failed to provide an example on your last point.
Also, honestly it’s really not that hard to explain and it’s not much more complicated to explain than if the under prizes were different amounts. This isn’t rocket science.
I have seen people calculate performance ratings that ignore the player’s rating and that throw out wins that lower the performance rating and losses that increase the performance rating.
Beating a 300, 800, 1300 and 1600 while losing to a 2500, 2100 and 1800 has the results against the 2500 <2100 performance>. the 300 <700 performance> and the 800 <1200 performance> removed while averaging the other two losses <1700 and 1400> and the other two wins <1700 and 2000> for a 1700 performance for four games - would have increased to 1780 for five games with the 2500 loss included or dropped to 1408 for six games with the bottom two wins added or ended at 1543 for seven games including all of them.
Remove any wins against players rated more than 800 points below the lowest rated opponent that beat the player and more than 400 points below the lowest rated opponent that drew the player (all but the top win if the player had all wins).
Remove any losses against players rated more than 800 points above the highest rated opponent that lost to the player and more than 400 points above the highest rated opponent that drew the player (all but the bottom loss if a player had all losses).
If you do that then you have zero games left if a player does something like lose to a 2500, beat a 600, lose to a 2000, beat a 900, beat an 1150. In such a case you could average the worst losing performance (1600) and the best winning performance (1550) to get 1575.
Note that any discarding of results ends up with fewer data points to calculate a performance.
Remember that all ratings are approximations and two 1200 performances could vary greatly as to how accurate their calculation really was.
See this is exactly why I’m reluctant to help you, and I don’t think I’m alone.
Me: this tournament is poorly designed.
You: what about the design do you think is poor?
Me: [details]
You: a lot of tournaments are designed poorly.
I get your point and I even agree with you. I just assumed from your body of work on this forum you wouldn’t be satisfied with an imperfect flyer, much less one that is poorly designed. If you really don’t care then you shouldn’t have asked. I further assume that you’re not involved in the organization of the flyer/tournament since, recall, you brought this up as an example of poor tournament practice. So I’m mystified as to why you are getting so defensive at the criticism.
Alex Relyea
I’m not getting defensive and I agree with some of the criticisms.
Give it a go Micah. Assume these are the results (where F indicates female). Do it first under the literal description of how the female prizes are to be handled (assuming score rather than TPR is used). Then with the female prizes properly incorporated. Show your work (and do it closed book) and then explain how the male master wins a share of prizes for women and U1900’s.
5.5 2330
5.0 2250
5.0 2205
5.0 2160
4.5 2210
4.5 2160
4.5 2090 (F)
4.5 2030
4.5 1950
4.5 1890
4.5 1830
4.0 1880 (F)
…
This helps, but averaging the opponents’ ratings (with or without the aforementioned wrinkle) does not really seem the most intelligent way to compute a new player’s rating (or anybody’s TPR).
The TPR could, instead, be defined as “that rating which, when run through the regular rating formula (the one used for established players), results in a rating change of zero”.
Of course, that requires successive approximation. Start with a ridiculously low rating, like 400 points below the rating of the player’s lowest-rated opponent, and a ridiculously high rating, like 400 points above the rating of the player’s highest-rated opponent. Call these the “lower bound” and “upper bound”. Run each of these through the regular rating system, to produce a new lower bound and a new upper bound. The lower bound should increase, and the upper bound should decrease. Keep doing this until the two bounds are equal (when rounded to the nearest integer). Bingo – you have your TPR.
(If the two bounds oscillate rather than converging, just average the two bounds at this point to declare the TPR.)
This procedure essentially averages the winning expectancies rather than the ratings, which makes a lot more sense to me.
Of course, there are special problems for players who lose all their games or win all their games. With successive approximation, either the lower bound will decrease indefinitely and the upper bound will never catch up, or the upper bound will increase indefinitely and the lower bound will never catch up. In these cases, it makes sense (sort of) to add 400 points to the rating of the highest opponent or subtract 400 from the rating of the lowest opponent.
If software that attempts to compute TPR does it the above way, a lot of problems would be solved.
Bill Smythe
I was referring only to place and under prizes, not including any bonus prizes.
The US Chess special rating formula basically does that using the linear winning expectancy rather than the non-linear one. There’s a reason for that. Obviously, as you note, there is a problem with all wins or all losses, where the mathematically correct values are +infinity and -infinity. But it isn’t just with all wins and all losses. As Jeff mentions, you can have a set of results which provide very weak information. 500W, 2200L, 2200L has, under the linear winning expectancy, a flat spot from 900 to 1800. We use the age-based prior (or whatever other rating information) to pick one of those: anyone 18 or younger will get 900, up to a 26 year old getting 1300. If you use the non-linear winning expectancy, you get a hard number of 1289. Except that basically any prior rating from roughly 1000 to roughly 1700 would lead to a trivial change in the rating (under a point). So the reality is that the tough cases are only superficially improved, at best.
OK. Then ignore the female prizes.
Note, BTW, that this is much simpler than what you will likely see on the Senior TD exam. However, the simplicity of the calculation is precisely why it is NOT easy to explain to somehow who isn’t familiar with computing prize distributions.
There was a time, back in the 90’s under earlier ratings formulas, when some people would pair a newcomer against much stronger players, in order to bring up that initial rating by averaging in those losses to high rated players. Then others would feed off those points in subsequent events. It is a bit less likely to happen today, but I’ve seen situations in which it appeared that was the goal.
It doesn’t do the new players any favors, though.
When I was running single-section Plus-Scores around 1994-2000, I tended to put unrated players in at around 1399 on the wall chart. That seemed more sensible than just putting them at the bottom.
Bill Smythe
+1
The trouble is that some of the “help” in this thread hasn’t been help at all. For example, most of the stuff Allen has said in this thread has been completely wrong.
It’s more than tedious, it’s illogical.
But it is the current rule. I have long felt that rule 32B (regarding prize distribution) is the Stupidest Rule in the Rulebook, especially 32B2, 32B3, and 32B4. Recent small changes in this area have not gone far enough to remove the Stupidity.
Let’s simplify the example a bit. Suppose there is a $100 prize for 3rd overall, and another $100 prize for 1st U1800. What happens if a player rated 1750 qualifies for both? According to the above, that player is awarded 3rd overall, and 1st U1800 is left over for a lucky player rated below 1800 with a lower score.
Now let’s make a small change in the prize structure. Change 3rd overall from $100 to $99, and 1st U1800 from $100 to $101. Now the 1750 gets the (larger) U1800 prize, and 3rd overall is left over for lucky players not necessarily rated below 1800 with a lower score.
A mere $2 change in the prize structure has likely resulted in a $50 change for each of two players in a lower score group. One way, a player below 1800 gets the full $100. The other way, the $100 is split between two players, only one of whom is below 1800.
This is what I call a Stupid Discontinuity. A small change in the prize structure has resulted in a large change in the prizes won by two players in another score group.
We all agree (probably) that, in the $99 vs $101 case, this player is due $101. But why does it have to be that $101?? Why couldn’t it just as well be the $99 overall prize plus $2 of the U1800 prize, leaving most of the U1800 prize intact for players in a lower score group?
I propose the rule be changed so that you always go with the higher ranked (i.e. more inclusive) prize (such as overall vs “under”), only dipping into the less inclusive prize when necessary to fill out the full amount the player is due.
I think this would remove the Stupid Discontinuity in all cases. But I’ll admit I’m not sure. At this point Tom Doan will probably jump in with his “Arrow Impossibility” theorem or something.
Bill Smythe