My son’s game results were as follows:
Round 1: Win vs. Mark O. Lynch (10286182) – 1608
Round 2: Loss vs. Reynald E. Dikit (15457574) – 1685
Round 3: Win vs. Daniel F. Trimbach (12412089) – 1746
Round 4: Loss vs. Jose J. Fernandez (12658033) – 1600
Round 5: Win vs. Edward Y. C. Tan (12402070) – 1691
Round 6: Half-point bye
Round 7: Half-point bye
Based on these five played rounds against higher-rated opponents, we ran several USCF rating estimation tools and consistently obtained post-event ratings around 1610. This appears significantly higher than the published 1577.
I notice the round 5 not showing opponents’s rating in the dropdown calculation table . One possible issue we noticed is that the Round 5 opponent (Edward Y. C. Tan) appears to have re-entered the event and shows two entries in the standings table (#63 and #88). I am wondering whether only one of those entries was counted in the rating calculation and whether round 5 was not counted at all, which may have affected the final result.
Could you please help verify whether the rating calculation is accurate and confirm whether the re-entry in Round 5 had any impact? Based on the five game results, the final rating of 1577 seems much lower than the estimates from the estimation tools.
I don’t think so, I think that’s a display issue. When a player has more than one pairing number, all of the games are aggregated for ratings purposes, but the crosstable display only shows the rating info for one of the pairing numbers. We’ve asked for that to be fixed, I’m not sure where that is in the stack of issues being worked on.
I’ve got a meeting starting shortly, afterwards I’ll look at the ratings detail to see what it showed the expected score to be.
Interesting. I thought the rating changes for a couple my students in this event were strange, but I had not investigated further until I saw this post.
Here is a back of the envelope calculation using Jonathan Dai’s results using pre-tourney ratings:
Round 1 - 1608 –> 76 higher –> expected 0.4
Round 2 - 1685 –> 153 higher –> expected 0.3
Round 3 - 1746 –> 214 higher –> expected 0.25
Round 4 - 1600 –> 68 higher –> expected 0.4
Round 5 - 1691 –> 159 higher –> expected 0.3
Total is expected score of 1.65. He scored 3.0. The difference is 1.35.
K for 1600 and 5-6 rounds is about 34. For 1532 should be slightly higher, say 35.
Without bonus points, this means a rating gain of 1.35 x 35 = 47.
Bonus threshold with B = 10 and m = 5 is 22, meaning bonus is an additional 25 points.
Rating gain is 47 + 25 = 72.
New rating should be around 1532 + 72 = 1604.
MUIR gave a new rating of 1577 for a gain of just 45 points.
The Ratings Estimator (updated January 2025) with pre-tourney ratings gave me 1612. Using post-tourney ratings gave me 1600. The recursive result should be somewhere in between. My back of the envelope calculation agrees.
Being a curious numbers-type person, I checked my students in the same section using the Ratings Estimator and pre-tourney ratings.
Student A: Actual gain 27. Estimated gain 27.
Student B: Actual gain 30. Estimated gain 46.
Student C: Actual gain 38. Estimated gain 22.
Student D: Actual loss 36. Estimated loss 38.
The Ratings Estimator roughly matches the MUIR calculations for two students, significantly underestimates a third and significantly overestimates the fourth.
Is it possible that the system did not give bonus points for the golden state open?
Calculating Jonathan Dai post rating I have
1600
1694
1700
1600
1600
Score of 3/5 (used players post tournament ratings) I get a post tournament rating of 1577 without awarding the bonus points (rating gain of 1532 to 1577)
If you add in the bonus points earned the rating goes to 1590 (Jonathan had 13 bonus points in the event)
This is just a theory, and I have no idea what is going on, but it is interesting that if bonus points are not awarded the post tournament rating is 1577 (could just be a coincidence though)
It sure looks like the bonus points calculation could be a factor. My back of the envelope calculation for Jonathan Dai shows:
Without bonus points, this means a rating gain of 1.35 x 35 = 47.
MUIR gave a new rating of 1577 for a gain of just 45 points.
This also makes sense for my students A, B and D. However, my student C significantly gained more points in MUIR than the online Ratings Estimator (and more than I expected for his 50% score). I just reentered student C’s numbers into the Ratings Estimator to confirm what I got earlier.
It is worth noting that 4 of 5 opponents wound up on their floor in the post-event ratings, one was only 8 points off the floor and the other’s pre-event rating was already on his floor.
That almost certainly means their initial estimate (step 4) rating was below their floor, and those are the ratings that are used for your opponents in step 5. How far they were below their floor in step 4 is not information I have access to in the current system. I’ve asked the developers if they keep that in their logs.
Try the estimator assuming all 4 of those players were 50 points below their floor at step 4 and you get a result much closer to the actual one.
It is well-known that floors are inflationary, this demonstrates why.
The floor only comes into play after all the ratings are computed, when a player whose post-event rating would be below their floor is raised back up to that floor. This is explained in the white paper that describes the ratings system, but it is not easy reading.
I have often wondered how low ratings would go if there were no floors, including the hard deck at 100. My guess is some people could wind up with 3 digit negative ratings.
I think us chess is working on an archive so all algorithm changes, revisions can be viewed . I think the rating algorithm archive project is low on the priority list though.
Would be neat if the archives ever got published though!
I’m not sure some archives are that useful, and that’s considering I’m a first-order data hoarder!
I don’t think the changes in the ratings document since 2020 would have impacted this issue, though.
The key line in this case is this one from step 4 on page 2:
If the resulting rating from Step 4 is less than 100, then change the rating to 100.
Note that this does NOT reference a player’s personal floor, only the 100 absolute floor. A player’s rating from step 4 can be below their personal floor. That’s taken care of after step 5. This is done to keep floors from being even more inflationary.
In Step 4, the document states: “In the calculations, use the opponents’ pre-event ratings in the computation (for players with pre-event ratings).”
If an opponent’s pre-event rating is 1609 and their post-event rating is reduced to the 1600 floor, should Step 4 still use 1609?
I believe the pre-event rating (1609) should be used. Using the post-event rating would be problematic, because it would create a circular dependency: my post-event rating would depend on my opponent’s post-event rating, while my opponent’s post-event rating is itself influenced by my result. It is unclear how such a calculation could be consistently carried out.
Correct, but step 5 uses the preliminary estimates from step 4 as your opponents’ rating, and it CAN go below that player’s personal floor.
The reason for the two-step process is that it improves the final ratings by taking into account an opponent who had a really good day (so your game against that person has a larger ratings differential, resulting in a lower expected score) and, conversely, a higher expected score against someone who had a really bad day, which was the case for 4 opponents here.
This is one of those (fortunately) rare cases where the two-step process had a big negative impact on the ratings, because 4 of the 5 players were, based on their results in this event, overrated. And the ratings estimator doesn’t have the information needed to do that two-step process, for that you need the full crosstable and ratings database.
I understand your point. I went through Step 5 and ran it locally—you’re right, the floor doesn’t affect the calculation at all.
My estimation comes out to around 1577–1579 using the intermediate values in Step 5. However this result does not include any bonus points. Do you think the bonus simply doesn’t apply in this case, or could it be missing somehow in the system?
