to USCF Ratings Data


I've seen a number of threads discussing to USCF ratings, but none of the threads actually answer the question by looking at the data. If you supply me with the data, I'll do the math to come up with a conversion that lets you estimate your USCF rating from your rating or vice versa.

If you have an established rating both on and USCF please post your ratings in the following format at the very top of your post: online rating (correspondence) : 2062 standare rating (OTB) : 2045

USCF: 1789

heinzie p1712

USCF: none


FICS has a feature where you type "Tell surveybot ratings USCF estimate", and it tells you your USCF rating estimate. You can also do it for an FIDE ratings estimateSmile


davidegpc: There is a correlation between your ability to play blitz and your ability to play longer games, so one can be used to predict the other. The correlation would certainly be stronger if you played similar time controls, but even with different time controls the two ratings are related. If people post both ratings, I can run a basic linear regression to get a rough estimate of the relationship between the two rating systems. 

heinzie wrote: p1712

USCF: none

this is not good.  if 1712 = 0, then my 1512 = -1022. 

edited to use linear regression between the two known data points.  so, that makes me a negataive class D player?


davidegpc: I have edited the post to ask people for both online rating and standard so that we can see if there is a difference. I'm new to and didn't realize when I first posted that it wasn't just correspondence chess. 

The quick answer to your question, without an introductory course in statistics, is that linear regression is like drawing a straight line through a set of points. The layman's version would be to plot rating on the x axis, USCF rating on the y axis, take a look at the dots, and draw the straight line through the dots that's as close to all the dots as a straight line can be. If whatever reason your dot is an outlier that's no where near the line you just drew, so be it. The more people who provide their ratings, the better able to see that your ratings are just a fluke when compared to the relationship between most people's ratings on and USCF. Most people are much closer to the line and we get a rough estimate of the relationship between the two ratings.


Finally, we don't have a random sample because we're asking people to volunteer their USCF ratings. Yes, this could bias our estimate if people who volunteer their USCF ratings are different from those who do not, but I wouldn't worry about it. We're just trying to ballpark it, not hit the head on the nail.


In principle, there's nothing wrong with trying to correlate two rating distributions. In practice, however, I think you will find that the distributions are too irregular to fit a good regression model.


Hehehe... Mathematicians know The Least Square Regression, computer guys know The GIGO (Garbage In Garbage Out) Smile

yusuf_prasojo wrote:

Hehehe... Mathematicians know The Least Square Regression, computer guys know The GIGO (Garbage In Garbage Out)

you know, sometimes it is us noncomputer people that find the GIGO and spends hours explaining that there must be a problem with the computer answer.  it can be fun sometimes. 


pathfinder416: we won't know unless people actually post their ratings (I realize you probably don't have a USCF being from Canada, but still).

yusaf_prasojo: Everyone is familar with garbage in, garbage out. But the two rating systems are presumably highly correlated, we're not trying to establish causality, all we want to do is roughly predict one from the other, so I'd even settle for a simple difference of means. If you'd like to use another method, by all means, go for it, but good look explaining it. I guess we could use a LOWESS and explain it as drawing a slightly squiggly line through the dots. And really, garbage in is no one posting their ratings.


The FICS convcersion is dodgy - estimated FIDE - 1940, estimated USCF 1687?!


You can compare this to the glicko system which says that a player is with 95% certainty his rating +- 2 RD.

What you could do with enough data is saying that for example a correspondence player would have a USFC rating between a lower boundary and a higher boundary with some certainty.

To give an easy example someone rated 1600 on this site, could be predicted to be between 1300 and 1700 USCF with 95% certainty and for example between 1400 and 1600 with 68% certainty. (the numbers aren't based on data, I took as mean here rating - 100)

The uncertainties due to different time controls, different pools of rating are then put into the uncertainty interval and the chance that the player is between those intervals.


So, currently I'm rated 1829 on I don't have an OTB rating because I only do correspondence chess. What would be my upper and lower bounds for a USCF rating?