Another Look at the Chess.com vs. USCF Ratings Debate - Chess Forums - Page 4

DrawMaster

Jul 14, 2014

0

#61

Every rating is 'a true rating' if enough games against the same pool are used to calculate it. Rather the question has been this: does a blitz rating from internet chess compare at all to an OTB chess rating at longer time controls? Long after this thread is done and gone, the argument will still rage, at least because to many it's an interesting question - even if meaningless in some sense of the word.

JM3000

Jul 14, 2014

0

#62

"There's a closer correlation between blitz/bullet ratings and OTB ratings than online correspondence and OTB ratings."

Many Time in Online correspondence i take few time to respond that in a serious tornament. I play semi-Rapid chess.

I do a registrer of my oficial national games in OTB Standard (90 minutes +30 seconds move), in my last 50 I have a 61% of punctuation, against a mean of 1988 National Elo.

When I'm playing OTB blitz my punctuations fall a lot. For exemple, in my worst rapid tournament I obtain 0 of 6 games. In my worst standard (90minutes + 30 seconds) tournament I obtain 4 of 9.

"So whether it's a "true" measure of something else or not, there's a strong correlation.

Here are links to FIDE's top 50 Rapid vs the top 50 Standard:"

In my opinion the top players are good in all forms of chess. But in low ratings there are people more good in blitz other more goods in slow times, and others with correlation in two systems.

kleelof

Jul 14, 2014

0

#63

DrawMaster wrote:

Every rating is 'a true rating' if enough games against the same pool are used to calculate it. Rather the question has been this: does a blitz rating from internet chess compare at all to an OTB chess rating at longer time controls? Long after this thread is done and gone, the argument will still rage, at least because to many it's an interesting question - even if meaningless in some sense of the word.

It is meaningless. The problem is the human mind is able to draw correlations, even when none exist.

DrawMaster

Jul 14, 2014

0

#64

The correlations exist; only math is needed to show that. Whether causation is at play is much less clear.

kleelof

Jul 14, 2014

0

#65

Yes math is quite good at proving something, wether it is true or not.

neznaika2012

Jul 14, 2014

0

#66

JM3000 wrote:

In my opinion the top players are good in all forms of chess. But in low ratings there are people more good in blitz other more goods in slow times, and others with correlation in two systems.

I guess, in fast time controls, all it takes is practice. When I started playing bullet chess here about a year ago, I could barely beat 1000-1100 level in 2min 1sec format (and I am sure I will never beat anything higher than 1200 in 1 min format) losing mostly on time, but I picked my opening portfolio and stuck with it, playing the same lines almost religiosly to moves 15-20 without much thinking, and within a year I reached 1400 level. It just shows that higher than average long time OTB players may need some time to adjust to fast pace games, and some may not be able to do so anyway. I am often playing old players (60+) who have 1800-2000 USCF standard ratings, and they suck in 5 min blitz, not because they are weak players, but because they can not think/move fast enough in the complex situations.

JM3000

Jul 14, 2014

0

#67

"The correlations exist; only math is needed to show that."

Sometimes the correlations are dubios.

For example if we study all school students and only take data about results in mathematics and size of hand the data displayed a correlation with them. This is because the older children know more mathematics that the younger, and older children are Larger. But the reality is that they know more mathematics because they go more time to school.

heister

Jul 14, 2014

0

#68

Everybody knows that 50% of the time it works every time.

DrawMaster

Jul 14, 2014

0

#69

My point, petrip. This subject will be argued till the cows come home, of course.

One must also remember that the ELO system, as created by Arpad Elo, was designed so that 200 ratings points would be a 1 standard deviation. So, if anyone is really trying to make a huge deal of rating difference that are much smaller than that, sample size must grow to much larger numbers.

I have no dog in this hunt, really - even if I did make one of the original posts on the subject. If one wants to know what their OTB rating should be, play a lot of OTB chess. If one wants to know what their blitz rating should be, play a lot of that. One gets a better idea by actually conducting the relevant test.

However, the question of the relevancy of ratings comparison will only go away about 15 minutes after the last person with a rating dies.

SmyslovFan

Jul 14, 2014

0

#70

Jcbutler wrote:

There appear to be two kinds of people at Chess.com: those who care about their ratings and wonder how well they correspond to “real” ratings like USCF ratings, and those who dismiss the importance of ratings and question how they could possible mean anything at all. Another point of contention has been whether Chess.com ratings are inflated or "deflated" relative to USCF ratings. Fully aware of the risks of entering into a debate that has become heated at times, I humbly offer my own insights below, along with selected summaries from people who have considered this question before.

A couple of years ago, for example, DrawMaster examined data from more than 100 Chess.com members, rated 1500 to 1699, who also reported USCF ratings on their profiles. His main finding was that these members had Chess.com blitz ratings that were, on average, 73 points lower than their reported USCF ratings (1592 vs. 1665). He also provided a graph showing a linear relationship between USCF ratings and Chess.com ratings. He concluded tentatively that online blitz ratings "generally, if only slightly" underestimate OTB playing strength.

AdamRinkleff also argued that Chess.com blitz ratings were generally lower than USCF ratings, but he stated that the discrepancy was quite a bit larger. According to his own observations of about 20 people who maintain active ratings in both systems, AdamRinkleff suggested that Chess.com ratings are consistently 200-300 points lower than USCF ratings. Unfortunately, this claim was put forth from a small sample and without supporting statistical analysis. This led to a substantial debate on the forum and no real consensus. Many posters claimed that the two sets of ratings were like “apples and oranges” and could not be meaningfully compared.

In an attempt to test AdamRinkleff's hypothesis, ShindouHikaru posted data from 54 Chess.com members who had USCF ratings that had been verified by official records. Unlike DrawMaster's data, this sample was well above average in skill level, with many of the players taken from a list of titled players on the website. ShindouHikaru concluded that although many players had lower Chess.com ratings than USCF ratings, there was no formula he could detect for explaining the relationship.

I got interested in the question at this point and decided to do some additional statistical analyses. I took ShindouHikaru's data, which he had kindly posted on the forum, double checked the USCF ratings and adjusted them for changes in the intervening months. I also added additional data to the sample by using my friends, friends of friends, etc. I only included members if they had active, nonprovisional ratings based on many games. I examined blitz ratings, online (correspondence) ratings, tactics ratings, and USCF regular ratings.

Obviously, this is a nonrepresentative, convenience sample and I would have preferred a large, random sample of chess players from both systems. Nevertheless, the data are still useful and allow us to answer some questions about how the variables relate to each other. In the end, I was able to gather data from 80 people, which I figured would be large enough to detect statistically significant effects. If any Chess.com staff are reading this and curious about these findings, I could do a much more extensive analysis with access to more data...

According to my results, the average Chess.com blitz rating for this sample was 1817 and the average USCF rating was 1945. This supports previous statements that Chess.com ratings tend to be lower than USCF ratings. The difference was 128, which is higher than DrawMaster’s estimate, but lower than AdamRinkleff's. All the variables in the analysis were highly correlated with each other, suggesting that they are all aspects of the same underlying entity of chess skill. USCF ratings correlated r = .83 with tactics ratings, r = .79 with online ratings, and r = .93 with blitz ratings, all p < .001.

Blitz ratings are strongly related to USCF ratings in these data. Honestly, I was shocked at the magnitude of the correlation. Rather than apples and oranges, the situation is more like apples and apples of the same variety, but from a different tree. Then again, the fact that the sample consisted of active, relatively high level players who were serious enough about their Chess.com ratings to put their names on their profiles may have influenced the results. I would expect the correlation to go down with a larger, more casual sample of players.

Because the ratings are so closely related, it makes sense to use regression to calculate an equation to predict one from the other. This produced the following formula: USCF estimate = (Chess.com rating * .93) + 283. Or, if you don’t like math, you could do almost as well with the simpler formula where you just add the difference: USCF estimate = Chess.com rating + 128. Remember these are just estimates based on generalizations. Your mileage may vary.

As a final note, I was perusing the discussion on this issue in the forums and found a post by Pegrin from several years ago. Using Google, he found 59 Chess.com profiles that contained USCF ratings. With the caveat that people’s self-reported ratings might not be accurate, he computed a Pearson correlation of r = .67 and a regression equation of USCF estimate = (Chess.com rating * .74) + 280.5. My data suggest that this would underestimate the USCF rating for serious chess players, but I’ll leave it to the reader to decide which formula works best for them.

So what's my conclusion, for those of you who just skipped to the bottom? According to all the available data, Chess.com ratings and USCF ratings are substantially and meaningfully related to each other. They are clearly the same type of fruit. Also, people generally have lower Chess.com ratings than their published USCF ratings. Finally, the pattern in the data is clear enough that you can get a quick and reasonably accurate estimate of your USCF rating from your Chess.com blitz rating. You can try one of the formulas above, of if you are math phobic, just draw a regression line over the scatterplot with your finger and see where your USCF rating should be.

It seems that some people have skipped to the last page of the discussion without reading the original post.

JCButler found that there is a considerably closer relation between blitz ratings and USCF standard ratings (r=.93) than there is between correspondence rating and USCF rating (r=.79).

For those who don't know how to read this, a score of 0 would mean there is no correlation. A score of .10-.70 is generally considered to show a weak correlation, and a score of .70-1.0 shows a strong correlation. A score of .93 is an extremely strong correlation.

Of course, the issue of the spuriousness of a relation between two numbers must always be considered. But here, these are both measures of chess strength. The issue of spuriousness is something of a red herring.

There is a very strong correlation between USCF and chess.com's blitz rating, but not as strong a relation between USCF and correspondence rating.

JM3000

Jul 18, 2014

0

#71

I'm agree with that. "All ratings are measuring chess skills anyway".

In My opinion blitz and tournaments controls are similar but not the same, somepeople can be significantly better in one type of chess that in other.

However, level in blitz is level in chess. If in a hypothesis We take 2 players and one was significantly better in blitz than the other but the other was significantly better in slow time. Both players have their merits and I do not think either of them is entitled to be considered a better player than the other.

To sum up if a person is good in blitz he is a good player. However this don't permit him to ensure that has X OTB Standard Elo.

neznaika2012

Jul 18, 2014

0

#72

JM3000 wrote:

I'm agree with that. "All ratings are measuring chess skills anyway".

In My opinion blitz and tournaments controls are similar but not the same, somepeople can be significantly better in one type of chess that in other.

However, level in blitz is level in chess. If in a hypothesis We take 2 players and one was significantly better in blitz than the other but the other was significantly better in slow time. Both players have their merits and I do not think either of them is entitled to be considered a better player than the other.

To sum up if a person is good in blitz he is a good player. However this don't permit him to ensure that has X OTB Standard Elo.

I agree, we probably should not correlate blitz ratings with long time control OTB. Judging from the experience, when I started playing blitz on chess.com I used 10/5 controls, and my rating was as high as 1800, but as soon as I switched to 5/5 or 5/0 it dropped to even below 1500. Now I brought it back to ~1600 in 5/0 format. So, what is my true blitz rating then? The range is huge: from ~1400 to ~1800. My standard rating (though based on only 30-40 games) in 15/10 format is 1680 on chess.com and I estimate my OTB (G60 or G75) rating in ~1700-1800 USCF - I have not played rated tournaments in over 2 years, but I regularly play with active USCF players in my club, and this estimate seems to be true. Where would I draw a correlation? I can attest that blitz and long time games are quite different: one of the basic rules in long games - if you see a good move, wait and look for a better one - does not apply in blitz, so naturally most of the blitz games are decided by better time management and simple conversion of small advantages (like a pawn up or exchange up), without much looking into increasing material gains or setting up a mating combination - these will eat a lot of your time (unless you've seen it before and know precisely what to do). It is funny, but just recently I've seen a 1600 USCF player losing to 1200 USCF player in 10 min OTB blitz - the 1200 player had a lost position, but it took too long for 1600 player to finish it off. So, in terms of quality of the game, long time OTB rating has more value, while blitz and bullet ratings show how good you know the openings and how good is your time management (how fast you can make a move that may be a mediocre one, but is not an outright blunder and does not lead to a simple checkmating sequence)

AblazePopRocket

Jan 12, 2019

0

#73

This is interesting. My USCF rating is somewhere just above 1300, but my online blitz rating is 1576. Is this a bad thing? I'm guessing I just haven't played enough games in real life to determine my real rating, because I've played a total of 109 games in real life (USCF), whereas I've played 785 blitz and rapid games combined on chess.com. (Plus, I did horrible in the last tournament I played and kind of butchered my rating.)

MellowDog

Nov 19, 2022

0

#74

Another approach is to look at the rating distributions. Most people will have a way lower chess.com rating than USCF.

Average Blitz on Chess.com (look at global leaderboard): 799

Average USCF Non Scholastic: 1198

Average USCF Including Scholastic: 1068

91st percentile on Chess.com = 1382

91st on USCF = 1700-2000

The link below takes you to the USCF (its dated but I assume it has not changed a whole lot)

https://www.uschess.org/archive/ratings/ratedist.php

VLaurenT

Nov 19, 2022

0

#75

MellowDog wrote:

Another approach is to look at the rating distributions. Most people will have a way lower chess.com rating than USCF.

Average Blitz on Chess.com (look at global leaderboard): 799

Average USCF Non Scholastic: 1198

Average USCF Including Scholastic: 1068

91st percentile on Chess.com = 1382

91st on USCF = 1700-2000

The link below takes you to the USCF (its dated but I assume it has not changed a whole lot)

https://www.uschess.org/archive/ratings/ratedist.php

It's not clear that people will have a way lower chess.com rating than USCF as the USCF pool is much stronger than chess.com's. In my experience, it's often the opposite : OTB players have higher ratings online because the competition is much weaker.