Excellent Work! For someone who understands at least a bit of statistics, thsis makes sense :). 2 quick queries:
1. What was the linear regression equation got for this data between Chess.com and OTB rating?
e.g. looking for something like Y(OTB Rating) = a +bX; X-Chess.com rating
2. would it be possible to do this with data of all those who have a claimed USCF/FIDE rating and a chess.com rating? Larger the sample size, more valid the inference.
Nicely done once again, very insightful.
Too confusing!
are blitzs ratings higher or lower than your OTB?
Lower.
Suspected it, interesting to see a largeish survey.
DrawMaster here, and I'll offer a bit a feedback for Arun_1986:
Question 1:
Fortunately, I was able to go back and find my original file of data from which the graph above was constructed. I used Excel to generate the slope and intercept, yielding the following correlation:
Predicted USCF OTB Rating = 0.678 * chess.com Blitz rating + 586
I suspect that the R-squared for the fit is not impressive, given how shallow the slope is. I didn't calculate that value.
So, for example, the typical player with a chess.com Blitz rating of 1600 would be predicted to hold a USCF OTB rating of about 1670.
Again, on an individual basis, this means nothing. It's a correlation for the 120 data points I was able to collect by a rather laborious hand search of the chess.com records of those claiming to sport a USCF rating.
Question 2:
Certainly is possible. Perhaps you'll be the hero, Arun, who combs through today's records to broaden the data set. I'd love to see it. I don't think I'll have the time to even consider doing so for another 9+ months, if ever. So, have at it!
Regards,
DrawMaster
Many USCF games are longer than Blitz. Have studies been done to compare Standard games vs. USCF ratings? Thanks!
Please, the same analisis but with FIDE rating ! (Is the USCF rating the nationnal USA rating by the way ?)
Some day someone will be able to study and compare USCF ICCF Correspondence FIDE over the board and all the different ratings on chess com. It would be very hard to do but maybe interesting?
Many other possible comparisons can be made (e.g., USCF versus chess.com Standard, FIDE versus chess.com, correspondence versus chess.com). I have only made the very limited comparison you see, for illustrative purposes. At this point, I have no intentions of gathering any additional data, leaving that up to some enterprising premium member. Perhaps smileyhb or ponz11 will grab the sword and charge the hill. 
Chess.com blitz and USCF OTB shouldn't be compared since that's like apples and oranges.
Apples and oranges are frequently compared: how sweet each is, how many calories each might contain, which one has more acid, which one gets overripe before the other. Because two things are not identical does not mean that they don't have similarities worth examining.
However, to corroborate your point, there's no guarantee that two ratings obtained under different time controls, against different pools of players or with different opponent selection criteria will serve as anything more than a gross index of comparison. Still, the question is asked every day and it's not unreasonable to offer a reply, even if a large "Buyer Beware" sign need be hung around the findings.
Have fun. 
FWIW, my USCF quick rating is 2057, but my blitz rating here is 1512. The main reason for it being so low here is that I don't go to OTB tourneys drunk and buzzed like I do here on online blitz games. Online blitz is like cotton candy; it's sweet and sticky, but does nothing for your nutritional needs. On the other hand, my quick rating is based on tournament games, where I'm not messing around. It's something to consider when doing the comparison, because I'm sure that I'm not the only one here who approaches these two formats differently.
Thus the wide spread seen in the first chart ... LOADs of variability. But when you average 120+ data points, a bit of info pops out, I believe. You can't buy a cup of coffee with it, however.
Unfortunately there are a lot of engine assisted players here although almost no premium members. That reduces the rating considerably. With my FIDE rating of over 2150 it is difficult to maintain a blitz rating over 1800!
I didn't say they can't be compared, but they shouldn't be.
Comparing blitz from one site vs OTB yields useless information because a person's blitz skills is no indication of a person's OTB skills, and vice versa. The reason is because other factors such as different skills and certain skills weighing more than others affects the person's ability to play the games. This is obvious when you notice how there are many people with a lower blitz rating, and there are also many others with a higher blitz rating.
A blitz rating and a standard chess rating do not differ much as a rule. Maximum 200 points I would say. A blitz rating is a very good indication of someone's OTB rating and vice versa. No 2700+ grandmaster will have a blitz rating under 2500.
GM Alejandro Ramirez used to have a blitz rating 100 points superior to his standard rating, but now that his standard rating continues going up, the gap has reduced.
http://ratings.fide.com/card.phtml?event=6500617
While i don't agree with blitz and standard having not much difference, let's go with the example. A 200 point difference means a person with 1600 blitz can have a standard rating between 1400 and 1800. How does that translate when you compare that to USCF rating? The OP claimed that blitz Chess.com ratings were 100 points higher than USCF OTB (I take that to mean standard.) That means a standard in Chess.com ranges from 100 points below to 300 points above USCF rating.
Now, that's only if blitz could be accurately measured against standard. If it is shown that blitz and standard vary even more, then the difference is greater and even more meaningless. A 300 point diffference means it ranges from 200 below to 400 above, 400 pt. dif. means 300 below to 500 above.
How does that information help the individual player? Does that blitz player think s/he has a better chance of winning (at standard) against someone with a USCF rating 100+ lower? What if that player is one of those that has a much better blitz rating?
Wouldn't a better comparison be one done with the same game and same time controls?
Sunshiny:
The actually difference measured was 73 rating points, even smaller than the 100 points you use in your discussion. One would not easily notice a difference in playing strength of only 73 points without a goodly number of games. Indeed, Arpad Elo designed the rating system to use 200 rating points as 1 standard deviation.
How does this information help the individual player? Not at all ... that is, if he/she is trying to make some claim personally based on my narrow study. The numbers only signify something for the group, not for a single individual. Obviously the variation between individuals is huge.
The original purpose of my limited scope study was to simply answer the question that most frequently is asked here: How does a chess.com rating (I chose to define that as Blitz due to its higher popularity at the time) compare to an over-the-board rating (I chose USCF because there were more data points for that than for FIDE)?
This work offers a tad of insight regarding the question, suggesting that (over the range of ratings examined and at the time of that examination) there might be a small but real bias in favor of the chess.com number understating its pool of players' performance in OTB USCF chess play.
Overworking the data to say more that that would be unfair, in my view.
Looking it over again, it said another site had the 100 point greater difference. Here, the difference is 73 points lower.
The graph itself shows the wide range of USCF ratings within the Chess.com blitz rating of 1500 and 1700. The highest USCF rating considered is 2112 while the lowest is 968. At an average of 1600, that's at least a 500 point difference. How can you compare when the 968 USCF rated person has a higher blitz rating than someone with over 2000 USCF rating?
(Personal opinion below, and does not necessarily represent the views of chess.com management or anyone else with authority or insight.)
Hardly a day goes by in Live Chess Main Hall chat without at least one member asking some form of the following question: "How does my (insert rating number here) Chess.com Blitz rating compare to an over-the-board or official USCF rating?"
Good question. The unsatisfying but truthful answer is that since the player pools, ratings formulas, and playing conditions are not the same, no universal and direct comparison is fair, even if possible. Nevertheless, we all know that strong players in the over-the-board world usually translate to strong players here at Chess.com as well. So, is there anything at all that can be said quantitatively?
Some years back, a not-to-be-named competitor conducted a survey amongst its players to try to put a numerical spin on this off-asked-but-seldom-satisfyingly-answered question. Results of that survey indicated that - on average - online Blitz ratings were roughly 100 points higher thanUSCF ratings.
So, the $64K question is this: Is that also true here at Chess.com?
In search of an answer, I recently gathered some data from the very wonderful chess.com Live Chess players statistics database here - one of the more esoteric benefits of my new-acquired Diamond premium membership. The chart below illustrates the results of the analysis of these data and makes it possible to infer - albeit loosely - some interesting findings:
These data represent an instantaneous snapshot, taken over Dec/16-17/2010, of the Chess.com Blitz ratings for a sampling of Chess.com members in the 1500-1699 rating cohort who also has a claimed USCF OTB (over-the-board) rating - roughly 120+ members. The mean Chess.com Blitz rating of these players was 1592, whilst their mean USCF rating was 1665. This difference of 73 ratings points was statistically significant. The three players whose Chess.com Blitz rating and USCF rating differed by more than 800 points were not included in determining those mean values. On the face of it - though this sampling is quite small compared to membership counts here or the survey counts 'there' - we seem to be seeing a similar phenomenon: online Blitz ratings generally, if only slightly, understate OTB strength.
Another possible inference from these data: Although you can expect on average that your next Chess.com Blitz challenge against a 1500-to-1699 rated player will be with someone roughly comparable to that rating in OTB, the range of strengths of possible opponents is extremely wide - ranging from below 1000 USCF-strength to well above 2000 USCF-strength. See the figure below.
One can easily see that many of the claimed USCF ratings are indeed within the study's range for Chess.com ratings (roughly 1500 to 1700), but even more of them are in the next higher rated bin from 1700 to 1900!
Some disclaimers should be mentioned straightaway. The sample here is relatively small and only covers a 200-point rating range (though this range brackets the usual adult mean USCF rating average). It assumes that the claimed USCF rating is in fact both real and represents current OTB strength. No minimum number of Chess.com Blitz games was required for inclusion.
In closing, it's doubtful that this little study will answer the $64K question for all time or for all members, but it's interesting nonetheless, I believe. It leads me to conclude, on at least a semi-quantitative basis, that there is a reasonable correlation between and not a huge difference between average ratings and strength here in Blitz and those in USCF OTB chess - at least in the heartcut range of 1500-1700 rating - though members' Chess.com ratings may slightly understate USCF-asserted playing strength. The closer one gets to 1200, the more likely the provisional starting rating will bias findings, I suspect. The closer one gets to the top end of the ratings the less data available and the more suspect any conclusions will be, from a sheer insufficient sampling point of view.
Hey, it was fun to do the study, whether it means anything or not.
This material is a repost of a blog by Helper DrawMaster