Initial rating 1200 too high

The reason for the gap has more to do with the fact that you can't get a FIDE rating below 1000 and in many places, it is hard to get a FIDE rating at all (mainly due to limited events and a limited number of players with established FIDE ratings.) Again, the pools and rating methods are different.

I would have to look up the dates, but at one point, the lowest possible rating was 2000 and they have lowered that  of a couple of times. That can also play into the distribution.

I'm also pretty sure your labels are wrong; chess.com should be the orange, FIDE the blue. 

It appears you may be right, when it comes to Glicko at least, that players with essentially the same RD values, will gain and lose the same number of points. However, that only happens with players that have well established ratings. If the RD values are different,  the magnitude of change for each will diverge, depending on the differences, so still not zero sum.

^Sounds quite plausible.

FIDE ratings I guess are only for OTB competition players who are probably already fairly dedicated chess players.

The population on chess.com will be much broader and include a lot more casual players who are just dipping their toe in the water.

I would expect the mean/median chess.com rating to be lower than the mean/median FIDE rating.

That would more likely reflect a difference in the population characteristics rather than a relative discrepancy in the rating method.

It took me a while to drop below 1000 when I started playing Chess.com 

 ^They should send out one of the award icons when you drop below a 1000 after joining at 1200!

Part of the initiation to chess.com

I also question your numbers. According to the Chess.com blitz leaderboard graph,  there are 793 players at 2500-2600. There are around 500 FIDE players in that rating range in the most recent blitz rating list. 

So, something is off on your graphs.

This comparison is  Blitz comparison not rapid, it compares Chess.com BLitz with a recent FIDE RBB rating list of 2018. I took the numbers directly from Chess.com and from the FIDE website.

. If you check your ELO using a different method the same result appears: 

WE ALL HAVE MORE THAN 500 ELO POINTS LESS - LET'S CELEBRATE WE ARE ALL SO MUCH BETTER

I just looked at blitz too on both. And there is no such thing as a real rating; your rating is only relevant within the pool of players the rating measures. There may be correlations but there is no way I'm 1800 Blitz, especially not FIDE. My chess.com blitz is within 200 points of my OTB blitz, and while both are based on a limited number of games, my max would be around 1600 OTB USCF, after a good event.

Again, your chart is for sure wrong (FIDE has a minimum of 1000, not Chess.com) and the numbers are suspect as well. I don't have the time to dig through the FIDE lists again but one example is the Chess.com 1800-1899 rated players; there are 38,183, so are doing some weird scaling or misrepresenting your numbers.

Also, another thing alluded to above, the pools of players are different. There are over 4 million rated players on chess.com and there are only 92,564 players in the FIDE February 2018 Blitz rating file.

These are the official numbers on FIDE and chess.com, it takes 15 seconds to check this here on chess.com. Click on your own stats and you will see. It is the Fide RBB rating that they make public every year, quarter, sometimes month. It is the blitz rating not the rapid. But that does not make a big difference in general: all players at chess.com ARE MASSIVELY UNDERRATED. 

The unbiased ELO tests say there is a difference of 300-600 ELO points. 

The normal distribution of ELO ranking suggests of 500-100 ELO points.

Look at the Percentile of players. If you are If you are better than 95% of players you should really have a higher ELO ranking than 1500-1600, which was initially the starting value 10 years ago. 

Downsizing the initial ELO value, and the conservation of ELO points in the systems, has led to the effect that EVERYBODY IS TOTALLY UNDERRATED on Chess Sites like Chess.Com.   

ALL GLICKO SYTEMS ARE WAY TO STUPID TO CORRECT FOR THIS.

There is a real skill level in ELO that can be measured on tournaments, using chess test system, which is like a IQ system for chess based on scientific measures, approved by top chess players and psychologists as a highly suitable and correct measure: it shows THAT THE REAL SKILL LEVEL AND ELO is very much higher than on chess.com. YOU MUST ADD HUNDREDS OF ELO POINTS TO YOU OWN RATING - EVERYBODY ON CHESS.COM - LET'S CELEBRATE THAT WE ARE SO MUCH BETTER PLAYERS THAN OUR UNDERRATED ELO WOULD SUGGEST!  

^The discrepancy in percentile placing could be accounted for by a difference in the population characteristcs.

chess.com has a much larger population with, presumably, many more very amateur/beginner members. It would be expected that the 50th percentile player on FIDE will be much better than the 50th percentile player on chess.com.

If what you are saying is true it should be verifiable by comparing the FIDE rating with the chess.com rating for a group of members who have a rating on both FIDE and chess.com which can be compared directly.

(Also the graph is mislabelled as Martin pointed out. The orange bars must be chess.com and the blue must be FIDE but the labels are the other way round. Also putting the range increasing from left to right rather than right to left would make it a little more intuitive to understand)

Could be but is not the reason for the underrating. 

These are chess-fans, hobby-players, chess-professional, and grandmasters that come to play here. There is close to grandmaster play found already at 1600 level - that is not normal!

They have studied the openings and play 90% like computers - that is a 2000 level not 1600. 

It is definitely NOT the demographics: there are 600 Million chess-players worldwide, and more than 20 Million members on Chess.com that play up to 1 Million games per day, and there are 360.000 tournament players and 1594 grandmasters of which only 2.2% are female, in fact, although it is free to play for everyone.  It is a sport for people who really want to win and are eager learners. 

Of 600 million chess players, wouldn't you assume that the top 200T players are the very first league that starts at 2000 ELO? They are only 0,03% of players that get a rating of 1500-1600 ELO!

THAT IS INSANE! THEY ARE ABOVE 2000!!!!!! WE ARE MASSIVELY UNDERRATED HONESTLY!!!!!

There is even a scientific paper about it coming up because millions of players are degraded and depressed by this false ELO calculation system. Imagine, millions of chess players suffering so much and never get to their real ELO that they deserve. It is life destroying for a chess player this strategic ELO robbery crime!

I downloaded the blitz rating list from thr FIDE site and pulled the numbers from that. And it did take me less than 15 seconds to check the distribution graph here and see something is obviously off with your numbers.

If someone is in the 95th percentile on chess.com, that means there are a lot of very casual players in with the more competitive ones. If a player here has a liw RD value, meaning a large sample of hames played, then their rating is fairly accurate in the chess.com blitz pool. It is possible to he over/under rated if a player is involved in a smaller subset/pool; for example, many of my rapid games here are in the context of one club and those are all USCF members, so my rapid rating is mainly reflective of that pool.

When you mention 1500, is that in reference to the old FIDE bottom rating? You do realize that if you play a FIDE rated event and your performance rating is not higher than the bottom rating, then you don't get a rating? They don't start the rating at that value like a starting rating at chess.com.

And you need to correct the labels on your graph, even if your balues are not off (which they likely are). Orange is Chess.com, blue is FIDE (since FIDE has a bottom rating of 1000).

Yes, you are totally right with the legend, this is the updated figure I was talking about.

But this does not change anything, because my argumentation was built on the updated figure, assuming it was the updated figure. 

I am talking about the PERCETILE that you and everybody can check in 10 seconds. 

It is of course not normal that only 0.033% of all chess players achieve a score of 1600 ELO. They are the worlds TOP PLAYERS - They are the Chess Community of chess experts - and you or CHESS.COM are giving them only 1500-100 ELO points although all scientific measures say it is 500 ELO higher.

Why don't you resist so long to acknowledge this clear-cut fact, although clearly everything shows this? A system that gives too low ELO starting values can by theory not work and will always underrate all players be default setting. Hence, WE ARE ALL 500 ELO POINTS HIGHER and these points were strategically deprived to downsize the chess abilities of the entire chess community!!!

^The only way to demonstrate for sure that chess.com ratings are 500 points lower than FIDE ratings is by showing that a significant number of individual players, who are active on both systems, have an average 500 point discrepancy between their FIDE vs chess.com rating.

Otherwise any difference can be put down to differences in the population characteristics/demographics.

Absolutely. @Dr_Roman_Anton, you realize that FIDE doesn't assign a fixed initial rating at all? They use a caclulation based on performance against opponents of known ratings.

Giving initial ratings, like the site does, can skew the distribution graphs a little, but isn't going to impact ratings the way you think it does.

Absolutely. They are better assessing the initial ELO value at FIDE-RBB-ratings and thus they do not have this problem. But your second conclusion is false: 

Initial ratings skew the distribution extremely, very much, by 400-900 ELO points! That is very much! 

My rating here was 1500 when I checked my rating on a scientific ELO measurement that is 2000 ELO. That is 500 ELO difference that is a very huge ELO bias - Chess.com is cheating on all of us! 

Knowing things about chess and applying that knowledge over the board are two very different things. Doesn't matter how well you test if can't apply that and gain the rating. 

^^Thanks SM and, yes, your initial response was useful.

However the discussion evolved in 2 related directions.

1. The unusually high default of 1200 may contribute to skew the average. My own experience is that when I played a new player recently (it was his fisrt game and he showed up with a score of 1200) I got something like 23 points for a victory in 2 moves. The other player only made 2 moves and lost his knight on the 2nd move and then gave up so obviously was a beginner. I don't think I deserved 23 points for that. I think new joiners should get the median rating (50th percentile) or less to avoid gifting points on their first game.

2. @dr_roman_anton says that ches.com percentile scores are too low compared with FIDE and that the initial rating error contributes to this problem. However I am not sure how the initial rating being too high can make the percentiles too low.

If chess.com (or FIDE) desires percentiles to be at target levels (e.g. 50th percentile should always be near 1200) then the only real way to achieve this is to apply a normalisation factor to everyone's rating at regular intervals to maintain percentiles at target points levels.

1) ELOs must be comparable in chess - undeniable

2) Chess.com and Online-Chess site ELO rating is 300-900 ELO points to low - undeniable

3) ELO distributions show that up to 1000 ELO points are missing all players

4) Scientific ELO-Tests show that up to 900 ELO points are missing per player

5) The chess puzzle and lesson rating on CHess.com are show that 400 ELO point are missing

6) The 1600 games in chess.com show 2000-level chess play in opening, mid game and ending

7) There is frequently a grandmaster-level game at the ELO of 1600 - with no or only minimal weak moves - this corresponds to a level of 2000 ELO not 1600 ELO

7) It doesnt matter how well you apply or would apply your chess knowledge - if there are not enough ELO points in the system you will always stay in the "15-16-hundred Tower" what the players call the phenomenon that they cannot explain 

>>>> BY UNARGUABLE EVIDENCE THE ELO OF EVERY PLAYER IS TOO LOW  <<<<

One feels so historic in this discussion - all chess players will thank it if you make that change.

This is absolutely clear. Many hundreds of ELO points are missing

Your points are utterly deniable. But believe what you will. The proof you are wrong is in the fact that most players that have OTB ratings do not have a lot higher ratings OTB than they have here. There are some players where that is the case, but it isn't true for everyone, and isn't even close for everyone.