What's my real rating?

It's been long believe that such an experiment would be unnecessary because well, how to put it... it's fking common sense my friend 

It is ... impossible to keep one's excellence in a little glass casket, like a jewel, to take it out whenever wanted. On the contrary, it can only be conserved by continuous and good practice.  -  Adolph Anderssen

It kind of works exponentially. For example:

If you have a 1000 blitz rating on this site, your OTB rating would be about 500.

If you have a 1500 blitz rating on this site, your OTB rating would be about 1200.

If you have a 2000 blitz rating on this site, your OTB rating would be about 2000.

If you have a 2200+ blitz rating on this site (assuming you got it legitimately), you're probably either an IM or a GM in real life.

edit: add or subtract about 200 rating to these guidelines to get an elo chess rating.

There really shouldnt be any published ratings below 1200 imo. If you are under that then you should keep playing as unrated until at least 1200. 

There is no reliable conversion formula for online ratings to OTB. If you really want to know what your OTB rating would be there is only one way to find out and that's to play OTB. 

As Reb says, there is no proper conversion. Although many player's ratings are much higher online, my turn-based rating (1954) is pretty close to my (converted from ECF) FIDE rating (1906). There will be different comparisons for different players.

"Common sense" ignores scientific method. No experiment, no cred.

After seeing chess.com 2400's giving serious cred to chess program assessments of openings, I think I'll deduct 1000. Minimum.

I never pay that much attention to ratings on chess.com compared to my other ratings.  They are never the same but I have never found them to be far apart.  And there is somewhat of a see-saw motion in that sometimes one is higher and sometimes lower but I've never noticed a variation of more than about 100 points.

I think that the only way to get you actual OTB ratings would be to play in tournaments that use that rating system.  When you come across a figure, compare it to your Chess.com rating and go from that.  You may find chess.com's ratings are inflated, but I seriously doubt it would be a figure as large as some I read here the difference would be proportionate to your rating.  In other words the higher your rating the greater the difference.  Your difference would be greater that that of a player with a smaller average.

What type of "Scientific" experiment would you like to see? The assumption that Glicko uses is that your chess rating is always in flux. You are either getting better or getting worse. Are we certain that this is correct? Well, maybe not necessarily over the course of one or two games, but over the last few games (or weeks/months if it's been a while since you've played), this is almost certainly true. Why? Well, ask anyone who is truly an expert in basically anything. Stagnation is simply not human nature. So, based on the assumption that your "true rating", (which is what Glicko tries to put a quantative value on) is ever mutating, then it stands to reason that the most recent games are the most valuable in terms of trying to find your relative strentgh.

Hmm...it is difficult to say, because a person's rating is relative to the average real playing strength of the 'group' the rating system covers.

And as a total 'group,' I would say that the OTB crowd probably takes the game far more seriously than the online players, simply because of the travel and money and time that one must invest to attend a weekend long OTB tournament (as opposed to just sitting at home in your bathrobe playing online CC chess).

And because of that increased seriousness, the average real strength of the OTB group would naturally be a bit higher, thus creating larger barriers to entry (stiffer competition) and making it tougher to get a high rating.

Glicko does not make this assumption at all. The glicko assumption is that the uncertainty of your rating estimate grows with time. This allows for the possibility that your rating is in flux, but does not demand it.

Incidentally, since I started this thread almost 18 months ago, I got back to playing rated OTB chess. My first ECF rapidplay grade is 145, with my long play grade likely to come in at around 150. Depending what conversion method you use, I believe that's likely to land me just north of 1800 FIDE. Which would make my chess.com bullet rating massively inflated, my turn-based rating a little inflated, and my blitz rating about right.

Unless I've got Glicko wrong, it gives more weight to your next game if you play it a month from now than it would give if you played it tomorrow. The assumption -- which I question -- is that the month off makes you either (1) a lot worse than you'd be after one day, or (2) a lot better than you'd be after one day, based on winning or losing. But ... and here's the logical issue ... it's still just one game in both cases.

Unsubstantiated hand-waving about Glicko's assumption being 'obvious' does not persuade me that it's statistically correct, and that's why I ask if a proper study has been performed.

Try play some OTB rated chess, and find out.

All rating systems are designed to show your strength relative to a population.  The reason that Glicko exists at all is because someone wanted to produce a better model of reality (to show how the individual fits on the mathematical curve).

 I'm not saying that it's completely counter-productive to discuss the merits of one system over another, but everyone should be aware that rating has no meaning outside of the population it's based on.

 I could tell you that I was rated 15,000, but until you knew that I was basing it on a scale of 1 to 30,000, you wouldn't know if that meant I was average or not.

@pathfinder

In the sense of a scientific experiment I don't quite get it.  What to use as a control group to judge the numbers gliko stats produce?

Mathematically (not surprisingly but still) I don't understand how you could go about proving a statistics formula either.  When building a best guess system, it's all about making logical assumptions.  Ratings will never reflect a true strength.  To assume rating uncertainty as a function of time seems more reasonable than not making that same assumption.  If you would like to substitute or modify that assumption, it would make for a productive discussion... and I understand wanting some "proof" but again I don't see how such evidence would be possible.

Bottom line, it's easy to sit back and poke holes in something, and I don't think you're giving the idea due consideration.

Arpad Elo studied match results and derived an empirical formula.

Glicko uses elapsed time to inflate ratings, but I can't find any empirical justification for it. Perhaps it was derived empirically, but until I see that it was, it's a "t" parameter without experimental basis (ie. analysis of match results). I would appreciate anyone pointing me to the empirical analysis.

Fair enough I guess.  Wish I knew of such analysis but I don't :)

Mark Glickman (inventor of Glicko) based the system on the Elo system, so everything that Elo does, Glicko does too.  Here is everything you need to know about how the system came about, and the formulas used in the analysis:

http://www.glicko.net/glicko/glicko.doc/glicko.html

It's interesting that while professor Glickman originally felt that Glicko would be a "special case" of the Elo system, it actually turned out that Glicko takes more variables into consideration, making it a better model of player behavior.  Therefore, "pure" Elo is now considered a special case (that doesn't take normal flucuations of rating into consideration).

If it's 1800 turn-based, I heard it would be about 1200 OTB.