Chess Engine ELO accuracy > Online rating accuracy?

I've seen countless posts talking about how your online rating is usually a couple hundred points higher than your OTB physical rating.  This made me wonder, how on earth can we assess our more realistic OTB rating without playing in countless USCF/FIDE tournaments?  Some of us either don't have the time, aren't in convenient locations, or don't want to throw down the extra cash when it could go towards other things, such as more books!

I have shredder 11 at home as my chess engine of choice, and I always find the computer to be far more difficult to beat than human opponents of the same elo.  It made me ponder whether my own chess engine was a more accurate gauge of my true rating.

If my win/loss ration against shredder 11 (set to an elo of say...1500) was 50%/50% over a span of say 10 matches, then I would ponder whether that was more telling than our online ratings.

These questions are, in part, due to the fact that I have always wanted a more true litmus test for gauging personal progress.  I find it difficult to do so against human opponents online through correspondence chess.  Sure, to some extent you certainly can gauge things, but there's also a lot of other variables involved, such as: Has your opponent reached their rating cap yet or are they relatively new to the site and still finding their "rating zone"?  Is my opponent possibly making use of more resources than I am for this match (books, opening explorer, or god forbid a chess engine)?

Has anyone else performed studies against their own chess engines to compare to what their ratings are online?

The ratings from online sites rarely match up exactly to OTB ratings, but I'd say they're usually accurate to within a class. Someone with a 1700 rating on this site would probably be a class B player in OTB games also.

The only real way you can find your accurate OTB rating is to play OTB games. The more games you play against more opponents, the more accurate it will be. That being said, if you rarely or never play OTB events then it shouldn't matter to you what your OTB rating is. If you play 99% of your games online, then just focus on that rating.

In my experience the ratings withing a chess program such as Fritz or Chessmaster are usually the least accurate in relation to USCF or FIDE ratings. Also I don't think the way dumbed-downed chess engines play. I think you'll always get better practice from real live opponents.

Here's a comparison of the current Chess.com and USCF ratings distributions:

Online ratings should be more accurate.  You are playing more games, and even though one may say - you have more time to think about your move, so does your opponent. 

Curious to hear from someone who plays both a lot of OTB and online, to see if their ratings seem to correspond at all. I always wonder what my OTB rating would be but I pretty much play only online. I played someone OTB who said he was rated around 1800 and got thrashed pretty consistently, although I find it harder to recognize simple tactics and such because I'm so used to a 2d board.

Ratings are relative, not absolute, and only accurate within the pool of players they apply to. So your chess.com is an accurate reflection of your performance here and will become more accurate as you play more games. Your rating when playing against a computer will be an accurate reflection of your performance against the computer but the two ratings do not relate to each other in any way.

Interesting comment IndridCold. and very interesting charts.

Of course, no rating is any more 'correct' than any other.  All ratings are relative to the pool of players that you are comparing yourself against.  If Erik added 200 rating points to everyone at Chess.com then the ratings would still be vaild relative to one another, but might be less similar to other ratings awarded by FIDE or national chess organisations like the USCF.

i dont agree, im 2025 OTB and only 1650ish at blitz, 1600 ish at long, and 2000ish at quick....if i was 200+ online i would be 2200+ in all these?

Gonnosuke wrote:

gumpty wrote:

i dont agree, im 2025 OTB and only 1650ish at blitz, 1600 ish at long, and 2000ish at quick....if i was 200+ online i would be 2200+ in all these?

With only 31 CC games under your belt, your chess.com rating isn't a very good indicator of your true playing strength.  As you play more games, I would expect your rating, which was 1780 at the time I looked, would gradually rise and end up more in line with your OTB ranking.  That said, to reach a rating that is higher than your OTB rating assumes that you're taking full advantage of the rules available to you and by that I'm referring to opening books, databases etc.  Not everyone does but those that do certainly place higher than they otherwise would.

So thats what im doing wrong hahahahaha. I havnt got a chessbase or pc program and no big openings books hahaha. I think your right though all that does give a player an advantage u dont get in OTB style playing. But overall i still think its a little inflated anway. But who cares unless ur entering OTB games and F.I.D.E. one is recongnized throughout the world wheres your local or country one is only important to the country ur in.And the chess.com one is only important on this site....

ljw5021 wrote:

I believe chess engine ratings must be more accurate than online ratings. My colleagues father is rated 2200 online, but only 1500 in real life.

A difference that large is unusual, from what I've seen. Maybe he chokes when faced with the pressure of a clock and a opponent sitting across the board.

i dont know about the ratings as i have never been rated anywhere in chess but what i do know is that, on this site, i can win most games with opponents ranked lower than me and i have difficulty with opponents who are ranked higher than me. as a relative indicator, i feel that it works pretty well with the exception of the timed out games where i dont think i should get any points when there are no decisive advantages (early moves still) and those games i have lost because i took advantage of a bottle of wine, a few coldies or shots before getting online at home to make some questionable plays (yes i also make questionable plays when i havent been in the sauce but it is less frequently obvious errors and i usually remember the logic i had used). nonetheless ratings or not, thanks for the fun games and great challenges.

You are a rather extreme case though, since the normal difference between online and OTB is more like 200-300 points, not 700.

But really, 15-20 rated OTB games a year isn't much compared with hours and hours of play every day online, so in this case it might actually be the OTB rating that is not representative of actual skill level.

Play a lot more rated OTB games, and my guess would be that his OTB rating would climb and get a lot closer to his CC rating.

for example,your average search epth is 17,and ur rating would be 1700,etc.

Your rating would probably be much higher but you may need some rest between games

I really dont think so.

I suspect a search depth of 11 would absolutely kill a 1100.

To the original question: Some have already remarked that rating is only relative to the pool. If you play only against your engine, that pool has only 2 players, I don't see how that could in any way be accurate. The larger the pool, the closer to the truth, I think. (Alway remembering that the truth is relative to other truths)

My friend had around 2700-2800 on playchess when he had account although he is rated 1800ish on OTB. (Actually he was close to 2000 when he was at his peak a few years ago when he was 17-18 years old, and he was only playing money prized tournaments with U2200players (And faster than 60m/game time control) for getting some extra money to spare)

He has nearly 1000 points difference, now lets look at reasons of this, i have only seen him playing 1min games (i wondered how he plays and let play 40-50 1min games on my icc that time (my bullet rating suddenly went up from 1600 to 2200 lol) , fun part was he was beating some ims and gms with low rating (like 2000-2100 bullet)) Never lost on time, yet he won %80-90 games with checkmating opponents with using less than half of his minute, only low rated opponents lost on time, higher rating opponents general checkmated or resigned because huge material difference.)

I have seen two reasons, first of all nearly all of his games he wasn't even care for his material, when opponent threaten his weakly protected pieces he wasn't care even sometimes doesn't bother even recapturing instead he was constantly busy damaging other players pieces, position, king. Nearly half of games were saved by his miraculous tactics seems like come from outer space, brutally hard to defend but not all of them %100 accurate. But most of these wouldn't happen even in 15min games against 2000+ players.

Second reason was incredible impatient, i even don't know how he standed 3 minutes games, i didn't see him playing these (may be he plays more solid in these i can't now)

Funny thing about his patience he could play online poker which demands lots of patience but it might be related to money might be giving him more motivation.

Anyway he might be an absurd example but it definetely points there are some other factors related with it. In my opinion if somebody plays with the same psycology and time controls as OTB in online than his/her online rating should be slightly higher in online. For two reasons, his opponents on average would be less caring than OTB (i know times when i played sleepless, drunk or nonstop for ages till i can't see where my monitor is), secondly due to more games played online than OTB it means higher the inflation is. Under the perfect circumstances up to 200 point difference is normal but if you play on faster time controls (anything faster than 15min/game) it makes difference is higher. (I believe gaining rating in blitz easier so it may differ like 400 points but some players may underperform on faster games but stronger they are , they have less problem with time trouble. (some complex endgame positions i played in 3min/game blitzes, required nearly half an hour by a friend approximately 200 points weaker))

There is no such things as accurate rating. It's just a relative number compared to the playerbase. In my opinion a rating number doesn't even mean exact skills.
E.g: if the playerbase consisted of 8 years old kids who just learned the rules you would probably get a high 2000ish rating quite fast.
Our only "OTB indicators" are IMs and FMs on the site. If you can beat them in a long game your OTB rating is probably around 2300-2400 :)

Houdini plays more active than fritz, it is more similar to the OTB 2000 players than fritz 8 (actually i didn't use it i am comparing it to fritz 11)

If you can beat houdini 2000 with good hardware and endgame tablebases, in slow enough 15-20 minutes or more slower game, more than you draw. You are definetely at least 1900 in my opinion.

But it is only my opinion, i was beating CM2100-2200s when i was 1600 but chessmaster is much weaker than fritz and houdini (and makes dubious sacrifices usually)

Chess24.com has a garbage rating system. I'm like 1400 on here and 1800 over there. 1600 lvl players losing their queen to a bishop on the fifth move LOL

I prefer this site way more

Well, part of the problem with that thought is that many players on here are FIDE players and there are plenty of people who cheat with computers.  This pushes chess.com-to-USCF comparisons down. 

My best gauge is blitz, which is why I play so much of it.  My 5-minute rating online or computed at unrated blitz events at the chess club is always 300-400 points lower than my classic time control OTB rating in the USCF.

Game rating of players accuracy depends on how many games they have played. If they play frequently their ratings will eventually hit a stable point and slowly improve.