Is a 100 point difference actually significant?

When you are playing on chess.com do you really feel there is a significance in 100 point rating higher or lower than your opponent?

I always felt like 900-1200 are players who often make blunders and can be similar in skill. 1200-1300 players make blunders a lot but not as significant. 1300-1400 players seem to mostly make more mistakes rather than blunders and usually lose due to one or two mistakes (or blunders) in a game.

I can't really comment on higher than 1400...

What I really wonder is, what are the differences between say a 1600 and an 1800 compared to a 2000 player. Is it a lengthy process for one to go from 1600-1800 or is that simply a matter of learning a few more positions?

I remember GM Mamedyarov (2763) talking about how another GM (~2600), should not have defeated him because he was playing the strength of a 2600. Is it really that sensitive at that high level?

It makes me really shocked when I find that a 2500 can never beat a 2700. Considering both are GMs, both have had years of training and experience. Both know soooo many openings and theories---and yet you can always bet on the 2700 to pull the win.

How would you rank all these Class A, B, C, D, E players????

This is something many players wonder.  At the base of it is the rating system (which is just a statistics formula) so it's helpful to know what the statistics actually are given a difference in ratings, regardless of if we're talking about 1400-1600 or 2200-2400 etc.

100 points is a 64% chance for the stronger player.  Meaning in a match of 10 games they would be expected to score about 6.5

200 points is a 76% chance
300 is 85%
400 is 92%
500 is 96%
600 is 98%

Now of course this is if the two ratings are reliable.  Take some scholastic kid rated 1100 and due to how quickly these kids learn, (not to mention the problems they have with scholastic ratings to begin with) his rating will most likely be less accurate than a 2350 life master.  (Also at that level ratings adjust up and down more slowly).

I think the difference between 1500-1600-1700 to 18,19,20 and beyond are all relatively equal, it's that a majority of people get stuck at particular places that makes having to deepen or widen our knowledge of the game necessary.  For example a natural talent with a fantastic memory may make it to 2400 in 4-5 years they way your or I made it to 1400, but I think you're on to something with the number of known patterns.  GMs use a lot of long term memory when playing, not logic.

The way you say 1200s lose to errors is the way 1700s talk about 1400s and the way players above me think of beating me, on up to the way Anand or Carlsen would think of when they crush a "lowly" 2500 GM.  The fact that there's such a large ladder of skill is one of the things that makes the game so interesting to me.

You say a 2500 will always lose to a 2700 but that's not true.  In fact we can see they win one out of four on average.  "First class players lose to second class players because sometimes second class players play first rate games."  Like you and the statistics show though, 100 points isn't that much in terms of ensuring a victory.  It's always best to take each game as seriously as you can regardless of the rating difference.

As for specifically how many openings or how complex of tactics they see, to how much endgame technique they may have, it really varies from player to player, even between two of the same rating.  Anyway hope that's a good start to an answer for you question.

woah statistics..

if you are talking about OTB ratings, then 100 points is very very significant at the GM level.

for chess.com, however, it is not.

Imagine each game against an equivalent opponent being worth around 20 points for a win or a loss.

If two fated opponents who are rated equally were going to face each other in 1 week and 1 of these players were to drop 100 points, we would arrive at the scenario you are discussing.

In order to drop 100 points, that's just 5 losses - that's 1 drunk night, 1 break up, 1 mouse slip, 1 ghost game, and 1 loss just for the heck of it.

In other words, 100 rating points is very likely just as it is: 100 rating points.

Well even chess.com (all online chess ratings really) use roughly the same system.  BTW a rating difference of 100 points between GMs is as significant as it is between a 1300 and 1400 player... so long as the ratings are accurate.

What you're talking about is bound to happen more online than OTB, so I can see your point.  However if you're an active player with many completed games the less likely you are to make big fluctuations like this... the more accurate your rating will be.  With a very large pool of players, you may even have some sections of 1100s (for example) who are much tougher than other 1100s.  If the users are diverse enough (say people from one hemisphere tend to seek games at different times than people from the other hemisphere) then situations this wouldn't be too hard to imagine.

Come to think of it, even USCF ratings have this discrepancy.  I've seen out of state players gradually lose or gain a good number of points after moving to my area.

It is indeed very interesting. And you mentioned how it is relative to perspective, but there must be a scientific way to determine the statistical likelihood that certain rated players must lack certain knowledge or patterns (but considering the amount of unknown variables, we can't be sure, but we can definitely get an idea).

I agree that online ratings aren't 100%. For example, I've beaten a friend who's rated 1800 on chess.com about 3-4 times (he's beaten me many more, but he has lost to me, and I'm 400-500 points below him). So perhaps anything can happen in live chess.

However, I still think I can tell when someone is playing at a 1300 level, and someone who plays at a 1500 level, major difference there. When I win against a 1300, it's because he lost a major piece somewhere--when I win against a 1500, it's because he lost an extra pawn. When I win against an 1800, it's because of some inaccuracy like a passed pawn ending up somewhere that it shouldn't.

The thing to keep in mind is that it's an expected score, not a chance of winning. So a 1300 vs a 1400 is going to have the 1400 win six or seven time and lose three or four. But a 2300 vs a 2400 the 2400 will get three or four wins and about six draws. All those draws make it seem less significant, but it's the same score.

A rating number tends to have less meaning and less accuracy the lower it goes.   A rating of 2500 is much more stable and accurate than a rating of 2000, with a rating of 1500 being much less stable and less accurate than a rating of 2000, and so on . . .

So two 2500 players facing off are likely to be evenly matched, while two 1500 players could actually have significant differences in their actual playing strength. 

There are a huge number of reasons for this and we could go on all day listing them.  Just be aware that for lower ranked players in particular,  actual playing strength can be significantly higher or lower than their numeric rating.   So that 1500 person you're playing might actually be performing at a 1400 level, or could be closer to a 1600 level. 

Note that the accuracy of a rating improves as more games are played, so you can feel more comfortable of the accuracy of a rating for a person who has played thousands of games than for a person who has only played a dozen games.   And don't forget that everyone has bad days which can badly reduce our performance, so even when a rating number is fairly solid it doesn't mean we are playing at that level every single game.

 This isnt true since as one goes up in rating it becomes harder to gain 100 points.   Its a lot easier to go from 1400 to 1500 than it is from 2500 to 2600 for example.  The higher you go the tougher those points are to come by and the easier they are to lose.

I once saw IM Boris Kogan win a tournament in Atlanta Georgia with 4.5 points in 5 games and he LOST rating points because his rating was so much higher than his opponents that the draw he gave away in the last round cost him more points than his 4 wins gained him !  Since then I understand that uscf implemented a rule to prevent this from happening again, as its ridiculous. If a player wins an event now he can NOT lose rating points. I dont know if FIDE has some way to prevent a similar incident or not.

 No, not having him lose points would be ridiculous. It's like saying "You won the tournament, therefore we will count all of your games as wins." That would lead to inflated ratings.

And rating doesn't measure how hard the points are to get, it measures relative performance. So a 100 point difference means the same thing at any level, no matter how hard the points are to get at that level.

 You are simply wrong here. He wouldnt gain points from the event in question but he shouldnt lose any either, if he won the event. If he gained points, then yes it would lead to more inflated ratings.

About the 100 points. I know from my own personal experience that it was easier to go from 1700 to 1800 than it was to go from 1900 to 2000 and those 100 points were even tougher to get from 2100 to 2200 because you are facing consistently better/tougher opposition as you go up through the ranks.

With FIDE ratings this holds even more due to the " K factor " used in FIDE ratings.....

"NOT lose rating points" based on ALL of his games in the tournament is  not the same as "count all games as wins" now, is it?

Completely agree that the 100 point climb gets steeper the higher you go up the mountain.  My coach (a 2150+ NM) felt nostalgic when I told him I jumped from 1620 to 1700 in a  4-round tournament and joked that he hadn't seen a jump for him like that for ages.

I dont know if the uscf rating formula has changed since I left but when I was there you could gain/lose a maximum of 32 points in one game. With my FIDE rating my K factor is 15 which means the most I can gain/lose in one game is 15 points. As your fide rating goes up the K factor gets smaller.....

Um...is it just me or is this totally illogical?  If it's one of those fake draws, it would suggest that the rating doesn't need to be dropped...but if it's a hard fought game against a much lower opponent and he draws, then it seems his rating should come down a tad having drawn against a lower opponent. 

You wanna punish him for not sharing your value system? 

Anyway, either way I don't care how they are doing it...so long as it reflects the real playing ability as much as possible.

I think the opposite of Schachgeek. If the game was one of those non-game draws just to clinch the tourney, it doesn't inform us about the relative ability of the players. So it doesn't provide any information that can improve our estimate of their ratings. Rating the game doesn't actually improve our estimate of the playing strengths of the players.

I think the current USCF rating system decreases the K value at 2000 and 2200.  I think it may drop to 24 and 16.

I'm not totally positive about this. That was my understanding about 10 years ago and things may have changed.

Does anyone know where I can find the current uscf rating formula ?

It used to be 16 points for a win plus or minus 4% of the difference in the ratings up to 350 points difference I believe......

 No, I'm not wrong. Based on his performance he should have lost points. If he doesn't lose points his rating is higher than it should be. Higher than it should be is basically the definition of inflated.

The ratings before the game predicted he would do better than a draw. He didn't do better than a draw. Therefore the prediction was incorrect. If the rating is not adjusted to account for the innaccuracy, it will be even less accurate in the future.

http://main.uschess.org/content/view/7875/400/

Unfortunately, the links that are supposed to point to the details of the rating formula aren't currently working. But there are some notes there regarding the K value.

not true

the most points you can gain from a game is DOUBLE the K value
winning gives you the K (while drawing is 0 and loosing is -k), then the rating differences is added as a number between -k and k (I think a difference of 400 points)...

and for the GM who lost points, if all the players he played against were more then 400 points below him then he got no points from 4 wins and lost 0-k for the draw, sounds resonable enough

if they were rated so much below him he shouldn't have drawn

FORGET WHAT I SAID
I think the USCF is using glicko ratings and not elo, which means the formula is very different

http://www.glicko.net/ratings/rating.system.pdf

I don't understand how this examples makes your point.

If a 1600 played in a tournament where everyone else was rated about 1000, he would also lose points for a score of 4.5/5.

The scoring rate of a 2600 against a 2500 is the same as the scoring rate of a 1600 against a 1500. The lower K value at 2600 means more games have to be played to gain the same number of rating points as the 1600, but it doesn't mean the relative strengths between people rated 100 points different changes with higher ratings.