solving time & rating

[uri blass]

continued and after 40 problems I have 29 out of 40 in these conditions.
80% seems to be too optimistic for me but at maximal 8 minutes per move(I do not believe that I can do the same with rating 2400-2500 so maybe it is more than 100 elo per doubling and I will do better with 1900-2000 standard problems at 8 minutes per move but it is certainly significantly less than 200 elo per doubling.

Note that in part of the problems I decided to guess before 15 seconds because looking always at the clock cause me to lose concentration so when I think that the 15 seconds is close I hurry to guess something because I am afraid that after I look at the clock I am going to lose on time or do a mistake and lose concentration.
There was only one problem that I practically lost on time on it but I also failed on it by guessing a wrong move.

Edit:Next 20 are even worse and I solved only 11 correctly fast enough(the last 2 problems I lost on time by a small margin(by 2 and 1 seconds so practically I have now 40 out of 60)

Edit next 20 are very good and solved 19 out of them correctly in less than 15 seconds for the first move and later less than 15 seconds per move so I have now 59 out of 80

It seems that at 15 seconds per move my level is near 1600 standard

[uri blass]

I got 17 out of 20 with my last 20 problems
total result is 76 out of 100 at 1400-1500 at 15 seconds per move including the first move(meaning that if I use more than 15 seconds for the first move or more than 15* number of moves for all the moves I fail) .

Note that one reason that it may be better to save time and not to guess in the last second is that the rules do not allow me more time when I get an alternative and if I guess some alternative to the first move in 10 seconds then it means that I have only 5 seconds to find the computer move to consider the problem as solved correctly.

[munich]

I can get an educated guess, too:
before I switched to blitz I was doing a lot of standard. I became better and better, and of course at the same time slower and slower.
When I reached ~2200 Standard rating I was using on average about half an hour. Sometimes more than an hour, sometimes 8 minutes. But on average it is probably o.k. to say I needed 32 minutes (I deliberatly round it to a 2^5 number).
Then I did a lot of Blitz. And I also improved in Blitz due to my special tag themed easy puzzle training.
Then I went back to standard, and did standard puzzles as fast as I did Blitz puzzles. I round that to 1 minute per puzzle.
The result that my standard rating dropped to about my CT Blitz rating: around ~1900 CT Standard rating.
That is a drop by ~300 points.

So increasing my thinking time from 1 minute to 32 minutes would result in appoximately a gain of 300 rating points in CT Standard rating, and this results in appox. an increase of about 60 elo points per doubling the time.
Who knows? maybe 400 would also be possible, because in between the 2200 measure and the 1900 measure I also became a bit stronger.
Then this would result in 80 rating points in CT Standard for each time doubling.   
And this would fit to the computer expectation: If you double the time of a program, then you get a gain of ~70 elo points.

As Richard points out, it could be a very individual thing. some people it will help much more, while others hardly will become stronger. I know for instance a chess club player, who is an excellent Blitz player, often winning the blitz tournament in the club, but in elo he is rather on the bottom list. He is also not using so much time during a long tournament game, usually he is hardly using more than 30 minutes for the whole game. But 30 minutes is still a huge increase in using time compared to his very fast blitz where he is often winning on time against stronger players using just 3 instead of 5 minutes on average. So that equates to using 10x more time if he uses "just" 30 minutes in a tournament game. I guess, that with some people it is like this: if they dont spot in 1 minute, they will hardly going to find the solution even if they take a day. Call impatience or a lack of concentration, but whatever the cause might be: I agree with richard, that this could well be a very individual thing.
For myself I'd say that the 70 elo increase is not only valid for programs, but also for me. Yes, I know, it could be 100 or 50 or so. But 200 per doubling indeed seems really a bit of a big jump (But could be for some individuals).

Contrary to Uri, I guess that the stronger you are, the less it is helpful to use more time. I take this assumption by watching GMs giving simultan matches against 32 players. They still play strong and and even though their opponents have 5-fold more time, these GMs win most of the games. The strength of a GM against 32 people might drop by as little as 200 elo points. not more then 300. Otherwise the result for an average GM (2550 elo) would not look like 28 wins, 3 draws and 1 loss against the crowd (often at least A-class players, and on average for sure around 2000 elo).

 

[uri blass]

I believe that the The GM practically do not get 32:1 advantage on time by playing 32 opponents because after they finish part of their game the time advantage is smaller so even if they play at a level that is 500 elo weaker then it is only in the beginning of the game and later their playing strength get higher.

I can add that we talk about simultan with no clocks so the GM is not in time pressure and practically can thinks when he needs to do it.

It means that you cannot get conclusions about the rating advantage that GM get from time based on simultans.

Note also that chess tempo rating and playing strength rating are different and I do not see a reason to assume that players earn the same number of elo points from doubling  time in games and from doubling time in problems.
 

[aoxomoxoa]

I think there might be an missunderstanding. Harder problems have usually a higher average_sec. So even a stronger player needs already more time for this problem.
You need to compare ( normalise ) to(by) the Av Time for the problems.

See for example my good friend thomra : http://chesstempo.com/chess-statistics/thomra

He is (was) ~1700 Blitz but he is/was ~2300+ in Standard. A short few shows me that the time he did to solve a tactics is 6? times longer than the "Av Time".
Many high value tacticians wich are doing Standard AND Blitz have ~~ about the same rating in Blitz as in Standard. I did a more intense comparison with several Players at my blog.

So if you try a low rated set with 15 sec each problem and compare it with a high rated set with 30 sec you still have to "normalise" because the AV Time of the high rated problems will be higher then it should be ( usually )

.
Hmps:  i hope i made myself clear.

The only "clean" way is to do problems of the same type in 2 different speeds. The doubleblindstuff we have to forget anyway.

[aoxomoxoa]

Uri i had a first look on your statistic
The set "1400 to 1500 never tried"
you have been about 3.6 times faster than average solvers of these problems. Thats the reason your rating on this set is only 1650 ( not stable now! )
Your strength using avereage Time is 2000?  

If so ( just for fun
you lose 350 point by a speed factor of 3.6

That would be not that far from my 200 Points per factor 2  

But we dont know your strength at this set 

[uri blass]

Uri i had a first look on your statistic
The set "1400 to 1500 never tried"
you have been about 3.6 times faster than average solvers of these problems. Thats the reason your rating on this set is only 1650 ( not stable now! )
Your strength using avereage Time is 2000?  

If so ( just for fun
you lose 350 point by a speed factor of 3.6

That would be not that far from my 200 Points per factor 2  

But we dont know your strength at this set 

I understand but I disagree with you.
I do not see the point of comparing with the average time.
Why do I need to care about what other do?

I care about the simple question how much rating I can earn in standard if I double the time that I use for problems.

if I use average time instead of 50% of the average time then it does not mean doubling the time that I use because the average time is not the same.

[aoxomoxoa]

It means that you cannot get conclusions about the rating advantage that GM get from time based on simultans.

Uri, live is not Mathematics. If there are no precise data available, you have to look for some work-arounds to calculate a  half-way realistc estimate. If you find several hints a value might be in the same range.. then you feel better. Thats a good and established strategy in natural-science and Engineering. Even the wildes calculations are  helpful if you have only "believes"..As long as there is no study with thousands of chessplayer, there will be never a proven relation between speed and strength.

[naat]

It is important what you will do in this time if you double the solving time.
If you just sit and look at the board without thinking, than it makes no any sense.

[aoxomoxoa]

I care about the simple question how much rating I can earn in standard if I double the time that I use for problems.

Emiprical Rabbit calculations and mine are about improvement of strength by speed. You need to keep all other faktors the same. If you work on 2 sets with different Av Time then doubeling the time in wich you solve the problems is not the same as doubleling the speed.

You need to have compareble sets.
Empirical Rabbit and i where only thinking of an improvement on "the same problems".
How would i perform on a problem i did not see before with twice as much time.

Or you simply calculate a different value.. Thats possible to but this value should be a little lower.
In a good experiment you try to change only one parameter and keep all others the same ( as good as possible ).

If you compare your ratings, on a set like this you use now,  with more time for each move that will do perfect !

If you want to compare your ratings, on a set with higher rating AND more time each move, then you change several things: Time you use, rating of problem , Average Av time and .. who knows...

[aoxomoxoa]

It is important what you will do in this time if you double the solving time.
If you just sit and look at the board without thinking, than it makes no any sense.

Naat wait till you know what the tactic is about, dont move if you did not find it, and then check, check, check, and THEN! do your move and you will gain 200 points in no times ( ehhhhh long times )

[richard]

Uri does appear to get a relatively small benefit from taking longer.

Looking at his standard problem attempts, he averages around 8-9 minutes over all attempts. If we split those attempts into two, those done under half is average solve

Thomra has an average solve time of about 24 and half minutes. If we split all his attempts into those solve under that average, and those that were solved over that average, and look at a performance rating for each set, he gets a large benefit from more time, with a 238 rating point performance rating advantage in the 'over average' group compared to the under average (   2016.4 versus 2255.2, 4:36 mins in short, 50:26 in long, not the long is likely exaggerated, as extreme values are not filtered out, so any times a problem is left overnight and solved the next day will be included here).

Uri on the other hand gets a much smaller benefit, with only a 21 performance rating improvement in the longer group. (20:22 mins av time in long group, 3:54 in short group,    2286.4 perf rating in short,  2308.1 long).

So clearly there are big individual differences. Munich is probably correct that the stonger the player, the less doubling the time helps, but again this will depend on whether they are being limited by raw calculation ability or impatience, if the time you are spending now allows you to see as far as your current calculation ability can see, then doubling time is likely not a big benefit. Aoxomoxoa is also correct that these types of comparisons still have a lot of confounding variables however, but the two examples indicate there can be wide differences in impact of increasing/decreasing time taken between users.

[aoxomoxoa]

Aoxomoxoa is also correct that these types of comparisons still have a lot of confounding variables however, but the two examples indicate there can be wide differences in impact of increasing/decreasing time taken between users.

Thats a very wild calculation too I like it

You can see at my blog that the differences are gigantic, maybe some make a coffee in between?



buuuut

Uri might take longer for some problems because he feels that he dont have the solution correct and he needs to look further. Maybe he is calculating only longer at more complex problems with more AV-time? And thomra mybe take less time for some problem becouse he is doing a job while solving and he has to stop sometimes "to" early because he is interruptet ?

Bla Bla Bla

So i think : easy explanation: uri just has a better "feeling" -if he found the solution already or if he has to continue searching-  as Thomra. So your interesting analysis might have give us a different result as wanted: Better player are more efficent with their time!

But shure, its possible that Uri is very bad in calculation. Maybe he can only calculate 3 moves? Then his performance should be worse at problems with many moves ( ond/or many pieces?) compared with problems with less moves, thats what the data seems to indicate eventually. And his Fide estimate for Standard problems should be "much"  lower then his fide estimate for blitz, a standard problem needs more calculation at av.
My Estimates had been "close" together so i think i am an average "calculator"
I think endgamestrength might be an indicator for calculationskill too. At least Khmelnitsky is saying this. But endgamestrength is not that easy to compare

[uri blass]

aoxomoxoa is right in his last post that
I take longer for some problems because I feel that I dont have the solution correct.

In other words I use more time in problems that are relatively harder for me(even if they are not harder for the average user) so comparing my performance in these problems with my performance in other problems is clearly misleading.

The only way to know how much I earn from time is simply to compare my performance with x seconds per move with my performance with 2x seconds per move for the same problems.

1400-1500 may be too easy (15 seconds per move is enough to solve most of the problems)

I guess that I may try to use 1600-1700 and see the percentage of problems that I solve at 15 seconds per move
when I am forced to guess if I do not solve correctly and compare it with 30 seconds per move and 60 seconds per move.

[richard]

Ok here is some stats from Uri for problems with rating 1900-1950, he did 100 problems in standard in that range with
average solve time 4:27 compared to 3:59 for the average user's solve time on those problems. So he is a little slower than average.

When we split the problems into those he took over his own average, and under his own average we have
81 problems under, and 19 problems over. His under time was 3:00 and his over time 10:29. His over time performance rating is 33 points below his under time performance, so 2315 for under time and 2282 for over. This indicates that what Uri is saying is accurate, the problems he spends longer on are the ones he finds hard, and this fact underestimates any help he may get from thinking longer across all problems.

Also, I wasn't saying the data showed Uri was bad at calculation, only that for him, his solve time might allow him to see most of what his calculation will allow him to see, and if that is the case, more solve time might not help that much. This is different to saying his calculation is bad, and is rather saying he might be choosing a sensible solving time based on his current calculation ability, so upward changes in his solving time might not help his rating too much (although decreases may well harm his rating).

Regards,
Richard.