What is your opinion on win rates?

In researching various openings, the only indication of an opening's performance is often the win rate. However, this rarely assists me (and I am guessing many others) as the win rates for openings would lead into a vast range of different playing styles.  Furthermore, these statistics are only from high-level games. Thus, in applying an opening to my style I can only practise it. I am fine with attempting the various lines, as it is a time investment I am willing to make given the opportunity. With consideration of my opinions on this, should I give much merit to win rates?

Hey there! 

Statistics are always statistics. In my opinion (and according my experience ) the best thing is to play something you are comfortable with. Maybe that line has a less win rate but you will playing relaxed, enjoying and knowing what are you doing.  

Thank you for that. I will look to build up from openings which I enjoy

Statistics have to be interpreted carefully if they are to be useful. For one thing, you have to have enough data points and even then you have to take into consideration the variance. For example, in an opening that shows an average 60% win rate for White, if the opening is very sharp (meaning high variance), then that win rate is not very useful because it's picking up a lot of noise because of the sharpness. Similarly, people might point to a rare opening or variation like the Cochrane Gambit of Petrov's Defense and claim that 6. d4 is the best continuation for White based on the statistics, but there might exist only 7 games on which those statistics are based. Disregarding variance, I would still consider seven data points to be a passable *starting indicator* of the strength of the opening, but I would not let it influence my move choice too much. Certainly I would consider 100 games to be a good indicator, though (disregarding any high variance).

I rely a lot on the feel of the opening, especially as compared to similar openings or variations I've used. For example, just last week I was doing some serious reevaluation of a defense I'd been using for a long time. I was concerned since I would be playing Black and there were only two moves shown in the database at one critical point, and the statistics were showing 100% win rate for White with one move, and a 50% win rate for White with the only alternative move, and when I looked farther into the database I discovered that both lines merged by transposition later on, which meant that the average win rate for White for that variation was around 75%. There were only 11 games total across both variations, but 11 games should start to be a reasonable indication of soundness, and a 75% win rate for White was unacceptable to me. Worse, the ending position felt all wrong from what I was used to, because the pawns that were exchanged where not the pawns that I was used to being exchanged, which highly altered the normal characteristics of the opening, and yet those were the only choices of moves listed in the database. It looked to me like that 75% win rate for White was about right, and in fact my guess was that line for Black was unsound and probably would have a 100% win rate for White if White were Stockfish. Something was wrong. Either I had to abandon my long-standing defense or find a refutation to White's attack. I did find what I believed to be a refutation to White's attack, one that put the game back into its normal character, but if I hadn't known the opening so well I would have just given up because the database didn't give me any more suggestions.

That's a good illustration of what I mean: The problems with statistics in that case was that there were so few data points (= games) that no one had yet tried the better move I later found so the database did not even consider that better move, the statistics did not reflect the fact that both lines would lead to a transposition, and the statistics did not reflect the fact that the character of the game had changed. I had to rely on my own experience and knowledge to realize that. Another problem is that statistics won't show if a single move has been found that refutes a variation, which negates all statistics about that variation. A much more common problem is that statistics won't tell you at which move(s) the recorded games were lost: just because White has a high win rate in one variation doesn't necessarily mean that the variation becomes unsound for Black at exactly the point where the variation starts; more likely it means that Black commonly misses a good move(s) later on in the game that is not even listed in the database because all the games fork off before that point.

There are other problems with statistics, too. For one thing, people often assume that the most popular move is also the best move, but "best move" depends heavily on your definition of "best": if you like drawish games, for example, then the "best move" will not be the most popular move in the database because databases typically show only master games, and masters tend to choose sharp, non-drawish continuations because they are fighting for that extra half-point rather than merely trying to survive. Therefore personal style is another concept that pure numbers cannot capture.

The upshot of all this is that ideally you need more information, essentially in text form, to augment statistics, since such textual information can convey important concepts that statistics cannot. I keep my own such notes in my own master repertoire file but no move databases I've ever seen do that.

[Did I just write all that?! ]

Very interesting text, @Sqod. I totally agree with you. 

I think the best way to improve and play good games is focusing in our chess style, playing according to it and little by little, improving it and creating our own opening/defense system.  Just with this I went from 800 to 1100+ at blitz.  

You can get statistics to say anything.  But i digress...

There is a HUGE difference between studying  master games, and basing your openings on the win rate of master games.  

Obviously studying master game is essential for improvement.  But deciding on an opening rep based solely on winning % of master games is ignorance.

We arent GM's.  As an example...I play the Benko gambit, and had to listen to people tell me i shouldnt play it because its busted at the GM level.  Im not a GM, I dont play GM's, Im playing USCF A/Expert players, and the benko works just fine at that level.  

Find an opening you enjoy playing, learn the piece placement, pawn structure, and middlegame ideas.  

And as far as win rates go.  A line may have a win rate of 62%, but hasnt been played since 2012.  Thats when you need to know "why"

I guess an opening's win rate is something to consider but I think it's far more important to play openings and defenses that best match your style of play.

Thank you @sqod. I see that it can be dangerous to rely on statistics which have a high variance or low sample size. The difference in objective and playing style is also important as you mentioned. Thus I think I shall look on win rates with more caution from now on.

@ItsOzzy and @FBloggs: Yeah I have mainly been considering my own style, so I'll keep that as the main focus

Thank you. To be honest, I am yet to begin any consistent chess study due to the demands of my university workload (hence 'given the opportunity'). Furthermore, my primary consideration has been towards lines that I enjoy playing (such as QGA for white) and will look into further. Thus, I think I shall begin with some of my chosen lines and integrate some study of master games.

Statistics are not lies, just not subjective. That's why I'm dubious

If GMs play a move for years, and then someone finds a refutation, and publishes it in New In Chess Yearbook or wherever, the win draw loss statistics for that line may be great, even though no GM would ever play it.  It is probably a good idea to analyze, play through the games, and read up on an opening if you want to know if a move is good or not.

... and somehow that statement manages to be all three!

• A coach can help you decide on an opening that fits your style.
• Openings that perform well at top levels are considered "more solid" or "better"
• If you improve at a single opening and follow up with tactics and positional play, you can THEN decide on an opening for yourself.

Yeah I will definitely research any opening I try, either in practise or in theory

Thank you

One part of the OP is wrong...365chess.com's free opening database has master & non-master games in one database.  But yeah, they are sometimes hard to decipher.  In addition to some of the other issues noted above, there are preparation/comfort/psychological effects that influence win rates.  For example, many white players are uncomfortable playing the main lines of the French Defense, and play the Exchange Variation in a misguided attempt to take black "out of theory" or "out of their comfort zone".  Although the Exchange Variation is symmetrical, and by definition slightly better for white, black actually wins most of these games.  Why?  Because it is super common, and all French Defense players are very experienced and comfortable with it.  Perhaps the white players who chose the exchange variation also tend to be weaker, on average.  Regardless, over 1000s of games, black wins more than 50% in this symmetrical position which by definition favors white.  

Its interesting information...just not the best way to pick an opening repetoire as there can be all sorts of unexpected and strange reasons for the percentages.  

Ok I wasn't aware that non-master games were included. But you are right - even though the information can be hard to apply at times, it is fascinating to look at

Another factor to consider is that strong GM's will play main lines against strong opponents, but may deliberately play weaker lines against weaker players in order to get them out of book.  Thus bad moves may end up with very good stats.  I believe you can see the opponent's average rating in explorers like chesstempo's or in Chessbase the program.

Yeah nice! I didn't think of that. Thanks for the sources too