# Meaning of winning percentage

I was wondering about the exact meaning of winning percentage (white/draw/black) in database systems and how to use them for studying.

1. How are move order variations taken care of? For instance, if I go to a certain position, the winning percentage is only calculated from games which had this exact position on the board, right?
2. Let's say I look at a position from an opening which had been popular for many years. So there are lots of games in the database from those years with more or less equal percentage for black and white. Then at some point a novelty was found which gave a clear advantage for white, making the opening basically unplayable for black. As a consequence people stopped playing this opening and there are very few games in the database which show this big winning percentage for white. So if I look at some position before the novelty move, the winning percentage at that position will still show more or less equal percentages for white and black (because many pre-novelty-games get averaged with few novelty-games)?

Is this correct? If yes, IMO, the winning percentage is of rather limited use and should not be used when assessing a single position, without traversing and analyzing all the lines originating from that position.

Have there been attempts, to create a better (more useful for actual use) parameter than the "winning percentage"?

Winning percentages are quite useless not only because of novelties, but also because of general distribution of results. For example there is no meaningful average for line which is branching into (0,100,0) and (60, 0,40) later. I saw even more brutal examples like (55,0,45) branching on next black move into something like (100,0,0) and (0,20,80) which in original position seems like standard white advantage :D There is another problem that some lines are virtually always played by much weaker player. That can be reason why exchange Slav or 4 knights may in some books have positive score for black.

Engine evaluation / usage in corr games is in my opinion generally much closer to truth and count "N" can help you as well, even if this runs into novelty problem you pointed out. It has many drawbacks for evaluating human chess with many mistakes, but results graphs tells nothing of any value.

The percentages of triumph are very important. They tell us what types of movement have "more strength" to win. However, in chess there is no absolute truth, for many reasons. The first that comes to my mind is that during the game the player who lost is the one who committed the "last mistake". There are also reasons external to the chess logic, such as the health of the player, his strength and his psychology, which make him value in a different way the risk of losing and the variants that appeal to him.

Another aspect implicit in the question is whether the percentage of success is an evaluation of the resulting positions. The clearest answer is given above: No. However, there is some of it if we consider the percentage of success as an evaluation when it has been qualified by the ELO of the opponents. We all know that two players with an ELO difference of 400 points, it is almost certain that the result of a tournament between both is 100% for one and 0% for the other and with players with a similar ELO, it is expected 50% for both. Therefore, the winning percentage is a "statistical assessment" of the resulting positions (plural); I consider that as information.

How can we take advantage of that?

First, finding the movement in which the variant is considered a loser, for example, due to a mistake or a very difficult strategy to implement and seek a solution to that problem. We can also use percentages to choose more table-like continuations if we are facing impatient or stronger opponents than ourselves. Thirdly, I have used it to choose variants with fewer satisfactory continuations for the opponent; It seems to me that when there are few suitable continuations to the opponent, there is more risk that he chooses a wrong move. In this, the frequency of use statistic is interesting.

Winning percentages are weighted with win=1 draw=1/2 and loss=0.

It does not tell you how often white wins outright. You need wins to win a tournament not lots of draws. So players will play moves to improve outright winning chances while risking losses. That skews the statistics some.

Basically they are meaningless until you get to the end of the chain and know the actual result. But they are interesting and useful as a general guide. But they are not absolute guarantees of anything.