Let's Play: Rank the Bad Opening!

Not if, for example, a new line is found which refutes an opening and renders all those previous games useless.

The cosmic irony of your post is that proponents of a critical line will use databases as evidence that the lines are refuting. That is a big piece of how we know what the critical lines are.

This is why I said 'from the past 20 years'. Opening theory is a constantly growing foundation of knowledge. Modifications crop up frequently--new refutations are found, then refutations to the refutations are found. Both sides get improvements over time. So using recent games is a good way to control for 'what we know'. It prevents an opening with a historically high win % that was refuted later on from having unrealistically high performance in the database.

If you're saying 'well what if a refutation was found yesterday', then we don't know it's a real refutation yet. It has to be tested in real tournaments by good players to be empirically valid, and by the time its strength has been verified it will have made an effect on the statistics for the opening.

The main flaw in database percentage is the small statistical sample. The odd lines are played when facing a weaker opposition, and this is why 1.Na3 score 90% for White in chess.com master database (out of 5 games, 4 wins and 1 draw).

The same effect occurs at a lesser extent to pretty much any line. For instance, I think we can all agree that 1.g3 is not absolutely bad, at least better than 1.Na3 (or 1.f3 that scores 55%). Maybe it has been played more often (13000 master games in chess.com), but as it is quite not as dubious, it can still score better than 1.Na3 would if played regularly ; so provided it is played only by GM against IM, IM against CM, CM against amateurs, it is easy to achieve a high win percentage when the opening has nothing to do with it.

This being said, HurricaneMichael's list doesn't look ridiculous. But claiming to be absolutely sure of it, and backing it with database percentage, that is ridiculous.

This is not a flaw when it comes to measuring commonly played openings. Tens of thousands of games make a scientifically useful sample. When you can count the number of games with the sodium attack on one hand, obviously there aren't nearly enough games to glean anything from the databases about it. Everybody knows the rare openings will just have noise.

Statisticians who analyze the databases can't just look at wins/draws/losses. They have to account for rating differences. This can be done by comparing how players fare relative to their ratings on both sides of an opening. For instance, if stronger players reserve an opening against weaker players only, you can still compare the W/D/L rates to what they would be against a main-line opening with a similar difference in ratings. If the strong player underperforms relative to how they normally do (for instance, 100 points worse than their rating implies), then you get something informative about the opening (that it is weaker).

Well, in the case of rating-compared games, I agree.

But again, you will find that it reduces considerably the number of "plausible" games in the database (1-0 for Carlsen vs. Patzer in the line 1.f3 will not be taken into account), so you will fall back on the problem of small statistical sample.

Take 1.g3 as suggested by Scottrf. 13000 games in chess.com database (ok, maybe not the best, but this is an example). This is the most common after the four principal ones.

Let's say there are 10 rating steps in the games where one of the players is a master : 1700-1800, 1800-1900, etc. 2600-2700. and that the games you keep are only the ones played inside the same rating step (of course real statisticians would use a smooth function of the difference of rating, but we are only looking or an order of magnitude)

Let's say, which is an underestimation because people tend to play this that they consider as inferior against weaker players, that of our 13000 games are distributed 1% for each of the 100 pairs of rating possible. This makes 10% of exploitable games (ie where opponents were of approximatively equal strength), so around a thousand games. (I know that the 1800-1800 games do not exist in the database, when they would be taken into account in my estimation, another underestimation).

This makes 1300 games, say 1000 with my underestimations. 1000 exploitable games is not enough to judge an opening.

A game of chess has roughly three phases: opening, middle game and endgame. The cause of the result can come from one of those three phases. If you want to judge an opening based on the games played with it, why should you base it on the effects of the middle game and endgame as well? The value of a chess game for the opening played is until the first big mistake or a sufficient number of small ones. After that move will nobody take a serious look to it again. Whatever the result of that game might be. If there are many games played within an opening, then you have an almost equal number of games which a student of that opening will not take a look at anymore after a certain move. He will not consider the rest of the game of any more importance for the evaluation of the opening. But it is only after those mistakes that the results are made. Wins, draws and losses are not closely linked to any opening but to the mistakes made in the game (unless the opening is really bad). If you would compare openings, then would it be more meaningful to compare evaluations of positions directly related to the opening then comparing them by the results of the games.

How often have you read sentences like 'and now the opening has ended?' From that move on is the game hardly of any value for the opening theory anymore. That is why that sentence is made in the first place. If result is not an indicator of the value of a game for the opening played, what might then be a relevant indicator that an opening is playable? I think the average number of moves played can be a first indicator how well an opening is. If an opening has a big average of moves for all the games played with it, then can you suppose that there is reasonable play for black and for white. If the average number of moves of an opening is low, then can you imagine that there is something strange happening. Might there be a big difference in force between the opponents? Is the opening itself bad or unbalanced? Are there not enough games played with it?
But imho ideally, one is not looking at results at all nor only at the number of moves, but to the evaluation of a position after a certain number of moves. Which move can be very different for each opening, but one should be able to make an evaluation of the game before the first big mistake (within the opening). The overall value of an opening can be judged by the addition of all those evaluations in combination with the number of moves and the standard deviation. If you would find those kind of numbers for openings, you might know which lines are good or not. I would expect that the average evaluation for a well established opening as the Ruy Lopez would be closer to 0 with a higher number of moves and a smaller standard deviation then for an opening like the OrangUtan. If you would take a look at the Englund Gambit, then might you find an evaluation significantly further from 0, lower number of moves and bigger standard deviation (more wins or loses, less draws) then other lines within the same opening.

Roughly three...Never mind.

lol, no I do not mind. English is definitely not my first language. I can say enough in it, but it will always be a rough translation of what I could say in Dutch. I tried to express that some games do not have an endgame or middlegame. :-)

While not an art form in thte typical sense, properly utilizing a database does have some complexity. The question I always find myself asking is "the database says X - but what does that really MEAN??" And of particular interest, what would I look at, and what would I expect to see, if I were looking to corroborate that?

sorry got confused i thought the parham is the defense to it were you sac a pawn lol

what about my opening (German)?

your opening (German) is bad, and it's not even your opening (german)

i dont really get the joke

fred defence is the worst

you have mentioned a part of my repertoire!

actually I have won lots of games with albin , cochrane, and traxlor and in my opinion they are very good opennings if you are not afraid of being down in material

Englund, Traxler, Cochrane, Parham, Halloween, Albin, and Budapest. 

I ranked this in terms of proximity to an equal evaluation with bonus points for variance (complications!). Worst to best:

1. Englund (-/+)

2. Halloween (=/+ to -/+, or maybe just -+ if you analyze it to death with a computer in some long complicated line that nobody plays!)

3. Cochrane (=/+)

4. Albin (+/=)

5. Budapest (+/=)

6. Parham (= to =/+, objectively close to equal, but I like Black's practical chances better)

7. Traxler (Unclear! Objectively, White may be better, but all the crazy complications push this to the top for me.)

I like Black's winning chances after a swift 2...Qxh4, hoping White does not notice he made an illegal move.

I have won lots of games with utter rubbish, it does not make it any better even to my own eyes.

I know a lot of people claim it is refuted. Taking into account the number of strong players who are willing to play it OTB, at the very least it deserves to be in the top three here. Of the openings listed, this would probably be the one I'd mostly likely play, but I tend to avoid lines with a ridiculous amount of do-or-die tactical decisions and forced draws. I am no expert in the Traxler and play it for neither side. Please register your complaints with Ris and Alterman. ;)