I studied this question in my ChessBase article "1.e4 - 'Best by Test'?" Part 1 and Part 2. We all know that the databases show a higher percentage score for 1.d4. What is less well known is that White tends to pick d4 when he is the favorite and e4 when he is the underdog. Part 1 demonstrated that once this tendency is accounted for, the gap between 1.e4 and 1.d4 vanishes - both moves perform equally well.
But in Part 2, we see that there is more to the story. The average of White's and Black's ratings also matters. At the higher levels, the advantage of playing White increases. After correcting for this, e4 ever so slightly outperforms d4. If you switch from d4 to e4 in your next 1000 games with White, you will score two points (we suspect that this result will not cause droves of d4 players to change their repertoire). To be careful, I also checked for transpositions using the ECO classification of the games in my database. The results were unchanged.
Many times on chess forums, players lament that opening theory is going too far and that chess is being "played out" - there will be less and less scope for originality and creativity. Usually this is followed by recommendations that we alter the rules or switch to Chess960/FischerRandom. My results did shed some light on this debate. If chess really were being exhausted, we would expect that opening theory would eventually nullify White's advantage, since a perfectly played game will be drawn. Thus, over time White's percentage score should move towards 50%. I did find some evidence of this: there is a statistically significant drop in White's performance over the last few decades. However, the rate is extremely slow; Black's chances improve by only 0.01% per year. I concluded that we still have another couple of centuries before chess is played out.
This is very interesting. I always thought 1.d4 was statistically superior to 1.e4 because of the Sicilian defense. However, these statistics are pretty much biased due to rating differences, as you point out in the Chessbase articles. If anything, these rating-gap tables show that it's easier to outplay a weaker opponent (or get outplayed by a stronger opponent) with 1.d4, so that's maybe why AlphaZero prefers 1.d4 over 1.e4. By the way, what are white's rating-adjusted scores with 1.e4 against 1...e5 and 1...c5?
ReplyDeleteAfter 1.e4 c5, White performs 23 points above his rating. Against 1...e5, the performance gap is 34 points. I've been telling myself to *not* focus on percentage scores in the database. Instead, look for the gap between White's average rating and White' performance rating.
ReplyDelete