In his discussion on game database statistics, he advises caution because 1-the number of games may be too small to have any statistical meaning 2-the stats don’t take into accounts later improvements 3-they don’t take into account the playing styles of the players. For example, playing Black against the King’s Gambit Accepted I would find stats meaningless because I never accept the gambit and try to avoid tactical melees. 4-the strength of the players is often omitted and 5-you must take into account the number of draws. A 50% score in a line where White wins half of the games and loses half is not the same thing as a line where all of the games were drawn! As he points out in a must win situation, the first scenario would be the way to go but if one is playing in a section to earn a master norm, the second scenario with a lot of draws may be the best course.
For stats to be meaningful he recommends having several hundreds of games available. Also, it is important to select openings that fit one’s preferred style of play. It’s also important to do your research to make sure that no recent developments have refuted your selected line of play. White may have won 75% of the games in the db but a latrer improvement may have completely reversed the situation. He advises that if there is not a sufficient number of games in the db then it is absolutely essential to play over the games to make sure the results weren’t influenced by mistakes later in the game.
In a recent CC game I faced a situation (I had Black in a King’s Gambit Declined) where we quickly reached a position with only three games in my database. The results were +1 -1 =1 and I had a choice of two different moves, both of which looked OK, but one line (where Black scored +1 -0 =1) lead to a position that was materially imbalanced and the engine was showing a small advantage to Black. Because I am always skeptical of engine evaluations in this type of situation the question was, is that true?
Analysis lead to a position where White was a P up and Black was threatening to regain it immediately by capturing the a-Pawn. The engines wanted to play either 13.Be2 or 13.Bg5 with a tiny edge for White. But I had to ask what if White played 13.a3 which the engine says is evaluated at 0.00. The critical line came after 13…Bf5 forcing White to play 14.Kf2 to get out of the pin. This is the position in question: