Problems with using double-dummy data
There are obvious problems using double-dummy data to determine hand evaluation and discuss bidding decisions.
Picking up queens
The first is that double dummy data undervalues queens.
Double-dummy,I can always make 3 tricks in the suit:
That's because, double-dummy, I always know where the queen is.
Similarly, when I hold:
Double dummy, I can make 3 tricks whenever east has the queen or the ten - about 3/4 of the time. (If east has the ten and west the queen, I can lead the jack from hand, forcing west to cover, then finesse east's ten.)
I can always make 3 tricks in this suit, double-dummy, by first leading low to the nine, then, if that loses, guess where the second honor is, double-dummy.
There are cases where other holdings are over-valued:
We get three tricks, doubly-dummy, if the suit splits 3-3, or if the ace is in the hand with the shorter suit. But any single-dummy play must choose to play east for the Ax holding if he wants to keep the chance for 3-3 split.
Is AKQJ9 really better than AKQJT?
One unusual case is the difference between the holdings AKQJT and AKQJ9. In my double dummy data, the latter is considered stronger by a slight amount, which is a bit odd. I suspect it is partly due to data sample size, but it also might be due to double-dummy bias.
If I hold AKQJ9, I can still usually pick up five tricks, double-dummy, unless LHO has the ten guarded by five or more cards in the suit. If RHO has five or more cards in the suit, I can usually finesse against his ten, when partner is not void. On the plus side, when partner has the ten, he has an additional entry to his hand.
In my data sample, the added possibility of an entry to partner's hand is worth slightly more than what's lost when you can't pick up five tricks in the suit.
But notice that picking up RHO's Txxxx in the suit definitely requires double dummy play. In single dummy play, you are about twice as likely to fail to pick up the fifth trick in the suit.
Other problems are just that certain types of questions aren't really suitable for double dummy analysis. You might ask, for example,"If partner opens 1NT, when should I invite with a natural 2NT response?" But the answer to that question is made trickier by the fact that the 1NT-2NT-3NT auction tells more to the defense about your partner's hand, and the defense knows on the opening lead that you are likely to have a borderline game. You've basically told the defense that you have about 24-25 points between your two hands.
Possible solution: Use Single Dummy Data
The possibility of using single dummy data is tantalizing, but it provides its own set of statistical problems.