How about using single dummy data?
All the data in my hand evaluation articles are based on double dummy data, which, as a source of data, is flawed. I've enumerated some of the flaws of using double dummy data here.
The obvious solution would be to use single-dummy data, real plays of the hand from real tournaments.
The most obvious source of lots of single-dummy data is one of the online services like Bridge Player Live or OKBridge. They have archived tons of plays of the hand, and you could mine these for tons of information, potentially.
So why not use single-dummy data?
A whole different set of problems with single dummy data
The problems with single dummy data, unfortunately, are difficult to untangle. They include:
- The level of the players involved
- If an expert wants to know how many tricks he can take with a certain type of hand, would he want to include in the data used for the analysis all hands played by beginners?
- The different bidding at the many tables
- Single-dummy play is affected by the bidding, because the information available to the defense is different. But single-dummy data rarely has the meanings of the bids in the auction spelled out.
- At IMPs, how many tricks we make is dependent on how many we bid
- There are plenty of hands where the field is in 3S, but we belong in 4S. Still, despite 4S being, say, 50%, in 3S, we can take a safety play to make our contract exactly. On the other hand, single-dummy you might make 10 tricks when you've bid 3S, but when you bid 4S, the defense takes its four tricks.
- Not enough data
- The number of deals you need to get any reliable data is actually fairly large. If I want to look at certain rare conditions, like 2NT openers, it is difficult to get enough "real world data" to get significant statistical results.
It's hard to tell in what direction these various factors will skew the data, but I can easily imagine it tending to over-value "standard practices." That's because the players who avoid standard practices tend to be either experts or beginners, with the beginners much out-numbering the experts. The beginners, in turn, will tend to be much worse at declarer play, so he will not as often make his contract. All together, this could lead to a bias towards standard practice.
In any event, single dummy data is not a panacea. It certainly can help provide an antidote to some of the problems with double-dummy data, but it comes with baggage of its own.