OK, here's a double-blind experiment proposal for the hypothesis that dowsing can detect 'some target'.
I’d get some dowsers together for a test and ask them if they can detect a target ‘thing’ (to be nominated and agreed) under a completely opaque plastic bucket marked by a bamboo pole with a small paper flag. I'd let said dowsers using any method they chooose, rods, forked twigs, waving hands etc. Each dowser's chosen method has to be used for all experimental runs. The subjects that remain will carry out the following experiment.
I’d make a test field up with 10 such buckets upended in holes about 6” deep. Let's pretend I have a big field and can keep them five meters apart. Each bucket would be marked with a small flag set about a yard from the bucket. This identifies the buckets without marking the buckets.
10 random placements would be generated of one such target item, using as close to a random number generator as could reasonably be found. The output would be used for the tests even if it then generated several identical positions.
The dowsers would be kept in a closed room until their turn. Each one would then be led, one at a time, by a chaperone experimenter to dowse the buckets and nominate the one they think the target is under. The chaperone will confirm the number the subject’s picked is the one they meant. The person acting as the chaperone for the dowser will not know where the object is.
The subjects are then led into a separate closed room with its own chaperone.
Then the second subject etc.
Repeat twelve times. Two of the runs will have no target under a bucket (a control).
No results may be changed or altered after any run is complete. The subjects would get to see the buckets being removed after each run so they know where the object really was.
I’ll need another experimenter to replace the target and all the buckets between each run when the subjects are back in their closed room.
I’ve chosen 10 buckets as over ten 'target present' runs everyone will probably get one right.
Two consecutive ‘hits’ would be 1:100. Three would be 1:1,000. Four would be 1:10,000. Be very easy to see if anyone is doing way better than chance, but even someone getting 3/10 overall would still not be that unlikely by chance. Statistics analysis in any event.
It might be worth using a set of ten targets as a second control.
That’d do it.
If the statistical analysis of these results showed that the null hypothesis being true was sufficiently unlikely (we’re supposed to use
p=0.05 as a significance line), I'd run the experiment a few more times to make sure that the 1:20 chance of a fluke result wasn’t just that. That might depend on the number of willing subjects. The more subjects, the greater the significance of the result in general.
If this replication supported the hypothesis that dowsing was a repeatable detection instrument (with presumably a margin of error) I might consider dowsing to detect something else.
Feel fee to run that protocol and write it up.