Oh boy, we can't really avoid going to the basics when conversing with truly educated, can we?
Lets set out to design a correct test protocol for audibility of speaker cables. Since this question is only relevant in the context of a sound reproduction environment we will need various items to test with - in a world of DOE they are called factors. These would be at a minimum amplifiers, speakers, front ends, musical styles, listening rooms, etc. Note the plural - there are many versions of each factor - in a world of DOE they are called levels. When we try to setup an experimental matrix we quickly discover that the number of tests required to answer our question is impractically large. But instead of giving up, we will attempt to limit both the factors and levels to the bare minimum that can still support a reasonable design.
Lets limit the factors to just three - amplifiers, cables and speakers. We will not determine the importance of others, but we will rationally stipulate that since these factors are further removed from the cables, perhaps their effect is less pronounced. Additionally, we will limit the levels to just two, in order to arrive at the minimal number of required tests. In this choice we will attempt to cover the design space of said factors by using the most unlike versions - a solid state and tube amplifiers and conventional box and planar electrostatics - with all requirements for power being satisfied, of course. For cables we will pick very different approaches - maybe a zipcord and a fancy silver job that costs big money. And we will not attempt to quantify cable length effect, which can be important, in order to make the most compact test matrix possible.
Thus we arrive at the most basic, somewhat lame, but at least straight-face defensible protocol - 3factorial design in a 2cubed format. The "workings" are described here:
http://reliawiki.org/index.php/Two_Level_Factorial_Experiments#The_23_Design
This has the beneficial property of orthogonality, requiring only 8 test runs per repetition to study 3 direct and 3 interactive (confounding) effects. This is so basic that anyone who is setting up experiments that attempt to actually mean something has to know this. It's definitional.
Next we will need repetitions to gain statistical confidence. Upthread, a number of 10 seems to be in favor.
So in order to determine audibility of speaker cables (with large limitations as described above) we will need 80 independent double blind listening tests, presumably with the same tired group of listeners. This is the minimum protocol that is still defensible.
To those advocating DBTs, is that what you do? Because in order to claim anything of significance, this is the minimum that has to be done.
