[Harry Lavo]

"vlad" wrote in message
...
Harry,

first of all, ABX does not need validation, it is a tool and you use
it when you need it. When I need to put nail in a wall I use hummer
without worrying if it is validated for this task or not.

Your 'monadic' test has one huge flow - it brings a lot more
variables into the test that are not under your control but they
definitely influence outcome of the test.

For instance, if the test takes more then one day then temperature
and humidity of the air will definitely affect physical abilities and
mood of your subjects and become parameters of the test too. Even the
weather itself becomes a parameter. In sunny weather people react to
the music differently then in a cloudy weather.The quality of the food
from day to day (you know, some of them can have problems with
indigestion) can become a real issue too. I am sure there are many
other things that are not under your control.

Another thing is that you have no control over random guesses vs.
real recognition. But I better not to open this can of warms :-)

my $.02 worth

vlad

I appreciate your concern, Vlad, but it is misplaced.

Research designers try to anticipate and take into account significant
possible intervening variables. In such a monadic test, no doubt the
variable would be changed from one session to the next so that any point in
time, the sampling would be roughly 50-50. Musical segments would be
rotated within samples so there is no order bias, etc. etc.

When you have a large sameple size and randomly chosen and matched samples,
you don't worry about a few random guesses. The fact is, their is a very
well developed set of statistical operations that take into account the
"degree" of difference between the ratings of the two samples. A different
standard applies depending on the number of scaler points used, whether the
scalars are symetrical or not, etc etc. And the significance level is
determined by sample size and the shape of the distribution curves as it
determines standard deviation and standard error.

And if you really want to worry about random guesses, worry about an ABX
test where a change in one sample can determine whether or not the test is
judged signicant, and there are NO controls against random guessing to
create a virtually guaranteed "null" effect.