[Audio_Empire[_2_]]

On Thursday, December 20, 2012 9:43:32 AM UTC-8, ScottW wrote:
On Dec 19, 7:25=A0pm, Scott wrote:
=20
On Dec 19, 9:41=A0am, wrote:
=20

=20
On Wednesday, December 19, 2012 6:39:10 AM UTC-5, Scott wrote:
=20
Well, I am so glad you asked.
=20

=20
Not sure why. Let's take a look at the two key elements of the "data"=
you present.
=20

=20
You quote Howard Ferstler saying, "Even though a 68% correct score lo=
oks like there may have been significant audible differences with the 17 ou=
t of 25 mindnumbing trials I did, that score does achieve a 95% confidence =
level, indicating that the the choices were still attributable to chance."
=20

=20
You quote John Atkinson saying, "In other words, your own tests sugge=
sted you heard a difference..."
=20

=20
Howard is correctly interpreting the statistics here. John is not. A =
confidence interval is a hard target, not a rough idea you only have to get=
close to.
=20

=20
Um no, Howard interpreted the data backwards. he took 95% confidence
=20
level to mean that it was a 95% likelihood that his results were due
=20
to chance. The opposite is true. Atkinson was right. Ferstler was
=20
wrong.
=20

=20
=20
=20
The quote implies Howard took more than one test (each "test"
=20
consisting of 25 trials). If he took 20 tests then it is quite
=20
likely that 1 of the 20 will indicate a false positive result when
=20
using 95% confidence as the conclusion.
=20
It wouldn't be the first time Atkinson "cherry-picked" some data.
=20
As far as just "punching a button"...I would agree the test requires
=20
an honest effort
=20
to discern a difference even if consciously...the subject does not
=20
believe one exists.
=20
That might disqualify some people as a useful subject for such a test.
=20
=20
=20
A relatively simple way to control for this is to use multiple test
=20
amps and to include in one test a control pair of amps that the
=20
subject agrees do sound different. The subject should know this
=20
pairing will occurr but not know when. Even if the subject decides in
=20
a test that they cannot hear a difference in the amps and start
=20
guessing does not invalidate the results. Failing to correctly
=20
identify difference in the control pair would.
=20
=20
=20
ScottW

I still believe that the only way to do a DBT/ABX of something as=20
subtly different as amplifiers should be done with each amp being=20
auditioned for as much as a half hour before switching to the other=20
amp. Use the same cuts from test CDs for each session, played in=20
the same order. Of course careful level matching and strict double-
blindness must still be maintained. I suspect that such a test might
uncover differences that short term, instantaneous switching doesn't
reveal.=20