Blind Listening:
Page 4 Overall, 19.9% of the listeners scored 2 correct or less out of 7, 53.8% scored 3 or 4 correct, while 27.9% scored 5 correct or more out of 7. The best group score was 82% correct identification of the Adcom/VTL difference in Session V with the Fauré Requiem movement, and the worst was only 23% correct identification of Adcom/Adcom in Session II on the King's Singers track.

You can see from the data in fig.1 that the proportion of KEOs ranged from 13.3% in Session VIII to 47.3% (!) in Session I. How come? Beats me. I've long since abandoned fruitless speculation, but there it is. No, it wasn't differences in the music; the selections were always the same, and almost always in the same order. It certainly wasn't impossible to get them all right—6 listeners did so (and one of these even got all the amplifiers correct!), and 37 got 6 out of 7 correct. However, even though the amp selection was to some extent loaded in favor of there being an audible difference, a shade over half the listeners didn't do any better than by chance, and 20% did worse. No, no pause for wondering "What about...?" now; that's later.

Did the group, when considered as a whole, identify the amplifiers by ear? Looking at fig.2, which shows the percentages of correct answers as a bargraph, it can be seen that what should be a symmetrical bell-curve is somewhat skewed over to the identification side. Table 2 shows the overall distribution of correct identifications, as well as the results session by session. Of the 3530 attempts to identify if there had been a change in amplifier or not, 1846 (52.3%) were correct. This may seem very close to what could be achieved by chance (1765 correct, or 50%), but in fact, with such a large number of responses, the probability that this could have occurred by chance computes as being reasonably improbable (footnote 8).

Overall scoring (505 listeners)

The next step in the analysis was to see if it made any difference in responses whether the amps being compared were the same or different. It sure did! The data in Table 3 show that a difference was correctly identified over 64% of the time, but that sameness was correctly picked only about 38% of the time. This difference in results, at these numbers of responses, is way, way beyond the possibility of chance ("highly significant," in statistical terms). Of further interest, Adcom first, followed by VTL, was slightly better identified than the other way around (see Table 2). However, when there was no difference, it didn't make any difference which was the same, Adcom/Adcom being misidentified as often as VTL/VTL. Although there was again some inter-session variation in the specific proportion of correct responses to the four possible amplifier pairings, none of it went much against the overall trend.

JA had wanted to know whether the listeners were able to discriminate amps better with one musical selection than another. This was examined in considerable detail, especially to gain knowledge for the future—it's an obvious waste of time to test for discrimination using test material that doesn't allow it. Table 4 shows the overall percentage of correct identifications of "Different" and "Same" for each piece of music, arranged in order of success, while Table 5 shows the detailed analysis for the drum recording.