Reviews with all double blind testing?


In the July, 2005 issue of Stereophile, John Atkinson discusses his debate with Arnold Krueger, who Atkinson suggest fundamentally wants only double blind testing of all products in the name of science. Atkinson goes on to discuss his early advocacy of such methodology and his realization that the conclusion that all amps sound the same, as the result of such testing, proved incorrect in the long run. Atkinson’s double blind test involved listening to three amps, so it apparently was not the typical different or the same comparison advocated by those advocating blind testing.

I have been party to three blind testings and several “shootouts,” which were not blind tests and thus resulted in each component having advocates as everyone knew which was playing. None of these ever resulted in a consensus. Two of the three db tests were same or different comparisons. Neither of these resulted in a conclusion that people could consistently hear a difference. One was a comparison of about six preamps. Here there was a substantial consensus that the Bozak preamp surpassed more expensive preamps with many designers of those preamps involved in the listening. In both cases there were individuals that were at odds with the overall conclusion, and in no case were those involved a random sample. In all cases there were no more than 25 people involved.

I have never heard of an instance where “same versus different” methodology ever concluded that there was a difference, but apparently comparisons of multiple amps and preamps, etc. can result in one being generally preferred. I suspect, however, that those advocating db, mean only “same versus different” methodology. Do the advocates of db really expect that the outcome will always be that people can hear no difference? If so, is it the conclusion that underlies their advocacy rather than the supposedly scientific basis for db? Some advocates claim that were there a db test that found people capable of hearing a difference that they would no longer be critical, but is this sincere?

Atkinson puts it in terms of the double blind test advocates want to be right rather than happy, while their opponents would rather be happy than right.

Tests of statistical significance also get involved here as some people can hear a difference, but if they are insufficient in number to achieve statistical significance, then proponents say we must accept the null hypothesis that there is no audible difference. This is all invalid as the samples are never random samples and seldom, if ever, of a substantial size. Since the tests only apply to random samples and statistical significance is greatly enhanced with large samples, nothing in the typical db test works to yield the result that people can hear a difference. This would suggest that the conclusion and not the methodology or a commitment to “science” is the real purpose.

Without db testing, the advocates suggest those who hear a difference are deluding themselves, the placebo effect. But were we to use db but other than the same/different technique and people consistently choose the same component, would we not conclude that they are not delusional? This would test another hypothesis that some can hear better.

I am probably like most subjectivists, as I really do not care what the outcomes of db testing might be. I buy components that I can afford and that satisfy my ears as realistic. Certainly some products satisfy the ears of more people, and sometimes these are not the positively reviewed or heavily advertised products. Again it strikes me, at least, that this should not happen in the world that the objectivists see. They see the world as full of greedy charlatans who use advertising to sell expensive items which are no better than much cheaper ones.

Since my occupation is as a professor and scientist, some among the advocates of double blind might question my commitment to science. My experience with same/different double blind experiments suggest to me a flawed methodology. A double blind multiple component design, especially with a hypothesis that some people are better able to hear a difference, would be more pleasing to me, but even here, I do not think anyone would buy on the basis of such experiments.

To use Atkinson’s phrase, I am generally happy and don’t care if the objectivists think I am right. I suspect they have to have all of us say they are right before they can be happy. Well tough luck, guys. I cannot imagine anything more boring than consistent findings of no difference among wires and components, when I know that to be untrue. Oh, and I have ordered additional Intelligent Chips. My, I am a delusional fool!
tbg
As I recall, statistics can be very useful.

Stat 101....Intro to Statistics
Stat 102....Statistic Applications (How to fool others using statistics).
Stat 201....Advanced Statistics (How to fool yourself using statistics).

Just kidding. In my work with balistic missile inertial guidance systems, such as the estimation of CEP (circular error probability) based on a couple of hundred modeled error sources, I have been exposed to the most arcane forms of statistics. One must always remain aware of the risk of fooling yourself, and be able to laugh about it.
I just think the proper hypothesis should be that a sample of people can hear a difference between cables or amps.

Well, that's one possible hypothesis. Another possible hypothesis is that one particular individual can hear a difference. That's the equivalent of testing the fairness of one particular coin. Note that the sample size isn't one. It's the number of listening trials/coin flips.

I am only concerned that the choice of the sample size may be determined by what the researcher's intended finding might be.

The choice of sample size isn't what's critical here. The statistical significance is. Granted, larger samples reduce the possibility of false negatives, but it's not as if there have never ever been any ABX tests with large sample sizes. The Stereo Review cables test had a sample size of 165. The possibility of a false negative is very low with a sample that big. (Since you teach statistics, I'll let you do the math.)

And if you think the reason these tests come up negative so often is sample size, you as a "scientist" ought to know how to respond: Do your own experiment. Complaining about other people's data isn't science.

I think it is a far more interesting hypothesis to suggest that those with "better ears" would do better.

Then test it. The SR panel was a pretty audio-savvy bunch, as I recall.

I don't think most audiophile would be convinced or should be convinced that all amps or wires sound the same.

Are you saying they're all close-minded?
Pabelson, frankly I don't care enough about this question to expend the time necessary to do such work. I am more concerned with find a great loud speaker.

I just do not understand the expectation that all individuals are the same in these tests. It is not statistical significance, it is improbability that you are talking about.

How do you know when you wrongfully reject the null hypothesis?