老外的一次双盲测试

灰色的天空

这儿有一次关于电源线双盲测试的文章，提供给一些认为盲听不科学的人批判，文章链接如下：
http://www.hometheaterhifi.com/volume_11_4/feature-article-blind-test-power-cords-12-2004.html

把结论部分贴一下：

Conclusions

To many in the engineering community, blind ABX is an accepted experimental design. Using the blind ABX protocol, we failed to hear any differences between an assortment of generic power cords and Nordost Valhalla. Therefore, we cannot conclude that different power cords produce a difference using the blind ABX protocol. However, we also cannot conclude that there are no differences. We simply failed to prove that differences can be detected to a statistically significant degree using a blind ABX protocol.

John Johnson, who comes from a scientific background, suggests that if there are differences between cords, they appear to be so subtle that a blind ABX test cannot discern them with small numbers of participants. Failure to discern them could be due in part to the time it took for cable changes, and the possibility that accurate auditory memory is shorter than that. It may be necessary to switch between cords in a much shorter time.

Unfortunately, as John notes, we don't know of a way of accomplishing fast power cable changes, since, unlike interconnects which can be simply switched between A and B with the equipment all still powered on and playing music, changing AC power cords requires turning the equipment off, switching the cords, and then powering them back on.

The test was a grand and noble experiment at best and a bust at worst. Make of it what you will.

Editor's Note: ABX tests are valid and do work. Here is a link to some ABX tests of various types of audio products. In many cases, statistically significant differences could be discerned by participants. In others, no differences could be discerned.

Now, of course, one can dissect an experiment and say, well these 4 people out of 10 participants had good scores, so they could hear the differences. But, no, you have to take all the data together. You can't just pick out the numbers that suit your hypothesis. This would be statistically invalid. Same thing with just looking at one music selection. With statistical random patterns, it is likely that there will be one selection where more participants score correctly than on other selections. If we had enough music selections, there would likely be one where all participants scored perfectly. But, you have to look at all the selections together. That is the purpose of statistics. You may remember the famous monkeys typing randomly concept. If you have enough monkeys, eventually one of them will type all of Shakespeare's works perfectly. To look at only that one monkey might suggest it knew how to type Shakespeare. But, we can't do that and claim good science.

ardylei

交易区新帖推荐

进来参观学习一下。

XLH1812

楼主,这里40岁以上的人,超过80%没高中学历,你贴这么大段英文,谁看得懂?  

zhjzhjzhj

[s:97][s:97][s:97]

mozart08

[s:14][s:14][s:14]

发烧盗版

交易区新帖推荐

XLH1812 发表于 2013-12-4 09:33
楼主,这里40岁以上的人,超过80%没高中学历,你贴这么大段英文,谁看得懂?

每个字母都看得懂啦，哈哈哈…拼起来就…不过有在线翻译，只是一些莫名其妙的词冒出来，自己整理一下就行了

Xy770509

草，这种屁话用英文写就比较有说服力吗？要是论文还有严谨点，有点意义，这种[s:14][s:14]和我们论坛上这些胡扯有什么区别？

发烧盗版

发烧盗版 发表于 2013-12-4 10:25
每个字母都看得懂啦，哈哈哈…拼起来就…不过有在线翻译，只是一些莫名其妙的词冒出来，自己整理一下就行 ...

有些音响名词让不熟悉音响的英语本科学生甚至是大学英语老师翻译，双盲可能翻译成双眼全瞎…哈哈哈

smsmmsms

XLH1812 发表于 2013-12-4 09:33
楼主,这里40岁以上的人,超过80%没高中学历,你贴这么大段英文,谁看得懂?

[s:97]

自由心情

XLH1812 发表于 2013-12-4 09:33
楼主,这里40岁以上的人,超过80%没高中学历,你贴这么大段英文,谁看得懂?

一语中的啊，还基本都不会乐器[s:97]

chncwk

好像没人关注老外到底讲了些什么：

结论

在工程领域的许多情况下，双盲测试是一个可被接受的实验设计。采用双盲测试方案，我们没有听出一般的电源线和 Nordost Valhalla 之间的差异。因此，我们不能得出使用双盲测试方案下不同的电源线会产生差异的结论。但是，我们也不能断定就没有差异。我们只是使用一个双盲测试方案未能证明统计学上可被检测到的显著程度的差异。

约翰·约翰逊，来自科学界的背景，指出：如果芯线之间有差异，它们似乎是如此微妙，一个双盲测试不能令小数量的参与者辨别它们。辨别它们的障碍某种程度上可能是由于准确听觉记忆的时间可能比用于切换电缆的时间要短，这提示可能有必进一步缩短在线之间切换的时间。

不幸的是，约翰指出，我们不知道实现快速改变电源线的方法，因为，不像连接器那样可以简单地在A和B之间切换，设备都仍处于开机和播放音乐状态，改变交流电源线要求打开和关闭设备，切换电线，然后再重新接通它们。

该测试是一个伟大，严肃的实验，在最坏的情况下，在好与差之间，按你的意愿去标识它们。

编者按：双盲测试是有效和可行的。这里是一个链接关于各类音响产品的一些双盲测试。在许多情况下，在统计学上显著差异可被参与者识别出来。另一方面，无差异也能被识别。

现在，当然，人们可以分析一个实验，然后说，好了，这4人超过了10名参与者，取得了好成绩，所以他们能听出该种差异。但是不行，你必须把所有的数据放在一起。你不能只是挑选出符合你的假设的数据，这在统计学上是无效的。该结论同样适用于只是在一首音乐样本寻找的情况。在统计学的随机模式下，很可能会出现一个样本下更多参与者的打分比其他样本更准确的情况。如果我们有足够多的音乐样本，那就可能比只有一个样本下的所有的参与者的打分更接近完美。但是，你必须考虑所有的样本在一起。这就是统计的目的。你也许还记得著名的猴子随机打字的假设。如果你有足够多的猴子，最终它们中的一个将完美输入所有莎士比亚的作品。只看那一只猴子的话，可能暗示着它是知道怎样输入莎士比亚的。但是我们绝不能这样做并称这是好的科学。

nsx-r

每次本着科学精神来到折腾（玄学）领域都是找死，不解释。