
Home  Forums  Rules  Articles  diyAudio Store  Blogs  Gallery  Wiki  Register  Donations  FAQ  Calendar  Search  Today's Posts  Mark Forums Read  Search 
The Lounge A place to talk about almost anything but politics and religion. 

Please consider donating to help us continue to serve you.
Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving 

Thread Tools  Search this Thread 
21st November 2018, 01:06 PM  #10011  
diyAudio Member
Join Date: Jun 2004
Location: germany

Quote:
And during my own test we followed the same procedure.... Addressing your question for the statistical part unfortunately needs a short recap of some basics. (for simplicity i neglect all philosophical differences between NHST, Fisherian experiments and the Neyman/Pearson approach) The null hypothesis is stated before (random guessing, H0: p = 0.5), so two different errors can be made when analysing the results of a listening experiments: The null hypothesis could be rejected although it is true (socalled alpha error) and the null hypothesis could _not_ be rejected although it is _false_ (socalled beta error) The guard against alpha errors is to choose a significance level that is sufficiently low; the usual SL = 0.05 means that in the long run the alpha error will be <= 5%. So this required SL decides which minimum number of samples is needed. Therefore 5 samples are sufficient (in a onesided test) as the probability to get 5 hits in 5 trials is 0.032, which is below 0.05 The guard against beta errors is the statistical power (1 beta), so if you want to keep the beta error as low as the alpha error, you need a statistical power of (10.05) = 0.95. The usual minimum required statistical power is 0.8, which means to accept a beta error of 0.2 , that is already 4 times higher as the alpha error. To calculate the statisticial power you have to assume something about the difference under test conditions. The calculations i´ve shown before therefore assumed p2 = 0.6 (instead of the p = 0.5 as assumed under the null hypothesis) . I hope that helps to understand why the requirements for low alpha and low beta errors are different. And i hope it shows why training under the specific test conditions helps, because it might (if a difference is detectable) the proportion of correct answers and therefore lowers the number of samples required. Last edited by Jakob2; 21st November 2018 at 01:16 PM. 

21st November 2018, 01:31 PM  #10012  
diyAudio Member
Join Date: Aug 2014

Quote:
Quote:
Quote:
Quote:


21st November 2018, 01:37 PM  #10013  
diyAudio Member
Join Date: Dec 2015
Location: Penrhyndeudraeth

Quote:


21st November 2018, 01:38 PM  #10014  
diyAudio Member
Join Date: Apr 2002
Location: Prague

Quote:
And you are asking a good question  unfortunately I am not able to give you similarly good answer .
__________________
Pavel Macura http://pmacura.cz/audiopage.html https://www.diyaudio.com/forums/ever...4samples.html 

21st November 2018, 01:58 PM  #10015  
diyAudio Member
Join Date: Aug 2014

Quote:


21st November 2018, 02:10 PM  #10016  
R.I.P.
Join Date: May 2007

Quote:
Quote:
Quote:
Everyone has a worldview; everyone tends to reject anything which conflicts with that worldview, so I am happy to admit that I am a fairly normal person. Those who think that they are different from this are merely better at fooling themselves. Quote:
Quote:
I will show another piece of bias: I suspect that your position is at least partly motivated by a commercial interest in audio. What form that takes I don't know, but my experience is that dogged persistance (and faith in sighted tests) is often a sign of commercial interest. My experience is also that people with a commercial interest are often reluctant to say so, until one day it spills out when they use commercial success bragging in lieu of reasoned argument. Quote:
Last edited by DF96; 21st November 2018 at 02:11 PM. Reason: extend 

21st November 2018, 02:25 PM  #10017 
diyAudio Member
Join Date: Aug 2014

@DF96, your pedantry is getting tedious for everybody, I suspect so I'l keep it short?
It's not a 'competition', just a debate on various flaws in listening tests So you proffer no definition of what is a "reasonable" unsighted test to you  you simply "added the "reasonable" adjective to avoid being accused of accepting any old junk as evidence." So tell us how you know the unsighted test isn't just comprising of false negative results or are you not concerned about that, only wishing to minimize false positives? 
21st November 2018, 02:44 PM  #10018 
diyAudio Member
Join Date: Dec 2015
Location: Penrhyndeudraeth

Oh, I don't know, most of his posts make me smile

21st November 2018, 02:46 PM  #10019 
diyAudio Member
Join Date: Apr 2003
Location: Sweden

mmerrill99, your are in the audio business  right? Did I recall correct that you have a website?
//
__________________
More distortion to the people! 
21st November 2018, 02:47 PM  #10020 
diyAudio Member
Join Date: Dec 2015
Location: Penrhyndeudraeth

Bet he won't answer

Thread Tools  Search this Thread 


New To Site?  Need Help? 