Would failing half of the foobar abx tests count as undectable?
Aug 17, 2016 at 7:49 AM Thread Starter Post #1 of 5

yepimonfire

500+ Head-Fier
Joined
Dec 6, 2010
Posts
586
Likes
28
Was trying to see if I could tell the difference between 80kbps he-aac and flac.
 
Aug 17, 2016 at 1:12 PM Post #2 of 5
A 50% pass/fail rate is as bad as random guessing. You need a pass rate of at least 95% to have a valid positive result.

That doesn't necessarily mean you are unable to detect a difference, just that you didn't during that test. With some practice, and knowing what to listen for, you may be able to pass it.
 
Aug 17, 2016 at 6:07 PM Post #3 of 5
A 50% pass/fail rate is as bad as random guessing. You need a pass rate of at least 95% to have a valid positive result.

That doesn't necessarily mean you are unable to detect a difference, just that you didn't during that test. With some practice, and knowing what to listen for, you may be able to pass it.

 
50% correct trials will never be significant, but you can have much less than 95% correct to be significant at the 95% level. For instance, 15/20 is only 75% correct but is significant at the 95% level.
 
Aug 17, 2016 at 9:11 PM Post #4 of 5
Was trying to see if I could tell the difference between 80kbps he-aac and flac.

 
That's the best mean rate for guessing. That said, did you use your own, properly ripped or downloaded from reputable sites FLAC files then ripped or converted to 80kbps? Because if you used FLACs downloaded from torrent sites, chances are you downloaded a bum file that was originally low bitrate MP3 then converted to FLAC. I downloaded one album that I couldn't buy locally before, and I thought the mastering was just crap (too warm, dark even) so I decided not to buy it, but then I listened to one of the tracks on Spotify and it sounded normal.
 
Now, if it was reliable FLAC files, did you note down which tracks you got right, and which one you got wrong? If not repeat the test. Then repeat the test again twice over with a different set of tracks, but with the same variety in terms of categories of music. I'm not going to prime you for expectation bias so I won't say why, but get me three data sets where you log which tracks you got right and wrong.
 
Aug 18, 2016 at 2:53 AM Post #5 of 5
   
50% correct trials will never be significant, but you can have much less than 95% correct to be significant at the 95% level. For instance, 15/20 is only 75% correct but is significant at the 95% level.


Yes, 15 of 20 means the probability of this result being random is only 5%.  If you can in fact hear no difference, then one expects overall you would get half right and half wrong.  Yet just from pure random guessing you will 5 out of 100 times just luckily guess 15 of 20 correct.
 
Guessing or not hearing an actual difference would tend toward 50 percent the more trials you have.  So for instance with 100 trials 60 of 100 correct would also be likely the result of random chance only 5% of the time.

With 1000 trials 532 of 1000 or 53.2% would happen only 5% of the time from random guessing.
 

Users who are viewing this thread

Back
Top