Propeller Head Plaza Technical and scientific discussion of amps, cables and other topics.

---

[ Follow Ups ] Thread: [ Display  All  Email ] [ Propeller Head Plaza ]

---

[ Alert Moderator ]

---

That's a good question.

Posted by real_jj on June 1, 2006 at 16:16:11

In Reply to: A question posted by Robert Hamel on June 1, 2006 at 15:55:55:

However, overdetection, for instance, is probabilistic. Ergo, in an ABX or ABC/hr experiment this will come out clearly as random-looking responses from the subject, even given the subject's impressions that might be weak or very strong.

The only way to tell anything in a subjective test is by analysis of the response.

Let's say we have an ABX test between two signals. Say we have, oh, 16 trials. If the subject is responding randomly (and overdetection is such a thing) then the results will fit a binomial distribution. Of course, for any given test you can only say with certainty x% that this was a response due to randomness, or that it was a response that shows that the subject is within y% of responding randomly. Hmm, that isn't so clear, is it?

There are two kinds of results that can happen that cause misinterpretation of the results.

The first is a probabilistic chance (which is always non-zero) that the subjects' response is due to randomness, even if say the subject gets all the answers right. In an abx, that happens one of every 2^n times, for n trials.

The second is the chance that the subject is responding with non-random, but still probabilistic, responses, i.e. p(success)>.5, less than 1. (which is the normal case), but that the results do not pass your criteria for detection.

All of these can ONLY be stated as "within x% of error". There is never, ever, any "certainty".

Of course, if you run 100 trials with a subject, and they get them all right, the chance of that performance happening by random chance is pretty blinkin' small.

But one thing you have to remember, say if I set a 5% confidence bound, is that if I run 20 subjects, 1 of them SHOULD fall inside the 5% bound, on the average, by purely random outcomes.

So one would re-run the positive respose under that scenario.

You're absolutely right that a subject has no idea if they are overdetecting, if they are responding to some expectation, or anything of that sort. Something I've had to explain over and over again is that NO MALICE IS REQUIRED. It's just how we work.

This post is made possible by the generous support of people like you and our sponsors:

---

[ Top of Page ] [ Contact Us ] [ Support/Wish List ] [ Copyright © 2025, Audio Asylum, all rights reserved ]

---

[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]

---