Re: Improving 2-channel recommendations!!!


"John Corbett" <corbett@xxxxxxxxxxxx> wrote in message
news:corbett-1911051934410001@xxxxxxxxxx
>
> Since for an ABX test pure guessing would correspond to a long-term
> success rate of 50% and perfect detection would be 100%, we can take
> threshold-level performance to be 75%. It is important to remember that
> these performance figures involve the long-term average success rate, and
> not merely the observed proportion in a single run of trials. Stimulus
> levels above threshold are those having at least a 75% probability of
> correct response, corresponding to an actual detection rate of at least
> 50%. In other words, the stimulus is more likely to be detected than
> missed.
>
>
> If by positive results you mean 14 or more correct in 16 trials, then your
> claim is essentially that at the 99% confidence level the difference
> between pure chance guessing (50%) and getting 97% correct answers in the
> long run is too small to worry about! Remember that 75% long-run
> performance is usually regarded as substantially better than chance. (If
> you know what a confidence level and a confidence set are, then you should
> be able to verify that based on data of 13 correct in 16 trials a
> one-sided upper 99% confidence bound for the true success probability is
> about 0.971.)
>
> To be 99% confident that an effect is below-threshold (i.e., that the
> effect size is no more than .75 correct long term) you have to see
> performance with no more then 7 correct in 16 trials. Even a score of 8
> correct in 16 trials is _not_ bad enough to rule out psychoacoustically
> significant effects.
>
> So at the 99% confidence level, a score of 14 or more is evidence that the
> subject is really doing better than 50% correct (long-term), and a score
> of 7 or less is evidence that his true long-run performance is either no
> better than chance or not enough better than chance to be worried about.
> But scores from 8 through 13 are inconclusive, as they are simultaneously
> consistent with pure chance and above-threshold performance.
>

They are more than inconclusive, they are a result that is
better than chance, it is more likely a result from
hearing differences than it is likely a result from chance.
In terms of consumer preference, that is sufficient, the consumer does
not need any more of a confidence level than
'the preponderance of evidence" in making a choice
based upon a preference of perceived sonic difference.


.

Relevant Pages

  • Re: Improving 2-channel recommendations!!!
    ... >> success rate of 50% and perfect detection would be 100%, ... corresponding to an actual detection rate of at least ... >> performance is usually regarded as substantially better than chance. ... >> you know what a confidence level and a confidence set are, ...
    (rec.audio.opinion)
  • Re: Meaningful vs. Statistically Siginificant?
    ... items off the assembly line, and find that 95 of them are within specs, that we would be confident that we would find that 950 of 1,000 items would be within specs if our original sample size of 100 ... don't have the patience to look up how to do the confidence level ... Your 95% is a measure of the expected success rate. ... There are, in fact, topics within statistics that answer questions ...
    (alt.usage.english)
  • Probability vs. Confidence Levels
    ... but ties it to a confidence level. ... The current process calculates the percent detection as ... being equal to the percentage of sampled faults that were ... 32.3% probability of 90% fault detection. ...
    (sci.stat.edu)
  • Re: The tainted logic of ABX=Maoists of the Cultural Revoulution
    ... >>> But we are not talking about scientific research. ... >>> The FACT is, if the confidence level for an individuals's result is, ... >>> result is not from chance. ... Excelllant comatery on thuh natcher of statisitcs. ...
    (rec.audio.opinion)
  • Re: _Kitty Goes to Washington_, Carrie Vaughn
    ... she has a 99% chance of surviving the fight, ... chance of dying in a week, over a 26% chance of dying in a month, and ... only a 2.6% chance of *surviving* a year. ... With even a 99% success rate, ...
    (rec.arts.sf.written)