Re: Experiments with large N (Robert Zatorre )


Subject: Re: Experiments with large N
From:    Robert Zatorre  <robert.zatorre@xxxxxxxx>
Date:    Tue, 4 Dec 2007 13:06:54 -0500
List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>

Massimo, thank you for this very interesting reference. I think the point here is not so much about large N per se, but about whether one is interested in characterizing the typical population (for which a more limited sample is sufficient), or whether one is interested in investigating the extremes of a population, which by definition will require one to go and look for the tails of the distribution. Semal and Demany did a smart thing: they pretested 68 people (not such a huge number to test, actually, especially in a group setting as was done here), and then did more intensive testing with a small group of people (N=9) who specifically did poorly on the initial screening. This a good way to proceed if you want to find the extremes of a population. Psychoacousticians tend to use a small number of highly trained listeners, but many labs including my own, do not follow this convention unless we are specifically interested in testing the effects of training. (Indeed, this sometimes gets us into trouble with reviewers who insist that we should only use experienced listeners or else the data are too noisy and not representative--actually it can be the opposite, in some instances, since training and experience have their own effects.) Regards Robert -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Robert J. Zatorre, Ph.D. Montreal Neurological Institute 3801 University St. Montreal, QC Canada H3A 2B4 phone: 1-514-398-8903 fax: 1-514-398-1338 e-mail: robert.zatorre@xxxxxxxx web site: www.zlab.mcgill.ca Massimo Grassi wrote: > Dear all, > >> Huge samples are very nice if you can get 'em, though such is not >> always the case, alas. > > snip! > > If I had a large (very large) N I would be more interested in have a > glance of the population distribution. And this is the reason why, I > think, large N studies are important. > > For example, last year Semal and Demany (JASA, 2006) showed that there > are listerers that are able to detect subtle pitch differences but, at > the same time, are not able to tell the direction of these pitch > differences (up or down). They showed also that there are listeners that > do not show such a dissociation. How is the distribution of this trait > among the population? Are the first listeners a small or a relatively > large amount of the population? > > Think about. Pitch perception is one of the most investigated topic in > auditory perception. Why do we discover this only in 2006? Is it because > the average study is run with few listeners only and these listeners > are, in the majority of cases, the "usual suspects" (e.g., authors, > people with a background of a minimum of 1000 hours of psycho-beeps > experiment, etc.). > > In textbooks we often read that JND for pitch is about 1%. After reading > Semal and Demany results I asked myself: is this figure real? Does it > reflect all the population? Which part of the population? For which > particular task? > > It is trivial to stress that, with large N, statistical power goes soon > very high. I think the goal of large N study is different: have, at > least once, a real glance of the population. > > m >


This message came from the mail archive
http://www.auditory.org/postings/2007/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University