[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: AUDITORY Digest - 5 Jun 1999 to 6 Jun 1999 (#1999-82)
Athanassios et al.
A note on the difficulties of correlational research on individual
differences in human capabilities: When we collect moderately large numbers
of performance measures on virtually any types of tasks and when the
individual measures are at least moderately reliable, all measures tend to
be statistically significantly correlated with each other. That is to say,
the estimation of "G" is a nasty but unavoidable problem. When only two
measures are collected, or two classes of measures, one perhaps conjectured
to be causally related to the other (e.g. temporal processing, or pitch
discrimination, and reading disorders), it becomes easy to confuse "G"
effects with specific causal associations between the variables of interest.
I realize that this is very well understood by anyone who knows even a
little about multivariate analyses, but a lot of us in sensory research get
caught ignoring this problem. Some recent work on cognition and individual
differences seems to support the importance of "G" as accounting for a
pretty substantial amount of the total variance in multi-task test
batteries. I heard Pat Killeen give a talk recently in which this point was
made rather convincingly. When he finished someone asked, "so what is "G"?"
With hardly a blink he responded, "well, it seems to be mostly working
memory." Maybe not the gospel, but worth some thought...
A slightly related thought is that correlations over an entire population
can be very misleading. We have recently been measuring a lot of variables
in entering first graders, attempting to identify predictors of early
academic failure, in reading but also in other subjects. With several
hundred children we find that a large number of measures of visual abilities
(corrected, peripheral and central) basically show either no, or extremely
slight correlations with reading measures. But of course we suspect that
there may be among these children a handful with very poor visual processing
abilities, for whom their visual problem really are the key to their poor
reading. The answers are not yet in, but we realize that they will not
likely be discovered purely by examining the overall correlations for the
total sample (about 480 children). One alternative is to combine overall
multivariate analyses with case-by-case studies of the children who fail.
Ugly way to have to do it...any suggestions are welcome. Naturally we have
already looked at cluster analyses, and will continue to try them.
> -----Original Message-----
> From: AUDITORY Research in Auditory Perception
> [mailto:AUDITORY@LISTS.MCGILL.CA]On Behalf Of Automatic digest processor
> Sent: Sunday, June 06, 1999 11:00 PM
> To: Recipients of AUDITORY digests
> Subject: AUDITORY Digest - 5 Jun 1999 to 6 Jun 1999 (#1999-82)
> There is one message totalling 73 lines in this issue.
> Topics of the day:
> 1. pitch discrimination
> Date: Sun, 6 Jun 1999 15:08:12 +0300
> From: Athanassios Protopapas <protopap@PANTEION.GR>
> Subject: Re: pitch discrimination
> I would certainly not doubt the number of subjects or the thresholds
> reported by Dr. Watson. My question pertains to the kinds of differences
> in stimuli, procedure, or threshold estimation that could have led to this
> difference: They have found that 90% of their subjects can tell 1000 from
> 1010 Hz and virtually all discriminate 1000 from 1030 Hz. Our subject
> pool, in which poor readers were purposefully overrepresented, included a
> sizeable proportion of people (maybe around 20%) who had trouble telling
> 1000 from 1050 Hz in the particular setting we used. I am not including
> people who did not establish stable thresholds; presumably the fact that a
> threshold was obtained far from the procedure floor (500 Hz) can be
> interpreted to mean that the subject at least understood the task.
> Performance at this same/different task was highly correlated with
> performance at the up/down task, which was what the original posting to
> the list\ referred to. Note that only half of our subjects (52) received
> the latter task because it was not included in our initial study (which
> was reported at the ASA meeting in 1997), so the correlation of .78 is
> only calculated on 52 points. However, one point I wanted to make was
> that even the subjects with thresholds on the order of hundreds of Hz for
> the up/down task could perform the task to some extent, ie understood what
> the "up/down" meant (possibly with the exception of a couple of very poor
> readers) since they reached stable thresholds in the adaptive procedure
> far from the procedure floor. My other point with respect to the original
> question was that poor performance on this task seems to reflect poor
> frequency representation and not inability to deal with the "up/down"
> labels (hence the strong correlation).
> Back to the topic of frequency discrimination. The correlation of
> performance on the same/different task with a long ISI between the two
> tones to be compared with performance on the same/different task with 0
> ISI between the two tones was only .57 (and the former correlates
> significantly with (non)word reading and word memory whereas the latter
> does not). This is why I used the term "frequency representation" instead
> of frequency discrimination above. It seems that at the level of
> detecting a frequency transition (is this the same as "frequency
> discrimination"?) one finds better and more uniform performance than in
> tasks where a frequency must be retained over a silent interval to be
> compared from memory. I wish I was better versed in the psychoacoustics
> literature to understand this a bit more. Still, I am intrigued by the
> apparently selective correlations between high-level cognitive tasks and
> relatively simple psychoacoustic tasks.
> I am well aware of the studies mentioned by Dr Watson on correlations of
> the same sort. Indeed, I recall being surprised that they did not make a
> big deal out of the fact that frequency discrimination was significantly
> correlated to reading tasks (albeit not as much as in our study)---but
> they had a "temporal processing" theory to test and it is understandable
> that they placed all emphasis on the temporal tasks of their
> psychoacoustic battery. Still, theirs and other findings indicate that
> there are important issues in need of further research regarding
> psychoacoustics and language (oral and written). The procedural
> differences that have led to the observed differences in frequency
> discrimination performance I mentioned above may be a good starting point
> and I would appreciate the input of Dr Watson or others knowledgeable in
> the area regarding possible sources of the discrepancy.
> Thanassi Protopapas
> Athanassios Protopapas, PhD Department of Educational Technology
> Phone: +30 1 680 0959 Institute for Language and Speech Processing
> Fax: +30 1 685 4270 Epidavrou & Artemidos 6, Marousi
> e-mail: firstname.lastname@example.org GR-151 25 ATHENS, Greece
> End of AUDITORY Digest - 5 Jun 1999 to 6 Jun 1999 (#1999-82)