Differential EEG

Glen M. Sizemore gmsizemore2 at yahoo.com
Sun Mar 14 07:56:59 EST 2004


> GS: Hmmm. Sounds intriguing. I would like to see the individual-subject
> data. Any chance you could provide them? For the record, in experiments
like
> this, it is unnecessary to average the data from individual subjects
> together, despite the widespread practice of doing so. If one has
plausible
> baseline data, then one may demonstrate the effect repeatedly in each
> subject (i.e., several replications right there) and repeatedly in the
other
> subjects (i.e., several replications within each subject as well as the
> across-subject replications). Data obtained and analyzed in this fashion
can
> be used to DIRECTLY DEMONSTRATE the reliability of the finding, and rarely
> prove to be spurious.

NMF: Although you can look at the individual scores for the time estimation,
the
problem with that method is that individual variability may cause
comparisons between subject difficult. The only thing you could conclude is
a trend in the data set.

GS: No. Or rather, it depends on what you mean. If you have adequate
baseline data for all of the subjects, then you have a direct measurement of
the range of variation under baseline conditions. If each subject in the
clockwise group (or whichever was the one that was effective) shows no data
outside of the baseline range of variation under conditions of field
intensity, then there is no behavioral effect (at least as measured by you)
of such manipulation. And if every subject in the counter-clockwise group
showed data points outside of their range of variation under various
conditions of field intensity then there was an effect. It is as simple as
that.

Now, I don't know what you're driving at when you say that "the only thing
you could conclude is
a trend in the data set." On the contrary, if you observe, in the individual
subject data, what I described, then you can pretty much take that effect to
the bank - of course, it still could be that there is some
experimenter-induced variable or other extraneous variable operating, but
this is not a problem with the analysis of individual subject data per se.
It is a myth that the use of inferential statistics is necessary or even
appropriate for the analysis of behavioral data, and other sorts (but not
all) of data. But, obviously, I know about the current view in many areas of
science concerning inferential statistics (most of what I have published
contains inferential statistics, but this is largely because the journals
simply do not publish individual subject data, except for primates), but I
am not going to make an extended argument at this time.

NMF: Moreover, because this was factorial design with
one level not repeated (field rotation) and one level repeated (field
configuration), one would not be able to make between group comparisons
using the method that you suggested. To reiterate, all subjects wereexposed
to the same fields but the rotation of the field (clockwise vs.
counterclockwise) was different.

GS: If there were no baseline data, and no manipulation of intensity (was
there - I don't recall) then what you say above would be correct. However,
as I said, since the effects (or lack thereof) of the fields may be directly
assessed in individual subjects, examination of them is critical. I admit,
however, that this is not the mainstream view. Personally, if I don't think
an effect is observable in individual subjects, I usually don't consider it
an effect. There are some questions to which such methods cannot be applied
as well. But this is not one of them.

NMF Unfortuantely, I do not have their actual raw data so I cannot really
comment regarding each individuals trend with time estimation, however,
considering that they reported a significant difference between subjective
estimates of time for the subjects exposed to the counterclockwise relative
to the clockwise presentations of the rotating electromagnetic field, one
can conclude that generally the subjects in this condition showed the
increased trend towards longer time estimations compared to the other grou
field rotation.

GS: No, I'm sorry to say that you cannot conclude this.

NMF: The reported standard deviations for the groups were
essentially the same and did not show large variability (i.e. the
distribution was not anisotropic and the presence of influential points
within the data set is highly unlikely).

GS: Consider the made-up data below. Let's say these are data from a
pre-test and a post test. There are 10 subjects:

Pre. Post

6.00 6.00

7.00 6.00

2.00 2.00

7.00 7.00

5.00 5.00

4.00 7.00

7.00 9.00

4.00 6.00

6.00 8.00

3.00 7.00

The mean of the first column is 5.1 and the sd=1.792.

The mean for the second column is 6.3 and the sd=1.889.

A paired t-test "reveals" that these data are statistically significant at
p<0.05. Yet, inspection of the data reveals that only ½ of the subjects
showed anything that looks like an effect. The data for 4 subjects stayed
the same, one subject's score decreased, and 5 subjects showed the increases
that produced the statistical significance.

There it is - half the subjects show nothing, and half show an increase.
Want to get it down below 50%? Just increase the N.


> Err....you may have mentioned, but I'm guessing that you are talking about
> humans here? Have you considered using non-humans? Is this done? What has
> been found? Off-hand, I can think of about 4000 experiments (assuming one
> can show some kind of effect).
>
> BTW, was the procedure supposedly free of experimenter-induced artifact?
> What was the nature of the "time estimation" task?

NMF: Yes we are talking about humans in this experiment. And yes they have
done
experiments using animals (primarily rodents and mice). I am by far more
familiar with the animal research then the human research, which was my
major focus when I was in the laboratory. However, I thought it would be
far more interesting to bring up the reference to the human study regarding
the time-estimation data and its implication to the 40 Hz temporal binding
frequency and consciousness. Unfortuantely, they have not done any studies
looking at time-estimation in animals exposed to these fields. The problem
with time-estimation in non-human primates is extremely difficult. However,
there has been studies that have investigated "subjective" time-estimation
in animals. I do not know how credible such studies are, however, one can
make the same argument regarding human studies. There has been studies
using these complex field configurations in animals in order to assess their
effects on learning and memory, analgesia, addiction, development,
aggression, epilepsy, etc. etc. The difficulty for affecting
time-estimation by the fields used in this study, in say a rodent model, is
that peak spectral cortical frequencies of the rodent is higher than man.
(which is pretty obvious). Thus, the field parameters would have to be
changed to accomodiate for this. The rational for the configurations that
they selected was that the data from Llinas and Ribary and thereotical
estimates from Edelman have sugggested that the temporal substrate for
conscious awareness may be recreated every approximately 20-25 msec (40 Hz)
by phase shifting transcerebral electromagnetic fields propagating in the
rostral to caudal direction throughout the cortical manifold. These phase
shifting fields were experimentally measured in humans previously (Llinas
and Ribary, 1993; Jeffreys et al., 1996); however, there has some of you
have suggested these processes may be artifactual of measurement.
(Ipersonally do not believe there is evidence that has shown that intrinsic
oscillations of cells are mere artifacts, especially when they measure these
cells individually and within a network. Both the in vivo and in vitro work
supports that these electrical characteristics are evident within neuronal
tissue. )

If one was to mae the appropriate calculation taking this into account
cortical frequency differences between the rodent brain and the human brain
and calculate the intrinsic binding frequency in the rodent brain that would
correlate with the 40 Hz binding range in the human, then one could change
the temporal parameters of the fields and do the study on rodents with an
appriopriate experimental design. This has been something that I have
thought about doing before. (Oh well... maybe in the future).

GS: Well, I don't know about the stuff that begins: "...data from Llinas and
Ribary and thereotical
estimates from Edelman have sugggested that the temporal substrate for
conscious awareness may be recreated every approximately 20-25 msec (40 Hz)
by phase shifting transcerebral electromagnetic fields propagating in the
rostral to caudal direction throughout the cortical manifold."

I'll be honest - it sounds like a load of crap (not that the phase shift
exists - if that's a fact it is a fact). Anyway, though, I am vaguely
familiar with timing studies in animals and many of them don't seem to
produce the temporal control required for a good experiment, but I'm pretty
sure one could come up with something. Plus, I'm not sure why you would have
to examine "timing." Wouldn't you be interested in almost any behavioral
effect of such fields? And if not, why not? And if you think that such
fields "affect consciousness" why not try to establish discriminative
control over responding by fields?

NMF: The method of time estimation they used was taken from Zakay et al.,
(1994)
from Memory and Cognition. 22: 344-51. The subjects are aware that temporal
judgement will be required. With this method attention of the subject is
directed in real time to information that is related to the passage of time.

GS: OK.

>
> GS: I know you weren't trying to be tricky. Anyway....I'd like to see the
> individual data; I can't emphasize this enough despite the minority status
> of the position. It is easily possible to "get an effect" in less than ½½
the
> subjects and still obtain statistical significance. Where
individual-subject
> data CAN be examined (as when each subject is exposed to the levels of the
> independent variable, and where the baseline is replicable - one mayreturn
> to the baseline conditions after each exposure, for example) they SHOULD
be
> examined. Again, if you can turn the effect on and off at will in, say, 4
of
> 4 individuals, then there is no need to "infer" the effect with
inferential
> statistics, the reliability is directly demonstrated and the generality
may
> be established in similar ways via systematic replication in other
> laboratories).
>
> Now, if one does keep obtaining a statistically significant effect that is
> not demonstrable in most of the individuals, then one must track down the
> source of this difference - of course, one only knows of this if one
> examines the individual-subject data.

NMF: Exactly. This is exactly what or presumably what a repeated measures
analysis is suppose to do. It is suppose to accomodate and explain more of
the variance associated with that error term placed in parameteric analyses.
Moreover, since pooled variance estimates are calculated to perform
parametric analyses the estimations are better than relying on statistical
assessments that would not take these parameters into account. In my
opinion, they analysed the data properly and correctly based upon the goals
that they setup to investigate in this study.

GS: I have already commented on what is widely held and what is actually the
case.

NMF: The analyses in this experiment were on the relative differences for
each
subject subject scores, which allow for more statistical control over
individual variability and accomodiate for individual differences. As I
stated before, the data does not seem to be skewed so the possibility of a
spurious result is highly unlikely. (I inspected the means and standard
deviations for each condition. If you like you can download the paper and
do the same: see Cook, Koren, Persinger (1999). Neuroscience Letters. 266:
61-64.). So it is unlikely, based upon looking at the disperision for the
relative mean time estimates, that the cluster of cases was hetergeneously
distributed. The means had very tight disperision variability.

GS: Obviously, I have already made comments relevant to this.

NMF: Also because the rotation direction of the field was different between
the
two groups one cannot use the approach that you suggested to compare.

GS: Right, but I modified it above (one can still compare two different
groups, as long as the effects of the independent variables can be assessed
in individuals). Second, the experiment should have been done otherwise,
with each subject exposed to all of the conditions.

NMF: Your
approach of just analysing the individual differences would be equivalent to
running a repeated measure analysis of variance with the repeated measure
being the field parameter. Thus, you would determine the difference in
subjective time estimations that way. That is essentially what they did and
they utilized symmetrical correlated or paired t-tests to assess the
differences. So your suggestion is exactly what they did. The only
addition is that they have another level to their design (i.e. field
rotation).

Perhaps I'm failing to see how you would analysis the data otherwise. One
cannot just look at the data and just because the estimates change or seem
to change infer there is a difference, obviously statistical measurements
are required.

GS: I disagree, and so do a small minority of scientists. If the baseline is
reliable (and one demonstrates this sometimes by simply showing that there
is no systematic trend) then anything outside the range of variability is an
effect. In this fashion, one directly demonstrates effects as well as their
reliability and generality. I am well aware of how you must perceive this
statement. There is a huge amount wrong with null-hypothesis statistical
testing, and I am not the only that thinks so. Though, admittedly, such a
position is the minority position.

NMF: Maybe you can tell me how you would analyse the data exactly.


GS: Well, I would have to do the experiment differently perhaps. Did they
observe "time estimation" across time under the "no field" conditions? I'm
too lazy to go get the paper. Anyway, it is essential that one observe
time-estimation under "no field" conditions and to ascertain the expected
range of variation under baseline conditions. The best way to do this is to
plot the dependent measure as a function of time. In the laboratory where I
work, and in many like it, daily experimental sessions are conducted and
various dependent measures are plotted daily (i.e. dependent measure on the
y and date, or ordinal session number on the x). The data are judged
visually (you know, those two things in our heads that we see with?) And
when there is no systematic trend one introduces the independent variable or
changes its level as the case may be. Some people use "stability criteria"
of various sorts, but these are usually derived by visual inspection anyway!
Once one has this, one introduces the field and continues to plot the
dependent measure(s). If the data points do not exceed the range, there is
no effect. If they do, either immediately or after exposure, and they stay
outside the range, then there is an effect. If the points go out and come
back into the range, then the effect is transient. Yes, I am dead serious.
> GS: I am definitely intrigued, but unconvinced there is an effect there. I
> am, however, completely open to the possibility that there is, and I think
> that it should be relatively easy to establish using a few non-humans in
> extended studies.

NMF: Agreed. That is why this field is a work in progress. Like all science,
one does not necessarly need to know the mechanism regarding how something
works in order to investigate the phenonmena. Acetaminophen is a prime
example. We only recently -actually very recently- determined how this
compound really works, despite it being routinely used for decades as an
analgesic. The point being is that one can investigate the effects of
electromagnetic fields on biological system without having a direct
framework regarding how these effects are mediated.

GS: Of course. And the facts uncovered at that level are NECESSARY for
formulating the "reductionistic mechanism" since any theory must be
consistent with these observations. I have made this point about the general
relation between the study of behavior qua behavior and the study of
physiological mechanisms ummm......elsewhere on USENET.

NMF: As stated earlier, experiments have been done on humans. However, since
I
used this example for a question regarding consciousness then non-human
extended studies become difficult. There are still some that believe,
perhaps ignorantly, that only humans can be conscious or self-conscious.

GS: Non-humans may be made self-conscious by arranging the proper
contingencies of reinforcement, and this is routinely done in drug
discrimination experiments (this is why I suggested experiments looking at
"field discrimination"). There are other experiments in which animals may be
made to "report" on their own "voluntary" behavior as well. But, as I said,
if you could detect an effect of fields on any steady-state operant behavior
in individual subjects then there would be no question about its behavioral
significance. Take four subjects, and obtain steady state performance under
an inter-response time greater than t s schedule (IRT>t; i.e., a response is
reinforced if it occurs longer than t s after a preceding response. Any
response preceded by another response by less than t s resets the "timer."
Then introduce fields. This gives you a nice baseline, and gets at your
"timing" a little bit (i.e., the "timing" is evidenced by the shift in the
distribution of IRTs and the fact that at least one mode falls just after t
s and this can be pushed around by changing t).

Cordially,

Glen

"NMF" <nm_fournier at ns.sympatico.ca> wrote in message news:DEP4c.11264





More information about the Neur-sci mailing list