32 seconds of alpha activity from site Pz, one subject. Data was analyzed using four different epoch lengths:
| Epoch Length | Magnitude | Power |
|---|---|---|
| 1 s | 2.12 | 5.18 |
| 2 s | 2.12 | 4.78 |
| 4 s | 2.12 | 4.57 |
| 32 s | 2.12 | 4.49 |
A Monthly Summary of News and Events
Vol. 6 No. 6 - June 2003
This newsletter is sponsored by EEG Spectrum International Intl, Inc.,
a leader in providing clinical service and training professionals.
Past issues are available at start.eegspectrum.com/Newsletter/
Information on how to subscribe or cancel a subscription appear at the end.
The opinions related in this newsletter reflect those of the author only.
Copyright (C) 2002 by EEG Spectrum International Intl, Inc. All rights reserved.
|
|
|---|
All links at: http://news.yahoo.com/fc?tmpl=fc&cid=34&in=science&cat=brain_research
Power is ever stealing from the many to the few.
-Wendell Phillips
In June 2001, I wrote an article for this newsletter about the use (or misuse) of the power transformation in quantitative analysis of EEG. The power transformation involves an entirely unnecessary squaring of the raw spectral output. It is an outdated technique adopted during the dawn of QEEG, when we didn't understand fully what we were looking at. Power is a great measure when all we care about is extreme values -- it accentuates them, stealing from the many (the mean) to give to the few (the outliers) as the quote mentions above. But surely nowadays we are confident enough to comb through the many and make distinctions there as well. Power analysis was developed by engineers for engineering tasks, not psychologists nor neuroscientists studying the brain.
Power analysis skews distributions and interacts poorly with a whole slew of linear analytical techniques. I doubt anyone can disagree to these mathematical facts. Perhaps a greater flaw in their application in this field is this: Most neurofeedback involves the training of sustained amplitudes -- essentially, magnitude values of specific frequency filters. So why would one evaluate on one spectrum (power) and train on another (magnitude) when one can very easily evaluate and train on the same spectrum (magnitude). Why add a layer of inconsistency in one's work?
Well, below is the article I wrote for WNIN, slightly revised, in 2001. I know for many clinicians this is a question below the radar screen, but it's a serious source of inconsistency and error, not to be lightly ignored. Error in evaluation, however slight, can lead us to make conditional decisions, unsubstantiated inferences, and eventually wrong choices in how we train clients.
New Year's Eve, 1984. New York City. Three hours before midnight. Thirty feet underground. As the subway doors opened to receive passengers at the 96th Street Station, the air erupted with sound. Someone was shouting on the platform. Or maybe they were singing. It was the same few sounds over and over again, that I was sure. Whoever it was, he was loud enough to attract all of our attention. I glanced up from my seat. From my angle it was hard to see what the disturbance was all about. Next thing you know, the doors slid shut, and we pulled away. As we entered the long tunnel between stations, the interior lights flick on and off. Business as usual. I could still hear the shouting, but it sounded faint, so whatever it was, would be miles away in a moment or two. But I was dead wrong.
Nine days before, Bernard Goetz had shot four teenagers on this very subway.
The shouting was getting louder. Two or three cars up, but coming this way. My sole escape route, if I needed one, would be through the flimsy door behind me. But it connected to the next car, so it wasn't much of an escape.
The far door crashed open and in they marched, in they stormed. Jackets and caps, flailing arms, pounding fists, angry faces. A many-faced hydra of leather, sound, and anger, spitting and screaming at every person they passed on the train:
Here I sat alone, in an empty corner of the train: white, male, college-educated, easily mistaken for the "power that be" -- in larval form.
So what is my beef against "the power" that brought this true story to mind. Well, power corrupts. Power transforms corrupt EEG analysis. Power transforms are a historical vestige, unsuited for current clinical needs. This technique served the initial investigators of brain waves well, when rulers and VAX machines were used to make sense of squiggly lines. Squaring spectral means emphasizes extreme data points in a sample; it launches fringe events far away from the central tendency of a sample's distribution so that they can be easily identified and circled with red ink by the engineer, I mean psychologist on duty. It is a technique to assist eyeballs, not computers. It is a geometric transformation that undermines the linear functions and assumptions of central tendency (i.e., averaging) and parametric statistics. Yet despite these negatives, people continue to leave the welcome mat out.
Power transformations are essential for many domains of analysis, and many forms of data sets, but psychophysiology is not one of them. They should be applied sparingly and after careful consideration of a sample's distribution. EEG data, if anything, should be geometrically compressed in order to make it correspond readily with psychological phenomena (cf Fechner's law) whereas power does the exact opposite; it geometrically expands the spectrum. Power analysis is often used when one doesn't know what one's looking for or only care about extreme states. In present-day QEEG analysis, neither of these conditions are met, thank goodness. We've had decades to refine our methods, learn about the complex cortical signal, and we care about subtle differences now.
Let's take a fresh look at what we are doing when we convert magnitude values to power values. Here are some examples:
Comparing squared values makes little sense in most endeavors. Let's look at what power transforms do to the simplest case in QEEG, the mean of two values:
| Raw data | Power | |
|---|---|---|
| Value A. | 1 | 1 sq units |
| Value B. | 2 | 4 sq units |
| Mean | 1.5 | 2.5 sq units |
| Converted back to Raw | 1.5 | 1.58 |
| Error | -- | 5.4% |
Always keep close to the raw data. It's a good rule of thumb for scientists. The farther one gets away from recorded measurements, the more likely error has crept in. This adage is doubly appropriate when statistics are to be used, because statistics can be misused almost as readily as they can be used. Statistics themselves are a step or two away from the raw.
Power derives from a geometric operation (squaring); means and averaging derive from linear ones (addition and division). When the two approaches are combined (e.g., mean power), results will vary unless the exact sequence of each operation is duplicated. This can be shown by merely altering epoch length. Mean power estimates will vary for the EXACT SAME DATA for different epoch lengths.
| Epoch Length | Magnitude | Power |
|---|---|---|
| 1 s | 2.12 | 5.18 |
| 2 s | 2.12 | 4.78 |
| 4 s | 2.12 | 4.57 |
| 32 s | 2.12 | 4.49 |
Notice how mean power changes across epoch lengths. Not good. Only when epoch size encompasses the entire dataset (32 s) is there no difference between magnitude and power (4.49 is 2.12 squared). Prior to this, we witness as much as 15% differences between mean power estimates.
Finally, power and normality don't mix. Alpha activity during eyes closed baselines from 275 subjects resembles the bell-shaped curve to a reasonably degree for magnitude values, as shown below, but power is horribly skewed and looks more like the area under the arc of a circle. A critical assumption for using parametric statistics is not met. (see start.eegspectrum.com/Newsletter/jun2003.htm for the figure)
Little more can I do to show the distortion inherent in power values, so I guess I can return to my night, now almost 20 years old.
When the mob descended on my lonely seat, I acted like any New Yorker would: I ignored them, pretended that this was an everyday occurrence in my jaded world (what did I know; perhaps it actually was). I never looked up from my book. The shouting men passed me by and moved onto the next car. Perhaps they realized that anyone riding the subway was unlikely to hold any power. Perhaps I had been perfectly safe all along. Perhaps not.
After the ball dropped at Times Square and the throng of thousands sang a drunken round of Auld Lang Syne, instead of returning to the subway, I conjured up some of that elusive power those men rebelled against. I avoided the subway entrance, pulled out a twenty, and hailed a taxicab home.
-DK
News & Reviews
NEW BOOKS
Right from the Start: Behavioral Intervention for Young Children With Autism
by Mary Jane Weiss, Sandra L., PH.D. Harris
Explains how intensive behavioral intervention can benefit young children with autism and related disorders.
--www.amazon.com/exec/obidos/ASIN/189062702X/top100
Cognitive Neuroscience: A Functionalist Perspective
by Roy H. Smith
An overview of to this field for the layperson. Reviews studies in perception, memory, motor function, language, executive function, emotion, and consciousness.
--www.amazon.com/exec/obidos/ASIN/0761823530/top100
The Emotional Revolution: Transform Your Life Using the New Science of Feeling
by Norman E. Rosenthal
Affective neuroscience and its impact on psychology, psychiatry, and everyday life.
--www.amazon.com/exec/obidos/ASIN/0806524472/top100
Principles of Learning and Memory
by Rainer Kluwe, Gerd Luer, Frank Rosler
Interdisciplinary perspective on learning and memory; useful textbook.
--www.amazon.com/exec/obidos/ASIN/3764366990/top100
Neurofeedback treatment for ADHD in children: a comparison with methylphenidate.
:
Both neurofeedback and methylphenidate were associated with comparable improvements on two attentional tests as were behaviors related to the disorder
Clinical correlates of aggressive behavior after traumatic brain injury.
:
Aggressive behavior was found in one-third of TBI patients and one-ninth without TBI 6 months after injury. Aggressive behavior was associated with depression, frontal lobe lesions, poor premorbid social functioning, and substance abuse.
Toward an understanding of bipolar disorder and its origin.
:
Role of genetic influences in bipolar disease is supported by family studies and high concordance rates among monozygotic twins. Substance abuse, common in bipolar disorder, interferes with diagnosis and can worsen the course of the disease.
Learning to control brain activity: EEG components for driving brain-computer interfaces.
:
To date, motor imagery has been the most commonly used task, but other cognitive tasks, including those used in imaging studies, may prove to be more effective.
Functional Neuroimaging Studies of Motor Recovery After Stroke in Adults
:
The damaged adult brain is able to reorganize to compensate for motor deficits, with recovery of motor abilities involving apparently enhanced activity in preexisting networks
Rightward cerebral asymmetry in subtypes of schizophrenia
:
Rightward asymmetry correlates with severity of negative symptoms (withdrawal, apathy) and suggests a distinct pattern of left hemisphere abnormality in a subgroup of schizophrenics.
Multisensory integration after traumatic brain injury
:
TBI patients often show great difficulty with integrating the tactile modality with auditory and visual information.
Salivary testosterone and EEG spectra of boys
:
Delta relative power decreased and alpha relative power increased in the higher salivary testosterone boys compared to low testosterone boys.
Drug use and prefrontal-associated traits.
:
Polysubstance users show greater dysfunction on disinhibition measures compared to non-polysubstance users, again supporting the link between prefrontal function and drug use.
Upcoming Courses
Prerequisites:
All Adv. classes require successful completion of the 4 Day Comprehensive Beta/SMR.
|
Conferences for Neurofeedback Clinicians & Researchers | ||
|---|---|---|
| CONFERENCE | LOCATION | DATES |
| SNR - http://www.snr-jnt.org | Houston, TX | Sep 18-21 |
SABA II was partly a training conference so much time was spent on case reviews, tutorials on artifacting, referencing, analysis, and protocol development, as well as basic data acquisition and analysis and other technical aspects.
www.skiltopo.com/html/saba2.htm