Biofeedback and Self-Regulation, VoL 1, No. 1, 1976
Conflicting Results in EEG Alpha Feedback Studies Why Amplitude Integration Should Replace Percent Time ~ James V. Hardt and Joe Kamiya Department of Psychiatry, Langley Porter Neuropsychiatrie Institute, University of California at San Francisco
Success or failure o f EEG feedback training for alpha enhancement can depend on how alpha activity is quantified and fed back. Alpha-enhancement failures usually employ a percent time (%) technique; successes typically use amplitude integration ( f ). To dramatize the differences between percent and integration techniques, we derived both measures simultaneously from left occipital (0,) and left central (C3) sites for 16 male subjects who were given 5. 6 hours o f integrated alpha feedback from the midline occipital (Oz) site. A t both the O~ and C3 sites the integrated and percent measures were not equivalent and not linearly related. Statistically significant differences in the (integrated, percent) correlation coefficients (z-transformed) were observed under the different recording conditions: alpha enhancement, alpha suppression, and baselines. Theoretical discussion o f integration and percent techniques is given and the adoption o f amplitude integration measures and feedback stimuli is strongly advocated. Currently there exists disagreement about the ability of human subjects to significantly increase their EEG alpha activity with the aid of feedback training. The reasons for the diversity of results may be found in methodological differences. A review of alpha feedback studies (Hardt, 1974) has yielded an itemized list of methodological considerations important in demonstrating significant voluntary control of alpha enhancement. This ~This study was supported by the following grants and contracts: National Institute of Mental Health (NIMH) Predoctoral Fellowship #1 F01 MH51704-01, NIMH General Research Support Grant #LPNI 185, and a Langley Porter Neuropsychiatric Institute Postdoctoral Fellowship (Interdisciplinary Training Program, NIMH #7082) to James V. Hardt, and by NIMH Research Scientist Development Award 2K02 MH38897, NIMH Research Grant #1R01 MH24820, Office of Naval Research (ARPA) Contract N00014-70-C-0350, and Instruction and Research Funds, Computer Center Accounts (UCSF) #1431 and #1437 to Joe Kamiya. 63 © 1976 Plenum Publishing Corporation, 2 2 7 West 17th Street, New Y o r k , N.Y. 10011. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any f o r m or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise, without written permission of the publisher.
64
Hardt and Kamiya
paper describes the importance of just one of the items from that list: the method of indexing or scoring the alpha activity, and the associated method of feedback of the alpha activity. The studies reporting weak or nonsignificant alpha enhancement by feedback training (Cleeland, Booker, & Hosokawa, 1971; Hord & Barber, 1971; Peper & Mulholland, 1971; Paskewitz & Orne, 1971, 1973; Honorton, Davidson, & Bindler, 1972; Mulholland, 1972; Walsh, 1972, 1974; Podlesney& Raskin, 1973) have all used a percent-time alpha-scoring index and a percent-time feedback signal (dichotomous, e.g., on or off) to the subjects, whereas reports of significant enhancement (Brown, 1970; Engstrom, London, & Hart, 1970; Kamiya, 1971; Gannon & Sternbach, 1971; Travis, Kondo, & Knott, 1974a, 1974b; and Hardt, 1974) have tended to use an integrated-amplitude alpha-scoring index and/or an integrated-amplitude (analogue or proportional) alpha-feedback signal. It is necessary to say "have tended to use" because a few of the percent-time feedback studies have demonstrated some degree of alpha enhancement, but only under one or more of some special conditions: (a) the combination of percent-time feedback with some additional feedback such as frequency feedback (Engstrom et al., 1970), (b) the use of a shaping procedure (Gannon & Sternbach, 1971), which involved carefully raising the percent-time threshold as the subject gained proficiency, and (c) the use of unique threshold settings for each subject (Travis et al., 1974a, 1974b), wherein threshold settings were taken as 50% of the peak alpha amplitude produced by each subject in a baseline condition. Except for these special cases, a simple fixed-threshold percent-time feedback signal and percent-time scoring technique does not generally result in significant enhancement of EEG alpha relative to pretraining baselines. It is the purpose of this report to contrast the two measures (integrated amplitude and percent time) and to present data obtained by simultaneously deriving both measures form each of two channels of EEG data. The results and the descriptive analysis of the two measures point strongly toward the replacement of the percent-time measure and its associated discontinuous feedback signal by the integrated measure and its continuously varying (infinite-valued) feedback signal. Typically in EEG feedback studies, a particular frequency band is extracted from the total electrical activity by means of filters. The problem is then one of quantifying the resultant signal for analysis and feeding back to the subject a stimulus which is somehow derived from the original signal. Since EEG alpha activity tends to occur in discrete trains or "spindles" composed of several alpha waves, an easy measure to derive is the ratio of the time during which spindles are present to the total time. This ratio is
Conflicting Results in EEG Feedback Studies
65
most commonly expressed as a percent and has been called by a variety of names: percent time, alpha abundance, alpha density, alpha index, and criterion alpha. This last name is suggestive in pointing out that the percenttime measure depends upon an arbitrarily chosen amplitude threshold (criterion), below which any signal amplitudes are defined as noise and ignored. All signal values above the chosen threshold are supposed to contribute identically to the percent-time measure regardless of whether the signal barely exceeds or greatly exceeds the threshold. Thus the percent-time measure throws away considerable amplitude information contained in the signal and yields only the binary information of "above threshold" or "below threshold." In most laboratories using this scoring method, the feedback stimulus is usually on or off and changes its state only when the threshold is crossed. Thus a signal barely above threshold produces the same feedback stimulus as a signal 10 or 100 times threshold. The situation is quite different with the integrated measures. This scoring index, derived by analogue or digital techniques, is the time integral, or area under the curve, resulting from the full-wave rectification of the filtered signal. The integrated score is proportional to the average amplitude and to the square root of the energy contained in the (EEG) signal, and it has the advantage of being a continuously variable measure with no arbitrary cutoff amplitude. A feedback stimulus controlled by an integrated signal varies in intensity as the amplitude of the signal varies--continuously from low or zero intensity for no (EEG) signal up through an intense feedback stimulus for the highest observed amplitude of the signal (often the fluctuations in the feedback stimulus are smoothed electronically). In short, integrated and percent measures, whether used for the feedback stimulus or in scoring the results, are not equivalent, and the greater information in the integrated measure would by itself be a sufficient reason to prefer it in studies of EEG feedback training.
METHOD
Subjects Subjects were 16 male college student volunteers who were paid $21.50 for their participation. The men were selected from a group of 100 male alpha-training volunteers who had taken a battery of personality tests including the Minnesota Multiphasic Personality Inventory (MMPI). The selected subjects were the 8 highest and 8 lowest scorers on the MMPI's Welsh A (anxiety) scale who also had valid lie scales (L < 9, F < 17, K < 23).
66
Hardt and Kamiya
Experimental Design All subjects were given one preliminary recording session, without feedback, to familiarize them with electrode attachment and recording procedures. Alpha training was given on seven days, consecutive except for Sundays. Each feedback session was divided into four recording conditions: (a) first baseline, 8 minutes with no feedback; (b) alpha feedback with instructions to enhance alpha activity, 32 minutes; a 10-15 minute break for filling out mood scales (no recording); (c) second baseline, also 8 minutes long with no feedback; (d) alpha feedback with instructions to suppress alpha activity, 16 minutes. Each subject received a total of 5.6 hours of feedback training. Integrated alpha feedback consisted of a tone whose volume was controlled by the amplitude of the full-wave rectified, smoothed alpha activity from midline occiput. Scoring periods were 2 minutes long. At the end of each scoring period, during feedback, the tone would shut off and the subject would then see an illuminated 3-digit score which was turned on for about 15 seconds. The score represented integrated midline occipital alpha (Oz). Data for the determination of the relationship between integrated and percent-time measures were obtained simultaneously from left occipital (0,) and left central (C3) sites.
Apparatus After the electrodes had been affixed, subjects were seated in a comfortable, upright, wooden armchair in a sound- and light-controlled room. The EEG scalp electrodes were referred to linked ears, and the forehead served as ground. The three channels of EEG activity were recorded on a Beckman Dynograph. All three EEG channels were filtered to extract alpha activity (8-13 Hz), which was integrated electronically and converted to digital-integrated alpha scores. Percent-time alpha scores with a 10-/~V threshold were derived electronically from the two channels not involved in feedback, O~ and C3.
Procedure After subjects were wired with electrodes and seated in the experimental room, the experimenter left, closing the door, and the subject remained seated in darkness. An intercom permitted continuous monitoring of the subject room. Instructions for the appropriate condition were played to the subject via tape recorder, and physiological recording commenced. Feedback of integrated alpha was provided only during alpha-enhancement
Conflicting Results in EEG Feedback Studies
67
and alpha-suppression conditions, but the subjects were instructed to avoid movements and to keep their eyes closed at all times and in all conditions, except for the brief periods when the illuminated 3-digit integrated alpha score was displayed (feedback conditions only). Data-A nalysis Methods Three different types of analysis were performed to reveal the relationships between integrated alpha and percent-time alpha measures: (a) correlation of integrated and percent alpha scores within days and within the four recording conditions for each subject; (b) analysis of variance (ANOVA) on the Fisher z-transformed correlations
z = 0.5,
ln(! + r~ W=T)
where r is the product moment correlation coefficient and In is the base e logarithm. In order to provide a more stable z score, the z scores for each condition (enhancement, suppression, and the two baselines) were averaged across each subject's 7 sessions before performing the ANOVAs; (c) scatter plots, correlation coefficients, and regression lines were generated from the combined enhancement and suppression data from all subjects in order to examine the overall relationship of integrated and percent-time scores. Three subregions of the scatter plots were also studied to demonstrate how the relationship of the two types of scores varied with different percent-time scores. The three subregions of the percent-time range were: 0-15°70, 15-85°70, and 85-100°70. Regression lines were also fitted to the three subregions.
RESULTS
Product-moment correlations of integrated and percent-t~me scores were computed within subjects, days, and recording conditions for each of the two recording sites (O, and C3). The ranges of the correlations by recording site and condition are given in Table I. It is important to our assessment of the equivalence of integrated and percent scores to note that some of the correlations are negative. The percentages of negative correlations, suitably broken down by conditions, are also in Table I. Overall, 2.9070 of the correlations from O, data were negative, while for C3 data, 3.4°7o were negative. A N O V A S on z-Transformed Correlations. An ANOVA on just the O, data showed that the correlations differed significantly among the four
68
Hardt and Kamiya Table I. Correlations .of Integrated Amplitude and Percent-Time Scores a Item Recording condition
Range of correlation
Negative correlations (percent of total)
Left occipital site First baseline Alpha enhancement Second baseline Alpha suppression
-.833-1.000 .066- .995 - . 9 2 2 - .999 -.920 .999
5.4 0.0 4.5 1.8
Left central site First baseline Alpha enhancement Second baseline Alpha suppression
- . 946-1.000 -.241 .997 -.933 .999 - . 824- . 997
3.6 2.7 4.5 2.7
aCorrelations were obtained within subjects, days, and recording conditions.
recording conditions [F(3,42) = 4.13, p < .02]. Although the condition effect in the C3 data was not as strong as the corresponding O1 effect, there was no significant overall site effect, as was shown by an ANOVA on the combined O1 and C3 data [F(1,14) = 2.20, p < .16]. Scatter Plots and Regression Lines. Scatter plots of integrated and percent-time alpha scores were generated by computer from the pooled feedback data of all 16 subjects. Each plotted point represents scores obtained from a 2-minute recording epoch during either alpha enhancement or alpha suppresSion feedback. Even though all (2-minute) epochs containing artifact and all baseline epochs are excluded from the scatterplots, the total recording time represented exceeds 73 hours. Figure 1 is the O1 scatter plot. A striking nonlinearity of the relationship of the two measures is apparent. The nonlinearity was also seen in the C3 data. Correlating the data in the scatter plots gives deceptively high correlations: for O1 data (Figure 1) r = .89, and for the C3 data r = .95. Regression lines were generated by covariance analysis with the variate (independent variable) taken to be the integrated alpha scores. For O~ data the slope of the regression line is rn~ = .039, and for the C3 data the slope is m2 = .072. However, the overall slope and the overall correlation are approximated only by the middle of the three subregions, where the percent-time scores range from 15°70 to 85o70. Table II gives the regression line slopes and the correlations for both sites, brol~en down by the range of percent-time scores. DISCUSSION The data clearly indicate that integrated alpha and percent-time alpha are quite different measures. Specifically, (a) they are not linearly related;
Conflicting Results in EEG Feedback Studies + . . . .
+ . . . .
+
. . . .
+
. . . .
÷
. . . .
+
. . . .
+
. . . .
+
. . . .
+ . . . .
÷ . . . .
69 ÷ . . . .
+
. . . .
+
. . . .
÷ . . . .
+ . . . .
1 1 1 L
[
1
+~tl
IIl[2312&IZIL
t i
iI
I
,
d
2
1
2 t
1
+ . . . .
÷ . . . . +
. . . .
+
:
100
11 ++ I 111 1 1
t 1
t 1
I
1
2
90
i
80
1
l
[ iIi tl 1 11
÷ . . . .
1111
1 11
211
e z,,!z t 3l+',17L1
51
[ llb ) llIll
d
i lllL1341Ilz[ <321ZI
1~1 , t~
i
iZ
572~I111
2231355~21 l,J 2q 25
'l 1,
+ . . . .
3 I 11 i 2 ILl 23 I 112241 ~Zl ~3155483 1,,13i23112 1gilt lit ~.lt, b O e 3 7 8 & 9 1 1 Z l t + q 7 8 4 8 7 8 3 9 e b a T 1 5 3 7 3 3~,7122~3z.¢.33 q~,z24A74555223t~lS12 11 t 9t~lZ?Zlt~zt3tz151z3~5 % :it z t [ 3Lhq45546332 22 1 312 2 1 11 12 ~)g~b134~Zl~Z43 Z311, ~ 1 2 It t 1 4,7z~&5 1[~+~ 2 2 z e z 2 1 1?12 1 ] 1 1
1
70
1
e
1 t
60 t
1 t , 4 . 1,' I ~ 1,,++
,,~ +
1+
(~
tt
,
I-I"
,
i 1 I ,
e ~ : c
l
I ,
11
ii
1
50
t
il
•
tl~+
l ,
1
J
1
,'1
1,~211
1
ltt
1
40
? . ~ : : Z L ,l
,' , ~ 3
Z
1 l
1 4 t Z
Z'2
l[~
I --
1
i I
30 + ~(. ~lt 17,915 Ic
~ ,+?:r7
I
200
l
I
20
l +7t,% i +3,'9+ 11 I,+ ?bu~'t ++ t /1493'~7/41 I + , ~+,~ 5 3 l
212
11![24
I
'.
10
6>C~L1
0
1151
600
1000
1400
1800
2200
2600
3000
3400
3800
4200
INTEGRATED O~ ALPHA (MICROVOLTSECONDS) Fig. 1. Scatterplot of left occipital (O0 integrated amplitude and percent-time alpha scores from 16 male subjects during feedback of O z alpha. Integrated amplitude scores, with units of microvolt seconds, were computed over 120-second intervals. The symbols in the scatter plots indicate point densities. If more than one data poinf is plotted at a given location in the scatterplot, numerals 2 - 9 are used to indicate the number of points, or density. Densities 10-36 are indicated by letters A - Z respectively. Table II. Regression Slopes and Correlations from the Scatterplots EEG site
Correlation
Slope of regression linea
Left occipital Left central
All regions: 0-100% alpha .894 .954
.039 .072
Left occipital Left central
Scoring region 1 : 0-15% alpha .651 .737
.025 .025
Left occipital Left central
Scoring region 2 : 1 5 - 8 5 % alpha .849 .874
.047 .070
Left occipital Left central
Scoring region 3 : 8 5 - 1 0 0 % alpha .464 .433
.004 .008
aThe integrated amplitude score was taken as the independent variable.
70
Hardt and Kamiya
(b) the scatter plot of the relationship between integrated and percent-time alpha is heteroscedastic, i.e., the amount of scatter about the regression line varies throughout the range; and (c) the two measures seem, in our data, to follow different sampling distributions. Integrated alpha seems to follow a Poisson distribution and percent-time alpha (with a 10-/aV threshold) seems to follow a rectangular distribution. The graphically nonlinear relationship of the two measures and the marked heteroscedasticity, with at least a 5:1 variation in scatter across the range, produce large differences in the (integrated, percent) correlations in different regions of the percent-time range, and can result in substantial errors when estimating integrated alpha scores from percent-time alpha scores. For example, in our data, correlations taken within days, subjects, and recording conditions (Ns = 4-16) ranged from above + .90 to below --.90. Even the correlations in the three subregions of the pooled scatter-plot data show disturbing variability. The existence of distinct subregions with different correlations and different regression slopes suggests that percent-time measures are particularly inappropriate to feedback studies, which often seek to induce the maximum possible change in the level of some given physiological activity. The more the activity changes, the more likely it is that the percent-time scores will be drawn from different regions of the distribution, and thus the more likely is a change in the relationship between the activity measured and the percent-time score. Table III. Dependence of Percent-Time Scores on Threshold and Signal Amplitude a,b Percent-time scores at selected amplitudes above and below reference Reference amplitude (/~V)
- 3 #V
+ 0 uV
+ 3 #V
+ 6 uV
+ 12 tzV
+ 24 tzV
68.40 (+14.86)
75.84 (+22.30)
83.49 (+29.95)
60.85 (+14.83)
68.94 (+22.93)
77.85 (+31.84)
55.36 (+14.39)
63.65 (+22.68)
73.24 (+32.27)
10-u V threshold 15
37.28 (-16.26)
53.54 (_)c
62.50 (+8.96)
15-#V threshold 20
31.19 (-14.82)
46.01 (_)c
54.77 (+8.76)
20-uV threshold 25
27.36 (-13.61)
40.97 (_)c
49.35 (+8.38)
aSignals used for computation were sine half-waves. bQuantities in parentheses are the changes produced in the percent-time scores when the signal amplitude is changed from its reference value to the new value specified in the column heading. Up through the + 6-#V amplitude change, the 10-#V threshold yields the largest changes in percent-time scores. But at the + 12-~V change the 15-#V threshold gives the largest scores; and by the + 24-~V change, the 20-~V threshold is yielding the largest score change from amplitude changes which are the same at each threshold setting. CNo score changes result when the reference amplitude is not changed.
71
Conflicting Results in EEG Feedback Studies
A Rubber Ruler. Not only does the percent-time measure fail to index any data below the threshold value, but even above threshold the scores are flawed by gauge variance, which means in practical terms that equal and opposite changes in the magnitude of the sinuosidal activity being measured (e.g., filtered EEG alpha) do not produce equal and opposite changes in the percent-time scores. It also means that equal changes in signal magnitude measured with different thresholds yield different changes in percent-time scores. Table III and Figure 2 are presented to show that the gauge of the percent-time measure is both amplitude and threshold dependent. Table III shows the changes in percent-time scores for three typical sine half-waves of three different amplitudes (15, 20, 25/aV--each chosen to be 5/aV above its own threshold) as the amplitude of each half-wave is changed by --3, + 3, + 6, + 12, and + 24/aV. Clearly, doubling the amplitude does not produce a doubling of the percent-time score. Moreover, equal amplitude changes measured from two different thresholds do not produce equal (or even proportional) changes in the two scores. In Figure 2 we see a graphic presentation of the nonlinear relationship of signal amplitude and percent-time 120
*
r
.
*
r
°
110
o > o ,.,,., r.)
100
90 v
<
80
..1
<
70
212
60 '!
50 J
40
<
30 s u o
10
20
30
40
50
20 60
70
80
90
100
PERCENT TIME SCORE (%)
Fig. 2. Threshold and signal-amplitude dependence of percent-time scores. Percent-time scores derived from a sine half-wave of various amplitudes (20-120/~V) are shown for three different thresholds: 10/~V (.); 15 gV (o); 20 ~V (*).
72
Hardt and Kamiya
scores. The three different curves show data derived from the three different thresholds of the half-waves of Table III (10, 15, and 20/~V). Figure 2 and Table III together suggest that the percent-time measure can be likened not only to a ruler with unequally spaced graduations, but worse, to a rubber ruler with unequal graduations which requires different degrees of stretch to account for different threshold settings. We could note here that the early, rapid progress of physics was aided by gauge-invariant measures of fundamental properties, and that the present new field could also benefit from gauge-invariant metrics. Integrated amplitude measures are linearly related to the amplitude of each alpha half-wave and can thus be related easily to the energy or power in the alpha (or other) frequency band. We suspect that EEG energy is more likely to be related to psychological processes than is the amount of time an arbitrary amplitude threshold is exceeded. Practical Examples. We can consider a bit more of the relationship of EEG and conscious experience and the superior effectiveness of continuously variable feedback signals by taking as examples two subjects receiving feedback for alpha enhancement. Before the feedback, both subjects show alpha spindles separated by troughs with no alpha activity. Consider a percent-time threshold set at some fraction of spindle amplitude, and assume that one subject responds to feedback by increasing alpha activity only in the trough regions. But the increases do not reach threshold and the experimenter misses them; the subject's percent-time feedback signal also fails to report the increase, so the unreinforced subject abandons a successful strategy. Consider the other subject who markedly increases spindle amplitude while slightly decreasing spindle length. More alpha energy is present, but the percent measure will indicate either no change or a decrease. The percent-time feedback will be as misleading as the percent-time score; so this subject will also abandon a successful but unreinforced strategy. Needless to say, the integrated alpha score and feedback would not similarly mislead experimenters and their subjects. In the case of the first subject developing alpha in the interspindle troughs, the full information of the integrated feedback would encourage the subject to continue whatever strategy was producing the increases and might even lead him toward an EEG of continuous high-amplitude alpha with minimal variability of amplitude (hypovariability). This pattern has been seen in Zen (Etevenon, Henrotte, & Verdeaux, 1972; Hirai, 1974), in Yoga (Anand, Chhina, & Singh, 1961), and in long-term practitioners of Transcendental Meditation (Banquet, 1973). We believe our methodological point has implicatiOns for the study of certain unusual EEG patterns such as the hypovariability of Zen, Yoga, and Transcendental Meditation. This pattern is often reported to be associated with altered states of conscious-
Conflicting Results in EEG Feedback Studies
73
ness which are of special interest for psychological study. If such patterns are trainable in nonmeditators, it would seem likely that integrated amplitude scoring and its associated continuous-valued feedback would be the best training method. In the second example of a subject training with feedback, the integrated amplitude score would not be fooled by a slight decrease in spindle duration into ignoring a large increase in spindle amplitude. Of course, if the opposing changes left spindle energy (swept area) unchanged, then the proper response would be to indicate no change of scores. In both examples, neither the experimenter nor the subject would be misinformed by the use of integrated scoring and feedback. We do not claim that the two examples define the only ways in which alpha can increase, but this qualification should not obscure the fact that integrated scores and feedback are sensitive to these changes and percent-time scores are not. An experimental comparison was made by Travis, Kondo, & Knott (1974c) of the effectiveness of the four common methods of feedback and scoring: (a) percent scoring and percent feedback, (b) integrated scoring and percent feedback, (c) percent scoring and integrated (analogue) feedback, and (d) integrated scoring and integrated (analogue) feedback. Results were entirely consistent with our analysis, because they found that only one method produced eyes-closed alpha feedback scores higher than an optimal baseline. Only integrated amplitude alpha scoring and its associated continuous-valued (analogue) feedback produced the desired results. Unfortunately the current acceptance of the percent-time technique by many manufacturers of commercial feedback equipment is likely to result in more studies built upon this measure. The existing hardware is being only gradually replaced or supplemented by equipment designed to provide scores and feedback based upon the integrated amplitude measure. But we can at least hope that investigators will be aware of the differences between integrated amplitude and percent-time techniques for scoring and giving feedback. We believe the integrated-amplitude technique will aid both the process of training and the statistical evaluation of results within and between various studies. To an investigator compelled to use percenttime equipment we make three suggestions: (a) Use more than one threshold setting in the pre- and posttraining baselines to maximize chances of detecting any changes. (b) Use a shaping procedure during feedback, adjusting the threshold (for each subject) for optimal sensitivity of the feedback signal, i.e., near the 50% time level (as in Travis et al., 1974a, 1974b). This procedure may help detect changes in both directions for each subject, but no attempt should be made to compare scores derived from different thresholds. (c) Use existing percent-time equipment to generate scores, but buy or build additional circuitry to provide subjects with the more accurate
74
Hardt and Kamiya
c o n t i n u o u s l y variable feedback. W i t h the superior feedback aiding subjects in p r o d u c i n g large changes in the physiological activity, the e x p e r i m e n t e r m a y be able to detect the changes with p e r c e n t - t i m e scores, p a r t i c u l a r l y if these are assessed as in (a) above. Investigators should bear in m i n d , however, that p e r c e n t - t i m e scores should be used with c a u t i o n in q u a n t i t a tive analyses a n d that p e r c e n t - t i m e feedback m a y c o n t a i n too little i n f o r m a t i o n to p e r m i t subjects to l e a r n c o n t r o l o f the e x p e r i m e n t a l p a r a m e t e r .
REFERENCES
Anand, B. K., Chhina, G. S., & Singh, B. Some aspects of electroencephalographic studies in Yogis. Electroencephalography and Clinical Neurophysiology, 1961, 13, 452-456. Banquet, J. P. Spectral analysis of the EEG in meditation. Electroencephalography and Clinical Neurophysiology, 1973, 35, 143-151. Brown, B. B. Recognition of aspects of consciousness through association with EEG alpha activity represented by a light signal. Psychophysiology, 1970, 6, 442-452. Cleeland, C. S., Booker, H. E., & Hosokawa, K. Alpha enhancement: Due to feedback or the nature of the task? Psychophysiology, 1971, 8, 262-263. Engstrom, D. R., London, P., & Hart, J. T. Hypnotic susceptibility increased by EEG alpha training. Nature, 1970, 227, 1261-1262. Etevenon, P., Henrotte, J. G., & Verdeaux, G. Approche methodologique de 6tats de conscience modifi6s volontairement. Revue E. E. G., Paris, 1972, 3, 232-237. Gannon, L., & Sternbach, R. A. Alpha enhancement as a treatment for pain: A case study. Behavior Therapy and Experimental Psychiatry, 1971, 2, 209-213. Hardt, J. V. Alpha EEG responses of low and high anxiety males to respiration and relaxation training and to auditory feedback of occipital alpha. Dissertation Abstracts, International, 1974, 35(4), Catalog No. 74-19309. Hirai, T. Psychophysiology of Zen. Tokyo: Igaku Shoin, 1974. Honorton, C., Davidson, R., & Bindler, P. Shifts in subjective state associated with feedbackaugmented EEG alpha. Psychophysiology, 1972, 9, 269-270. Hord, D., & Barber, J. Alpha control: Effectiveness of two kinds of feedback. Psychonomic Science, 1971, 25, 151-154. Kamiya, J. Self regulation as an aid to human effectiveness. Progress report on Office of Naval Research Contract No. N0014-70-C-0350, Washington, D.C.: May 15, 1971. Mulholland, T. B. Can you really turn on with alpha? The R.M. Bucke Memorial Society Newsletter Review, 1972, 5(1&2), 32-40. Paskewitz, D. A., & Orne, M. T. The effect of cognitive tasks on the feedback control of alpha activity. Psychophysiology, 1971, 8, 263. Paskewitz, D. A., & Orne, M. T. Visual effects on alpha feedback training. Science, 1973, 181, 360-363. Peper, E., & Mulholland, T. B. Methodological and theoretical problems in the voluntary control of electroencephalographic occipital alpha by the subject. In Biofeedback and SelfControl 1970. Chicago: Aldine-Atherton, 1971. Podlesney, J. A., & Raskin, D. C. Biofeedback: A failure to enhance alpha EEG. Proceedings of the Society for Psychophysiological Research, Galveston, October 1973. Travis, T. A., Kondo, C. Y., & Knott, J. R. Alpha conditioning: A controlled study. The Journal of Nervous and Mental Disease, 1974, 158, 163-173. (a) Travis, T. A., Kondo, C. Y., & Knott, J. R. Personality variables and alpha enhancement: A correlative study. British Journal of Psychiatry, 1974, 124, 542-544. (b) Travis, T. A., Kondo, C. Y., & Knott, J. R. Parameters of eyes-closed alpha enhancement. Psychophysiology, 1974, 11, 674-681. (c) Walsh, D. H. Effects of instructional set, reinforcement and individual differences in EEG alpha feedback training. Proceedings of the Biofeedback Research Society, Boston, November 1972.
Conflicting Results in EEG Feedback Studies
75
Walsh, D. H. Interactive effects of alpha feedback and instructional set on subjective state. Psychophysiology, 1974, 11, 428-435. (Original received October 10, 1974)