Biofeedback and Self-Regulation, Vol. 2, No. 1, 1977
Effecl of Feedback Contingencies on the Control of Occipital Alpha Thomas Mulholland and Peter Eberlin Psychophysiology Laboratory, Veterans Administration Hospital, Bedford, Massachusetts 01730
Ten color transparencies were presented 30 times each to ten normal adults in response to changes in their occipital-parietal EEG's. By means of a feedback paradigm, detected alphas caused each slide to flash and stay on as long as alpha persisted, and then to turn off when alpha attenuated, according to four different contingency conditions. For half of the feedback trials, stimulus presentation depended on alpha detection in only one hemisphere and was not influenced by changes in the simultaneously recorded contralateral EEG. For the other feedback trials, stimulus presentation was bilaterally contingent. These contingency configurations were compared with sham feedback, a noncontingent condition during which slide presentation was controlled by a prerecorded tape. For both alpha and no-alpha, the ratio of mean duration over standard error (X/SE) was used as a quantification o f the EEG response to visual stimulation. It was assumed that larger ratios indicated increased control of the EEG. Compared with the sham condition, all feedback contingencies produced greater X / S E ratios, and hence, improved control of the EEG. The highest ratios were obtained during unilateral feedback from the EEG in which the occurrence o f alpha elicited the visual stimulus. The results show that contingency between a visual stimulus and the EEG is an important parameter with regard to experimental control of the EEG. Biofeedback methods are usually employed to facilitate the process of acquiring voluntary control over a physiological response. They also have a role in basic electrophysiological method. Just as feedback can improve the control of the output of a nonliving system (e.g., tracking radar), biofeedback can improve the control that a visual stimulus exerts on the resulting physiological response so that the response itself becomes less variable rela43 © 1 9 7 7 P l e n u m P u b l i s h i n g Corp., 2 2 7 West 1 7 t h Street, N e w 'York, N . Y . 1 0 0 1 1 . T o prom o t e freer access to published material in t h e spirit of t h e 1 9 7 6 C o p y r i g h t L a w , P l e n u m sells reprint articles f r o m all its journals. This availability underlines t h e f a c t t h a t n o part of this publication m a y be reproduced, stored in a retrieval system, or t r a n s m i t t e d , in a n y f o r m or by a n y means, electronic, mechanical, p h o t o c o p y i n g , microfilming, recording, or otherwise, w i t h o u t w r i t t e n permission of the publisher. S h i p m e n t is p r o m p t ; rate per article is $ 7 . 5 0 .
44
Mulholland and Eberlin
tive to its average value or major trend (Eberlin & Mulholland, 1976; Lewis & McLaughlin, 1976; Mulholland, McLaughlin, & Benson, 1976). This kind of improved control results not from practice or training, but from the unconditioned effects of the feedback stimulus on the physiological response in conjunction with different contingencies of feedback. Such an improvement in the average-to-variance ratio translates into a better method for studying the parameters relevant to the stimulus-response process because systematic variation due to the experimental conditions is more clearly differentiated from unsystematic, random variation (Mulholland, 1977). Improved control of an EEG response is observed when the occurrence and nonoccurrence of occipital alpha bring about the occurrence and nonoccurrence of a visual stimulus by means of an external path of electronic devices. Visual stimuli typically evoke an unconditioned suppression of occipital alpha or a reduction of alpha amplitude. If the stimulus is contingent on the EEG so that it is presented when alpha occurs and is removed when alpha attenuates, a regenerative, self-stimulating system results: alpha occurs, the stimulus goes on; alpha is suppressed (no-alpha), the stimulus goes off; alpha returns, etc. (Mulholland, 1973). Despite the fact that the internal functional connections between the visual stimulus and the EEG alpha are loosely coupled, i.e., not perfectly predictable, the external feedback causes an improvement in the reliability of the measured EEG response to the stimulus. One useful way of expressing this improvement is to compare the ratio of the EEG central trend, e.g., the )7, to the scatter or variance of raw data around that central trend in feedback vs. nonfeedback conditions. In previous studies it has been shown that the ratio of the mean duration of alpha and no-alpha intervals to the standard error (SE) (or to the standard deviation) is significantly increased with feedback stimulation compared to nonfeedback conditions. With ,~/SE as a measure, it has been shown that the contingency between the feedback stimulus and the physiological EEG response is an important parameter. For instance, if stimulus presentation depends on changes detected in only the left (O1-P3) EEG, the ratioX/SE is greater for the left EEG alpha and no-alpha durations compared with the simultaneously recorded right EEG; if display of the stimulus depends on changes in only the right (O~-P4) EEG, the X/SE is greater for the alpha and noalpha durations of the right EEG compared with the left EEG (Eberlin & Mulholland, 1976). The experiment below examines the effect of two unilateral, two bilateral, and one sham feedback contingency in order to identify the optimal contingency, i.e., the one associated with the greatest X/SE ratio. It also presents a demonstration of the effects of feedback relative to nonfeedback conditions in terms of improved control of the durations of alpha and no-alpha.
EEG Feedback Contingencies
45 METHOD
Ten right-handed students from a local community college (ages 18-37) volunteered as subjects for the experiment. Five were men and five were women. All were right-eye dominant and had 20-20 vision or an acuity corrective to at least 20-20. Glasses were worn during the experiment if needed. None of the students was acquainted with experimental details, especially that one's EEG actually controlled the stimulus presentations. Although no monetary compensation was offered, all subjects were extremely cooperative and highly motivated. Each person exhibited abundant alpha rhythms with eyes closed in the dark and showed the classical alpha "block" when looking at a visual stimulus.
Apparatus The general purpose of the EEG feedback system was (1) to detect and quantify intervals of alpha and no-alpha, and (2) to present or remove stimuli contingent upon the presence or absence of alpha during feedback. The major components of the apparatus (Boudrot, 1972) were (a) a light-tight, soundproof, electrically shielded chamber, (b) a Grass Model 7 polygraph for recording brain rhythms and event marks, (c) a variable frequency bandpass filter with roll-off of 24 dB/octave, and (d) an amplitude threshold detector including state relay outputs which controlled an electronic shutter (minimum cycle time _ 5 msec) attached to the lens of (e) a slide projector. Recordings were obtained from standard left (O1-P3) and right ( O : P , ) occipital-parietal electrode placements. A PDP-12 computer served to collect, store, and analyze the data (Goodman, 1973). Alpha was defined for each individual: (a) the bandpass filter limits were adjusted to within __.2Hz of the dominant alpha frequency; and (b) the amplitude threshold was set at 25% of his or her eyes-closed-in-the-dark "resting" alpha. The minimum duration of a detectable alpha "burst" was set at .25 sec; minimum computer resolution of alpha and no-alpha intervals was to the nearest. 1 sec. When the criteria for alpha were met in the left, right, or both EEG channels according to the prearranged contingencies described in Procedure (below), the state relay would go on, opening the electronic shutter and projecting the slide stimulus; subsequently, when alpha was "blocked" or attenuated following the stimulus, the relay would go off, in accordance with the contingencies, closing the shutter and removing the stimulus. When alpha returned, this rapid sequence would be repeated. The EEG's from each hemisphere were calibrated independently to compensate for the initial individual and interhemispheric differences in the
46
Mulhoiland and Eberlin
amplitude of baseline alpha, thereby reducing amplitude differences at the input to the external path. If differences among individuals or between left and right recordings occurred under experimental feedback conditions, they would more likely be due to the dynamics of the feedback condition than to inherent, individual differences in alpha amplitude unaffected by visual stimulation. The visual stimuli consisted of ten 35-mm color slides of pictorial subject matter (scenes, people, etc.) and one blurred square of light, the dimensions and brightness of which approximated those of the colored slides. All transparencies were projected through a glass window from behind onto a screen 6 ft in front of the subject. The purpose of the blurred square, which was presented before the series of color slides on a noncontingent schedule, was to habituate each person to the effects of a flashing, indistinct patch of light following the 5-10 min of dark adaption which occurred during the calibration procedures.
Procedure After electrodes had been attached and procedural instructions administered (essentially, to "watch a series of slides, each of which will flash on and off several times..."), the subject was seated comfortably in a soundproof room. The recording and detecting equipment were calibrated according to that individual's alpha characteristics. The subject was then informed via the two-way intercom that the experiment would begin. During baseline conditions 30 events (an event was defined as an alpha and the succeeding no-alpha interval) were collected by the computer with the subject's eyes closed and again with his eyes open in the dark. Next, the blurred square and each of the ten stimulus slides were presented in the following manner: (1) at least 10 events were collected with eyes open in the dark (Before Feedback); (2) feedback was then turned on and approximately 30 events (and hence, at least 30 presentations of each slide) were collected from both recording channels under the appropriate contingency (During Feedback); and (3) computer collection and feedback were simultaneously turned off, and the subject's alpha was permitted to return in darkness to "resting" baseline levels. This intertrial interval lasted about 1 5 - 2 0 sec. Then computer collection/slide presentation was restarted under whatever contingency condition was next scheduled. After the blurred square and the entire series of slides had been shown, 30 events were collected with eyes closed and again with eyes open in the dark.
EEG Feedback Contingencies
47
Four different contingencies controlled presentation of the slide stimuli during feedback: (1) Left Contingent (LC): the stimulus was presented only for as long as alpha was detected from the left EEG recording, and was removed when no-alpha was detected. Alpha detected from the right EEG did not elicit the stimulus. (2) Right Contingent (RC): presentation occurred only for alpha detected from the right EEG, and was removed when no-alpha was detected. Alphas in the left E E G did not cause the stimulus to go on. (3) Either-Or Contingent: the stimulus was presented as long as alpha was detected from either the left, right, or both EEG's and was removed when no-alpha was detected simultaneously in both EEG's. (4) Both-And Contingent: the stimulus was presented only when alpha was detected simultaneously from both the left and the right EEG's and removed when no-alpha was present in either EEG. For the blurred square and sham feedback (noncontingent) conditions included as experimental controls, stimulus presentation was independent of alpha detection, viz., " s y n c h " marks on a prepared tape opened and closed the electronic shutter. The stimulus on-off intervals approximated the durations of alpha and no-alpha, respectively, which were typically recorded for the contingency conditions. Each of the four contingency conditions as well as sham feedback were used twice in a mirror-image Latin-square design where the slide presentation order was systematically varied across all ten subjects.
Data Analysis The trends for alpha and no-alpha durations were computed for each condition using the method of linear regression and Least Squares (see Fig. 1). A straight line was fitted to the alpha time durations (Ata) and a hyperbola' to the no-alpha time durations (Atna) before and during feedback (Mulholland, 1973). In the following equations At is a time duration, N is the number of events, SE is the measure of unpredictable variance or
'Although the hyperbola is a curvilinearfunction, linear regression can still be applied after iV, the independent variable, has been transformed into its own reciprocal, I/N. Thus, C/N becomes C'(1/N) and is, therefore, comparable to A "N in Equation 1 (see Fig. 1 inset).
48
MuihollandandEberlin g |
"•
g
6.() ~
6.0
g
•
Linearlzed
~
Hyperbola
4.0
x .
.
.
.
--S.0
Y
1.9
i 4.0 ":
1
i
~1 i1
1- ~
o
_N
a 3.0
:(C/N + D) -4- SE
Idd
•
;
! 2.01.0~; .
oil
°
. . . . . t. . .
I
I
•
----:--"
I
I
o "---~
at
II
A 5 10 15 20 25 30 / FEEDBACK SERIAL POSITION OF EVENT BEG,, (N) Fig. 1. Schematic illustration of parameters of the best-fit functions for alpha (filled circles) and no-alpha (open circles) durations. N = serial position of any event from 1 to 30. For the straight-line function to derive durations of alpha (Ata):A = slope or rate of change of alpha intervals; B = Y-intercept in seconds. For the hyperbolic function to derive durations of no-alpha (Atna): C = slope or rate of change between any two durations of noalpha on linearized hyperbola (see inset); D = asymptote of hyperbola for N = oo. In this example Ata = .023N + .72 and Atna = 4.2/N + 1.9. " s c a t t e r " o f the raw d a t a a r o u n d the best-fit f u n c t i o n s , a n d A , B, C, a n d D are the c o m p u t e d p a r a m e t e r s o f the best-fit f u n c t i o n s .
+ B) +_ S E
(1)
h t n a = ( C / N + D) +_S E
(2)
Ata
= (AN
F o r the straight-line t u n c t i o n s fitted to alpha intervals, A indicates rate o f c h a n g e (slope) in d u r a t i o n (usually n e a r zero d u r i n g feedback) a n d B is the e x t r a p o l a t e d a l p h a time in seconds (Y-intercept) w h e n N = zero, before the first event. W h e n the v a l u e o f A is zero, the intercept B equals the m e a n a l p h a d u r a t i o n for 30 events. F o r h y p e r b o l i c f u n c t i o n s fitted to
EEG Feedback Contingencies
49
no-alpha intervals, the magnitude of C measures the rate of change (slope) of no-alphas from the initial reaction to the first slide presentation to When the curve levels off at D, the asymptote, which is an estimate of the no-alpha durations in seconds after an infinite number of stimulus events. The functions fitted to data obtained during feedback are quantitative estimates of initial response and habituation. The statistics of best-fit functions obtained for each contingency condition were treated as a "score" and analyzed by ANOVA. The statistics analyzed were A, B, and SE for alpha; C, D, and SE for no-alpha; the average durations (X') of alpha and no-alpha in seconds over the 30 events of each condition; and the ratio of X'/SE for both alpha and no-alpha durations. Since the main hypothesis concerned bilateral comparisons of EEG control under five different contingency conditions, X/SE was emphasized in the data analysis. However, the parameters of the linear and hyperbolic functions, A, B, C, and D, were also analyzed to evaluate the possibility that differences in SE or X/SE might parallel differences in the parameters of the best-fit functions, i.e., where "scatter" was partially correlated with central trend.
RESULTS
The following is an outline of the presentation of this section. Statistics of both left and right EEG best-fit functions computed for each contingency and for alpha and no-alpha durations are presented in Table I. Descriptive statistics for the baseline conditions are shown in Table II. The relationships described in Tables I and II are compared in Fig. 2. The ANOVA's which were computed for A, B, C, D, SE, and X/SE are summarized in Table III. The ratio X/SE was computed from the time series of alphas and no-alphas for each trial; the corresponding average ratio was based on the pooled single-trial ratios, not on the ratio of the average X and the average SE that are presented in Table I. All average statistics were calculated from at least 20 data points. Three separate ANOVA's were computed for each statistic: (1) comparing all contingencies; (2) comparing just feedback contingencies (excluding-sham); (3) comparing only unilateral contingencies (to test the replication of previous results). In each ANOVA, variance was partitioned among Subjects, Contingencies (LC, RC, Both-And, Either-Or, Sham), Sides (Left EEG vs. Right EEG), Trials (first presentation of a particular condition vs. second presentation of that same condition), and the interactions of these factors. A pooled residual obtained by subtracting the par-
50
Mulholland and Eberlin
Table 1. Average Statistics of Best-Fit Functions (all values below are derived by pooling across trials) Contingencies and side LC L
RC R
L
R
Either-Or
Both-And
L
L
R
R
Sham L
R
Alpha A B
SE
~/SE 2/SE
.002 .002 .001 .001 .001 .002 . 0 0 1 .36 .34 .32 .32 .31 .30 .37 .39 .36 .35 .35 .33 .33 .39 .19 .19 .15 .13 .15 .14 .19 1.98 2.90 2.23 2.53 2.10 2.47 2.05
(pooled sides)
X/SE (unilateral noncontingent) (contingent)
2.38 2"26L
t
.001 .39 .40 .19 2.04
2.07
.004 .005 .48 .48 .55 .55 .39 .36 1 . 4 7 1.70 1.59
2"44
2.02 2.69~ No-Alpha
C D 2-
SE X/SE ~/SE (pooled sides)
X/SE (Unihtera1 noncontingent) (contingent)
3.18 2.11 2.41 1.59 1.87
3.39 1.96 2.31 1.81 1.52
3.35 2.42 2.82 2.05 1.50
l0l]
4.08 1.98 2.47 1.72 1.89
3.46 2.26 2.71 1.91 1.49
2.49 2.10 2.39 1.77 1.54
2.92 1.90 2.22 1.63 1.63
2.96 1.93 2.20 1.62 1.66
4.23 1.85 2.45 2.15 1.18
4.37 1.66 2.18 1.85 1.25
1.52
1.65
1.22
1.95
1.86
1.40
1.51 1.88~
X/SE (pooled sides, alphas, no-alphas) 1.98
2.07
titioned variance f r o m the total variance was used as the d e n o m i n a t o r for F tests. The results o f 30 A N O V A ' s are difficult to present compactly. In Table III, only significant ( p < .05) F ratios are presented with the relevant d fe and MSe'tO the right. T h o u g h the description o f results will be in terms o f Fig. 2, and Tables I and II, statistical verification can be f o u n d in Table III. All a posteriori multiple c o m p a r i s o n s alluded to below were examined statistically t h r o u g h the N e w m a n - K e u l s m e t h o d .
EEG Feedback Contingencies
51
Table I1. Average Statistics for Baselines for Before Feedback, Eyes Closed, and Eyes Open (all values below are derived by pooling across trials) Before feedback RC
LC L
L
R
Either-Or R
L
B oth-And
R
L
R
Sham L
R
Eyesclosed Eyes open L
R
L
R
Alpha
SE
Y/SE ~/SE Z/SE
.83 .88 1.02 1.06 1.07 1.20 .77 .87 1.13 1.31 .83 .87 1.06 1.04 .64 .73 .87 .99 .90 .90 .60 .79 .91 .82 .85 .85 1.02 .93 1.53 1.43 1.30 1.40 1.54 1.72 1.34 1.33 1.57 1.62 1.17 1.17 1.16 1.18 (pooled Before)
L: 1.46
(pooled sides)
R: 1.50
(pooled EC, EO)
1.48
L: 1.17
R: 1.17
1.17 No-Alpha
1.33
1.11 1.80
1.86 2.24
1.88
1.37 1.12 1.42 1.05 2.39 2.10 1.24 1.04 .94 1.06 2.37 2.35 1.24 1.14 1.30 1.31 1.23 1.22 1.08 1.11 1.11 1.18
SE 1.00 1.23 1.63 1.49 2.35 1.82 1.40 1.08
~/SE
1.40 1.29 1.32
X/SE (pooled Before)
%/SE
1.36 1.35
1.26
L: 1.32
R: 1.29
(pooled sides)
(pooled EC, EO)
L: 1.10
1.30
1.12
1.39
1.14
R: 1.15
X/SE (pooled sides, alphas, no-alphas)
EEG-Stimulus Contingencies Control. The degree of control (X/SE) was least for the baselines "eyes open" and "eyes closed," greater for Before Feedback, and greatest for the stimulus contingencies. The Sham feedback showed the smallest increase in EEG control over baseline values; the contingent feedback conditions were associated with larger increases in control compared with baseline. Combining X/SE over left and right EEG's, for alpha durations the Either-Or contingent (2.38), Right contingent (2.44), and Left contingent (2.26)--all statistically equal--had the highest control values; Both-And contingent (2.07) was less than the Either-Or contingency (Q[135] = 3.74, p < .05); and all four contingencies showed greater control of the EEG compared with noncontingent, Sham feedback (Q[171] i> 5.14, p < .01). For no-alpha durations the differences among contingencies were less. BothAnd (1.65), LC (1.70), and RC (1.69) were all about the same; the smallest ratio was obtained for the Either-Or contingency (1.52). Although these contingency differences for no-alpha were not significant, as with alpha all contingencies exhibited statistically greater control compared with Sham (Q[171] >I 4.74, p < .01).
52
Mulholland and Eberlin
3.e-
['-'~ A L P N A S
[]
IIIO-AL PNA8
|.0 o
L
•
LC
•
RC
R
L II I[ITlUlI011
L | ImTNAND
L
I
~- CONTINGENCIES -~4NAIII~
Fig. 2. X/SE for alpha and no-alpha durations as a function of the four EEG contingencies and Sham; baseline averages for Before Feedback, eyesclosed (EC), and eyes open in the dark (EO) are included for comparison. Pooling .X/SE across both Side and alpha/no-alpha intervals, the two unilateral contingencies (LC: 1.98; RC: 2.07) and the Either-Or condition (1.95) showed the largest ratios; Both-And (1.86) was slightly (though not significantly) less; and all four contingencies had greater control than Sham (1.40) (Q[315]/> 7.71, p < .01). Variability. Combining left and right EEG data, the standard error (SE) for alpha durations was greatest for the Sham feedback (.38), less (Q[171] = 12.02, p < .01) for the Both-And contingency (.19), and least (Q[ 135] I> 2.83, p < .05) for the unilateral ( L C : . 17; R C : . 16) and Either-Or (. 145) contingencies. For no-alpha durations, although SE tended to be highest for Sham feedback (2.00) and least for the Both-And contingency (1.625), none of the differences among contingencies or even between any one contingency and Sham was significant. Central Trend. For alpha durations, mean (X') for Sham was greater than for any of the contingencies (Q[171] ~> 2.96, p < .05). All intercontingency mean comparisons were significant (Q[135] i> 3.54, p < .05), except
EEG Feedback Contingencies
53
Table IlL Summary ANOVA (F, df) for Each ANOVA, Each Statistic, and Each Variance Partition (p < .05) Contingency
Analyses
1. (all contingencies plus Sham) 2. (Sham excluded)
3. (unilateral only
Side
CXS
Trial
Alpha A 2.8, 4 B 11.7, 4 2- 27.6, 4 SE 36.5, 4 X/SE 13.7, 4 5.5, 1 A B 7.4, 3 2- 12.8, 3 SE 9.9, 3 X/SE 3.5, 3 4.6, 1
4.9, 1
A B X
SE X/SE
4.4, 1
MSe
dfe
0.320 0.158 0.110 0.010 7.2, 4 0.343
171 171 171 171 171
0.164 0.005 0.002 12.4, 3 0.002 10.9, 3 0.299
135 135 135 135 135
5.1, 1 0.003 38.9, 1 0.002 31.0, 1 0.290
63 63 63
No-Alpha 1.
C D X
2.6, 4
SE X/SE 10.3, 4 2.
C D J(. ~2.8,3
SE .Y/SE 3.
C D X
SE X/SE
15.5, 5.4, 17.7, 12.5,
1 1 1 1
16.700 0.510 0.637 0.670 4.4, 4 0.161
171 171 171 171 171
11.8, 7.1, 20.6, 14.2,
1 1 1 1
17.200 0.510 0.540 0.660 5.4, 3 0.174
135 135 135 135 135
23.950 0.430 8.6, 1 0.570 6.3, 1 0.680 11.4, 1 0.246
63 63 63 63 63
4.6, 1
b e t w e e n L C a n d RC. F o r n o - a l p h a d u r a t i o n s n o c o n t i n g e n c y m e a n was statistically d i f f e r e n t f r o m S h a m . The o n l y significant i n t e r c o n t i n g e n c y comp a r i s o n was b e t w e e n R C (2.65) a n d B o t h - A n d (2.21) (Q[135] = 3.79,
p < .05). L e f t E E G vs. Right E E G Control. F o r alphas, p o o l i n g all c o n t i n g e n c i e s , the ratio, X / S E , was slightly greater o n the right (2.25), c o m p a r e d with the left (2.05). N o l e f t - r i g h t differences were s i g n i f i c a n t for n o - a l p h a s .
54
Mulhoiland and Eberlin
Variability. No significant left-right differences were found for alpha and no-alpha durations. Central Trend. No significant left-right differences were found. Interaction of Contingency and Side Control. The ratio, X/SE, for the unilateral contingencies was greater in the EEG which elicited presentation of a stimulus slide: for LC, control was greater in the left EEG for both alpha and no-alpha durations; for RC, control was greater in the right EEG. Pooling across contingencies, for alpha durations the contingent EEG's (2.69) showed 33% more control than the noncontingent EEG's (2.02); for no-alpha the ratio was 24.5% greater for the contingent (1.88) than the noncontingent (1.51) EEG's. Combining X / S E ratios across alpha and no-alpha durations as well as Side, the overall increase in control for the contingent (2.28) over the noncontingent EEG (1.76) was 29.5o/o. Left-right differences were less for the other contingencies with no consistent pattern of interactive differences. Variability. For alpha intervals during LC and RC, SE was less in the EEG controlling stimulus presentation. Left-right differences were much smaller for the other contingencies. For no-alpha durations no significant interaction between contingency and side was found. Central Trend. No significant interaction of contingency and side was found for mean (27) alpha or no-alpha intervals. Trial The changes from the beginning to the end of the experimental sessions were consistent with habituation. Alpha durations were shorter (X = .39 sec) in the first half of the session compared to the second half (X = .41 sec). No-alpha durations were longer in the first half (.,Y = 2.65 sec) and briefer in the second half (X = 2.17 sec). These differences are especially prominent and consistent for the no-alpha durations (See Table III: Trial).
Individual Differences Though not shown in Table III, differences among subjects were significant for all analyses and contributed to the largest proportion of total variance. This proportion was smallest for the statistic Y,/SE.
EEG Feedback Contingencies
55
Baselines Alpha and no-alpha durations were measured before feedback began in each trial. The EEG record was monitored visually, and when alpha was occurring the collection of Before Feedback data began. There was no reason to expect a trend over the course of the Before Feedback condition; therefore, the mean was taken as the best estimate of the durations of alpha and no-alpha. Before feedback, alphas are longer and no-alphas briefer than during feedback (see Tables I and II). Increased control during feedback can be determined by comparing the ratio X / S E before feedback with the same ratio during feedback. In all cases, pooling left and right EEG's, feedback contingencies were associated with increased X / S E during feedback compared with before feedback. The differences between before and during visual stimulation for the Sham condition were small compared to those obtained in the feedback conditions. The ratio .~/SE was least for the eyesclosed and eyes-open baseline conditions compared to the other conditions. For a more comprehensive description of results, the reader can compute the best-fit functions using the values of A, B, C, D, and SE given in Table I. The resulting habituation functions will give an estimate of the initial disturbance and recovery of the EEG following the onset of stimulaItion.
DISCUSSION The principal finding of this experiment is that the contingency between the EEG and visual stimulus is a relevant variable for the control of the alternation between alpha and no-alpha intervals. Control is measured by the ratio of X/SE. The EEG is best controlled bilaterally with the EitherOr contingency, and unilaterally for that EEG connected to the stimulus. This latter finding confirms results previously reported (Eberlin & Mulholland, 1976). Control is less for the Sham, noncontingent stimulation than for any of the contingent feedback stimulation conditions. The control of al_pha durations is greater than the control over noalpha durations, i.e., X / S E is greater for alpha compared with no-alpha durations During Feedback and for Sham conditions. For Before Feedback, and the eyes-open and eyes-closed baseline conditions, the differences between the control of alpha and control of no-alpha are much less and not consistent. Under bilateral feedback it should be noted that while alpha durations show greater control for the Either-Or contingency than for Both-And, with
56
Mulholland and Eberlin
no-alpha intervals the relationship is reversed; control is better for BothAnd than for Either-Or, though the difference is not as large as that for alphas. This reversal can be tentatively explained by contingency differences between stimulus onset and offset. For the two bilateral contingencies the stimulus was turned on when the requirements listed in Procedure were satisfied. However, the stimulus was removed for Either-Or when no-alpha was detected in both EEG's simultaneously and for Both-And when no-alpha occurred in either EEG. Thus, during Either-Or conditions, offset of the stimulus (and hence, the beginning of each no-alpha interval) was actually controlled by a Both-And contingency; for Both-And, stimulus offset was controlled by an Either-Or contingency. Ratios for this study do show the EEG to be somewhat better controlled bilaterally when stimulus onset is under Either-Or feedback. However, the experiment was designed to demonstrate that any contingency will improve EEG control over a noncontingent (Sham) condition, not necessarily to detect the subtle differences observed post hoc between any two types of contingency. Continued research is important to determine if these differences between Either-Or and Both-And increase when both ends of the visual stimulus presentation are under equivalent control. The noncontingent EEG provides a rigorous test for the effect of contingency. The cortex on the noncontingent side is receiving the identical stimulus as the cortex on the contingent side, and is connected to the "contingent cortex" by internal paths. Thus, the noncontingent EEG is a yokedcortex comparison condition for contingency. The results show that improved feedback control is due to feedback contingency rather than some special temporal pattern of intermittent stimulation (Eberlin & Mulholland, 1976). Apparently, within the baseline conditions, two different levels of control exist, both of relatively low magnitude. The X / S E ratios for Eyes Closed (EC) and Eyes Open (EO) are statistically equal, and average across left and right EEG's to 1.17 for alphas and 1.12 for no-alphas. The respective ratios for the ten event Before Feedback baselines are 1.48 and 1.30, both substantially higher than the eyes-closed and eyes-open baselines. We suspect that this increase is due to (1) a "carry-over" entraining effect from the previous During Feedback condition (despite the 20-sec interval between conditions), or (2) a generally higher level of subject alertness during the series of slides (note the similar ratios for Sham feedback; for alphas: 1.59 and for no-alphas: 1.22), or an interaction of both. An additional increase in X / S E is evident when the respective ratios for Before Feedback EEG's are compared with the alpha (2.02) and noalpha (1.51) ratios for the "noncontingent" EEG's during unilaterally contingent feedback. We suggest that although alpha intervals recorded from
EEG Feedback Contingencies
57
only the contingent EEG are eliciting the stimulus, because of the internal neural pathways between hemispheres, some degree of control must be exerted on the contralateral EEG. Thus, it might be better to refer to the "noncontingent" EEG as being "less reliably contingent." The response to the stimulus and the habituation of the response as measured by the best-fit trends for alpha and no-alpha are similar for the feedback contingencies. However, the "scatter" relative to the best-fit trend is different. The best-fit line is a "better fit" for the unilateral, contingent EEG data and for the bilaterally contingent (Either-Or) EEG data compared with the other contingencies. It follows that comparison of the trend lines is more accurate and precise when the trends have large ratios of X/SE. This improved control of alpha and no-alpha durations by feedback (perhaps analogous to enhanced signal-to-noise) is a basis for the utility of feedback methods in research. Additional experimentation is needed to examine other parameters of external feedback, e.g., time delay, or continuous vs. dichotomous feedback, to further optimize the control of this EEG response by feedback.
ACKNOWLEDGMENTS The authors thank Anne Davidson of Tufts University for her assistance with the statistical analyses. This research was supported in its entirety by the Veterans Administration Research Program MRIS 5890. REFERENCES Boudrot, R. An alpha detection and feedback control system. Psychophysiology, 1972, 9, 461-466. Goodman, D. ALFIE: Collection of EEG alpha under feedback control using time series analysis. Psychophysiology, 1973, 10, 437-440. Eberlin, P., & Mulholland, T. Bilateral differences in parietal-occipital EEG induced by contingent visual feedback. Psychophysiology, 1976, 13, 212-218. Lewis, C. N., & McLaughlin, T. J. Baseline and feedback EEG correlates of interview behavior. Psychophysiology, 1976, 13, 302-306. Mulholland, T. Objective EEG methods for studying covert shifts of visual attention. In F. J. McGuigan & R. A. Schoonover (Eds~.), The psychophysiology of thinking. Academic Press: New York, 1973, pp. 109-151. Mulholland, T. Biofeedback as scientific method. In J. Beatty & G. Schwartz (Eds.), Biofeedback theory and research. Academic Press: New York, 1977 (in press). Mulholland, T., McLaughlin, T. J., & Benson, F. Feedback control and quantification of the response of EEG alpha to visual stimulation. Biofeedback and Self-Regulation, 1976, I, 411-422. (Revision received December 21, 1976)