1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
Back to Biofeedback... / Home PageChapter 4, In L. White and B. Tursky (Eds.), Clinical Biofeedback: Efficacy and Mechanisms. Guilford, pp. 75-108.
Clinical_bio82.doc
See also: Comments by Thomas B. Mulholland, and
Reply by John Furedy and Diane Riley
CLINICAL BIOFEEDBACK:
EFFICACY AND MECHANISMS
Edited by
LEONARD WHITE
Long Island Research Institute
Department of Psychiatry and Behavioral Science
State University of New York at Stony Brook
and
BERNARD TURSKY
Laboratory for Behavioral Research
Department of Political Science
State University of New York at Stony Brook
THE GUILFORD PRESS
New York, London
4 CLASSICAL AND OPERANT
CONDITIONING IN THE ENHANCEMENT OF BIOFEEDBACK:
SPECIFICS AND SPECULATIONS
JOHN J. FUREDY AND DIANE M. RILEY
More general formulations of the approach to biofeedback enhancement that is employed in the University of Toronto laboratory have been provided previously (Furedy, 1977, 1979). In this chapter we first describe this approach in a detailed way, wherein we provide “specifics.” In the first section, we lay out the ways in which our approach differs from conventional biofeedback approaches, and then provide some testable hypotheses derived from this approach, together with brief descriptions of the methods of testing those hypotheses. In the second major section we move beyond such empirical considerations to more theoretical and historical ones. These considerations arc necessarily more speculative and subjective than arc the empirical, essentially descriptive, considerations raised in the first section. Nevertheless, such speculations axe important, if only because they often form a powerful influence on the manner in which a research program is carried out. In such contexts as those of grant proposals or papers directed at a large group of generalists, these
John J. Furedy and Diane M. Riley. Department of Psychology, University of Toronto. Toronto, Ontario, Canada.
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
speculative aspects are seldom laid out systematically, the result being that they arc not subjected to adequate critical examination. A specialized group of contributors such as that represented in this volume can provide the detailed criticism that will allow for improvement of these formulations, particularly with respect to their explicitness and clarity.
SPECIFICS: A RESPONSE-LEARNING APPROACH TO THE CONTROL OF STRESS-ELICITED PHASIC AUTONOMIC CHANGES1
BACKGROUND AND THE RESP0NSE-LEARN1NG APPROACH
Modern life is full of stressful events that produce the “light or flight” reflex in humans. Components of this reflex are short-term or phasic autonomic changes, such as heart rate acceleration. In some cases, even “normal” responses to stress can be undesirable. One strategy for eliminating such reactions is that of eliminating the stressors themselves. However, this strategy is impractical because people vary so much in what they find stressful. Thus, to take an extreme but illuminating example, consider the coronary patient resting in a quiet ward designed to minimize the occurrence of stressful stimuli. A new doctor comes to visit the patient, and it happens that the doctor’s face is similar to the face of another person with whom the patient has had a very emotional disagreement. As a result, the new doctor’s face serves as a stressor, and it elicits an undesirable heart rate acceleration (HRA), which can itself precipitate another debilitating cardiovascular episode. It will be noted that the strategy of eliminating such stressors from the environment is unlikely to work, because there is no way of predicting what, for a given patient, will be stressful.
An alternative to the strategy of external environmental control is that of internal behavioral control. In this strategy, persons axe taught to control their reactions to whatever events happen to be stressful to them. So, to control stress-induced HRA, a patient might learn to produce the opposing heart rate deceleration (HRD) at the time that the HRA-inducing stressor appears. The particular tactic for such behavioral control that has been most generally used adopts the approach of providing information about the autonomic nervous system’s functioning through biofeedback. For example, in applying this “informational-biofeedback” approach to the teaching of HRD, the procedure is to have the person informed (using expensive computerized polygraphic equipment) about the rate of his or her beating heart, on the assumption that this information will enable the person to learn to decelerate his or her heart rate.
1For some of the ideas formulated in this section, we arc indebted to. Arabian. R. Heslegrave, and E. Tulving.
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
Recent research has indicated that this assumption is largely a “hope” that is “unfulfilled” (Blanchard 8c Young, 1973), at least with regards to the control of such autonomically mediated changes as HRD. To draw this pessimistic conclusion is not to deny the few successes that some workers have had in producing large HRDs and control of cardiac arrhythmias (e.g., Engel, 1972). What has continued to fail to emerge is statistically significant evidence (1) that such control is indeed attributable to the provision of information, and (2) that it varies as a function of the adequacy of that information. Recent strong support for this pessimistic conclusion is the fact that one of the most influential and thorough investigators of the informa-tional-biofeedback approach to HRD has counseled clinicians to use the less expensive relaxation methods to control anxiety, in preference to the more expensive computerized polygraphic biofeedback arrangements (Lang, 1977). The basis of this advice includes evidence from Lang’s own studies where, for example, a group given precise information about their heart rate produced less HRD than did a group of subjects who were simply told to repeat a “mantra*’ on a regular (and presumably relaxing) basis.
The alternative tactic or approach that we have adopted in our studies of HRD learning is based on the assumption that the desired response should be elicited first, before any attempt is made to teach it. This “response-learning” approach (Furedy, 1979) begins with events such as an instructed breath hold (e.g., Furedy &Poulos, 1975, Exp. I) or a negative body tilt (e.g., Furedy &Poulos, 1976, Exp. I). These events reliably elicit a short-term or phasic HRD response of large magnitude (over 30 beats per minute). In terms of Pavlovian conditioning, the events are unconditional stimuli (USs) that elicit the HRD response as the unconditional response (UR). Moreover, at least on the face of it, this sort of UR appears to be a good candidate for controlling the phasic HRA responses elicited by stressors. That is, if the subject can learn to produce a significant portion of the HRD UR in the absence of the US but in the presence of the stressor, then this response-learning tactic may be an effective way of learning behavioral control over stress.
Accordingly, the next step is to determine whether pairing a relatively neutral stimulus (such as a tone) as the conditional stimulus (CS) with the US will produce some HRD response learning to the CS alone in a Pavlovian-conditioning paradigm. This has in fact been shown to occur both with the instructed breath-hold US (Furedy & Poulos, 1975, Exp. II) and the negative-tilt US (Furedy &Poulos, 1976, Exp. II). Moreover, it appears that the most promising form of this Pavlovian paradigm 2s the “imaginational” one (Furedy, 1977), wherein part of the CS is an instruction to the subject to imagine the US. This imaginational paradigm, which has produced the largest HRD responses, is still Pavlovian inasmuch as the target anticipatory HRD response does not affect the tilt US. The procedure, however, is quite complex (for details, see Furedy &Klajner, 1978), with the role of imagery in particular being apparently important but very difficult to specify.
We have also extended this complex but still Pavlovian paradigm to an oper-
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
ant-conditioning procedure with discrete trials. In this arrangement (see Furedy, 1979. pp. 210-211), the HRD response elicited by the CS through Pavlovian conditioning is now shaped or contingently reinforced by verbal praise (“good,” “very good,” or “excellent”); the larger the HRD response in the presence of the CS (now a discriminative stimulus in an operant-conditioning procedure with discrete trials), the more praise the subject gets. This Pavlovian-operant procedure has produced some encouraging results (e.g., Furedy, 1977, 1979) both in terms of HRD and in terms of probable sympathetic withdrawal as indexed by the T-wavy component of the electrocardiogram (sec Fig. 14.1 in Furedy, 1979).
The two approaches that we have characterized respectively as “informational-biofeedback” and “response-learning/* are not completely incompatible. For example, it will be noted that the operant-conditioning paradigm described above, which has stemmed from our response-learning approach, is a form of biofeedback. Nevertheless, the response-learning approach does differ from the conventional biofeedback approach in ways that have been indicated elsewhere (Furedy, 1979) and are summarized in an injunction to “remember the response.” Among the important differentiating characteristics (elaborated in the cited study) of the response-learning approach are (1) the focus on phasic rather than tonic changes; (2) initial elicitation of the target response before attempting to teach it, and (3) the relative reemphasis of such purely informational aspects of learning as the exact stage of the target behavior (e.g., the precise heart rate) and the predictive or sign-significant relationship (in Pavlovian conditioning) between the CS and the US. What the Furedy study (1979) docs not provide, however, arc a number of testable hypotheses that can be generated from the response-learning approach, and that are not derivable from or even compatible with the informational-biofeedback approach. We now present four such hypotheses.
HYPOTHESES TO BE INVESTIGATED Hypothesis 1 (Hi)
The Pavlovian HRD conditioning preparations that we have developed through the response-learning approach arc premised on a stimulus-substitution or response-transfer account of the process. For example, on this account, the HRD response initially elicited as a UR by the tilt US comes through stimulus substitution to be elicited to a lesser extent by the CS as a result of CS-US pairings; the HRD response, that is, is partially “transferred” from the US to the CS. It follows from this account that an important factor determining the strength of conditioning is the temporal contiguity or closeness between the CS (the future elicitor of the HRD response) and the UR (elicited by the US). Because the UR is closely time-locked to the US, this CS-UR contiguity b closely related to the temporal interval between the onsets of the CS and the US, which is commonly termed the “interstimulus interval” (ISI).
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
Accordingly, H1 is as follows: It is the ISI that is important, rather than information given to the subject about the relationship between the CS and UR or about his or her heart rate following CS presentation. It will be noted that this ISI (H1) hypothesis denies significance to informational “cognitive” factors that are statable in terms of propositional information,either about the CS-US relationship or about the state of some autonomic function. (For further elaboration of the distinction between propositional and response processes, sec Furedy, 1979, pp. 211-212.)
Hypothesis 2 (H2)
There is, as we have noted previously (Furedy & Poulos, 1976, p. 95), another aspect of the tilt-conditioning procedure that makes this form of human cardiac conditioning more consistent with a stimulus-substitution, response-transfer view than are the other, more conventional procedures, which use noxious stimuli (such as shocks and loud noises) as USs. This aspect is the topographical similarity of the CR and UR, both being decelerative. In contrast, the conventional cardiac conditioning procedures produce an acceleratory UR, but a CR that has been variously identified as multiphasic (Hendrick & Graham, 1969; Zeaman, Deane, & Wegner, 1934), decelerative (Wood & Obrist, 1964) or accelerative (Zeaman &Smith, 1965). Indeed, it is from this human cardiac conditioning area, with its failure to yield clear evidence for UR-CR topographical similarity, that much of the support has come for the currently dominant cognitive view of Pavlovian conditioning. On this cognitive view, the mechanism involved is the learning of (propositional information about) CS-US relationships and not the transfer of responses from US to CS. For the present purposes, our interest is not in Pavlovian conditioning in general (for which issue the distinction between propositional and response processes offered by Furedy, 1979, may turn out to be relevant), but in the more circumscribed question of what mechanism of learning is involved in the tilt-conditioning preparation. The clear decelerative CR-UR similarity found in that preparation tends to support a response-learning view; however, a more compelling source of support for this view would be obtained if the following, more “daring” hypothesis (H2) were confirmed: Reversing the direction of the UR in the tilt-conditioning preparation will reverse the direction of the CR.2
Hypothesis 3 (H3)
The next two hypotheses concern the operant-conditioning extensions that we have developed for HRD response learning. Although, as indicated above, biofeed-back terminology is applicable to these operant-conditioning procedures with dis-
2This hypothesis was recently formulated by T. Matyas, and its test will, we hope, be carried out in collaboration with him.
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
crete trials, the emphasis on response learning or response shaping yields a view that is different from the conventional, informational-biofeedback approach. That approach emphasizes the quality of the propositional information given to subjects about their target behaviors (here, HRD). From a quality-of-information point of view, the emphasis is on such aspects as the degree to which the feedback is continuous and hence contains accurate information about the target behavior, in contrast to the cruder information transmitted by “binary” feedback (i.e., presence vs. absence of some contingent reinforcer). On the other hand, in operant response-conditioning terms, what is being shaped is a response. From this vantage, the important factor is not the amount of information contained in the reinforcement, but its temporal contiguity or closeness to the target response that it is meant to reinforce. In other, more Hullian, words, it is the minimization of the delay-of-reinforcement gradient that would be emphasized by a response-learning approach.
Accordingly, H3 is as follows: In the operant decelerative paradigm, what is important is the immediacy of reinforcement (i.e., the continguity between the instrumental or target response and the reinforcer), rather than the amount of information provided in the reinforcement about the target responses.
Hypothesis 4 (H4)
The final hypothesis is also focused on the response-reinforcement link rather than on information about the response, but the aspect of the link in this case is not temporal contiguity but the degree of similarity of afferent inputs associated with the two (response and reinforcement) elements. The possible importance of this aspect was first noted by Tursky (sec Furedy, 1979, p. 215). We have been able to demonstrate that the tilt can be used as an effective reinforcer (see Furedy, 1979, pp. 215-216). However, H4 is as follows: In the operant decelerative paradigm, a reinforcer that shares afferent pathways with the target response will be more effective in an operant procedure than one that does not.
PROPOSED HYPOTHESIS-TESTING EXPERIMENTS
We regard the testing of the above hypotheses as a way of evaluating the fruitfulness of the response-learning approach. Here, we outline the proposed experimental tests that we have devised in the University of Toronto laboratory. Methodological criticisms of these tests are especially welcome at a time when these tests have not actually been carried out.
To test the ISI aspects of Hi, levels of HRD performance will be compared between groups of subjects conditioned with ISIs of 1, 5, and 10 seconds, respectively. It is predicted that conditioning will be a negative function of ISI, with little or no conditioning shown in the 10-seconds ISI group. As a control for conditioning, a fourth group will be run in which the CS and US arc presented in a random relation-
1
CLASSICAL AND OPERANT CONDITIONING IN BIOFEEDBACK
ship (Rescorla, 1967). Then, to test the aspect of Hi that relates to information about the relationship of the CS to the US, a fifth and a sixth group will be run under the random CS and US arrangements, but instructed, respectively, that the CS is an imperfect predictor of US presence (excitatorily instructed) or absence (inhibitorily instructed). A postexperimental continuous measure of subjective CS-US contingency (e.g., Furedy Sc Schiffmann, 1971) will be used to assess the subjects’ beliefs about the CS-US relationship in the three physically random groups. If, as expected on the basis of past evidence, this measure indicates the instructional manipulation to have been effective, then the question will be whether the three physically random groups differ appropriately with respect to their HRD responding to the CS; according to H1 they will not, nor will there be a correlation between degrees of belief in the CS-US relationship and magnitude of HRD responses. Finally, to test the aspect of H1 relating to information about heart rate, all six of the above groups (the three ISI and the three random groups) will each be divided into two equal subgroups. The “informed” subgroups will be given accurate information about their HRD response to the CS, whereas the “misinformed” subgroups will be given inaccurate information. Past evidence suggests that in this phasic HRD situation subjects cannot tell that they are in a “false” feedback condition. If this is confirmed in the study, then the question will be whether the accuracy of feedback affects HRD performance; according to Hi, it will not.
To test H2, the “reversal” hypothesis, we shall contrast conditioning between head-up (positive-tilt) and head-down (negative-tilt) USs. The negative tilt will be 45 rather than the 90 tilt used in our previously published studies, and will commence from a horizontal plane. Earlier studies have indicated that this negative-half-tilt procedure does produce a reliable HRD UR (about half the magnitude of the UR elicited by the “full” tilt), as well as learned HRD conditioning. Completely uninvestigated, however, is the effect of using the positive-tilt US. This will also begin from a horizontal starting point, and will be a 45 head-up change in body position with the same speed as that used for the negative half tilt—that is a duration of about 1.2 seconds. The initial studies will explore the URs elicited by these positive- and negative-tilt USs, with the expectation that the USs will produce changes in heart rate that arc topographically similar except for direction; the negative- and positive-tilt USs arc expected, respectively, to produce deceleration and acceleration. Topographical analyses of the UR will follow the method used by Furedy and Poulos (1976, Exp. I). These initial steps will aim at developing negative- and positive-tilt USs that produce opposite but otherwise equivalent topographical heart rate changes. Tone stimuli will then be paired with each US in a Pavlovian-conditioning paradigm. If conditioning is demonstrated in the positive-tilt as well as in the negative-tilt groups, then the question will be whether the CR under the former (acceleratory-inducing US) condition is itself acceleratory, as required by H2.