Like heart-rate changes, conditioned suppression does not reflect a mediating fear reaction in a manner completely consistent with two-process theories. However, there are two alternative interpretations of the CER experiments which might make the results compatible with a two-process description of avoidance learning. First, it may be that the role of the fear reaction in maintaining avoidance behavior is different from its role in the establishment and extinction of avoidance behavior. Thus the failure of a CS for a well-learned avoidance response to produce conditioned suppression may indicate that the CS is not producing fear in the avoidance situation and that the traditional two- process account of avoidance learning does not apply to maintained avoidance behavior. A second possibility is that the CER experiment is not an adequate index of the conditioned fear reaction. After all, there does not exist a closely reasoned account of the fact that the CER procedure produces suppression of the appetitively maintained operant. Why should we not instead find rate increases?

3. Physiological manipulations. Another method can be used to examine the interrelations between Pavlovian CRs and instrumental responses. If we suspect that a specific set of CRs is mediating instrumental behavior, we can simply eliminate those CRs and observe the effects upon the instrumental behavior. Solomon and his co-workers have pursued this line of research for various classes of CRs.

Using sympathectomized dogs, Wynne and Solomon (1955) have demonstrated that although removal of peripheral autonomic impairs avoidance learning, it does not prevent it; nor does such removal facilitate extinction of avoidance. Sympathectomy after avoidance learning does not impair performance in dogs. Presumably, sympathectomy combined with vagal blocking as used by Wynne and Solomon eliminates cardiac and blood- pressure changes. Following the same strategy, Auld (1951) usedtetraethylammonium (TEA) to block sympathetic autonomic nervous system (ANS) reactions, and found results perfectly paralleling those of Wynne and Solomon. There followed a long series of experiments, too numerous to describe here, in which barbiturates, autonomic blocking agents, stimulants, and tranquilizers were used to study the relationship between autonomic CRs and avoidance behavior. They did not importantly affect the conclusion that autonomic CRs are not necessary mediators of avoidance behavior.

Similarly, the transfer of Pavlovian fear conditioning from the curarized to the normal state, demonstrated by Solomon and Turner (1962) shows that peripheral skeletal CRs are not required for avoidance behavior. (These experiments are described in detail in the next section.) Both the sympathectomy and the curarization preparations eliminate broad classes of peripheral CRs as necessary mediators of avoidance behavior.

In summary, we have not yet identified any peripheral CRs which are necessary to mediate avoidance behavior. From this review of both aversively and appetitively motivated behavior, the simple idea that some peripherally observed CR is essential in the mediation of operant behavior seems implausible. In no case that we have studied does a peripheral CR seem to bear the required strong relation to the instrumental behavior. However, the two alternative views of the mediational process, (a) that a complex of autonomic and skeletal CRs is the mediator, and (b) that the peripheral CRs are simply indexes of central events, still seem to be reasonable. Both views permit some slippage between instrumental behavior and CRs. That we are not measuring complex enough peripheral behavior is difficult to refute; on the other hand, it is a relatively unattractive position because it suggests little that is new by way of experimentation except the recording of a larger number of CR measures.

However, consider a third view, (c), that what concomitance we do observe between instrumental behavior and peripheral CRs is due to mediation by a common central state. Then the concurrent measurement of instrumental behavior and Pavlovian CRs is not the optimal experimental strategy. Indeed, it becomes an irrelevant strategy.

Accepting the third view, then to find that a particular CR does not control operant behavior is hardly a refutation of a general two-process approach; indeed, it would be surprising if we should be able to select from the complex instrumental situation the few controlling CRs. Rather, the essential postulate of two process theory will then be that manipulation of Pavlovian conditioning procedures should have important effects upon instrumental behavior. Although we may not be able to identify the precise Pavlovian CRs which affect instrumental behavior, we can demonstrate that Pavlovian conditioning procedures exert strong influences over instrumental behavior in the absence of changes in instrumental contingencies. Such studies are reviewed in the next section.

MANIPULATION OF INSTRUMENTAL

BEHAVIOR BY SEPARATELY

CONDUCTED PAVLOVIAN

CONDITIONING PROCEDURES

It is one matter to lift Pavlovian concepts out of Pavlovian theory and experiments, as Hull and Spence did, and use them to explain instrumental behavior. It is quite another matter to employ the procedures of Pavlovian conditioning in order to influence already established, or to-be-established, instrumental behavior. The latter strategy is generated by two-process theory, because it assumes that Pavlovian conditioning and instrumental learning are two distinct processes, each governed by its own appropriate sets of operations and laws, and it is typified by the experiment of Solomon and Turner (1962). They avoidance- trained dogs in a panel-pressing apparatus with a visual Sd. The dogs were then completely paralyzed by d-tubocurarine, and were subjected to purely Pavlovian, discriminative conditioning procedures, with tone CSs and a shock US. When Ss were later tested in the panel-pressing apparatus, they retained their avoidance response to the visual Sd. In addition, they showed reliable panel- pressing responses to the tone paired with shock (CS+) during Pavlovian conditioning, but these responses were weak or absent when the tone not paired with shock (CS—) was presented. Thus, the postulated two processes were seen to interact in a particular way, such that the instrumental panel-pressing was immediately elicited by the introduction of a Pavlovian CS+, even though S had never before pressed a panel in the presence of CS+. This experiment can be analyzed in terms of two propositions of two- process learning theory: (a) Pavlovian association processes precede the acquisition of emotional reactions to previously neutral stimuli; and (b) these emotional reactions have motivational properties that can influence instrumental responding. It follows that any empirical or theoretical law of Pavlovian conditioning has profound implications for the control of instrumental responding when the two processes are interactively combined by E’s procedures.

What are some of these Pavlovian laws (see Pavlov, 1927), and how would they be expected to reveal their impact?

We have reviewed in some detail a few of the major laws of Pavlovian conditioning. What does two-process theory do with such laws? First, it assumes that these laws of Pavlovian conditioning of the salivary reflex are probably the laws of emotional conditioning or laws of acquired drive states. Second, it assumes that conditioned emotional states change S’s motivation level and thus can serve either as motivators or rein forcers of instrumental responses.

Table 1 provides a convenient summary of the variety of ways in which Pavlovian conditioned emotional states can interact with instrumental learning to produce changes in instrumental responding.

Note that Table 1 accomplishes three purposes. First, it allows us to classify and organize the existing knowledge about interactions between the two hypothetical processes. Second, it dramatizes the absence of certain kinds of knowledge, thus pointing to new experiments which need to be done. And third, it raises new and interesting theoretical questions concerning the outcomes of the experiments in the table.

The Solomon and Turner (1962) experiment, which we have already described, can be located in Table 1 in the following manner. In this experiment, the US for fear conditioning was aversive (shock) as was the reinforcer for avoidance behavior. In addition, the Pavlovian conditioning was discriminative, and the Ss were tested with independent presentations of CS+ and CS—. The instrumental training was discriminative, since Ss learned to respond by panel-pressing in the presence of Sd. This allows us to insert the results of the Solomon and Turner experiment in Cells 15 and 16. We arbitrarily use a (j’) sign in Cell 15 to indicate that the Pavlovian CS+ produced an enhancement of instrumental responding. CS—, on the other hand, had little or no effect, and so we have inserted a (?) in Cell 16.

The traditional CER experiment falls into Cell 3. The S is trained to perform some instrumental response reinforced by some appetitive stimulus. Pavlovian conditioning is carried out with an aversive US. The S is tested with presentations of CS+ while performing the instrumental response. The usual finding is that the response rate is suppressed by CS+, and so Cell 3 contains a (~).

Now we can examine the laws of Pavlovian conditioning as they are reflected in the interactions contained in Table I. In all cases, the dependent variable is some measure of instrumental responding. Yet the independent variables, the influence of which is being tested, are those contained in the Pavlovian conditioning experiment.

EXCITATION AND INHIBITION

1. Differential Excitation and Inhibition.Rescorla and LoLordo (1965) gave dogs discriminative conditioning with an aversive US (shock). A Pavlovian law predicts that the stimulus not paired with shock (CS—) will become a differential inhibitor—that is, it will have the capacity to suppress actively whatever emotional reflex pattern is usually elicited by CS+. Thus, CS— should be able to inhibit “fear of shock.” Prior to discriminative fear conditioning, the dogs had been trained so that they reliably responded on a Sidman avoidance schedule at a rate of seven jumps per minute in a dog shuttlebox. While Ss were jumping, short presentations of CS+ and CS— were given in some random sequence.

Two-process theory leads to the following expectations. If CS+ is a Pavlovian excitor, then conditioned fear should be augmented by its presence, and the instrumental responding rate should increase above the normal rate. In contrast, if CS— is a Pavlovian differential inhibitor, then it should actively suppress conditioned fear and the instrumental responding rate should decrease below the normal rate.

Rescorla and LoLordo (1965) found that the presentation of the CS+ resulted in a doubling of the Sidman jumping rate, but the presentation of the CS — resulted in a large reduction in the Sidman jumping rate. It seems clear that CS— had acquired inhibitory properties. The Rescorla and LoLordo experiment can be inserted in Cells 11 and 12 in Table 1. They represent the intersection of aversive, discriminative Pavlovian conditioning preceded by unsignalized (Sidman) aversive training, with no Sd.

Another example of differential Pavlovian conditioning combined with instrumental training is the experiment by Ray and Stein (1959) who trained hungry rats to bar-press for milk on a VI—2 schedule. Then, when the rats attained a steady response rate, an 1800 cps tone was presented for 5 minutes and it terminated with shock to the feet. A contrasting 200 cps tone was presented at other times, but it terminated without shock. Eventually, the 1800 cps tone acquired the capacity to suppress bar- pressing completely. In contrast, presentation of the 200 cps tone (CS—) often produced increases in the bar-pressing rate above the normal base-line rate, though this difference was not a large one (see Cells 3 and 4, Table 1).

Hoffman and Fleshier (1964) have carried out a series of experiments similar to that of Ray and Stein. In general, their findings paralleled those of Ray and Stein with regard to the CER suppression effect of CS+ presentations. However, unlike Ray and Stein they did not find the enhancement of instrumental responding in the presence of CS—. Hammond (1966) repeated and extended the results of Ray and Stein. He further showed that enhancement of bar-pressing rate by presentation of CS— occurred only when the base-line responding rate was below normal.

These interesting findings raise the possibility that a Pavlovian differential inhibitor established by aversive conditioning can enhance instrumental responding established by an appetitive reinforcer. But what could the finding, that an inhibitor of fear enhances appetitively maintained behavior, mean? One possibility is that reflex interrelations exist between appetitively and aversively based incentive states. The occurrence of an elicitor or inhibitor of a fear state may reflexly depress or enhance positive incentive motivation. On the other hand, it may be that an inhibitor of fear has no effect upon an appetitive motivation state. Instead, there may be a general level of fear produced by the experimental situation; since such fear presumably reduces the response rate, an inhibitor of that fear would lead to a rate enhancement. The Hammond experiment strongly supports such an interpretation. These findings are of considerable importance to a theory of incentive motivation.

So far, we have concerned ourselves with aversive Pavlovian conditioning, and have traced the effects of CS+ and CS—, in their roles as differential excitors and inhibitors, on both aversive and appetitive instrumental responding. There is, in addition, a series of experiments in which appetitive, discriminative conditioning procedures are combined with appetitive instrumental training procedures. These experiments come out of the Skinner- Ian tradition. A prototype experiment is that of Estes (1948), and it was probably the first of its kind to be successful in showing that a CS+, previously paired with food presentations, could enhance, during extinction, the rate of an operant previously reinforced by food presentations. As is the case in all of the experiments subsequently using the Estes paradigm, CS+ presentations are in reality being paired not only with food presentations, but also with magazine approach responses. This is inevitable, because the conditioning procedure in these experiments is done through magazine training. In order to get the food, Ss must make instrumental responses. Even though this procedure is not “pure” Pavlovian conditioning, it resembles it to the extent that the approach response cannot produce the food US; only the CS+ can produce it. Neither can the approach response produce the CS+. (The procedure can be described in Skinnerian terminology as a “discriminated operant.”)

Morse and Skinner (1958) trained pigeons to approach a magazine for food in the presence of one color (CS+), but the food never was presented in the presence of a contrasting color(CS—). The behavior of S was irrelevant for the occurrence of magazine operation. Then, in the second stage of the experiment, Ss learned to peck at a white key for food. Finally, extinction was instituted, during which there were test periods in which CS+ and CS— were alternately presented. The pecking rate was higher in the presence of CS+ than it was when CS— was present. There was, however, no control for normal extinction in white light alone, so we cannot tell whether CS— was inhibitory, or the CS+ excitatory, or both. We know CS+ showed differential excitatory properties when compared to CS—, and this confirms Estes’ (1948) findings, but it leaves in doubt the proper sign to put in Cells 1 and 2 of Table 1. It is our guess that Cell 1 should contain a (~) sign, and Cell 2 a (~) sign, but in the absence of the proper controls we cannot be sure. A recent experiment by Bower and Kaufman (1963) confirms the finding by Morse and Skinner of a difference between the effects of CS+ and CS—.

Most of the earlier experiments showing the differential excitatory and inhibitory effects of CS+ and CS— on instrumental responding have used extinction responding as a base line. Recently, however, there have been several experiments exploring the effects of differential Pavlovian conditioning procedures on subsequent learning of discriminative instrumental responses. For example, Bower and Grusec (1964) used thirsty rats, conditioning them by having tone S1 paired with water reinforcements and tone S2 occurring without water reinforcements. The rats had previously been trained to press a lever to get water, but this early training was not discriminative (no Sd). Then, in a third stage of the experiment, the rats were given discriminative instrumental training. One group was trained with its CS+ from the conditioning phase now the Sd for the operant, and its CS— as S~, while in contrast, another group had its CS+ from the conditioning phase now the S~ for the operant and its CS— as Sd Thus, in one group the CS/Sd relationship was consistent, but in the other group the CS/Sd relationship was inconsistent. Bower and Grusec found that the acquisition of the Sd—S~ discrimination was enhanced for the consistent group but was interfered with in the inconsistent group. There was, however, no way of ascertaining whether or not CS— had true differential inhibitory properties because there was no control group that learned the Sd-S~ discrimination without prior conditioning with CS+ and CS—.

We can speculate, however, that learning was interfered with whenever CS— inhibited conditioned appetitive reactions in the presence of Sd, and learning was facilitated whenever CS + facilitated excitatory appetitive reactions in the presence of Sd.

More recently, Trapold (1966) has shown that inconsistent differential conditioning can actually facilitate reversal learning of a discriminative instrumental response. Rats were first trained, in an operant discrimination, to press a lever for food in the presence of S1 but to refrain from pressing in the presence of S2. Then the lever was removed, and S2 was paired with food presentation while S1 was paired with absence of food. Later, when the rats were presented again with the lever, they were required to reverse the original operant discrimination and press in the presence of S2. They learned this reversal more rapidly than a group that had not received pairings of S2 with food presentation. Thus, Pavlovian contingencies were powerful enough to assist the instrumental discrimination reversal.

2. Conditioned Inhibition. Rescorla and LoLordo (1965) trained dogs in our laboratory on a Sidman avoidance contingency in the shuttlebox, until the dogs performed the avoidance response at a stable rate. Then they subjected some of the dogs to a Pavlovian fear conditioning procedure in which CS+ was followed 2—8 seconds later by shock on one-half of the trials, but on the other half of the trials CS+ was followed 2—8 seconds later by CS— and no shock. Other dogs, after learning the Sidman avoidance response, were subjected to a Pavlovian fear conditioning procedure in which CS+ was followed by shock on one-half of the trials, but on the other half of the trials CS— was inserted 5 seconds prior to CS+ and no shock followed. CS — in both procedures was shown to have inhibitory properties. Test presentations of CS — reduced the Sidman avoidance response rate significantly. Rescorla and LoLordo infer that conditioned fear was inhibited by their CS—. In contrast, fear was aroused by CS +.

3. Inhibition by Temporal Delay. Rescorla (1967a) trained dogs in our laboratory in a Sidman avoidance response in the dog shuttlebox. When the dogs hadacquired a stable jumping rate, they were subjected to a Pavlovian fear conditioning procedure in which only a long-delay CS+ was used. A 30-second tone was followed by shock on all conditioning trials. Then later, when the dogs were performing their avoidance response in the shuttlebox, the 30-second tone was presented from time to time. Rescorla found that the onset of the tone produced a decrease in jumping rate, and the rate thereafter increased until, at about 20-second duration, the jumping rate went above normal, increasing steadily to the end of the interval. Cessation of the tone produced a decrease in jumping rate to a below-base-line level, followed by slow recovery to the base-line rate. Here is a case where onset of a “danger signal” resulted in a temporary decrease of avoidance responding. This is what one would expect from Pavlovian experiments on long duration CSs. CS onset is never closely paired with shock, and so it serves as a CS—, an inhibitor of conditioned reflexes. In the Rescorla experiment, we can infer that CS+ onset inhibited fear.

Recently, Trapold, Carison, and Myers (1965) have shown how Pavlovian inhibition by temporal delay can operate in an appetitive situation. They conditioned rats with either fixed or variable temporal delay intervals between food US presentations, in the absence of any CS (temporal conditioning). Then they gave the rats Fl training with food reward in a bar-press situation. They found that when the temporal interval between US presentations previously used in Pavlovian conditioning was the same as that used in the subsequent Fl training, the development of a sharp Fl “scallop” was facilitated. Evidently the delivery of a US can serve as a stimulus which temporarily inhibits an appetitive state that mediates bar-pressing for food. The temporal pacing of instumental behavior, as in FT contingencies, can be “sharpened” by Pavlovian conditioning treatments of the proper sort. This is evidence quite compatible with that of Pavlovian experiments.

4. Induction. Not much is known about the operation of the Pavlovian induction phenomenon as it influences instrumental responding. Rescorla (1967a) and Rescorla and LoLordo (1965) found that termination of an aversive CS+ during instrumental avoidance responding reduced the response rate for a few seconds. This, however, is not the only way of showing induction. One might explore the increase in jumping rate produced by a CS+ presentation that followed a CS— presentation, as compared with one that followed another CS+ presentation, in order to measure positive induction. This has not been done. Neither has negative induction been studied in such a way. In order to do this, one would compare responding to CS— when it has recently been preceded by a CS+ presentation, as compared with being preceded by a CS— presentation.

5. External Inhibition and Disinhibition. The effects of extraneous stimuli upon CSs imposed on ongoing instrumental behavior have been little studied. It would be of interest to see whether a novel stimulus could disrupt the effect which a CS+ has upon instrumental behavior (external inhibition). Likewise, we would like to know whether we can remove the inhibitory effect which a CS — has upon instrumental behavior by presentation of a novel stimulus (disinhibition). Preliminary experimentation with dogs (Rescorla, l967a) suggests that disinhibition of inhibition of temporal delay can be produced with fear conditioning; however, no evidence was found for external inhibition.

We have covered many of the experiments showing how Pavlovian procedures, interacting with Thorndikian (or Skinnerian) procedures, can influence instrumental responding. In doing so, it became clear that many of the cells in Table 1 are empty. Some of the empty cells are as interesting as the filled ones. For example, take Cell 9. This cell would require the testing of the effect of an appetitive CS+ on unsignaled avoidance responding. What would happen? Would the “irrelevancy” of the fear that mediates avoidance to the appetitive CRs elicited by CS+ mean that avoidance rate would be unaffected by CS+? Or is there a matrix of interrelationships among emotional-motivational states that requires that appetitive CRs interfere with or inhibit all aversive states to some extent? Perhaps, instead, the Spence (1956) view of “generalized D” would be correct, and any excitatory CS+ would enhance any response mediated by any excitatory state in the presence of Sd. CER experimentation(see Cell 3) would argue against this expectation, but perhaps it is too early to be certain. Certainly, much work is needed in this area of ignorance.

Another interesting cell is No. 10. Suppose a dog has acquired a stable Sidman avoidance response in the shuttlebox. He is then given test presentations with a CS— previously established in an appetitive Pavlovian conditioning procedure. What might be expected to happen? Would CS — be “irrelevant” for fear level, and therefore leave jumping rate unaffected? Or would CS—, being a signal for the nonoccurrence of an appetitive US, be “disturbing” in some way, thus “adding to” fear level and increasing the avoidance response rate? Is there a dimension of “pessimism” established in the emotional life of laboratory animals, such that a signal that says “food won’t come,” although it may superficially seem irrelevant for fear motivated avoidance responding, nevertheless is a “bad” event, just as shock is also a “bad” event? Amsel (1965) and Brown and Wagner (1964) have recently shown the generality of a “perseverance” attribute for laboratory Ss given special types of partial reinforcement experiences. Perhaps “pessimism” can be similarly established by appropriate Pavlovian and Thorndikian treatments, such that all CS—’s for appetitive differential conditioning, and all CS+’s for aversive differential conditioning, can enhance all instrumental responses reinforced by the avoidance of aversive USs of any type. It certainly would be valuable to know how these interactions work. The techniques for getting this information have already been worked out.4

There are other cells in Table 1 that command attention, but the reader can now generate his own experiments to fill them. We should note, however, that Table 1 does not exhaust the possibilities for analysis of interactions between Pavlovian and Thorndikian processes. The Pavlovian conditioning procedures can either precede, or be preceded by, the instrumental training procedures (thus focusing attention on order effects in the interaction of the two processes). Or the Pavlovian conditioning procedures can be carried out either within the situation used for instrumental training or in another situation as distinctively different as possible from the instrumental training situation. This variation focuses attention upon situational stimuli as a factorin the interaction of the two processes. Intrasituational Pavlovian conditioning can, in turn, either be carried out “on the base line,” that is, while the instrumental behavior to be influenced is occurring; or, in contrast, it can be carried out “off the base line,” that is, when instrumental responding is impossible.

Up to this point we have talked only of variations in the Pavlovian conditioning parameters. Another strategy is to vary the instrumental training parameters while holding the Pavlovian conditioning constant. For example, would Pavlovian conditioned stimuli based on a shock US have the same effect on an S during asymptotic, reinforced, operant performance as it would have during the first moments of extinction of that operant? Such a question has implications for any theory that argues for the importance of emotion in the control of instrumental responding, since presumably different emotions are present in training and extinction. It seems clear that systematic variations of both the Pavlovian and instrumental operations in Table 1 would be valuable in extending our understanding of the emotional control of instrumental behavior.

The many possible variations in procedure will complicate Table 1 and expand it to almost unmanageable proportions. Nevertheless, such variations must be kept in mind, because we already know that they are important. For example, the effects of the order of Pavlovian conditions and instrumental training are subtle and interesting. Overmier and Leaf (1965), working in our laboratory, found that the discriminative control of avoidance responding by a Pavlovian CS+ and CS— was poorer when the conditioning preceded avoidance training than when the reverse order was used. However, there is nothing in two-process theory that predicts an order effect. This result indicates that two-process theory is rapidly generating new data requiring further refinement and extension of such theory while not requiring the abandonment of the approach.

We can conclude that, following one strategy suggested by two-process theory, Pavlovian conditioning procedures can readily be used to control instrumental responding. Furthermore, it might very well turn out that instrumental responding is as sensitive, or perhaps even more sensitive, a measure of the effects of Pavlovian conditioning procedures than are the traditionally measured conditioned visceral or motor reflexes themselves. If this should turn out to be true, it would constitute a major heuristic, albeit somewhat ironic, contribution of two-process learning theory.

Finally, we point to the success achieved in controlling instrumental responding by means of a wide variety of Pavlovian procedures, contrasted with the failure to establish definitive relationships between CRs (as mediators) and concurrent instrumental responses. Such success gives support to the version of two- process theory postulating that the concomitance we do observe between CRs and instrumental responding is mediated by a common central state, and the changes in that state are subject to the laws of Pavlovian conditioning.

NOTES

Anxiety (Drive) Level and

Performance in Eyelid Conditioning1

KENNETH W. SPENCE, University of Iowa

Studies from the Iowa laboratory and elsewhere that have involved a comparison of the eyelid conditioning performance of Ss scoring at the extremes of the Taylor Manifest Anxiety (MA) scale are reviewed. In 21 of 25 independent comparisons, differences between groups were in favor of the high anxiety (HA) Ss, with the majority being statistically significant. Although these data provide substantial confirmation of the implication of the drive interpretation of MA scale that HA Ss should exhibit a higher level of performance than LA Ss, an attempt was made to ascertain what factors might be responsible for failure of the difference to occur in some studies. The major factors appeared to be small numbers of Ss and the presence of “voluntary form” responders in the samples. Significant differences appear to be related to the degree of experimental naiveté of the Ss and the extent to which the experimental situation is designed to arouse some degree of apprehensiveness.

Recently King, Kimble, Gorman, and King (1961) reported failure to find a significant relation between level of emotionality as measured by the Taylor Manifest Anxiety (MA) scale and performance in aversive (eyelid) conditioning. In the light of the previous evidence, this finding was rather surprising and the authors were able to conclude only that emotionality or anxiety must be an interacting variable that is related to conditioning under certain as yet unknown conditions and not under others. The purpose of the present article is (a) to review the data bearing on the relation between the MA scale and aversive conditioning, only a small fraction of which was cited by King et al., and (b) to examine this evidence in detail in an attempt to suggest what the factors are that might be responsible for the failure of the relationship to appear in some studies.

In presenting these data, the results of studies comparing performance in simple conditioning of Ss preselected on the basis of extreme MA scale scores will first be summarized. The second group of studies to be reviewed also involves comparison of the performance of preselected Ss but in differential conditioning. Thirdly, the results of a number of previously unpublished comparisons based on simple conditioning data from studies conducted in the Iowa laboratory for other purposes will be presented. The Ss serving in these latter studies were unselected with respect to the MA scale. Since their scores on the test were available, it was possible to determine the differences in the conditioning performance between Ss who scored at the extremes of the scale.

STUDIES INVOLVING PRESELECTED

EXTREME GROUPS

Simple Conditioning.

The findings of the previously published Iowa studies and one unpublished study directly concerned with comparing the conditioning performance of preselected high and low anxiety (HA and LA) Ss are summarized in Table 1. The criteria of se- lection of Ss in these studies varied slightly, ranging from raw scores of 20 to 24 as the lower limit of the HA Ss and from 7 to 9 as the upper limit of the LA Ss. These scores roughly mark the upper and lower twentieth to twenty-fifth percentiles of the distribution of MA scale scores made by students in the introductory course in psychology at Iowa. Shown in the successive columns of the table are: (a) the reference studies, (b) the number of trials over which the conditioning measures were obtained, (c) the use or nonuse of a ready signal, (d) the strength of the UCS (air puff) employed, (e) the total number of Ss in the two groups, (f) the mean percentage of CRs made by the HA group of Ss, (g) the mean percentage of responses made by LA Ss, (h) the difference in percentage of CRs given by the two groups, and (i) the significance level of the differences between the groups expressed in terms of p (two-tailed test).

As may be seen, the six Iowa studies provided nine independent comparisons of HA and LA groups at UCS intensities ranging from very weak (.25 psi) to moderately strong ones (2.0 psi). In every study HA Ss responded with a higher percentage of CRs over the conditioning period than did the LA Ss and in each the difference was significant at the .05 level or better. Since the conditioning curves of the HA and LA groups tended to diverge, it is apparent that the differences in their performance at the end of training were even larger than those shown in Table 1. Thus in the case of these six studies the differences were approximately 20 percent larger over the last 20 conditioning trials than for all trials. These larger differences also tended to be slightly more significant. Attention should also be directed to the fact that the numbers of Ss on which the significance values obtained in these studies were based are relatively large, ranging from a minimum of 45 in studies with two groups to a maximum of 120 when four groups were used.

In turning to the non-Iowa studies in Table 2, it will be seen that not a single one of the comparisons involved as many as 50 Ss. The study (No. 3) with the largest number of Ss (36), it is worthy of note, did give results in line with those obtained in the Iowa laboratory. In this study, conducted at Kent State University, HA Ss responded at a significantly higher level (.01) than LA Ss. The remaining comparisons in this portion of the table, however, failed to support such a finding, that is, there were no significant differences between groups. The first two of the studies probably should not be taken too seriously as they involved only 10 Ss in each extreme group. The variability of conditioning performance among individual Ss is so great that such results would not be unexpected with so few Ss. However, the final three comparisons, all of which were reported by King, Kimble, and their students (1961) present a more formidable array of evidence that is quite contrary to the findings of the Iowa laboratory. Not only did they fail to obtain a significant difference in favor of the HA Ss, but two of the differences, though not significant, were actually in favor of the LA Ss. While only 30—32 Ss were involved in each experiment, the results as a whole are clearly opposed to the Iowa findings.

In addition to the experiments contained in Tables 1 and 2, the findings of three further published studies that bear on the relation between the MA scale and performance in simple conditioning should be mentioned. In each of these investigations unselected Ss were run in the experiment and comparisons were subsequently made between Ss who scored in the upper and lower halves of the distribution of scores on the MA scale: Two of these studies were primarily interested in the effects of administrating shocks on the level of conditioning performance. In the first, Spence, Farber, and Taylor (1954) found that HA Ss (N = 15) responded at a significantly higher level in Trials 1—40 than LA Ss (N = 25) under the shock condition (p < .01). Under no shock HA Ss also gave more CRs than LA Ss, but with only 10 Ss per group the difference was not significant. In a partial replication of this experiment carried out at Peabody College, Caldwell and Cromwell (1959) compared Ss, 30 from the upper and 30 from the lower halves of the distribution of MA scale scores and obtained a significant relation (.05 level) with conditioning performance over Trials 1—40. In each of these experiments, however, conditioning performance over Trials 41—80 was not related either to shock-no shock or to anxiety level. The third and final study that compared above and below average Ss on the MA scale was conducted with students from the evening classes of Northwestern University (Spence & Taylor, 1953). Over the 80 conditioning trials, 22 HA Ss averaged 32.2 percent CRs, 21 LA Ss, 18.2 percent. The difference was significant at the .04 level.

Differential Conditioning.

Table 3 presents the data from two reported investigations of the relation between the MA scale and level of performance to the positive CS in differential conditioning. As in the studies of Table 1, HA and LA groups represent roughly the upper and lower quartiles on the MA scale. Spence and Farber (1953, 1954) conducted two separate experiments differing slightly in conditioning methodology. The data presented in the table show the level of response to the positive CS over the last 20 reinforced trials (31—50). As may be seen, in the first experiment, the HA group gave significantly more CRs than the LA group. The second experiment of Spence and Farber, as originally reported, included Ss on the basis of a forced-choice version of the scale or on the basis of the MA scale. The data presented here are based only on Ss who were identified as high or low on the MA scale. Again, as may be seen from the table, HA Ss responded at a higher level than LA Ss. The difference was not significant at the .05 level by a two-tailed test, but was on the basis of a one- tailed test.

Using a factorial design, Prokasy and Whaley (1962) investigated whether the relation between performance on the MA scale and differential conditioning was a function of the presence or absence of a ready signal in the experimental procedure. Table 3 presents the percentage of CRs given to the positive CS on the last 20 reinforced trials (26—45). As may be seen, when a ready signal was employed HA Ss gave a significantly larger number of CRs than did LA Ss. In the absence of a signal, however, the superiority of the HA group was small and not significant.

UNPUBLISHED DATA INVOLVING

UNSELECTED SUBJECTS

The third set of studies providing relevant evidence is presented in Table 4. Included in this group are the findings of two previously published studies (No. 1 & 2) that were concerned with investigating among unselected Ss the relation between conditioning measures and physiological indices of emotionality (e.g., heart rate changes, GSR, and muscle action potential). As the scores on the MA scale were available it was possible to examine the conditioning performance of Ss who scored in the upper and lower quartiles of the distribution of scores on the scale. As the table shows, HA Ss gave a significantly greater number of CRs than LA Ss in both of these studies.

The remaining data presented in Table 4 come from three as yet unpublished studies from the Iowa laboratory. Each is identified by a word or label describing the primary interest of the study. In each of these investigations a fairly large number of Ss, the MA scale scores of whom were available after completion of the experiment, were conditioned under identical circumstances. Again, the findings reported are for Ss who scored in the upper and lower quartiles of the MA scale distribution. As may be seen, the three comparisons (No. 4, 5a, & Sb) provided by the last two experiments once again show that HA Ss responded at a higher level than LA Ss. Two of these differences were significant, while the third (that involving the smallest number of Ss) was not. Attention should be directed to the fact that a ready signal was not employed in these latter studies. The relation of the use or nonuse of a ready signal to the findings of these studies will be discussed in a later section.

The final item of note in Table 4 is the negative difference found in the study entitled “Individual Differences” (No. 3). This is the only instance among a total of 17 comparisons of HA and LA Ss provided by our Iowa studies in which LA Ss gave a larger number of CRs than HA Ss. The small difference is not, of course, significant.

DISCUSSION

Looking at the findings of the studies included in Tables 1, 2, 3, and 4 as a whole, a number of characteristics may be noted. First, the proportion of instances in which the conditioning performance of the HA group was higher than that of the LA group is much greater than one would expect by chance. Thus, the results of 21 of 25 comparisons were in this direction. If there actually were no relation between the MA scale and conditioning performance, the probability of obtaining such a percentage of differences (84 percent) in the same direction by chance is less than .01.

Secondly, it is clearly evident that the studies with relatively large numbers of Ss tended to provide significant differences in favor of HA Ss, whereas those with smaller numbers did not. Thus, in the case of comparisons involving 36 or more Ss, 65 percent (11/17) of the comparisons were significant at the .05 level or better on the basis of a two-tailed test. On the other hand, in studies involving fewer than 36 Ss only two of the eight comparisons provided a significant result. Moreover, all of the 13 significant differences were in the direction of higher conditioning performance on the par) of the HA Ss, whereas none of the four obtained differences in favor of the LA Ss were significant.

As was mentioned earlier, intersubject variability in eyelid conditioning is exceedingly great. In our experiments the percentages of CRs given by individuals customarily range all the way from zero to very high values. The standard deviations of these percentage measures vary roughly between 20 and 25 percent. Under such circumstances it is readily apparent that fairly large numbers of Ss must be sampled if an adequate test as to whether the manipulated variable produces a difference is to be made. This is particularly the case if the size of the difference is not large relative to the variance of the measures. Apparently, degree of emotionality, as specified in terms of extreme scores on the MA scale, is a relatively minor factor or variable among all those contributing to the intersubject variance of conditioning performance. In this connection, it is interesting to note that in two studies in which emotionality differences and differences in UCS (puff) strength were both variables, the differences between the conditioning performances of the HA and LA Ss were of about the same order of magnitude as those obtained between relatively weak and strong puff intensities (.25 vs. 1.5 psi and .6 vs. 2.0 psi). In the case of both types of comparison, individual performances ranged from very low to high levels in the groups being compared and the overlap was considerable.

In the light of these considerations, it is not at all surprising to find such divergent findings among studies that employ such small numbers of Ss, nor is it difficult to understand why the Iowa studies, with their relatively larger samples, have tended to be more consistent and to give a higher proportion of significant differences in favor of the HA Ss.

A third characteristic of the data provided by these studies is that the relation between the MA scale and conditioning does appear to vary from one laboratory to another. This fact is clearly revealed in Table 5, which presents an analysis of the data in terms of their source. As may be seen, we have divided the studies into three groups, those carried out in the Iowa laboratory and those from other laboratories. The latter, in turn, have been broken down into two subgroups, primarily on the basis of the strikingly different findings obtained in them.

It is clearly evident from the data of Table 5 that the Iowa studies have most consistently demonstrated a relation between conditioning performance and emotionality (MA scale). Thus 16 (94 percent) of the 17 comparisons gave differences in favor of the HA Ss, with 10 (60 percent) of these differences being significant at the .05 level or better. Contrasting most sharply with these findings are those obtained in the subgroup of non-Iowa studies from Duke and North Carolina. These studies, carried out or supervised by Kimble and his associates, failed to find a significant difference in three comparisons of the HA and LA Ss and, moreover, found that LA Ss responded at a higher level than HA Ss in two of their three experiments. Falling in between, but closer to the results of the Iowa studies, are the findings from the four other non-Iowa laboratories. Thus, in four of five comparisons, HA Ss gave more CRs than LA groups, with two of these being significant. Recalling that two of the studies in this latter group had only 10 Ss per group, we see that two of the three experiments that had groups of 36 or more obtained significant differences.

In summary, we see that when studies with a reasonably large number of Ss (36 or more) are considered, the experiments conducted in the Iowa laboratory and those from laboratories other than Duke and North Carolina tend to be in close agreement with each other in pointing to a genuine difference in the conditioning performance of the HA and LA Ss. The studies of Kimble and his associates, on the other hand, do not.

In an effort to ascertain what factors might be responsible for these contrasting findings a comparison was made of the experimental conditions under which the studies were conducted. For this purpose a more detailed description of the experimental situation, procedures, and experiences of the Ss was obtained through correspondence with the investigators involved.

With respect to the conditions under which the experiments were conducted at Iowa, we were greatly influenced by the writer’s theoretical interpretation as to why HA Ss could be expected to perform at a higher level than LA Ss. According to this theory, the differences in conditioning performance of these two groups of Ss reflect differences in their level of generalized drive (D), which, in turn, are assumed to be the result of differences in their level of emotional reactivity to the experimental situation and procedures. Accordingly, a deliberate attempt was made in the Iowa studies to provide conditions in the laboratory that might elicit some degree of emotionality. Thus, the experimenter was instructed to be impersonal and quite formal in greeting S and in giving the necessary instructions. On coming into the experiment, S at first saw an impressive array of electronic recording equipment and was then led into an adjoining room in which was located an isolated, screened cubicle. The latter contained a dental chair (sic) in which S was seated in a reclining position, while a headband was placed on his head and a plastic piece was fastened to his upper eyelid. After completing the instructions, the illumination in the cubicle was reduced to a low level of semidarkness and S was informed that, if the need arose, he could get in touch with E by means of a microphone placed on a stand within his reach. The door to the cubicle and the door leading to the adjoining room in which E worked were then closed and S was left in isolation.

To say the least, these conditions were unusual and strange for Ss. Furthermore, in order to maximize the likelihood that they would have a tendency to arouse some degree of apprehensiveness, only individuals who had no previous experience as an S in psychological laboratory experiments were used in all but one of our studies in Tables 1, 2, and 3. It has been our experience that students are much more likely to be concerned and apprehensive in the first experiment in which they serve. After one or two experiences they become much less fearful and, all too often, quite bored and blasé. Some experimental evidence, that the amount of such previous experience is a factor in experiments comparing the performances of the HA and LA Ss, has been provided by an experiment reported by Mednick (1957). The latter found that experimentally naive HA Ss showed more stimulus generalization of a response than LA Ss, whereas no such difference was obtained in the case of Ss who had served in from two to three previous psychological experiments.

It is evident from this account that the conditions under which our experiments were conducted were designed to arrange for some degree of situation-aroused anxiety or emotionality. While one version of our theory hypothesized that differences in emotionality of the HA and LA Ss might be chronic, an alternative possibility was that it was dependent upon some degree of stress being present in the situation. We were primarily interested in testing the theory under conditions that maximized the likelihood of differences in the emotional reactions of the two groups of Ss.

In the light of the above discussion, it is interesting to compare the conditions and procedures employed by Kimble and King with those employed at Iowa. First, with respect to physical conditions, there are a number of differences. The two Duke experiments were conducted in a cubicle within a room in which E was located. However, instead of being semidark, the cubicle was rather brightly lighted (23 apparent foot-candles). The experiment at the University of North Carolina was also conducted under bright illumination with S and E being in the same room on the opposite sides of a partition. In place of a dental chair, Ss in the Duke experiments were seated on a secretarial-type chair, with chin in a headrest. At North Carolina, S sat in a straight desk chair with head free. In addition to these physical differences, most Ss in the Kimble-King experiments had already been in one or two experiments prior to serving in the conditioning experiment.

Considering the differences in the two laboratories in terms of the use of different degrees of isolation of S from E, semidarkness versus well-lighted room, dental chair (emotionally conditioned cue) versus neutral-type chair, and experimentally naive versus sophisticated Ss, it would seem quite plausible to suspect that the differences in the findings could, in part at least, be due to the fact that the experimental conditions at Duke and North Carolina were not as emotion arousing as those employed in the Iowa experiments. Further experimentation is needed, of course, to check on the role of such factors for there are other differences between the two sets of experiments that might have played important roles. One important one that should be investigated is the possibility that cultural backgrounds of southern and northern students may lead to a difference in the manner in which they respond to the different items in the MA scale.

Still another variable is the E-S interaction. Our Es were not acquainted with Ss and we have attempted to instill in them an attitude of being quite impersonal at the beginning of the experiment and to avoid making any particular attempt to put S at ease or allay any expressed fears. Undoubtedly Es differ greatly in the degree to which they are able to achieve and maintain this objective. It is interesting to note that the person who served as one of two Es in the single Iowa study that gave negative results (LA> HA) has recently completed a doctoral dissertation in which there again was little difference between HA and LA Ss. Unfortunately, there were only a small number of the HA and LA Ss under each of the different conditions of this experiment and the proportion of male and female Ss was not equated so the results are not very helpful. This is, nevertheless, a potentially important variable and should be investigated further, possibly by deliberately manipulating the behavior of F. It has recently been shown experimentally that one can markedly increase the conditioning performance of Ss by emotion producing instructions (Spence & Goldstein, 1961).

A final variable to be considered is the presence or absence of a ready signal. In a recent article Prokasy and Whaley (1962) have proposed that the use or nonuse of a “ready-blink” signal is a possible factor with which emotionality might be interacting in ‘these conditioning studies. In support of this notion these investigators found, as is shown in Table 2, that in differential conditioning, their HA Ss gave a significantly larger number of CRs to the positive CS than LA Ss when a ready-blink signal was used, whereas only a small, insignificant difference was obtained in its absence. However, before this study even appeared there was evidence against the notion that the difference in conditioning performance of the HA and LA Ss was a function of the presence or absence of a ready signal. As Table 1 reveals, Baron and Conner (1960) had reported a highly significant difference (p < .01), while Spence and Weyant (1960) also obtained a significant F (<.05 level) for HA and LA Ss in experiments that did not employ a ready signal. Further evidence against this view that a ready signal is a necessary condition for the difference to occur is provided by the findings of recent experiments from our laboratory presented in Table 4. As the last three items in this table show, three comparisons between HA and LA Ss in which there was no ready signal all gave differences in favor of the HA group with the two comparisons that involved more than 35 Ss being significant at the .05 level.

Unfortunately, a serious methodological error is present in the early studies that did not employ a ready signal. As Hartman and Ross (1961) have recently demonstrated the latency criterion employed in the Iowa laboratory (Spence & Ross, 1959; Spence & Taylor, 1951) to identify and eliminate “voluntary” responders from the data sample is not applicable when a ready signal is not employed. The use of this criterion was originally based on the finding in experiments using a ready signal and a CS-UCS interval of 500 milliseconds that the latency distribution of responses judged to be similar in form to voluntary eyelid closures did not overlap to any great extent with that for CRs, which are very different in form. Thus, it was possible to employ this more convenient latency property as the index of whether a response was a voluntary response or a CR. As employed in these studies all responses with latencies between 300 and 500 milliseconds were counted as CRs, while responses whose latencies fell between 150 milliseconds (tone CS) or 200 milliseconds (light CS) and 300 milliseconds were counted as voluntary responses. The Ss who gave 50 percent or more responses in the voluntary category were eliminated from the sample data.

The reason for so doing was that the behavior of such Ss appeared to obey quite different laws from those holding for Ss who gave few such voluntary responses. Among the differences importantly related to our present concern are (a) the tendency to respond at an extremely high level (approaching 100 percent) once the first such anticipatory voluntary response is made and (b) the lack of any relation whatever between the frequency of responses given by such Ss and the level of the MA scale score. The effect of both of these tendencies is clearly that of reducing the likelihood of obtaining a significant difference in the conditioning performance of the HA and LA Ss, the first by greatly increasing the variance of the conditioning scores and the second by reducing the difference between groups that contained a number of such Ss.

Unfortunately, investigators who did not use a ready signal took over the latency criterion to identify voluntary responders without checking its appropriateness, with the consequence that all such Ss probably were not eliminated from their samples. The Prokasy and Truax (1959) study is especially suspect in this regard as it involved a very strong puff (3.0 psi), a condition that produces a high incidence of such voluntary responders. When one considers the small number of Ss (10 per group) in this experiment, the negative difference obtained is not too surprising. The presence of one or two more high responding voluntary Ss in the LA group than in the HA group could easily have produced this result. The findings obtained with the no-signal groups in the later study of Prokasy and Whaley (1962) and one of the Duke studies also may be suspected of being affected by the presence of a number of voluntary responders who were not detected by the inappropriate latency criterion that was employed.

An example of the effects of including the data of voluntary responders in a sample is provided by the recent studies (Table 4) from the Iowa laboratory that did not employ a ready signal. Having confirmed the findings of Hartman and Ross (1961) that latency does not differentiate voluntary form responses from CRs when a ready signal is absent, the identification of voluntary Ss in these studies was made in terms of the form of the eyelid responses. If, instead, the latency criterion had been employed none of the voluntary responders detected by ‘the form criteria would have been identified and thus would have been included in the samples. It is worthy of note that the inclusion of six voluntary responders (three in each group) in the case of the comparison involving the male Ss of our most recent experiment (Table 4, No. 5a) would have reduced the reported difference of 14.2 percent between HA and LA Ss to 11.0 percent. Correspondingly, the significance of the difference would have been reduced from the .05 level to one of .16. In other words, the addition of only six voluntary Ss to the sample of 67 would have changed the conclusion from one that the difference is significant to one that it is not.

In concluding this discussion, attention should especially be directed to the point that the behavior of human Ss in classical eyelid conditioning, while simple in form, is complexly determined. Recent research in this and other laboratories has revealed that higher mental processes (inhibitory and facilitatory sets, etc.) play a much more prominent role in human conditioning than has sometimes been realized. Moreover, our lack of knowledge concerning these factors has precluded our controlling them in any satisfactory manner with the result that the intersubject variance of conditioning performance is extremely large. This state of affairs has unfortunate consequences for the testing of simple theories of behavior which specify only associational and motivational constructs (e.g., H and D) as determiners of excitatory strength of the conditioned responses. Obviously, the additional uncontrolled variables minimize the role of these basic theoretical factors, tending to hide their effects. Under such circumstances it is necessary to have reasonably large samples so that the effects of these confounding variables are more likely to be equalized in the comparison groups. On the basis of the writer’s experience with conditioning data, experiments of the type discussed here should be required to have groups of at least 25 Ss. Unhappily, as we have seen, this has all too frequently not been the case.

NOTE

1. This study was carried out as part of a project concerned with the influence of motivation on performance in learning under Contract Nonr1509 (04), Project NR 154—107 between the University of Iowa and the Office of Naval Research.

Paired-Associate Learning as a Function of Arousal

and Interpolated Interval.1

LEWIS J. KLEINSMITH AND STEPHEN KAPLAN,

University of Michigan

This experiment tested the hypothesis that due to the phenomenon of perseverative consolidation, a pattern perceived under high arousal should show stronger permanent memory and weaker immediate memory than a pattern accompanied by low arousal. While recording skin resistance as a measure of arousal, 48 Ss were presented 8 paired associates for learning. The Ss were tested at various time intervals: 2 min., 20 min., 45 min., 1 day, and 1 wk. The results confirmed the hypothesis (p = .001) - Paired associates learned under low arousal exhibited high immediate recall value and rapid forgetting. High arousal paired associates exhibited a marked reminiscence effect, that is, low immediate recall and high permanent memory.

There is growing evidence that perseverative consolidation of the memory trace can last over a considerable period of time. Glickman (1961) provides an extensive review of the literature in this area. Still more recently, Pare (1961), by the use of drugs, has shown that arousal plays an important role in the consolidation process. A stimulant administered immediately following a learning trial increases retention, while a depressant has the opposite effect.

As Pare has indicated, reverberating neural circuits (Hebb, 1958) provide a mechanism to explain the effect of arousal on consolidation. Under conditions of low arousal, relatively little nonspecific neural activity will be available to support the reverberating trace, resulting in little consolidation and poor long- term retention. On the other hand, under conditions of high arousal the increased nonspecific neural activity will result in more reverberation, and thus retention should be better.

While reverberation is taking place, however, one might expect the trace to be relatively unavailable to the organism, resulting in poor recall of the consolidating material during this interval. Such unavailability follows from a consideration of the difficulties in refiring a neuron which is already firing repeatedly in a reverberating circuit. Hodgkin (1948) has shown that neurons are sharply limited in their maximum rate of firing. Thus under high arousal, perseverative consolidation should be effective both in strengthening the memory trace and in making the memory difficult to evoke until the perseveration terminates.

In the present study Ss were asked to learn a set of paired associates composed of words as stimuli and single-digit numbers as responses. The words were chosen to differ in their arousal value (as measured by galvanic skin response). In terms of the hypothesis, word-number pairs of high arousal value should be recalled poorly at first, but should be recalled well at a later time. Low arousal pairs, by contrast, should be remembered better at first and should show a gradual decay (forgetting) with time.

METHOD

Subjects.

The Ss were 48 University of Michigan undergraduates obtained from introductory courses in psychology. They were run in six subgroups of 8 Ss each.

Procedure.

The Ss were given a single learning trial with a list of eight word-number pairs. Eight words expected to produce different arousal levels were chosen as stimulus words. These were KISS, RAPE, VOMIT, EXAM, DANCE, MONEY, LOVE, SWIM. The response items were single digits from 2 to 9.

A slide projector with an exposure time of 4 sec. was used to present the stimuli. During the training trial S first saw the stimulus word alone, and then saw the word repeated with a single digit response. To separate the arousal effects from one set of pairings to the next, two slides containing four colors each were inserted between the paired associates, and S was instructed to name the colors. (Red, green, blue, black, yellow, and orange were used randomly on these slides.) The S was instructed to “concentrate carefully on both colors and word-number pairs” as he called them out loud, but to avoid rehearsal S was not specifically told that he would be tested for recall.

In order to determine the arousal effects of each stimulus word, skin resistance was recorded during learning. The electrodes were of the zinc variety described by Lykken (1959). To insure constancy of conditions, electrodes and recording apparatus were also used during the recall session.

During the recall session S was instructed to indicate the correct number for each stimulus word as it appeared. The correct numbers were not repeated. Colors were used as an interpolated task as before.

Design.

The six groups were tested one at each of the following recall intervals: immediate (about 2 min.), 20 min., 45 min., 1 day, and 1 wk. Two groups were run at the first recall interval.

To correct for serial order effects eight different training lists were used for each group, designed so that each of the eight words appeared in each position in the list once (Fisher & Yates, 1938). The order of the recall lists was varied in the same manner.

Data analysis.

Any drop in S’s skin resistance which occurred within 4 sec. of presentation of a given word was considered an arousal deflection; these were converted to percent deflections. Each S’s eight GSR deflections were then ranked. (In case of ties, absolute skin resistance level was used to break them. A deflection occurring at a low level of absolute skin resistance was considered higher arousal than a similar deflection occurring at a higher absolute level.) The three highest deflections for each S were designated “high arousal learning,” and the three lowest were designated “low arousal learning.”

RESULTS

Figure 1 illustrates the relationship between high and low arousal recall as a function of time. At immediate recall, numbers associated with low arousal words are recalled five times as often as numbers associated with high arousal words. The capacity to recall numbers associated with low arousal words decreases as a function of time in a normal forgetting curve pattern. On the other hand, the capacity to recall numbers associated with high arousal words shows a considerable reminiscence effect. After 20 min., the increase is more than 100 percent, and after 45 min. it has increased 400 percent. This high capacity for recall of high arousal pairs persists for at least a week—the longest interval employed.

Analysis of variance (Lindquist, 1953) for correct responses confirms the trends in the figure at p < .001 (Table 1). The significant interaction is primarily attributable to the effect of the immediate condition, F(1, 15) = 11.36, p < .005, and the 1-wk. condition, F(1,7) = 11.67, p < .025. Thus at immediate recall, low arousal learning is significantly greater than high arousal learning; after 1 wk., the situation has reversed and high arousal recall is significantly better than low arousal recall.

As can be seen in Table 2 the distribution of items at different times sampled forms no systematic patterns that might account for the results obtained. This table shows the proportion of each of the eight items at each time sampled for both high and low arousal learning.

DISCUSSION

The increase in the capacity to recall items learned under conditions of high arousal in contrast to items learned under low arousal conditions provides further support for the theory of reverberating neural circuits. A somewhat simplified physiological explanation of the processes involved may be pictured as follows. When a person perceives a pattern, a closed, reverberating neural circuit is set up in his brain corresponding to this pattern. The more nonspecific neural activity or arousal present, the greater the number of times which the trace is likely to reverberate. And the greater this perseverative consolidation of the neural trace, the stronger the permanent memory.

However the apparent paradox is that while perseveration is taking place, recall ability is poor. This follows from the hypothesis that at any given instant the neurons involved in the perseverating trace are either already in the process of firing, or are in an absolute refractory state, or may be in a state of slowly developing subthreshold activity (Hodgkin, 1948); thus the trace would be relatively unavailable to the organism. With greater arousal there will be increased perseveration and thus poorer immediate performance.

This explanation coincides with the empirical findings. High arousal learning showed poor immediate recall due to the hypothesized period of unavailability of the trace, while recall was better under low arousal, for this immediate period. The long-term effects are, as predicted, in the opposite direction. There was significantly poorer long-term retention under low arousal and superior long-term retention under high arousal. This is in keeping with the expectation that there would be more consolidation and thus better learning under high arousal.

Reminiscence has become so rare in recent years that Underwood (1953) has wondered if it might not be a prewar phenomenon. This study suggests that methodological considerations might be an important factor in the return of reminiscence. The innovation in this experiment consisted of measuring arousal during learning and thus independently determining in advance which items should show reminiscence effects and which should not. When recall as a function of time is plotted for all items at once, little reminiscence is evident.

NOTE

1. This investigation was supported in part by a research grant (M4239) from the National Institutes of Health, Public Health Service. The Ss in this study were run by Robert D. Tarte, research assistant on this project.

The Misbehavior of Organisms

KELLER BRELAND and MARIAN BRELAND,

Animal Behavior Enterprises, Hot Spring, Arkansas

There seems to be a continuing realization by psychologists that perhaps the white rat cannot reveal everything there is to know about behavior. Among the voices raised on this topic, Beach (1950) has emphasized the necessity of widening the range of species subjected to experimental techniques and conditions. However, psychologists as a whole do not seem to be heeding these admonitions, as Whalen (1961) has pointed out.

Perhaps this reluctance is due in part to some dark precognition of what they might find in such investigations, for the ethologists Lorenz (1950, p. 233) and Tinbergen (1951, p. 6) have warned that if psychologists are to understand and predict the behavior of organisms, it is essential that they become thoroughly familiar with the instinctive behavior patterns of each new species they essay to study. Of course, the Watsonian or neobehavioristically oriented experimenter is apt to consider “instinct” an ugly word. He tends to class it with Hebb’s (1960) other “seditious notions” which were discarded in the behavioristic revolution, and he may have some premonition that he will encounter this bete noir in extending the range of species and situations studied.

We can assure him that his apprehensions are well grounded. In our attempt to extend a behavioristically oriented approach to the engineering control of animal behavior by operant conditioning techniques, we have fought a running battle with the seditious notion of instinct.1 It might be of some interest to the psychologist to know how the battle is going and to learn something about the nature of the adversary he is likely to meet if and when he tackles new species in new learning situations.

Our first report (Breland & Breland, 1951) in the American Psychologist, concerning our experiences in controlling animal behavior, was wholly affirmative and optimistic, saying in essence that the principles derived from the laboratory could be applied to the extensive control of behavior under nonlaboratory conditions throughout a considerable segment of the phylogenetic scale.

When we began this work, it was our aim to see if the science would work beyond the laboratory, to determine if animal psychology could stand on its own feet as an engineering discipline. These aims have been realized. We have controlled a wide range of animal behavior and have made use of the great popular appeal of animals to make it an economically feasible project. Conditioned behavior has been exhibited at various municipal zoos and museums of natural history and has been used for department store displays, for fair and trade convention exhibits, for entertainment at tourist attractions, on television shows, and in the production of television commercials. Thirty-eight species, totaling over 6,000 individual animals, have been conditioned, and we have dared to tackle such unlikely subjects as reindeer, cockatoos, raccoons, porpoises, and whales.

Emboldened by this consistent reinforcement, we have ventured further and further from the security of the Skinner box. However, in this cavalier extrapolation, we have run afoul of a persistent pattern of discomforting failures. These failures, although disconcertingly frequent and seemingly diverse, fall into a very interesting pattern. They all represent breakdowns of conditioned operant behavior. From a great number of such experiences, we have selected, more or less at random, the following examples.

The first instance of our discomfiture might be entitled, What Makes Sammy Dance? In the exhibit in which this occurred, the casual observer sees a grown bantam chicken emerge from a retaining compartment when the door automatically opens. The chicken walks over about 3 feet, pulls a rubber loop on a small box which starts a repeated auditory stimulus pattern (a four-note tune). The chicken then steps up onto an 18-inch, slightly raised disc, thereby closing a timer switch, and scratches vigorously, round and round, over the disc for 15 seconds, at the rate of about two scratches per second until the automatic feeder fires in the retaining compartment. The chicken goes into the compartment to eat, thereby automatically shutting the door. The popular interpretation of this behavior pattern is that the chicken has turned on the “juke box” and “dances.”

The development of this behavioral exhibit was wholly unplanned. In the attempt to create quite another type of demonstration which required a chicken simply to stand on a platform for 12—15 seconds, we found that over 50 percent developed a very strong and pronounced scratch pattern, which tended to increase in persistence as the time interval was lengthened. (Another 25 percent or so developed other behaviors—pecking at spots, etc.) However, we were able to change our plans so as to make use of the scratch pattern, and the result was the “dancing chicken” exhibit described above.

In this exhibit the only real contingency for reinforcement is that the chicken must depress the platform for 15 seconds. In the course of a performing day (about 3 hours for each chicken) a chicken may turn out over 10,000 unnecessary, virtually identical responses. Operant behaviorists would probably have little hesitancy in labeling this an example of Skinnerian “superstition” (Skinner, 1948) or “mediating” behavior, and we list it first to whet their explanatory appetite.

However, a second instance involving a raccoon does not fit so neatly into this paradigm. The response concerned the manipulation of money by the raccoon (who has “hands” rather similar to those of the primates). The contingency for reinforcement was picking up the coins and depositing them in a 5-inch metal box.

Raccoons condition readily, have good appetites, and this one was quite tame and an eager subject. We anticipated no trouble. Conditioning him to pick up the first coin was simple. We started out by reinforcing him for picking up a single coin. Then the metal container was introduced, with the requirement that he drop the coin into the container. Here we ran into the first bit of difficulty: he seemed to have a great deal of trouble letting go of the coin. He would rub it up against the inside of the container, pull it back out, and clutch it firmly for several seconds. However, he would finally turn it loose and receive his food reinforcement. Then the final contingency: we put him on a ratio of 2, requiring that he pick up both coins and put them in the container.

Now the raccoon really had problems (and so did we). Not only could he not let go of the coins, but he spent seconds, even minutes, rubbing them together (in a most miserly fashion), and dipping them into the container. He carried on this behavior to such an extent that the practical application we had in mind— a display featuring a raccoon putting money in a piggy bank— simply was not feasible. The rubbing behavior became worse and worse as time went on, in spite of nonreinforcement.

For the third instance, we return to the gallinaceous birds. The observer sees a hopper full of oval plastic capsules which contain small toys, charms, and the like. When the SD (a light) is presented to the chicken, she pulls a rubber loop which releases one of these capsules onto a slide, about 16 inches long, inclined at about 30 degrees. The capsule rolls down the slide and comes to rest near the end. Here one or two sharp, straight pecks by the chicken will knock it forward off the slide and out to the observer, and the chicken is then reinforced by an automatic feeder. This is all very well—most chickens are able to master these contingencies in short order. The loop pulling presents no problems; she then has only to peck the capsule off the slide to get her reinforcement.

However, a good 20 percent of all chickens tried on this set of contingencies fail to make the grade. After they have pecked a few capsules off the slide, they begin to grab at the capsules and drag them backwards into the cage. Here they pound them up and down on the floor of the cage. Of course, this results in no reinforcement for the chicken, and yet some chickens will pull in over half of all the capsules presented to them.

Almost always this problem behavior does not appear until after the capsules begin to move down the slide. Conditioning is begun with stationary capsules placed by the experimenter. When the pecking behavior becomes strong enough, so that the chicken is knocking them off the slide and getting reinforced consistently, the loop pulling is conditioned to the light. The capsules then come rolling down the slide to the chicken. Here most chickens, who before did not have this tendency, will start grabbing and shaking.

The fourth incident also concerns a chicken. Here the observer sees a chicken in a cage about 4 feet long which is placed alongside a miniature baseball field. The reason for the cage is the interesting part. At one end of the cage is an automatic electric feed hopper. At the other is an opening through which the chicken can reach and pull a loop on a bat. If she pulls the loop hard enough the bat (solenoid operated) will swing, knocking a small baseball up the playing field. If it gets past the miniature toy players on the field and hits the back fence, the chicken is automatically reinforced with food at the other end of the cage. If it does not go far enough, or hits one of the players, she tries again. This results in behavior on an irregular ratio. When the feeder sounds, she then runs down the length of the cage and eats.

Our problems began when we tried to remove the cage for photography. Chickens that had been well conditioned in this behavior became wildly excited when the ball started to move. They would jump up on the playing field, chase the ball all over the field, even knock it off on the floor and chase it around, pecking it in every direction, although they had never had access to the ball before. This behavior was so persistent and so disruptive, in spite of the fact that it was never reinforced, that we had to reinstate the cage.

The last instance we shall relate in detail is one of the most annoying and baffling for a good behaviorist. Here a pig was conditioned to pick up large wooden coins and deposit them in a large “piggy bank.” The coins were placed several feet from the bank and the pig required to carry them to the bank and deposit them, usually four or five coins for one reinforcement. (Of course, we started out with one coin, near the bank.)

Pigs condition very rapidly, they have no trouble taking ratios, they have ravenous appetites (naturally), and in many ways are among the most tractable animals we have worked with. However, this particular problem behavior developed in pig after pig, usually after a period of weeks or months, getting worse every day. At first the pig would eagerly pick up one dollar, carry it to the bank, run back, get another, carry it rapidly and neatly, and so on, until the ratio was complete. Thereafter, over a period of weeks the behavior would become slower and slower. He might run over eagerly for each dollar, but on the way back, instead of carrying the dollar and depositing it simply and cleanly, he would repeatedly drop it, root it, drop it again, root it along the way, pick it up, toss it up in the air, drop it, root it some more, and so on.

We thought this behavior might simply be the dilly-dallying of an animal on a low drive. However, the behavior persisted and gained in strength in spite of a severely increased drive—he finally went through the ratios so slowly that he did not get enough to eat in the course of a day. Finally it would take the pig about 10 minutes to transport four coins a distance of about 6 feet. This problem behavior developed repeatedly in successive pigs.

There have also been other instances: hamsters that stopped working in a glass case after four or five reinforcements, porpoises and whales that swallow their manipulanda (balls and inner tubes), cats that will not leave the area of the feeder, rabbits that will not go to the feeder, the great difficulty in many species of conditioning vocalization with food reinforcement, problems in conditioning a kick in a cow, the failure to get appreciably increased effort out of the ungulates with increased drive, and so on. These we shall not dwell on in detail, nor shall we discuss how they might be overcome.

These egregious failures came as a rather considerable shock to us, for there was nothing in our background in behaviorism to prepare us for such gross inabilities to predict and control the behavior of animals with which we had been working for years.

The examples listed we feel represent a clear and utter failure of conditioning theory. They are far from what one would normally expect on the basis of the theory alone. Furthermore, they are definite, observable; the diagnosis of theory failure does not depend on subtle statistical interpretations or on semantic legerdemain—the animal simply does not do what he has been conditioned, to do.

It seems perfectly clear that, with the possible exception of the dancing chicken, which could conceivably, as we have said, be explained in terms of Skinner’s superstition paradigm, the other instances do not fit the behavioristic way of thinking. Here we have animals, after having been conditioned to a specific learned response, gradually drifting into behaviors that are entirely different from those which were conditioned. Moreover, it can easily be seen that these particular behaviors to which the animals drift are clear-cut examples of instinctive behaviors having to do with the natural food getting behaviors of the particular species.

The dancing chicken is exhibiting the gallinaceous birds’ scratch pattern that in nature often precedes ingestion. The chicken that hammers capsules is obviously exhibiting instinctive behavior having to do with breaking open of seed pods or the killing of insects, grubs, etc. The raccoon is demonstrating socalled “washing behavior.” The rubbing and washing response may result, for example, in the removal of the exoskeleton of a crayfish. The pig is rooting or shaking—behaviors which are strongly built into this species and are connected with the food getting repertoire.

These patterns to which the animals drift require greater physical output and therefore are a violation of the so-called “law of least effort.” And most damaging of all, they stretch out the time required for reinforcement when nothing in the experimental setup requires them to do so. They have only to do the little tidbit of behavior to which they were conditioned—for example, pick up the coin and put it in the container—to get reinforced immediately. Instead, they drag the process out for a matter of minutes when there is nothing in the contingency which forces them to do this. Moreover, increasing the drive merely intensifies this effect.

It seems obvious that these animals are trapped by strong instinctive behaviors, and clearly we have here a demonstration of the prepotency of such behavior patterns over those which have been conditioned. We have termed this phenomenon “instinctive drift.” The general principle seems to be that wherever an animal has strong instinctive behaviors in the area of the conditioned response, after continued running the organism will drift toward the instinctive behavior to the detriment of the conditioned behavior and even to the delay or preclusion of the reinforcement. In a very boiled-down, simplified form, it might be stated as “learned behavior drifts toward instinctive behavior.”

All this, of course, is not to disparage the use of conditioning techniques, but is intended as a demonstration that there are definite weaknesses in the philosophy underlying these techniques. The pointing out of such weaknesses should make possible a worthwhile revision in behavior theory.

The notion of instinct has now become one of our basic concepts in an effort to make sense of the welter of observations which confront us. When behaviorism tossed out instinct, it is our feeling that some of its power of prediction and control were lost with it. From the foregoing examples, it appears that although it was easy to banish the Instinctivists from the science during the Behavioristic Revolution, it was not possible to banish instinct so easily.

And if, as Hebb suggests, it is advisable to reconsider those things that behaviorism explicitly threw out, perhaps it might likewise be advisable to examine what they tacitly brought in— the hidden assumptions which led most disastrously to these breakdowns in the theory.

Three of the most important of these tacit assumptions seem to us to be: that the animal comes to the laboratory as a virtual tabula rasa, that species differences are insignificant, and that all responses are about equally conditionable to all stimuli.

It is obvious, we feel, from the foregoing account, that these assumptions are no longer tenable. After 14 years of continuous conditioning and observation of thousands of animals, it is our reluctant conclusion that the behavior of any species cannot be adequately understood, predicted, or controlled without knowledge of its instinctive patterns, evolutionary history, and ecological niche.

In spite of our early successes with the application of behavioristically oriented conditioning theory, we readily admit now that ethological facts and attitudes in recent years have done more to advance our practical control of animal behavior than recent reports from American “learning labs.”

Moreover, as we have recently discovered, if one begins with evolution and instinct as the basic format for the science, a very illuminating viewpoint can be developed which leads naturally to a drastically revised and simplified conceptual framework of startling explanatory power (to be reported elsewhere).

It is hoped that this playback on the theory will be behavioral technology’s partial repayment to the academic science whose impeccable empiricism we have used so extensively.

NOTE

1. In view of the fact that instinctive behaviors may be common to many zoological species, we consider species specific to be a sanitized misnomer, and prefer the possibly septic adjective instinctive.

Phyletic Differences in Learning1

M. E. BITTERMAN, Bryn Mawr College

One way to study the role of the brain in learning is to compare the learning of animals with different brains. Differences in brain structure may be produced by surgical means, or they may be found in nature—as when the learning of different species is compared. Of these two approaches the first (the neurosurgical approach) has been rather popular, but the potentialities of the second still are largely unexplored. Students of learning in animals have been content for the most part to concentrate their attention on a few closely related mammalian forms, chosen largely for reasons of custom and convenience, which they have treated as representative of animals in general. Their work has been dominated almost from its inception by the hypothesis that the laws of learning are the same for all animals—that the wide differences in brain structure which occur in the animal series have a purely quantitative significance.

The hypothesis comes to us from Thorndike (1911), who more than any other man may be credited with having brought the study of animal intelligence into the laboratory. On the basis of his early comparative experiments, Thorndike decided that however much animals might differ in “what” they learned (which could be traced, he thought, to differences in their sensory, motor, and motivational properties), or in the “degree” of their learning ability (some seemed able to learn more than others, and more quickly), the principles which governed their learning were the same. Thorndike wrote:

If my analysis is true, the evolution of behavior is a rather simple matter. Formally the crab, fish, turtle, dog, cat, monkey and baby have very similar intellects and characters. All are systems of connections subject to change by the laws of exercise and effect [p. 280].

Although Thorndike’s hypothesis was greeted with considerable skepticism, experiments with a variety of animals began to turn up functional similarities far more impressive than differences, and before long there was substantial disagreement only as to the nature of the laws which were assumed to hold for all animals. As acceptance of the hypothesis grew, the range of animals studied in experiments on learning declined—which, of course, was perfectly reasonable. If the laws of learning were the same everywhere in the animal series, there was nothing to be gained from the study of many different animals; indeed, standardization offered many advantages which it would be foolish to ignore. As the range of animals declined, however, so also did the likelihood of discovering any differences which might in fact exist.

It is difficult for the nonspecialist to appreciate quite how restricted has been the range of animals studied in experiments on animal learning because the restriction is so marked; the novelty of work with lower animals is such that two or three inexpressibly crude experiments with a flatworm may be better publicized than a hundred competent experiments with the rat. Some quantitative evidence on the degree of restriction was provided about 20 years ago by Schneirla, whose conclusion then was that “we do not have a comparative psychology [Harriman, 1946, p. 814].” Schneirla’s analysis was carried further by Beach (1950), who plotted the curves which are reproduced in Figure 1. Based on a count of all papers appearing between 1911 and

1948 in the Journal of Animal Behavior and its successors, the Journal of Comparative Psychology, and the Journal of Comparative and Physiological Psychology, the curves show how interest in the rat mounted while interest in submammalian forms declined. By the ‘30s, a stable pattern had emerged: about 60 percent of papers on the rat, 30 percent on other mammals (mostly primates), and 10 percent on lower forms. The set of points at the extreme right, which I have added for the decade after 1948, shows no change in the pattern. You will note that these curves are based on papers published only in a single line of journals, and on all papers in those journals—not only the ones which deal with learning; but most of the papers do deal with learning, and I know of no other journal which is a richer source of information about learning in submammalians or which, if included in the tabulation, would alter the conclusion that what we know about learning in animals we know primarily from the intensive study of a small number of mammalian forms.

How widespread is the acceptance of Thorndike’s hypothesis by contemporary theorists and systematists may be judged from a set of writings recently assembled by Koch (1959). Skinner is quite explicit in his assumption that which animal is studied “doesn’t matter.” When due allowance has been made for differences in sensory and motor characteristics, he explains, “what remains of . . . behavior shows astonishingly similar properties [Koch, 1959, p. 375].” Tolman, Miller, Guthrie, Estes, and Logan (representing Hull and Spence) rest their perfectly general conclusions about the nature of learning on the data of experiments with a few selected mammals—mostly rat, monkey, and man— skipping lightly back and forth from one to another as if indeed structure did not matter, although Miller “does not deny the possibility that men may have additional capacities which are much less well developed or absent in the lower mammals [Koch, 1959, p. 204].” Harlow alone makes a case for species differences in learning, pointing to the unequal rates of improvement shown by various mammals (mostly primates) trained in long series of discriminative problems, but he gives us no reason to believe that the differences are more than quantitative. While he implies clearly that the capacity for interproblem transfer may be absent entirely in certain lower animals—in the rat, he says, it exists only in a “most rudimentary form [Koch, 1959, p. 505]”—sub- mammalian evidence is lacking.

Although I have been considering thus far only the work of the West, I do not think that things have been very different on the other side of the Curtain. The conditioning has been “classical” rather than “instrumental” in the main, and the favored animal has been the dog rather than the rat, but the range of animals studied in any detail has been small, at least until quite recently, and the principles discovered have been generalized widely. In the words of Voronin (1962), the guiding Pavlovian propositions have been that

The conditioned reflex is a universal mechanism of activity acquired in the course of the organism’s individual life . . . . [and that] In the course of evolution of the animal world there took place only a quantitative growth or complication of higher nervous activity [pp. 161—162].

These propositions are supported, Voronin believes, by the results of some recent Russian comparisons of mammalian and submammalian vertebrates. On the basis of these results, he defines three stages in the evolution of intelligence which are distinguished in terms of the increasing role of learning in the life of the individual organism, and in terms of the precision and delicacy of the learning process. He hastens to assure us, however, that there is nothing really new even at the highest stage, which differs from the others only quantitatively.

The results of the experiments which I shall now describe support quite another view. I began these experiments without very much in the way of conviction as to their outcome, although the formal attractions of the bold Thorndikian hypothesis were rather obvious, and I should have been pleased on purely esthetic grounds to be able to accept it. I was convinced only that the hypothesis had not yet received the critical scrutiny it seemed to warrant, and that it was much too important to be taken any longer on faith. With the familiar rat as a standard, I selected for comparative study another animal—a fish—which I thought similar enough to the rat that it could be studied in analogous experiments, yet different enough to afford a marked neuroanatomical contrast. I did not propose to compare the two animals in terms of numerical scores, as, for example, the number of trials required for (or the number of errors made in) the mastery of some problem, because such differences would not necessarily imply the operation of different learning processes. I proposed instead of compare them in terms of functional relations—to find out whether their performance would be affected in the same way by the same variables (Bitterman, 1960). Why I chose to begin with certain variables rather then others probably is not worth considering—the choice was largely intuitive; whatever the reasons, the experiments soon turned up some substantial differences in the learning of fish and rat. I shall describe here two of those differences, and then present the results of some further experiments which were designed to tell us what they mean.

One of the situations developed for the study of learning in the fish is illustrated in Figure 2. The animal is brought in its individual living tank to a black Plexiglas enclosure. The manipulanda are two Plexiglas disks (targets) at which the animal is trained to strike. The targets are mounted on rods set into the needle holders of phonograph cartridges in such a way that when the animal makes contact with one of the targets a voltage is generated across its cartridge. This voltage is used to operate a set of relays which record the response and control its consequences. The targets are illuminated with colored lights or patterns projected upon them from behind; on any given trial, for example, the left target may be green and the right one red, or the left target may show a triangle and the right one a circle. The reward for correct choice is a Tubifex worm discharged into the water through a small opening at the top of the enclosure—the worm is discharged from an eye-dropper whose bulb is compressed by a pair of solenoid-operated jaws. When a worm is dropped, a magazine light at the rear of the enclosure is turned on for a few seconds, which signals that a worm has been dropped and provides some diffuse illumination which enables the animal to find it. All of the events of training are programmed automatically and recorded on tape.2

I shall talk about two kinds of experiments which have been done in this situation. The first is concerned with habit reversal. Suppose an animal is trained to choose one of two stimuli, either for a fixed number of trials or to some criterion level of correct choice, and then the positive and negative stimuli are reversed; that is, the previously unrewarded stimulus now is rewarded, and the previously rewarded stimulus is unrewarded. After the same number of trials as were given in the original problem, or when the original criterion has been reached in the first reversal, the positive and negative stimuli are reversed again—and so forth.

In such an experiment, the rat typically shows a dramatic improvement in performance. It may make many errors in the early reversals, but as training proceeds it reverses more and more readily.

In Figure 3, the performance of a group of African mouth- breeders is compared with that of a group of rats in a series of spatial reversals. (In a spatial problem, the animal chooses between a pair of stimuli which differ only with respect to their position in space, and reinforcement is correlated with position, e.g., the stimulus on the left is reinforced.) The apparatus used for the rat was analogous to the apparatus for the fish which you have already seen. On each trial, the animal was offered a choice between two identically illuminated panels set into the wall of the experimental chamber. It responded by pressing one of the panels, and correct choice operated a feeder which discharged a pellet of food into a lighted food cup. The fish were trained in an early version of the apparatus which you have already seen. For both species, there were 20 trials per day to the criterion of 17 out of 20 correct choices, positive and negative positions being reversed for each animal whenever it met that criterion. Now consider the results. The upper curve of the pair you see here is quite representative of the performance of rats in such a prob1cm—rising at first, and then falling in negatively accelerated fashion to a low level; with a little more training than is shown here, the animals reverse after but a single error. The lower curve is quite representative of the performance of fish in such a problem—there is no progressive improvement, but instead some tendency toward progressive deterioration as training continues.

How is this difference to be interpreted? We may ask first whether the results indicate anything beyond a quantitative difference in the learning of the two animals. It might be contended that reversal learning simply goes on more slowly in the fish than in the rat—that in 10 or 15 more reversals the fish, too, would have shown progressive improvement. In fact, however, the training of fish has been carried much further in later experiments, some animals completing more than 150 reversals without any sign of improvement. I invite anyone who remains skeptical on this point to persist even longer in the search for improvement.

Another possibility to be considered is that the difference between fish and rat which is reflected in these curves is not a difference in learning at all, but a difference in some confounded variable—sensory, motor, or motivational. Who can say, for example, whether the sensory and the motor demands made upon the two animals in these experiments were exactly the same? Who can say whether the fish were just as hungry as the rats, or whether the bits of food given the fish were equal in reward value to those given the rats? It would, I must admit, be a rare coincidence indeed if the conditions employed for the two animals were exactly equal in all of these potentially important respects. How, then, is it possible to find out whether the results obtained are to be attributed to a difference in learning, or to a difference in sensory, or in motor, or in motivational factors? A frank critic might say that it was rather foolish to have made the comparison in the first place, when a moment’s thought would have shown that it could not possibly have any meaningful out-come. It is interesting to note that neither Harlow nor Voronin shows any appreciation of this problem. We may doubt, then, whether they have evidence even for quantitative differences in the learning of their various animals.

I do not, of course, know how to arrange a set of conditions for the fish which will make sensory and motor demands exactly equal to those which are made upon the rat in some given experimental situation. Nor do I know how to equate drive level or reward value in the two animals. Fortunately, however, meaningful comparisons still are possible, because for control by equation we may substitute what I call control by systematic variation. Consider, for example, the hypothesis that the difference between the curves which you see here is due to a difference, not in learning, but in degree of hunger. The hypothesis implies that there is a level of hunger at which the fish will show progressive improvement, and, put in this way, the hypothesis becomes easy to test. We have only to vary level of hunger widely in different groups of fish, which we know well how to do. If, despite the widest possible variation in hunger, progressive improvement fails to appear in the fish, we may reject the hunger hypothesis. Hypotheses about other variables also may be tested by systematic variation. With regard to the question of reversal learning, I shall simply say here that progressive improvement has appeared in the rat under a wide variety of experimental conditions—it is difficult, in fact, to find a set of conditions under which the rat does not show improvement. In the fish, by contrast, reliable evidence of improvement has failed to appear under a variety of conditions.

I cannot, of course, prove that the fish is incapable of progressive improvement. I only can give you evidence of failure to find it in the course of earnest efforts; and the point is important enough, perhaps, that you may be willing to look at some more negative results. The curves of Figure 4 summarize the outcome of an experiment in which the type of problem was varied. Three groups of mouthbreeders were given 40 trials per day and reversed daily, irrespective of their performance. In the visual problem, reinforcement was correlated with color and independent of position, which varied randomly from trial to trial; e.g., red positive on odd days and green positive on even days. In the confounded problem, reinforcement was correlated both with color and position; e.g., red always on the left, green always on the right, with red-left positive on odd days and green-right positive on even days. The Riopelle problem was like the visual problem, except that each day’s colors were chosen from a group of four, with the restriction that there be no more than partial reversal from one day to the next; i.e., yesterday’s negative now positive with a “new” color negative, or yesterday’s positive now negative with a “new” color now positive. The upper curves show that there was no improvement over days in any of the three problems (the suggestion of an initial decline in the confounded curve is not statistically reliable). The lower curves show that there was a considerable amount of learning over the 40 trials of each day in each problem and at every stage of training, but that the pattern of improvement over trials did not change as training continued. Negative results of this sort now have been obtained under a variety of conditions wide enough, I think, that the burden of proof now rests with the skeptic. Until someone produces positive results, I shall assume that the fish is incapable of progressive improvement, and that we have come here upon a difference in the learning of fish and rat.

Experiments on probability learning also have given different results for rat and fish. Suppose that we train an animal in a choice situation with a ratio of reinforcement other than 100:0; that is, instead of rewarding one alternative on 100 percent of trials and the other never, we reward one alternative on, say, a random 70 percent of trials and the other on the remaining 30 percent of trials, thus constituting what may be called a 70:30 problem. Under some conditions, rat and fish both “maximize” in such a problem, which is to say that they tend always to choose the more frequently reinforced alternative. Under other conditions—specifically, under conditions in which the distribution of reinforcements is exactly controlled—the rat continues to maximize, but the fish “matches,” which is to say that its distribution of choices approximates the distribution of reinforcements: In a 70:30 problem, it chooses the 70 percent alternative on about 70 percent of trials and the 30 percent alternative on the remaining trials.

Figure 5 shows some sample data for a visual problem in which the discriminanda were horizontal and vertical stripes. In the first stage of the experiment, response to one of the stripes was rewarded on a random 70 percent of each day’s 20 trials, and response to the other stripe was rewarded on the remaining 30 percent of the trials—a 70:30 problem. In the second stage of the experiment the ratio of reinforcement was changed to 100:0, response to the 70 percent stripe of the first stage being consistently rewarded. The curves shown are plotted in terms of the percentage of each day’s responses which were made to the more frequently rewarded alternative. The fish went rapidly from a near-chance level of preference for the 70 percent stimulus to about a 70 percent preference, which was maintained from Day 5 until Day 30. With the beginning of the 100:0 training, the preference shifted rapidly upward to about the 95 percent level. The preference of the rats for the more frequently reinforced stimulus rose gradually from a near-chance level at the start of the 70:80 training to about the 90 percent level on Day 30. In the 10 days of 100:0 training, this preference continued to increase gradually, as it might have done irrespective of the shift from inconsistent to consistent reinforcement. Some further evidence of the close correspondence between choice ratio and reward ratio, which is easy to demonstrate in the fish, is presented in Figure 6. The upper portion shows the performance of two groups of mouthbreeders; one trained on a 100:0 and the other on a 70:30 confounded (black-white) problem, and both then shifted to the 0:100 problem (the less frequently rewarded alternative of the first phase now being consistently rewarded). The lower portion shows what happened when one group then was shifted to 40:60 and the other to 20:80, after which both were shifted to 50:50.

Two characteristics of these data should be noted. First, the probability matching which the fish curves demonstrate is an individual, not a group phenomenon—that is, it is not an artifact of averaging. All the animals in the group behave in much the same way. I make this obvious point because some averaged data which have been taken as evidence of matching in the rat are indeed unrepresentative of individual performances.3 Second, the matching shown by the fish is random rather than systematic. The distribution of choices recorded in the 70:30 problem looks like the distribution of colors which might be obtained by drawing marbles at random from a sack of black and white marbles with a color ratio of 70:30—that is, no sequential dependency is to be found in the data. While the rat typically maximizes, it may on occasion show a correspondence of choice ratio and reward ratio which can be traced to some systematic pattern of choice, like the patterns which are displayed in analogous experiments by human subjects. For example, a correspondence reported by Hickson (1961) has been traced to a tendency in his rats to choose on each trial the alternative which had been rewarded on the immediately preceding trial. Quite the opposite tendency, which also tends to produce a correspondence between choice ratio and reinforcement ratio, has been found in the monkey—a tendency to avoid the rewarded alternative of the preceding trial (Wilson, Oscar, & Bitterman, l964a, l964b). The matching shown by the fish, which I shall call random matching, is a very different sort of thing.

Here then, are two striking differences between rat and fish. In experiments on habit reversal, the rat shows progressive improvement while the fish does not. In experiments on probability learning, the fish shows random matching while the rat does not. These results suggest a number of interesting questions, of which I shall raise here only two: First, there is the question of how the two differences are related. From the point of view of parsimony, the possibility must be considered that they reflect a single underlying difference in the functioning of the two animals—one which has to do with adjustment to inconsistent reinforcement. Inconsistency of reinforcement certainly is involved in both kinds of experiment, between sessions in reversal learning and within sessions in probability learning. It also is possible, however, that the results for reversal learning reflect one functional difference and the results for probability learning quite another. A second question concerns the relation between the observed differences in behavior and differences in brain structure. We may wonder, for example, to what extent the cortex of the rat is responsible for its progressive improvement in habit reversal, or for its failure to show random matching. In an effort to answer such questions we have begun to do some experiments, analogous to those which differentiate fish and normal rat, with a variety of other animals, and with rats surgically deprived in infancy of relevant brain tissues.

I shall describe first some results for extensively decorticated rats (Gonzalez et al., 1964). The animals were operated on at the age of 15 or 16 days in a one-stage procedure which resulted in the destruction of about 70 percent of the cortex. Two sample lesions, one relatively small and one relatively large, are shown in Figure 7. The experimental work with the operates, like the work with normals, was begun after they had reached maturity— at about 90 days of age. From the methodological viewpoint, work with a brain-injured animal is perfectly equivalent to work with a normal animal of another species, and rats operated in our standard fashion are treated in all respects as such, with systematic variation employed to control for the effects of sensory, motor, and motivational factors. The substantive relation of the work with decorticated rats to the work with normal animals of different species is obvious: We are interested in whether extensivecortical damage will produce in the rat the kinds of behavior which are characteristic of precortical animals, such as the fish, or of animals with only very limited cortical development.

The results for decorticated rats emphasize the importance of the distinction between spatial and visual problems. In a pure spatial problem, you will remember, the two alternatives are identical except for position in space, and reinforcement is correlated with position, e.g., the alternative on the left is reinforced. In a pure visual problem, the two alternatives are visually differentiated, each occupying each of the two positions equally often, and reinforcement is correlated with visual appearance— e.g., the green alternative is reinforced independently of its position. The behavior of the decorticated rat is indistinguishable from that of the normal rat in spatial problems, but in visual problems it differs from the normal in the same way as does the fish.

The criterion-reversal performance of a group of decorticated rats trained in a spatial problem is shown in Figure 8 along with that of a group of normal controls. There were 20 trials per day by the correction method, and the criterion of learning was 17 out of 20 correct choices. As you can see, the performance of the two groups was very much the same in the original problem. In the first 10 reversals the operates made more errors than did the normals, but (like the normals) they showed progressive improvement, and in the last 10 reversals, there was no difference between the two groups. The results for two additional groups, decorticated and normal, trained under analogous conditions in a visual problem (a brightness discrimination) are plotted in Figure 9. Again, the performance of normals and operates was much the same in the original problem. In the subsequent reversals, the error scores of the normal animals rose at first and then declined in characteristic fashion, but the error scores of the operates rose much more markedly and showed no subsequent tendency to decline.

In spatial probability learning the performance of the operates was indistinguishable from that of normals, but in visual probability learning the operates showed random matching. The asymptotic preferences of operates and normals, first in a 70:30 and then in a 50:50 brightness discrimination, are shown in

to maximize in the 70:30 problem. The two whose preferences came closest to 70 percent adopted rigid position habits (CP) in the 50:50 problem, while one of the others also responded to position, and two continued in the previously established preference. In both spatial experiments, then, the decorticated rats behaved like normal rats, while in both visual experiments they behaved like fish.

On the Generality

of the Laws of Learning’

MARTIN E. P. SELIGMAN, Cornell University

That all events are equally associable and obey common laws is a central assumption of general process learning theory. A continuum of preparedness is defined which holds that organisms are prepared to associate certain events, unprepared for some, and contraprepared for others. A review of data from the traditional learning paradigms shows that the assumption of equivalent associability is false: in classical conditioning, rats are prepared to associate tastes with illness even over very long delays of reinforcement, but are contraprepared to associate tastes with footshock. In instrumental training, pigeons acquire key pecking in the absence of a contingency between pecking and grain (prepared), while cats, on the other hand, have trouble learning to lick themselves to escape, and dogs do not yawn for food (contraprepared). In discrimination, dogs are contraprepared to learn that different locations of discriminative stimuli control go—no go responding, and to learn that different qualities control directional responding. In avoidance, responses from the natural defensive repertoire are prepared for avoiding shock, while those from the appetitive repertoire are contraprepared. Language acquisition and the functional autonomy of motives are also viewed using the preparedness continuum. Finally, it is speculated that the laws of learning themselves may vary with the preparedness of the organism for the association and that different physiological and cognitive mechanisms may covary with the dimension.

Sometimes we forget why psychologists ever trained white rats to press bars for little pellets of flour or sounded metronomes followed by meat powder for domestic dogs. After all, when in the real world do rats encounter levers which they learn to press in order to eat, and when do our pet dogs ever come across metronomes whose clicking signals meat powder? It may be useful now to remind ourselves about a basic premise which gave rise to such bizarre endeavors, and to see if we still have reason to believe this premise.

THE GENERAL PROCESS

VIEW OF LEARNING

It was hoped that in the simple, controlled world of levers and mechanical feeders, of metronomes and salivation, something quite general would emerge. If we took such an arbitrary behavior as pressing a lever and such an arbitrary organism as an albino rat, and set it to work pressing the lever for food, then by virtue of the very arbitrariness of the environment, we would find features of the rat’s behavior general to real-life instrumental learning. Similarly, if we took a dog, undistracted by extraneous noises and sights, and paired a metronome’s clicking with meat, what we found about the salivation of the dog might reveal characteristics of associations in general. For instance, when Pavlov found that salivation stopped occurring to a clicking that used to signal meat powder, but no longer did, he hoped that this was an instance of a law, “experimental extinction,” which would have application beyond clicking metronomes, meat powder, and salivation. What captured the interest of the psychological world was the possibility that such laws might describe the general characteristics of the behavior acquired as the result of pairing one event with another. When Thorndike found that cats learned only gradually to pull strings to escape from puzzle boxes, the intriguing hypothesis was that animal learning in general was by trial and error. In both of these situations, the very arbitrariness and unnaturalness of the experiment was assumed to guarantee generality, since the situation would be uncontaminated by past experience the organism might have had or by special biological propensities he might bring to it.

The basic premise can be stated specifically: In classical conditioning, the choice of CS, US, and response is a matter of relative indifference; that is, any CS and US can be associated with approximately equal facility, and a set of general laws exist which describe the acquisition, extinction, inhibition, delay of reinforcement, spontaneous recovery, etc., for all CSs and USs. In instrumental learning, the choice of response and reinforcer is a matter of relative indifference; that is, any emitted response and any reinforcer can be associated with approximately equal facility, and a set of general laws exist which describe acquisition, extinction, discriminative control, generalization, etc., for all responses and reinforcers. I call this premise the assumption of equivalence of associability, and I suggest that it lies at the heart of general process learning theory.

This is not a straw man. Here are some quotes from three major learning theorists to document this assumption: