Superstitious Behavior in Children: An Experimental Analysis1

SUPERSTITIOUS BEHAVIOR IN CHILDREN: AN EXPERIMENTAL ANALYSIS Michael D. Zeiler EMORY UNIVERSITY

I. SUPERSTITION AND THE REINFORCEMENT PROCESS . . . . . . A. THE IMPLICATIONS O F SUPERSTITIOUS BEHAVIOR . . . . B. REINFORCEMENT DEPENDENCIES AND CONTINGENCIES

2 2 4

11. DELIBERATE AND ADVENTITIOUS REINFORCEMENT . . . . . . A. PROBABILITY O F RESPONSE . . . . . . . . . . . . . . . . . . . . . . . . . . . B. TEMPORAL PATTERNS OF RESPONDING . . . . . . . . . . . . . . . C. STIMULUS CONTROL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. REINFORCEMENT AND TEMPORAL CONTIGUITY . . . . . .

6 6

10 13 18

CONTROL O F MULTIPLE RESPONSES . . . . . . . . . . . . . . . . . . . . . . A. CONCURRENT RESPONSE-INDEPENDENT AND RESPONSE-DEPENDENT REINFORCEMENT . . . . . . . . . . . . . . . . B. SOME COMMENTS ON MEDIATION AND INDIVIDUAL DIFFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. A NOTE ON REINFORCING NOT RESPONDING . . . . . . . . .

21 22

IV. OTHER EFFECTS O F RESPONSE-INDEPENDENT REINFORCEMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. ELICITATION O F PREVIOUSLY PROBABLE RESPONSES . . B. ELICITATION O F A NEW RESPONSE: AUTO-SHAPING . .

23 23 25

CONCLUDING COMMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

REFERENCES

28

111.

V.

.............................................

19

19

1 Preparation of this paper was facilitated by Research Grants HD-04652 and GB-25959 from the National Institute of Child Health and Human Development and the National Science Foundation respectively. The three previously unpublished experiments described in Sections IIB, IIC, and IVD were conducted at the Institute of Child Behavior and Development of the University of Iowa with support by Research Grant HD-02845 from the National Institute of Child Health and Human Development.

1

2

Michael D.Zeiler

I. Superstition and the Reinforcement Process A. THE IMPLICATIONS

OF SUPERSTITIOUS

BEHAVIOR

In 1948, Skinner brought the study of superstition into the realm of science. In so doing, Skinner’s work initiated fruitful analyses into how behavior is controlled by its consequences and generated concepts which may explain much complex behavior. Skinner’s experiment was simple: He presented food to pigeons every 15 seconds independently of their behavior. After a number of these food presentations, Skinner observed the behavior and found that each bird displayed consistent responses. One turned in a counterclockwise direction several times between each food delivery, another repeatedly tossed its head, a third hopped from one foot to the other, and others showed still different behavior. Since these responses were not prevalent prior to the first food presentation and stopped when it was discontinued, the delivery of food was the responsible agent. Yet the responses did not influence food presentations. The parallel to the kinds of behaviors classified as superstitious outside the laboratory led Skinner to describe his experiment as involving superstition in the pigeon. In either the natural or the experimental setting, the term “superstition” refers to behaviors which are emitted as if they have environmental consequences, but in fact do not. Herrnstein ( 1966) has pursued the relation between experimental analogues of superstition and folklore critically and in detail, and points out some provocative similarities and differences. The present discussion concentrates on laboratory situations and the relevance of experiments on superstition to an understanding of how reinforcing stimuli control behavior. Skinner’s experiment has important theoretical implications. Apparently a particular behavior predominated because it happened to occur in close temporal contiguity with the presentation of food. Food was a reinforcing stimulus even though the correlation with behavior was adventitious. Once the frequency of some behavior increased, it became even more likely that the same behavior would be contiguous with the next food presentation so that its frequency would increase still further. This circular relation eventually made the behavior so probable that it became apparent during the intervals separating food presentations. Is there a difference between this phenomenon and the typical operant conditioning situation in which a response increases in frequency when it produces a reinforcing stimulus? Perhaps not, except that in the superstitious situation the response that will be contiguous with the stimulus is unspecified by the experimenter, whereas in the typical operant conditioning

Superstitious Behavior in Children

3

situation this response is preselected by the experimenter. In either case, the behavior immediately prior to the delivery of the reinforcing stimulus increases in frequency of occurrence, is thereby even more likely to precede the stimulus in the future, and eventually becomes predominant over all other behaviors. Superstitious behavior serves as the clearest example that the temporal relation between responses and reinforcing stimuli is the critical one. Setting aside philosophical considerations about the nature of causality, superstition indicates that the cause and effect relationship maintained when reinforcing stimuli depend on a response is not essential. Although it is plausible to assume that reinforcing stimuli affect the frequency of responses via the temporal relations, a position maintained by most if not all reinforcement theorists, it is an unproven hypothesis that the essential relation is in fact temporal. The reason for the hypothetical status is that experiments involving a causal relationship between a response and a reinforcing stimulus have revealed other effects besides that of increasing the probability of the response. Until these other effects can also be observed in the absence of a causal relationship, the inference that the essential relationship is temporal rather than causal remains plausible but hypothetical. Therefore, in order to maintain that the temporal correlation is the necessary and sufficient one, it is essential to demonstrate more than that a stimulus delivered without reference to behavior increases the frequency of the temporally contiguous response. For example, response-dependent presentations of reinforcing stimuli influence the way in which responses are distributed in time depending on the schedule according to which the reinforcing stimulus is presented. That is, the pattern of responses varies depending upon whether reinforcing stimuli follow the first response occurring at regular time intervals (fixed-interval schedules), or the first response occurring at irregular time intervals (variable-interval schedules), or whether they depend on the execution of a constant or varied number of responses (fixed-ratio and variable-ratio schedules). Also, the responsedependent presentation brings behavior under the control of exteroceptive stimuli if the availability of the reinforcing stimulus is correlated with the presence or the absence of certain discriminative stimuli. To the extent that the temporal relation of behavior to reinforcing stimulus presentation is the only essential condition, superstitious behavior should mimic all of the effects observed when reinforcing stimuli occur dependent on a response. As seen in Section 11, superstitious behavior in children does reveal effects on probability of response, patterns of responding, and stimulus control when temporal rather than inevitable causal relations are scheduled by the experimenter.

4

Michael D.Zeiler

B. REINFORCEMENT DEPENDENCIES AND CONTINGENCIES Several critical words, phrases, and procedures must be defined at the outset, because historically they have had multiple meanings and consequently are subject to confusion. A prime candidate is reinforcement and its derivatives. Catania (1969, p. 845) pointed out that “the vocabulary of reinforcement includes at least three nouns: reinforcer as a stimulus, reinforcement as an operation, and reinforcement as a process or as a relationship between an operation and a process.’’ The definitions used in the present paper follow from his discussion and thereby correspond with those used in the major discussions of operant behavior and reinforcement schedules (Ferster & Skinner, 1957; Morse, 1966; Skinner, 1938). By process o f reinforcement is meant an observed increase in responding, and a reinforcing stimulus or reinforcer is a stimulus which increases the frequency of the response that it follows. Reinforcement is confined to its operational sense; it refers to the presentation of a reinforcing stimulus. A schedule of reinforcement refers to the arrangement according to which reinforcing stimuli are presented. None of the definitions involves inferred events or states of the organism; all have reference to stimulus presentations and changes in response frequency. Contingency is another word that has had multiple meanings in psychology. In addition to explaining present usage, a discussion of the meaning of contingency helps to clarify the significant relationships treated in this paper. Webster’s New World Dictionary ( 1968) defines contingent as: “1. that may or may not happen; possible. 2. happening by chance; accidental; fortuitous. 3. dependent (on or upon something uncertain) ; conditional. 4. [Archaic], touching, tangential. 5. in logic, true only with certain conditions or contexts; not always or necessarily true. n. 1. an accidental or chance happening.” Other noun uses are given, but these refer to another class of meanings. Contingencies, then, properly refer to events which are not specified as necessities but which may occur. In the study of behavior, contingencies often refer to the relation between responses and reinforcement. In one use, contingency describes an independent variable, the conditions of reinforcement: The reinforcer will be presented if certain conditions are met and will not be presented otherwise. Thus, reinforcement is not a certainty but is contingent upon the occurrence of other events. In this usage, contingency is synonymous with schedule, that is, it states the conditions under which reinforcement will occur. There are two general types of condition. On the one hand, there are those requiring a specified response; these include ratio schedules which prescribe the number of responses required for reinforcement, interval schedules which specify the time that must elapse since the last reinforcement before the

Superstitious Behavior in Children

5

next response is reinforced, and schedules in which reinforcement depends on the time that has elapsed between successive responses. On the other hand, the second condition involves no response requirement; these are time schedules in which reinforcement occurs after a specified period of time elapses. Schedules involving response requirements can be referred to as involving contingent reinforcement without doing violence to the dictionary definition of contingent, since the reinforcement may or may not occur depending on whether the criterion response does or does not occur. However, although noncontingent reinforcement has been used to describe time schedules, this does not accord with the dictionary definition and, in fact, is confusing or even contradictory if taken literally. Since reinforcement is scheduled in relation to time, it is contingent on something (time), although it may be noncontingent with respect to responses. It is important to distinguish between schedules requiring responses and those that do not, but contingent and noncontingent are not the best way of doing so. The distinction is unambiguous if the classes of schedules are referred to as response dependent and response independent. A different usage of contingency refers to a dependent variable by expressing the relation between responses and reinforcement with the emphasis on the response. In this vocabulary, contingent is not synonymous with schedule but refers to the behavior that happens to occur prior to reinforcement. Although every response-dependent schedule at least requires that the specified response occur and may make it likely that other responses occur as well, none prescribes all of the behavior that may precede reinforcement. Response-independent schedules leave all of the behavior free to vary. Contingencies refer to the complex behavior that happens to occur prior to reinforcement whether the behavior is required or not. Reinforcement cannot be noncontingent because some form of behavior must occur and thus has a contingent relationship to reinforcement. Contingency in this sense is shorthand for scheduled and adventitious effects of reinforcement on behavior. The distinction between dependency and contingency derives from the efforts of Catania (1968) and Reynolds (1968) who use dependency to refer to events which must occur if reinforcement is to occur and contingency to refer to events which do not have a perfect probability of occurring prior to reinforcement. Their definitions (which a survey of recent volumes of the Journal of the Experimental Analysis of Behavior shows as being adopted with increased frequency) coincide with the present suggestion that dependencies relate to schedules and contingencies to the responses which may precede reinforcement. In summary, schedule and dependency describe how the experimenter arranges his apparatus to dispense reinforcement in relation to (or independent of) responses, and con-

6

Michael D . Zeiler

tingencies describe the behavior that is affected by reinforcement. A schedule may or may not involve a response dependency, but all involve contingencies. Students of superstitious behavior deal with contingencies in that they study behavior that need not occur. Sometimes the contingencies refer to responses, sometimes they refer to antecedent discriminative stimuli which may or may not precede reinforcement as well. Contingencies also arise in response-dependent reinforcement because behavior other than that required may also occur in relation to reinforcement. Basically, the concern is with experiments using response-independent, response-dependent, stimulusindependent, and stimulus-dependent reinforcement schedules to explore contingencies.

11. Deliberate and Adventitious Reinforcement A. PROBABILITY OF RESPONSE Although Skinner’s (1948) paradigm is the most straightforward one for revealing how response-independent reinforcement affects the probability of responses, it places severe demands on the experimenter. The experimenter has no way of knowing what response to observe, instead he must wait to see if some behavior emerges and then find some way to measure and record it objectively. A solution has been to circumvent the problem by first generating the response via response-dependent schedules and then to observe how the behavior is maintained when the response is no longer required. It must be noted though that this solution places the emphasis on the maintenance rather than on the acquisition of behavior, and thus is not the strongest test of the hypothesis that response-independent and response-dependent reinforcement are equivalent with respect to response probability. It seems fortunate, therefore, that there have been several instances in which experimenters have observed that a response-independent schedule did establish a new response as well as maintain an old one. Both initial establishment and maintenance of a response have been observed with children. Zeiler (1970) first exposed four- and five-year-old children to a fixedratio schedule. The child received candy whenever a total of 30 responses had occurred to two response keys; these responses could be distributed in any proportion to the two keys. One key was blue and the other red. Stable behavior consisted of equal responding to each key. The children either executed an entire ratio on one key and then switched to the other, or alternated keys with successive responses. Following the development of

Superstitious Behavior in Children

I

stable behavior, the schedule was changed so that responses to the red key had no effect (extinction), while not responding to the blue key for periods which were increased up to 60 seconds produced candy ( D R O 60-second schedule). This combined schedule was a concurrent extinction DRO 60second schedule. The change from the fixed-ratio to the concurrent schedule had rapid and striking effects. Responses to the blue key slowed and then stopped in accordance with the dependency between candy presentations and the absence of responses. Simultaneously, though, most of the children maintained a high rate of responding to the red key even though these responses actually were irrelevant to candy deliveries. That the responses to the red key depended on the candy deliveries was apparent in that these responses stopped when the DRO schedule was changed to extinction so that candy deliveries ceased. The behavior on the red key was similar to that observed when children received reinforcing stimuli for some response according to a fixed-interval schedule (e.g., Long, Hammack, May, & Campbell, 1958). Even though the candy was independent of responses to the red key, it maintained or increased the probability of responding. It was not necessary that the particular form of behavior emitted prior to reinforcement be pressing the red key; whatever behavior happened to occur at that time could increase in frequency. The behavior of two other children in fact did reveal different forms. Their behavior was reminiscent of Skinner’s ( 1948) pigeons. Neither of these children maintained presses of the red key during the concurrent schedule phase. One of the children instead sat quietly and watched the tray. picking up a piece of candy when it appeared. This subject had alternated successive responses between the keys during the fixed-ratio schedule phase, and early in the concurrent schedule phase had received all of the candy during rest periods following periods of sustained responding. The other child explored the room and floor between candy presentations during the concurrent schedule rather than press the keys. Early in the concurrent schedule phase, he received several pieces of candy via the DRO schedule while searching on the floor for a dropped piece. Subsequently, hc spent the time crawling around the floor, reaching up to get each piece of candy when it was discharged into the tray. These characteristic behaviors of each child also stopped when the DRO schedule was changed to extinction. It appears, therefore, that the particular forms of behavior that a child emitted prior to reinforcement became predominant whether that behavior was key-pressing or something else. Some of these data, then, indicated that a response was established that had no scheduled relation to reinforcement, and the rest indicated that the rate of a previously established response could be maintained or increased.

8

Michael D . Zeiler

Another experiment also showed the ability of response-independent reinforcement to maintain responding in children. In two experiments, Weisberg and Kennedy (1969) first trained two- to five-year-old children to press a lever by delivering snacks according to either a variable-interval or a variable-ratio schedule (Phase 1) . They then changed to a schedule which delivered the snacks independent of responses at either variable or fixed time periods (Phase 2). The remaining children were shifted from the response-dependent schedules (Phase 1 ) to extinction (Phase 2 ) . The behavior of the groups in Phase 2 provided between-group comparisons of the ability of response-independent and extinction schedules to maintain responding. Additional manipulations provided a within-subject comparison. The children who had been given the response-dependent schedule in Phase 1 and the response-independent schedule in Phase 2 were returned to the response-dependent schedule (Phase 3 ) and then were changed to extinction (Phase 4). Those who had the response-dependent schedule of Phase 1 followed by extinction in Phase 2 were returned to the responsedependent schedule (Phase 3 ) and were then changed to a responseindependent schedule (Phase 4). A comparison of Phases 2 and 4 revealed the effects of response-independent reinforcement and extinction on the behavior of each child. Weisberg and Kennedy therefore provided both within- and between-subject comparisons of the ability of response-independent and extinction schedules to maintain a previously established high probability response. The results were unequivocal. Under extinction conditions the rate of lever pressing quickly dropped to a low level. With response-independent food presentations, however, responding persisted for a much longer time before dropping to a low level. In fact, three of the four children who were shifted from the response-dependent variable-ratio schedule of Phase 1 to the response-independent fixed-time schedule of Phase 2 pressed the lever throughout the ten experimental sessions. One child showed no decrease, a second dropped to a low stable level, and the third responded at a higher rate in the response-independent schedule phase. All three subjects subsequently stopped responding after being shifted from the Phase 3 variableratio schedule to extinction. The data suggested that the durability of responding during the responseindependent phase might depend on the rate of the response when that phase began. Subjects differed in their response rates in the variable-timc schedule were ordered just as they were under variable-interval: The higher the rates during variable-interval, the higher the rates and the more persistent the behavior under variable-time. The variable-ratio schedule established still higher response rates than did the variable-interval, and responding seemed even better maintained with the shift from the ratio than

Superstitious Behavior in Children

9

from the interval to the response-independent schedules. Prevailing higher response rates imply the greater probability of a given response occurring in close temporal contiguity to reinforcement under a response-independent schedule, so that it is consistent that behavior should be better maintained if it is occurring at a high rate. These data show that stimuli which reinforce behavior when presented according to response-dependent schedules also reinforce behavior when presented independently of responses. Thus, the data support the hypothesis tkat response-independent and response-dependent presentations have similar effects with respect to developing and maintaining behavior. The only difference seems to be that the response-dependent case guarantees that a specified response will be contiguous with reinforcement, while the response-independent case leaves the particular response free to vary. Why did the rate of responding decline in the response-independent schedule phase with many of the children in the Weisberg and Kennedy (1969) study? Similar rate decreases also occur when nonhuman subjects are shifted from response-dependent to response-independent reinforcement (Appel & Hiss, 1962; Herrnstein, 1966; Zeiler, 1968). The explanation is straightforward. Skinner ( 1948) noted that when reinforcements occur independently of responses, it is likely that although a certain form of behavior is most probable that precise behavior will not always immediately precede reinforcement. Although a reinforcing stimulus does strengthen behavior that is not exactly contiguous with it, the strongest effects are on the immediately preceding events (Dews, 1960). If a behavior incompatible with the response being measured should have the closest contiguity with reinforcement (e.g., withdrawing the hand from the key after pressing is incompatible with pressing in that both cannot occur simultaneously), the measured response eventually should be replaced by the other. Thc new behavior could be of any form based on exactly what did occur at the moment of reinforcement. Skinner observed that the nature of superstitious rituals emitted by pigeons changed over time. This drift in behavior probably is due to the sort of events just described, and illustrates an important difference between response-independent and response-dependent schedules. In the response-dependent case the response requirement precludes drifting by maintaining the close temporal relation between reinforcement and the prescribed response. The recognition of drift raises the question: Why should a certain form of behavior ever persist indefinitely with response-independent reinforcement? Weisberg and Kennedy's results showed that for some children the response did persist, and Herrnstein and Morse (Herrnstein, 1966) reported similar results. Informal observations by several experimenters have suggested that a previously required behavior can continue indefinitely

10

Michael D.Zeiler

under a response-independent reinforcement schedule. Why this should occur is a puzzling and unsolved question. All that can be said at this time is that apparently behavior which has been established strongly can be maintained, perhaps permanently, under certain as yet unspecified conditions.

B. TEMPORAL PATTERNS OF RESPONDING

In nonhuman subjects, response-dependent reinforcement schedules have effects on behavior other than affecting its overall probability. According to the particular schedule in force, the distribution of responses in time (the pattern of responding) varies. Thus, with fixed-ratio schedules of moderate or large size, a pigeon pauses after reinforcement and then quickly changes to responding at a high rate which continues until the next reinforcement. Fixed-interval schedules also produce a postreinforcement pause, but then responding gradually accelerates to a high rate. Variable-interval and variable-ratio schedules generate stable rates without long pauses throughout the periods between successive reinforcements. To the extent that response-dependent and response-independent reinforcement are the same, response-independent schedules should also produce characteristic patterns. Schedules involving temporal specifications are of necessity the meeting point of response-dependent and response-independent schedules. Fixedand variable-ratio schedules specify only response number as the requirement for reinforcement; by definition, there can be no response-independent analogues of ratio schedules. Fixed- and variable-interval schedules, in contrast, involve both time and response requirements: either reinforcements follow the first response occurring after fixed periods or after variable time periods. Response-independent schedules can mimic the temporal aspect of interval schedules. I n this respect, fixed-time schedules correspond to fixed-intervals and variable-time schedules correspond to variableintervals. Research with nonhuman subjects indicates that fixed- and variable-time schedules do indeed generate distinctive patterns, and, in fact, produce patterns like those occurring with fixed- and variable-interval scHedules (Appel & Hiss, 1962; Herrnstein & Morse, 1957; Zeiler, 1968). There is no published work with children on this problem; however, one might anticipate difficulty in obtaining differences between fixed- and variable-time schedules because children respond similarly on fixed- and variable-interval schedules. Both produce fairly steady response rates with occasional pauses in responding (e.g., Long et al., 1958). In a previously unpublished study, Zeiler and Orr investigated the pat-

Srcperstitious Behavior in Children

11

terns of responding established in four- and five-year-old children by candy presented according to either fixed- or variable-time schedules. For some children the initial schedule was a fixed interval in which the first response occurring after 30 seconds produced candy; for others, it was a variable interval having a mean interreinforcement interval of 30 seconds. The purpose of these schedules was to generate substantial rates of pressing the response key. After there were at least three sessions and responding had stabilized, the schedules were changed to provide candy independent of responses. The children who had been on the fixed-interval schedule now received candy on a variable-time schedule : Candy appeared at irregular intervals averaging 30 seconds. The children who had been on the variableinterval schedule now obtained candy on a fixed-time schedule providing candy every 30 seconds without reference to behavior. For one child the change from variable interval to fixed time occurred in the middle of a session, but for the others changes were made at the beginning of a session. Although there were substantial differences among the five children in the average response rate in the first phase, these differences were unrelated to whether the schedule was fixed or variable interval. Cumulative records revealed similar patterns of responding with the two schedules: All of the children responded at a fairly steady rate although pauses did occur during some interreinforcement periods. Pauses did not often occur immediately after candy deliveries but were distributed throughout the intervals. This sort of behavior is typical of children under fixed- and variable-interval schedules. The record shown in segment A of Fig. 1 is for a child given a variabletime schedule. The behavior continued much as it had been under fixed interval, with responses emitted at a fairly stable and high rate. The change from the fixed-interval to the variable-time schedule had little effect on behavior. The change t o the fixed-time schedule had rapid effects. Segment B of Fig. 1 shows the behavior on the first day of the fixed-time schedule for one of the children. The behavior up to the first two candy presentations was like that observed previously under the variable-interval schedule. Subsequently, responding became more erratic, and then was marked by periods of pauses followed either by a n abrupt shift to responding at a high rate or by positive acceleration. Pauses after candy presentations were evident. Still later in the session responding was often first positively, and then negatively, accelerated in each period so that the records were S-shaped. The next session showed similar behavior. Segment C of Fig. 1 shows the behavior of the child changed from the variable-interval to the fixed-time schedule midway through a session. Although the prevailing rate was low under both schedules, the patterns of

12

Michael D . Zeiler

I

8 Minutes

-I

Fig. 1. Cumulative records obtained by Zeiler and Orr. ( A ) A full session o f the variable-time schedule, ( B ) a frill session of the fixed-time schediile, and ( C ) a sessioii in uBhich the schedule n'as changed from variable interval to fixed time. The record is broken at the point of transition. Deflections of the response pen indicate candy preserr tafiorrs.

responding were like those of the other children. The variable-interval schedule generated a rather erratic pattern : In some intervals responding was steady and in others it was positively accelerated. With the switch to the fixed-time schedule, responding was positively accelerated in nearly every interval. Two other children were first trained with a variable-interval schedule and were then shifted to fixed interval to determine whether the change in temporal patterning of reinforcement from variable to fixed was responsible for the change in the pattern of responding observed for children switched from the variable-interval to the fixed-time schedule. However, the change from variable- to fixed-interval produced no noticeable changes in behavior. These data revealed that the response-independent schedules did control distinctive patterns of response. In fact, the difference in the behavior under fixed-time and variable-time schedules was much greater than that between fixed-interval and variable-interval schedules. It remains unclear as to why the two types of interval schedule usually have generated similar behavior in children but not in nonhuman subjects. What is particularly puzzling is why both forms of variable schedule as well as the fixed-time schedule generate behavior in children like that observed with other species, while fixed-interval schedules differ. One possibility is that there is no research

Superstitious Behavior in Children

13

with children involving fixed intervals with substantial temporal requirements: Positive acceleration in pigeons does not always occur with short intervals (Ferster & Skinner, 1957). Perhaps fixed-time schedules will show positively accelerated responding at temporal parameter values too low to reveal such behavior under fixed intervals in human subjects. Weisberg and Kennedy ( 1969) studied transitions from variable-interval and variable-ratio schedules to variable-time and fixed-time schedules. The only data provided on patterning was on fixed-time following variable-ratio training. These data confirmed Zeiler and Orr's finding that fixedtime schedules engender positively accelerated rates of responding. Two of the children displayed positively accelerated patterns during the variableratio schedules but the pattern was enhanced by the fixed-time schedule. The third subject had a fairly stable rate under variable-ratio reinforcement and then displayed positive acceleration with the fixed-time schedule. These limited data on patterns of responding under response-independent schedules indicate that variations in the response-independent conditions do produce different patterns of response. The data confirm those of nonhuman subjects given fixed- and variable-time schedules. It may be, however, that children differ from nonhumans in that fixed response-independent and response-dependent schedules (fixed-time and fixed-interval) produce different effects. Since the data are scanty, all that can be said now is that the results are equivocal with respect to the similarity of the two forms of fixed schedule; obviously, final conclusions await further research.

C. STIMULUS CONTROL One important property of response-dependent reinforcement is that when the availability of reinforcement is differentially related to environmental stimuli, organisms usually come to respond differentially with respect to the stimuli. Stimulus control, therefore, is observed under experimental paradigms in which a response ( R ) is followed by a reinforcing stimulus (SR) in the presence of a certain antecedent stimulus (S). In one familiar instance of the S-R-S" paradigm, responding is reinforced in the presence of one stimulus and not reinforced in the presence of another; other common cases involve reinforcement in the presence of two or more stimuli with a distinctive schedule correlated with each. In any event, stimulus control is evidenced when subjects' responses differ in some respect (e.g., rate or pattern) depending on the prevailing stimulus. These typical situations involve a two-way dependency : Reinforcement requires the presence both of a certain stimulus and of a certain response. Discriminative behavior should also occur with response-independent reinforcement if the response-dependent and independent cases involve the

14

Micliuel D . Zeiler

same essential processes. However, the general problems encountered in studying response-independent reinforcement in nondiscriminative situations also arise, and, in fact, seem evcn more burdensome in the discrimination situation. What response should be observed? Will the experimenter notice it if and when it occurs? In addition, how can the experimenter rigorously evaluate whether the observed superstitious behavior is differentially related to the prevailing stimuli? One possible solution is the one already described, namely, generate a high probability response by prior training and then shift to a response-independent schedule while manipulating stimuli. Although such a procedure would be of interest and might well prove effective, it apparently has not been used. Instead, a different procedure, one first reported by Morse and Skinner (1957), has proven valuable and provides related information. It consists of a maintained response-dependent schedule with changes in stimuli that have no scheduled relation to the availability of reinforcement. The procedure consists of having reinforcement be stimulus-independent while continuing to be response-dependent. Differential responding under these conditions indicates that the three-term S-R-SR dependency is not necessary for obtaining stimulus control, but is perhaps effective only because it guarantees certain temporal relationships between discriminative stimuli, responses, and reinforcement. Thus, the procedure provides a way of breaking up part of the three-term dependency to determine if in fact it is essential for stimulus control to occur. Morse and Skinner found that pigeons pecking a key which produced food according to a variable-interval schedule had a different response rate in the presence of each of two colors, although the colors were only adventitiously related to the reinforcement schedule. Thus, in pigeons, different scheduled consequences for responding to two stimuli are not necessary to establish discriminations; pigeons responded differently to two stimuli when each provided the same outcome. Tt is appropriate to refer to this phenomenon as a “superstitious discrimination” since differential responding occurred in the absence of differential requirements. The experiment reported here investigated further the occurrence of discriminations in the absence of experimenter-scheduled differential reinforcement. The research had two main purposes: ( 1 ) to investigate the generality of the Morse and Skinner effect with children and ( 2 ) to determine whether discriminations would develop with a reinforcement schedule other than variable interval. The first purpose, assessment of generality, stemmed from the absence of any data bearing on this problem with children. The second arose because previous research used only variable-interval schedules. Both variable-interval and fixed-ratio schedules were used to determine

15

Supersiiiioiis Behavior in Children

if the superstitious discriminations occur only with irregular temporally determined reinforcement presentations. The subjects were 12 children ranging in age from four to five years. During the 10 experimental sessions, they received candy for pressing a key according to either a variable-interval schedule having a mean interval of 30 seconds or a fixed-ratio schedule maintained at 15 responses per candy presentation. Six children were exposed to each schedule. The significant aspect of the experiment was that key pressing produced candy according to the schedule in force independent of the color of the key. The color alternated between red and blue, staying red for 30 seconds and changing to blue for 6 seconds. The occasional interruptions of one color by another was the same general procedure used by Morse and Skinner, although their intervals involved the presentation of the second stimulus for 4 minutes once per hour and a baseline schedule of variable interval averaging 30 minutes between reinforcements. In the first session most of the children responded at nearly equal rates in the presence of the two stimuli. Differences in rate began to develop by the end of the session. With additional sessions, 9 of the 12 children had a substantially higher rate in the presence of one stimulus than the other, and several had more than a 10-fold rate difference. Figure 2 shows the largest differences obtained in any session for children exhibiting a higher rate in red, for those having a higher rate in blue, and for those with n o clear rate difference. Figure 3 presents cumulative records showing a higher response rate when the key was red under both the variable-interval and the fixedratio schedules. The rates during each stimulus can be analyzed by com-

A

B

C

D

E

F G SUBJECT

H

I

J

K

L

Fig. 2. The largest rate differences nttained to the two Jtinirtli b y each child. The figure is segmented io show higher rates in red, higher rates irz hlite, and equal rates to boih stimuli. Thc solid bars show the rate during the blue stitnrrlus, and the striped bars show the rate during the red stimulus.

16

Michael D. Zeiler

I

1 0 Minutes

I

Fig. 3. Citmrtlafive records indicating a higher response rate in red than in blue. T h e key was red f o r 30 seconds and changed fo blue during the 6-second periods ihai the response pen W ~ deflected. S Cirrve A shows an FR 15 schedule; curve B shows a VI 30-second schedule. T h e o c c ~ t r r e ~ ~ofc ecandv deliveries is riot shown.

paring the slopes in the records when the pen was in its normal position (during red) and when it was deflected (during blue). The children responded at a substantial rate (as fast as 2.00 responses per second) when the key was red, but virtually stopped pressing during the 6-second periods when it was blue. Counters and observations by the experimenter indicated that all candy was obtained either during the red periods or immediately after the color changed to blue. The overall rate of responding did not differ consistently with the two types of schedule, except that there was some pausing after candy deliveries with the fixed ratio. Figure 4 presents records showing a higher response rate in blue than in red. Again there were no marked schedule effects. With the fixed-ratio schedule children began responding during the 6-second blue periods and continued to respond until the delivery of candy. Then they paused until the next blue period. Candy

k

10 Minutes

I

Fig. 4. Cirmirlative records indicating a highrr response ratr in bliir than in red. Recording was described for Fig. 3. Cirrve A shows an F R 15 schedrrle; cirrvr B shows a V l 30-second schedule.

Superstitious Behavior in Children

17

presentations usually occurred in red following a burst of responses initiated by the change from red to blue. With the variable-interval schedule, they either did not respond at all in red or responded at a low rate. Almost all candy presentations occurred in blue. Different children exhibited various time courses of the discrimination. F o r some, the stimuli maintained different rates of responding over the 10 experimental sessions. Others showed a discrimination in the first four o r five sessions, but then came to respond equally during both stimuli. For others, stimulus control developed after several days of training. This sort of variability and changes in differential responding typically occurs in experiments on discriminations appearing in the absence of scheduled differential reinforcement in subjects other than children (Lander, 1968; Morse & Skinner, 1957). Also, the failure of a superstitious discrimination to occur for some subjects is not unusual in experiments of this type. Morse and Skinner pointed out that, although the stimulus which is present at the time of reinforcement is accidental, the reinforcement may increase the future rate of responding during that stimulus. If reinforcement now should happen not to occur when the second stimulus is present for any of a number of possible reasons, responding to that stimulus is likely to decrease. Now, given a greater likelihood of responding in the presence of one of the stimuli, the subject is more likely to meet the schedule requirements and to obtain additional reinforcements during the first stimulus. This sort of circularity would eventually result in a substantial degree of differential responding with respect to the two stimuli. The only difference between the superstitious discrimination and discriminations occurring when differential reinforcement is scheduled explicitly is that the superstitious case allows the stimulus which is correlated with reinforcement to vary. Because of that, either stimulus could control the higher response rate, and, because reinforcements may occur during the stimulus correlated with the higher rate (unless the subject completely stops responding), the discrimination either could disappear or even reverse its direction over a number of sessions. (Reversals in the stimulus which controlled the higher rate were not observed in the children.) These hypotheses would seem to be reasonable accounts of the nature of superstitious discriminations; as such, they demonstrate the similarity in process for the events responsible both for superstitious and experimenter-determined discriminations. Apparently, the prevailing reinforcement schedule is not critical since similar behavior occurred with both fixed-ratio and variable-interval schedules, perhaps because both schedules produced similar rates and patterns of responding in children. It seems likely, however, that supmtitious discriminations probably require at least a moderate degree of intermittency in

18

Michael D.Zeiler

the response-reinforcement relation. Continuous reinforcement or schedules having little intermittency probably would establish such a high probability of reinforcement in the presence of both stimuli that rate differences and the ensuing circular relationship between ongoing behavior, reinforcement, and consequent behavior would not develop. One new observation was that children did not require a strict correlation between reinforcements and stimuli to establish the superstitious discriminations. Some subjects had a higher rate during the blue stimulus even though the candy presentations occurred during red. Since the run of responses always did begin in blue, it seems as if the stimulus present when responding began was more important than the one present at the moment of reinforcement. This may be a difference between children and nonhuman subjects; however there are insufficient data to evaluate this possibility.

D. REINFORCEMENT AND TEMPORAL CONTIGUITY The preceding review demonstrated the close similarity between adventitious reinforcement, be it response-independent or stimulus-independent, and the effects of reinforcements scheduled to occur only under specified and experimenter-controlled stimulus and response conditions. Although there are differences which limit final conclusions, they may be more apparent than real because of the absence of sufficient data concerning the behavior of children. At the present time, the similarity in the effects of adventitious and deliberate reinforcement appears compelling. A reinforcing stimulus seems to strengthen ongoing responding even though the reinforcing stimulus and the responses and the prevailing discriminative stimuli are only temporally related. Furthermore, the behavior is essentially the same as when reinforcements require the presence of certain responses and discriminative stimuli. The temporal relation describes both the necessary and the sufficient conditions for modulating behavior. The critical events, therefore, are the same in both the adventitious case and the case in which reinforcement is delivered dependent upon the occurrence of a response and/or the presence of a specified discriminative stimulus. The only difference between superstitious behavior and the ordinary operant behavior observed with response- and stimulus-dependent reinforcement is that the dependent cases specify the behavior contiguous with reinforcement while the adventitious cases leave the behavior free to vary. Thus, behavioral control is identical for reinforcements given for specific responses and reinforcements having an accidental relation to responses and discriminative stimuli; reinforcement has the same temporal relation to behavior and stimuli in either case.

SupersfifioiisBehavior in Children

19

111. Control of Multiple Responses A. CONCURRENT RESPONSE-INDEPENDENT A N D RESPONSE-DEPENDENT REINFORCEMENT A number of experiments (e.g., Antonitis, 195 I ; Skinner, 1938) have shown that reinforcement affects several forms of behavior simultaneously. For example, when a rat obtains food dependent on a bar press, the probability of bar pressing increases. In addition, other aspects of behavior, e.g., the duration o r the topography of the bar press, become stereotyped even though the reinforcement occurs independently of these aspects of behavior. (Except, of course, that the duration must exceed the minimum value specified as a n acceptable response by the programming equipment, and that the lever must be pressed in some fashion.) These data suggest that the response-dependent presentation of a reinforcing stimulus not only influences the behavior scheduled to precede it but concurrently operates on the behavior which accidentally precedes it. Some data with children also support this conclusion. Bruner and Revusky (196 I ) studied the behavior of adolescent boys when a response to one telegraph key earned five cents whenever at least 8.2 seconds but less than 10.25 seconds elapsed since the preceding response to that key. This is a differential-reinforccment-of-low-rateschedule ( D R L 8.2-seconds) with a limited hold of 2.05 seconds. In addition to the telegraph key correlated with this schedule, there were three other keys. Responses to these other keys had no consequence with respect to the delivery of nickels. The complete schedule, involving all of the keys, was a concurrent [ ( D R L 8.2 seconds limited hold 2.05 seconds) extinction extinction extinction] schedule. Subjects were instructed to press one key at a time and were permitted to use only one hand. T h e four boys all had a higher rate of pressing the irrelevant keys than the one correlated with the delivery of nickels: Three of the four confined the high rate responses to one of the extinction keys, while the fourth boy responded about equally to two of the three keys. Responses to the DRL key were spaced sufficiently in time so that nearly every such response was followed by the presentation of a nickel. When the schedule on the D R L key was then changed to extinction so that nickels were no longer available, responding became more erratic. Apparently, the responses to various keys in the D R L phase were being maintained by the nickels even though presses of all but the one key were irrelevant with respect to the reinforcing event. Bruner and Revusky interpreted their data as indicating a chain of behavior which served to bridge

Michael D.Zeiler

20

the time gap between successive presses of the DRL key. What is more important is that the response-dependent delivery of reinforcement according to one schedule also generated stable behavior which had a responseindependent relation to deliveries of the reinforcing stimulus. Such effects are not peculiar to DRL schedules. In Zeiler’s (1970) study referred to in Section IIA, four- and five-year-old children began with candy presented according to a fixed-ratio schedule. There were two keys, one red and one blue, and the ratio requirement could be met by a total of 30 responses distributed in any proportion between the two keys. Every child established a systematic pattern of responding. Some alternated keys with every response, and others alternated keys after each candy presentation, Thus, the fixed-ratio schedule: ( a ) increased the probability of keypressing, the behavior on which candy presentation depended, and ( b ) fixed the way in which the responses were emitted even though such stereotypy was superstitious in that it was irrelevant to reinforcement deliveries. Tritschler’ has found similar stereotyped and often even more complex patterns of response in college students required to execute a fixed ratio on any of five telegraph keys. In the next phase of Zeiler’s (1970) study, responding to one of the two keys had no scheduled consequence and not responding to the other key for a specified period of time resulted in candy (concurrent DRO Extinction schedule). Two forms of behavior resulted. The children stopped pressing the DRO key. In addition, the DRO schedule maintained behavior which did not influence the delivery of candy. Either the children pressed the extinction key at a high rate or they consistently performed other behavior such as crawling on the floor or sitting and staring at the candy tray. Again, a reinforcing stimulus influenced both required behavior and behavior which had only an adventitious relation to the stimulus presentations. Catania and Cutts (1963), studying college students, demonstrated that the close temporal contiguity of the superstitious behavior to reinforcement maintains the behavior. Their subjects had two push buttons with responses to one reinforced on a variable-interval schedule and responses to the other having no scheduled consequences (concurrent variable-interval extinction schedule). The subjects maintained a high rate of responding to both buttons, and one subject actually responded at a much higher rate to the extinction button than to the variable-interval button. Catania and Cutts then added a new requirement: A press on the variable-interval button could not be reinforced if it followed a press on the extinction button within a certain time. The imposition of this change-over delay greatly reduced the superstitious responses to the extinction button; without the change-over 2

J. Tritschler, personal communication, 1970.

Sirpersririoirs Brhuvior in Childreri

21

delay, superstitious responses were maintained because they could occur in close temporal contiguity to the response deliberately reinforced on the variable-interval schedule. I t seems reasonable to hypothesize that the superstitious behavior in Bruner and Revusky’s and Zeiler’s experiments was maintained in the same way. The obvious conclusion is that reinforcing stimuli influence several forms of behavior simultaneously even when they have a response-dependent relation to one and a response-independent relation to the other.

B. SOMECOMMENTS O N MEDIATION AND INDIVIDUAL DIFFERENCES When superstitious behavior occurs during DRL or DRO schedules which require periods of not performing the criterion response if reinforcement is to occur, it may be tempting to attribute a mediating function to the superstition. The implication would be that the behavior serves to facilitate the acquisition of reinforcement by filling the time interval. The possible advantages of such behavior, though, should not obscure the fact that the particular behavior that occurs could only be strengthened because it accidentally preceded reinforcement delivery and is maintained because it continues to have this temporal relation. To pose an extreme position, perhaps mediation refers to nothing more than collateral adventitiously reinforced behavior. Although superstitious behavior would seem to have no beneficial function with schedules such as fixed ratio or variable interval where the subject can respond continuously, similar collateral behavior does occur with those schedules. Hence, the collateral behavior need not be useful for it to occur and be maintained; it occurs and is maintained because it continues to bz reinforced. Tt is possible that the same kind of independence characterizes human verbal and motor behavior. namely, that these are separate response systems which may be simultaneously affected by reinforcement. This is a different notion than the common one that verbal mediation is instrumental in producing certain kinds of motor behavior (e.g., choices in discrimination learning) ; perhaps the motor and correlated verbal responses d o not always serve a functional role with respect to each other and are related only in the sense that both can be established by the same reinforcing event. T h e concurrent establishment and maintenance of deliberately and adventitiously reinforced responses may provide some insight into the basis of individual differences. Herrnstein ( 1966) pointed out that when reinforcements depend on a certain response the specification usually involves only some of the dimensions of the response. For example, the experimenter specifies the location and minimum force of an acceptable response but does not also specify its precise topography, rate. maximum force, or dura-

22

Michael D . Zeiler

tion. Yet if these aspects of behavior are conditionable, whatever values they assume will be correlated with reinforcement and will thereby become more probable. Since the values are unspecified, they are likely to differ among individual subjects. That the laboratory situation may be analogous to the natural environment in this respect does not seem farfetched. In the natural environment children generally receive reinforcements for similar behaviors, for learning to care for themselves, for doing well in school, for obeying adults, etc. Despite this general homogeneity, many other aspects of behavior can vary. A child may answer a question in a low or a loud voice, and the reinforcement given for answering correctly may also increase the probability of speaking softly or loudly in the future. Could similar accidents play a role in creating different behavioral styles? Although speculative, it is an alternative to usual explanations of individual differences.

c. A NOTEON REINFORCING NOTRESPONDING In a DRO schedule, a reinforcement occurs whenever a specified response has not occurred for a specified period of time because reinforcements depend on the subject’s not emitting a certain response. These schedules have proven useful in rapidly eliminating behavior and as controls in a number of contexts. The delivery of a reinforcing stimulus following not responding, though, has led to some curious interpretive problems. These apparently stem from reluctance to treat the nonemission of a specified response as a functional response. Instead, starting with Reynolds ( 1961 ), DRO schedules have been defined as “differential-reinforcementof-other-responses,’’ the implication being that the reinforced behavior is not the absence of a response but rather is the occurrence of some other, although unobserved, behavior. Unfortunately, this definition of the schedule involves a theoretical account of its method of action instead of an objective description of the dependency. It may also obscure the possible fact that not emitting a particular response may actually have functional response properties. In any case in which a given behavior is occurring with some frequency, if its probability decreases following consequences for its nonoccurrence, the absence of the response meets the requirements of a functional response. To attribute the behavior to the strengthening of some other behavior is an unsupported inference even if other behavior should be observed to occur, and certainly should not be involved in the name of the schedule. It is interesting that Lane (1961 ) also used the initials DRO to describe the schedule almost simultaneously with Reynolds but translated the initials as “differential-reinforcement-of-no-responding.” Thus, the two earliest uses of

Siiperstitious Behavior in Children

23

the initials defined them differently. [At almost the same time, Kelleher ( 1961 ) used the schedule and called it DRP: “differential-reinforcementof-pausing.’’ Kelleher’s provides an objective description, but his terminology was not adopted subsequently.] It does turn out that DRO schedules can strengthen some behavior other than that prescribed as prerequisite for reinforcement (not emitting a certain response). However, the strengthening of behavior in addition to that required is true of every other reinforcement schedule as well; there is nothing unique about DRO in this respect because adventitious strengthening of responses appears to be a general property of reinforcement. Therefore, it would be appropriate to describe any schedule as involving reinforcement of “other” behavior. The concurrent reinforcement of several aspects of behavior has played an important part in theoretical explanations of how reinforcement schedules establish their distinctive effects on behavior. Thus, first Skinner (1938) then Ferster and Skinner (1957), and then Morse (1966) hypothesized that the combination of response-reinforcement dependencies and response-reinforcement contingencies operate on different schedules of reinforcement. Although we do not yet have an adequate theory of reinforcement schedules, it seems likely that a comprehensive theory will be unable to ignore the relation of reinforcement to required and unrequired behavior in all schedules.

IV.

Other Effects of Response-Independent Reinforcement

Reinforcing stimuli are defined by their ability to affect the probability of responses that precede them; however, such stimuli also have other effects on behavior. These effects do not involve a dependency between the behavior and the reinforcer and thereby qualify as response-independent effects. Staddon and Simmelhag (1971) consider such behavior to be of such fundamental importance thJt it justifies (in fact requires) revised thinking about the nature of reinforcement. Commentary on their provocative arguments go beyond the scope of the present paper, however, a number of the effects they consider important have been observed in children. A. ELICITATION OF PREVIOUSLY PROBABLE RESPONSES

One property of a reinforcing stimulus presented independently of behavior is that it may elicit a response which previously had a high probability but presently is not occurring. Reid (1957) reported such an effect. He found that when pigeons, rats, or humans had been trained to make a

24

Michael D . Zeiler

response by following it with the presentation of a reinforcing stimulus, subsequent discontinuation of reinforcement deliveries eliminated the behavior. After several days of the extinction procedure, Reid delivered the previous reinforcing stimulus once without reference to behavior. The subjects immediately resumed their previously trained response. Instead of the response-independent stimulus delivery having its major effect on the behavior which immediately preceded it, the effect was on the behavior which had occurred the last time the reinforcement had appeared. Skinner ( 1938) had described somewhat similar results. Reid suggested that the stimuli provided by reinforcement delivery or by consummatory responses may have been the first part of a behavioral chain and that the free food delivery may have reinstated responding by providing the early components of the chain which had never been affected by the extinction procedure. Although this hypothesis has not received further study, the basic finding has been replicated and extended. Spradlin, Girardeau, and Horn (1966) delivered tokens to retarded adolescents after every fiftieth pull of a plunger. All of the subjects attained a high rate of responding under this fixed-ratio schedule and emitted more than 1500 responses in the last four 20-minute sessions. The schedule was then changed to a variable DRO averaging 2 minutes: A token was delivered when the subjects did not pull the plunger for periods ranging from one to four minutes. An additional degree of intermittency was added by requiring that the no-response criterion be met twice for each token delivery. This complex schedule is a second-order schedule in which the behavior under the DRO schedule is treated as a unitary response and is reinforced according to a fixed-ratio schedule: In second-order schedule notation, the schedule is F R 2 (variable DRO 2-minutes). This schedule insured that the subjects were not operating the plunger in close temporal contiguity with token deliveries. All of the subjects pulled the plunger in the period following prescntation of a token. Five of the six subjects did decrease these responses with maintained exposure to the complex DRO schedule. Even after 25 scssions, however, the remaining subject continued to pause after receiving a token, then operated the plunger at a high rate, and then paused until receipt of the next token. These data indicated that the plunger pulls were elicited by the tokens evcn though they could never precede token delivery by less than one minute. This last subject was changcd to a schedule providing tokens every 30 seconds independently of behavior (fixed-time 30-second schedule ’) . The behavior became distinctly fixed-ratio-like in that most interreinforcemcnt periods showed a pause after token presentation followed by a high response rate which usually continued until the next token appeared. This last behavior is most parsimoniously considered as maintained by adventitious

Siiperstitiorrs Behavior in Cliildren

25

correlations with token deliveries rather than as elicited by a reinforcing stimulus. Other data (e.g., Zeiler, 1970) suggest that responding maintained for long periods with D R O schedules is unusual. Typically, as with the majority of the children in the Spradlin et a[. experiment, responding stops. This suggests that the ability of a reinforcing stimulus to elicit a response which once preceded it may disappear. There is no obvious explanation for the extremely persistent responding of the one child in the Spradlin et al. study. The only obvious difference between her and the other subjects was that she was the only subject with a history of response-dependent reinforcement under a differential-reinforcement-of-low-rate ( D R L ) schedule, although it is unclear as to why or whether this history should be important. O F A N E W RESPONSE:AUTO-SHAPING B. ELICITATION

In 1968 Brown and Jenkins reported the novel finding that responseindependent but stimulus-dependent presentations of food resulted in key pecking in pigeons that had not previously been shaped to peck the key. Just presenting stimuli in certain ways elicited the key-peck, a response which previous investigators developed with the explicit differential reinforcement involved in response shaping. Their procedure was as follows. A trial consisted of an 8-second period with a white key and terminated with the response-independent delivery of food. In the intertrial interval, which varied from 30 to 9 0 seconds, the key was dark. Within 160 trials all of the pigeons pecked the key during the white period. Control procedures revealed that what was essential was that the key have some distinctive stimulus during each trial and that food presentation occur after the trial stimulus came on. Brown and Jenkins ( 1 9 6 8 ) called this phenomenon “auto-shaping.” Rachlin ( 1969) observed auto-shaping when the trial stimulus was followed by the termination of electric shock; it seems, therefore, that the emergence of pecking directed at the key is the outcome of the correlation between antecedent stimuli and the occurrence of reinforcement. N o response rcquirement is involved. Some previously unpublished data show that auto-shaping also occurs with children. Four- and five-year-old experimentally naive children were given no instructions other than to stay in a small room and d o whatever they pleascd. Experimental events were programmed and recorded automatically; however, the experimcnter also observed the children through a onc-way vision window. The front wall of the booth contained a response key and a tray into which pieces of candy could be dispensed. The key was transilluminated with either red or green light; the lights alternated from

26

Michael D . Zeiler

red to green, staying red for 30 seconds and changing to green for 6 seconds. The green light terminated simultaneously with the delivery of a piece of candy into the tray. Pressing the key at any time resulted in the immediate presentation of a piece of candy. Each child pressed the key during either the first or the second session of training, with the first press occurring somewhere in the middle of the session. One girl never pressed the key with sufficient force for the apparatus to record a response but instead touched the key on each trial. The key-manipulating behavior emerged much as it did in Brown and Jenkins’ study. In the early trials the children did not appear to be looking at the key. With additional trials they seemed to orient more toward the key and began to approach it when it became green. Finally, they touched or pressed it during the green periods. All of the pressing and touching began during green, but later occurred during the red periods as well. The essential finding, though, was that the pressing emerged from the responseindependent but stimulus-dependent presentation of candy. Of course, once the child began to press the key the dependency between pressing and candy deliveries undoubtedly was responsible for the increase in response rate. Research with infants has shown a similar finding, namely, that responses initially established to a stimuli via stimulus pairings become even more probable if they are followed by reinforcement. PapouSek (Reese & Lipsitt. 1970) rang a bell and then elicited head-turning either by tactile stimulation of the cheek or by manually turning the infant’s head. Each head turn was followed by reinforcement (milk). This combination of stimulus pairings which elicited the response and reinforcement following the response produced a substantial frequency of head turns. Williams and Williams ( 1969) demonstrated the noninstrumental nature of the auto-shaped key-peck by finding that auto-shaping occurred in pigeons even when key-pecking was negatively correlated with the presentation of food. They used the Brown and Jenkins paradigm with the added consideration that a peck during the stimulus that was correlated with food presentation cancelled the food delivery and reinstated the intertrial interval ( a DRO schedule). Here, where pecks eliminated reinforcement, the pecks occurred anyway. The peck was elicited and maintained by the stimulus dependencies when the consequences of the peck precluded food presentation. These results imply that auto-shaping may not involve operant behavior but rather an eliciting property of certain stimulus arrangements correlated with the response-independent presentations of food. The auto-shaping procedure involves the straightforward use of the Pavlovian conditioning paradigm : Two stimuli are paired without reference to behavior. The second stimulus (food) depends on the presence of a

Supersfirioics Behavior in Children

27

certain antecedent stimulus ( a red key) but is independent of any response. Auto-shaping does, however, differ from what is commonly considered as Pavlovian conditioning (and PapouSek’s procedure) in that the response which comes to occur during the first stimulus has no necessary counterpart in an unconditioned response to the reinforcing stimulus (cf. Rachlin’s experiment) and in that the response is directed toward the antecedent stimulus. Thus, auto-shaping is an effect of the Pavlovian conditioning paradigm that differs from the classical effects observed by Pavlov and later investigators. Auto-shaping is not the only behavior of this type (cf. Rescorla & Solomon, 1967).

V. Concluding Comments The preceding review of response-independent reinforcing effects showed that reinforcement has multiple influences on the behavior of children. Some types of behavior are elicited by response-independent reinforcing stimuli, other forms of behavior are strengthened because a responseindependent reinforcing stimulus follows them. It is this latter behavior that is referred to as superstition since the term “superstition” usually refers to behavior that occurs as if it influences environmental consequences but in fact does not. The emphasis on superstition, though, should not obscure the fact that reinforcement has multiple effects on behavior; it controls behavior scheduled to precede it, it controls behavior that is not scheduled but happens to precede it, it produces stimulus control over behavior both when discriminations are deliberately arranged and when they occur adventitiously, and it elicits various types of behavior. The complexity of the reinforcing event becomes even more evident when it is recognized that all of these different effects may occur simultaneously. The operant conditioning paradigm describes a dependency between a specified response and a consequent stimulus, and operant discrimination describes a dependency between an antecedent stimulus, a specified response, and a consequent stimulus. But superstitious responses occur when a reinforcing stimulus is presented without reference to either an antecedent stimulus or a response, and superstitious discriminations occur when a reinforcing stimulus occurs independently of antecedent stimuli. What superstitious behavior illustrates is that in the absence of a response specification, responses which happen to precede reinforcement increase in probability, and that in the absence of stimulus specification, the stimulus present when reinforcement occurs develops control over responding. Superstition shows that the essential process involved in operant behavior is the temporal contiguity among response and stimulus events.

28

Michael D . Zeiler

REFERENCES Antonitis, J. J. Response variability in the white rat during conditioning, extinction, and reconditioning. Journal of Experimental Psychology, 195 1, 42, 273-28 1. Appel, J. B., & Hiss, R. H. The discrimination of contingent from noncontingent reinforcement. Journal of Comparative and Physiological Psychology, 1962, 55, 37-39. Brown, P. L., & Jenkins, H. M. Auto-shaping of the pigeon’s key peck. Jorrrnal of the Experimentai Analysis of Behavior, 1968, 11, 1-8. Bruner, A., & Revusky, S. H. Collateral behavior in humans. Journal of the Experimental Analysis of Behavior, 1961, 4, 349-350. Catania, A. C. Glossary. In A. C. Catania (Ed.), Contemporary research in operant behavior. Glenview, Ill.: Scott-Foresman, 1968. Pp. 330-33 1. Catania, A. C. On the vocabulary and the grammar of behavior. Journal o f the Experimental Analysis of Behavior, 1969, 12, 845-846. Catania, A. C., & Cutts, D. Experimental control of superstitious responding in humans. Journal of the Experimental Analysis of Behavior, 1963, 6, 203-208. Dews, P. B. Free-operant behavior under conditions of delayed reinforcement. I. CRF-type schedules. Joirrnal of the E.uperimenta1 Analysis of Behavior, 1960, 3, 221-234. Ferster, C. B., & Skinner, B. F. Schedules of reinforcement. New York: Appleton, 1957. Herrnstein, R. J. Superstition: a corollary of the principles of operant conditioning. In W. K. Honig (Ed.), Operant behavior: Areas of research and application. New York: Appleton, 1966. Pp. 33-51. Herrnstein, R. J., & Morse, W. H. Some effects of response independent positive reinforcement on maintained operant behavior. Journal of Comparative and Physiological Psychology, 1957, 50, 461-467. Kelleher, R. T. Schedules of conditioned reinforcement during experimental extinction. Journal of the Experimental Analysis of Behavior, 1961, 4, 1-5. Lander, D. G. Stimulus bias in the absence of food reinforcement. Journal of the Experimental Analysis of Behavior, 1968, 11, 71 1-714. Lane, H. Operant control of vocalizing in the chicken. Journal of the Experimental Analysis of Behavior, 1961, 4, 171-177. Long, E. R., Hammack, J. T., May, F., & Campbell, B. J. Intermittent reinforcement of operant behavior in children. Journal of the Experimental Analysis o f Behavior, 1958, 1, 315-339. Morse, W. H. Intermittent reinforcement. In W. K. Honig (Ed.), Operant behavior: Areas of research and application. New York: Appleton, 1966. Pp. 52-108. Morse, W. H., & Skinner, B. F. A second type of “superstition” in the pigeon. American Journal of Psychology, 1957, 70, 308-3 1 1 . Rachlin, H. Autoshaping of key-pecking in pigeons with negative reinforcement. Jorrrnal of the Experimental Analysis of Behavior, 1969, 12, 521-531. Reese, H. W., & Lipsitt, L. P. Experimental child psychology. New York: Academic Press, 1970. Reid, R. L. The role of the reinforcer as a stimulus. British Journal of Psychology, 1957, 49, 202-209. Rescorla, R. A., & Solomon, R. L. Two-process learning theory: relationships between Pavlovian conditioning and instrumental learning. Psychological Review, 1967, 74, 151-182.

Sirperstitiotrs Behavior in Children

29

Reynolds, G. S. Behavioral contrast. Journal of the Experimental Analysis of Behavior, 1961, 4, 57-71. Reynolds, G. S. A primer of operant conditioning. Glenview, Ill.: Scott-Foresman, 1968. Skinner, B. F. The behavior of organisms. New York: Appleton, 1938. Skinner, B. F. “Superstition” in the pigeon. Joitrnal of Experimental Psychology, 1948, 38, 168-172. Spradlin, J. E., Girardeau, F. L., & Hom, G. L. Stimulus properties of reinforcement during extinction of a free operant response. Journal of Experimental Child Psychology, 1966, 4, 369-380. Staddon, J. E. R., & Simmelhag, V. L. The superstition experiment: A reexamination of its implications for the principles of adaptive behavior. Psychological Review, 1971, 78, 3-43. Webster’s new world dictionary. (College ed.) Cleveland: World, 1968. Weisberg, P., & Kennedy, D. B. Maintenance of children’s behavior by accidental schedules of reinforcement. Journal of Experimental Child Psychology, 1969, 8, 222-233. Williams, D. R., & Williams, H. Auto-maintenance in the pigeon: sustained pecking despite contingent non-reinforcement. Joitrnal of the Experimental Analysis of Behavior, 1969, 12, 5 1 1-520. Zeiler, M. D. Fixed and variable schedules of response-independent reinforcement. Journal of the Experimental Analysis of Behavior, 1968, 11, 405-414. Zeiler, M. D. Other behavior: consequences of reinforcing not responding. Joirrnal of P s y ~ h o l ~ g 1970, y, 74, 149-155.