Response Elimination Using Response-Contingent and Response-Contiguous Schedules

The experiment collected both molar and molecular data to compare the effect Differential Reinforcement of Other Behavior (DRO), Fixed Time (FT), Variable Time (VT), and Extinction (EXT) schedules of reinforcement have on decreasing the frequency of a trained response in rats. The purpose of comparing these schedules was to investigate the relationship between responses and reinforcements in order to determine whether contingency theory or contiguity theory explains the differential effects frequently reported in . the literature. Collecting data at both the molar and molecular level allowed a more conclusive theoretical explanation of the suppressive effects seen with these schedules. The experiment consisted of an eight session acquisition phase during which all animals were exposed to a FI 20 sec schedule of reinforcement. During the 15 session treatment phase the subjects were separated into ten groups consisting of a Fr 10 sec, Fr 20 sec, Fr 30 sec, EXT, VT 10 sec, VT 20 sec, VT 30 sec, DRO 10 sec, DRO 20 sec, and DRO 30 sec conditions. Finally, during the reacquisition phase all treatment groups were again exposed to a FI 20 sec schedule for 30 minutes. Molar analyses of the data showed that during the treatment phase the greatest response suppression was seen for the ORO and EXT treatment groups with only limited response elimination effects for the FT and VT treatment groups. A molar analysis of response-reinforcement interval data showed an increase for the FT and VT treatment conditions whereas a molecular interpretation of response-reinforcement intervals showed an increase in contiguity for the ORO animals. During reacquisition recovery of responding to pre-treatment levels was evident for all groups with the slowest resurgence of responding observed in the EXT , then DRO and finally the Fr and VT animals. The benefits and implications basic research has for applied settings were discussed. Acknowledgements At the end of my seemingly unending career as a graduate student, I come to realize that a multitude of others has had a significant effect on shaping my behavior as an experimental psychologist. First and foremost, Dr. Nelson Smith, my major professor for these five years, has been not only my advisor, but a good friend. His support has always been evident. Whether this consisted of academic advice on my graduate work, getting the necessary equipment to pursue my studies, or finding the necessary financial support for me to continue working in the lab, he has never let me down . I value the knowledge I have gained from him and know that my future as a researcher and professor is due to his shaping. Others that have provided the contingencies and therefore modified my behavior are Dr. Jerry Cohen and Dr. Dom Valentino. As a methodologist and statistician Dr. Cohen has provided me with the guidance to complete this dissertation. When I first started teaching psychology as a TA, Dr. Cohen put his trust in me by allowing me a great deal of freedom while at the same time providing me with feedback that now makes me confident in my teaching abilities. I have had more courses with Dom Valentino than anyone else at URI, and although he joined my dissertation committee late, it seems that he has always been there to listen to this study or some other. Dr. James Loy, though alwa:YS admittedly out of his area of expertise, has always been able to make me realize that we need to go beyond our rats. The same is true for Dr. Patricia Morokoff, who as a clinician , has made me aware that eventually these techniques will be used on humans and that therefore my conclusions had better be correct. There are fellow students with whom I have shared my office, my weekends, my ups and downs, my lunch, and many gripes and laughs. PJ Martasian and Paige DiBiasio know what it has taken to come this far. I thank them for their continued support and wish them all the best with


Abstract
The experiment collected both molar and molecular data to compare the effect Differential Reinforcement of Other Behavior (DRO), Fixed Time (FT), Variable Time (VT), and Extinction (EXT) schedules of reinforcement have on decreasing the frequency of a trained response in rats.
The purpose of comparing these schedules was to investigate the relationship between responses and reinforcements in order to determine whether contingency theory or contiguity theory explains the differential effects frequently reported in . the literature.
Collecting data at both the molar and molecular level allowed a more conclusive theoretical explanation of the suppressive effects seen with these schedules. The experiment consisted of an eight session acquisition phase during which all animals were exposed to a FI 20 sec schedule of reinforcement. During the 15 session treatment phase the subjects were separated into ten groups consisting of a Fr 10 sec, Fr 20 sec, Fr 30 sec, EXT, VT 10 sec, VT 20 sec, VT 30 sec, DRO 10 sec, DRO 20 sec, and DRO 30 sec conditions. Finally, during the reacquisition phase all treatment groups were again exposed to a FI 20 sec schedule for 30 minutes. Molar analyses of the data showed that during the treatment phase the greatest response suppression was seen for the ORO and EXT treatment groups with only limited response elimination effects for the FT and VT treatment groups.
A molar analysis of response-reinforcement interval data showed an increase for the FT and VT treatment conditions whereas a molecular interpretation of response-reinforcement intervals showed an increase in contiguity for the ORO animals. During reacquisition recovery of responding to pre-treatment levels was evident for all groups with the slowest resurgence of responding observed in the EXT , then DRO and finally the Fr and VT animals. The benefits and implications basic research has for applied settings were discussed.

Acknowledgements
At the end of my seemingly unending career as a graduate student, I come to realize that a multitude of others has had a significant effect on shaping my behavior as an experimental psychologist.
First and foremost, Dr. Nelson Smith, my major professor for these five years, has been not only my advisor, but a good friend. His support has always been evident. Whether this consisted of academic advice on my graduate work, getting the necessary equipment to pursue my studies, or finding the necessary financial support for me to continue working in the lab, he has never let me down . I value the knowledge I have gained from him and know that my future as a researcher and professor is due to his shaping.
Others that have provided the contingencies and therefore modified my behavior are Dr. Jerry Cohen and Dr. Dom Valentino. As a methodologist and statistician Dr. Cohen has provided me with the guidance to complete this dissertation. When I first started teaching psychology as a TA, Dr. Cohen put his trust in me by allowing me a great deal of freedom while at the same time providing me with feedback that now makes me confident in my teaching abilities. I have had more courses with Dom Valentino than anyone else at URI, and although he joined my dissertation committee late, it seems that he has always been there to listen to this study or some other. Dr. James Loy, though alwa:YS admittedly out of his area of expertise, has always been able to make me realize that we need to go beyond our rats. The same is true for Dr. Patricia Morokoff, who as a clinician , has made me aware that eventually these techniques will be used on humans and that therefore my conclusions had better be correct. Dr. Stuart Vyse, without whom I would never have become interested in computer programming , and without whose equipment which I inherited, has surely saved me from a few years at URI.
No graduate student could ever complete these between subjects design studies without the help of an armada of rat runners. Gina LaFauci and Debbie Cohen have run rats in my other studies so that I could spend more time on my dissertation, while at the same time patiently listening to problems unrelated to their work . Peter Russo and Kathleen Collins-Pucino ran early DRO studies while I tried to convince them that their study would be the last and, of course, a major breakthrough.
Stephanie Villari helped collect the reams of data presented in the following dissertation and checked each data point twice . She has been more help than she will ever know.
Two friends outside of URI have been of great assistance to me these past years . Dr. Lee Doerries and Dr. Jack Colby, both former students of Nelson Smith have given me the confidence to pursue what I am now doing. Jack provided me with the opportunity to teach at Providence College and indicated his trust in my ability when he hired me as a grant consultant. Lee, although further away, has been a collaborator on other .
ongoing research projects and has been the impetus for me to submit my own postdoctoral proposal. He has gone out of his way to help me while I was looking for my first faculty position .
The final person that had more to do with my reaching this point than anybody else is my wife, Sheila. Every time another exam came up she stood by me when I became a "grouch ." Every time another deadline came up she patiently reminded me, and reminded me that it was time to get going. I think that, after successfully shaping several hundred rats, she has learned that reinforcement can both increase and decrease the probability of a response . She therefore provides me with the support and IV love that one can only hope for from a spousal unit. Thank-you again Sheila , and it does get better after this.  Table 13   Table 14   Table 15   Table 16   Table 17   Table 18   Table 19 Table 20                    (Latta!, 1981) without the use of aversive stimuli.

V
The effectiveness of reinforcement has commonly been explained by either of two theoretical positions. Contingency theory focuses on the dependence of reinforcement on the occurrence of a response.
Contingency embodies three features; it is generally studied using one instrumental response , involves a short delay between response and reinforcer, and in the absence of responding no reinforcement is delivered .
In this view , reinforcement strengthens the behavior on which it is dependent. However, contingency theory is a procedural not a behavioral process . That is, contingencies must act through some mechanism.
mechanism by Contiguity theory of reinforcement focuses on that ensuring that behavior is strengthened through the response-reinforcer temporal relationship.
While the terms contiguity and contingency are -often used interchangeably they need to be differentiated . A contiguity between response and reinforcer implie s only a temporal relationship between the two .
A contingency between response and reinforcer implies a necessary causal relationship between the two. The expressions "responsedependent" and "response-independent " imply that a response is either essential or non-essential for reinforcement to occur. In this way schedules that are used to achieve a high stable state of responding such as FI and FR schedules involve a contingency as well as temporal contiguity between responding and reinforcement. If the animal does not make a response he will not be reinforced, and if a response is made then a reinforcer is delivered immediately although not necessarily for 100% of the responses. A similar type of relationship is true for ORO schedules which necessitate a longer temporal contiguity between responding and reinforcement and imply a dependency between not responding and reinforcement.
In ORO schedules then, there is a dependency between not making the targeted response and reinforcement and a delayed contiguity or none at all between responding and reinforcement. Finally. responseindependent schedules such as FT and VT allow the contiguity between response and reinforcer to vary and at the same time do not specify a dependency or contingency. The four types of appetitive schedules presented so far may be characterized by Table 1.
Contingency arrangements can be described in terms of a contingency space.
Using such a space one can depict all the possible contingency relations between responding and reinforcement. In this way t~e abscissa (the horizontal axis) in Figure 1 represents the probability of the organi sm receiving a reinforcer given that no response was made, and the ordinate (the vertical axis ) reflects the probability of reinforcement in the situation in which a reinforcer is dependent on a response . The diagonal between the two axes represent a situation in which the two probabilities are equal, that is regardless of whether the organism makes a response or not a reinforcer is delivered , with the same probability . It is therefore true that all possible contingencies fall within  Diagram of the response-reinforcer contingency space.

Probabiltiy (Reinforcer I No Response )
this space . In terms of the schedules described in Figure 1 the noncontingent schedules of FT and VT fall along the diagonal, no matter whether or not a lever response is emitted, a food pellet is delivered with equal probability. DRO schedules would fall below the diagonal, as the absence of lever responding will increase the probability of pellet delivery. The typical appetitive schedules fall above the non contingent diagonal as lever pressing will increase the probability of food delivery. The Extinction schedule (EXT) would be represented by the zero point of both axes as by definition in this type of schedule no reinforcers are delivered, so these would be neither contingent nor contiguous.
In the past, one of the ways in which stable responding was decreased or eliminated involved response dependent contingencies and aversives.
The use of punishment to decrease responding generally involves the presentation of an aversive stimulus contingent on making a response. In general, results from studies in which the presentation of punishment is contingent and immediately contiguous with responding, show the greatest decrease in responding (Azrin & Holz, 1966;Church, 1963;Church, 1969).
Punishment that is delivered independent of responding differs from reinforcement in the direction of its effect.
In studies involving response independent punishment, the effect of punishment is generally studied on some ongoing behavior. Azrin (1956) presented electric shock to pigeons on a FT schedule while the animals were responding on a VI schedule of reinforcement.
Responding was maintained but at a lower rate while under these conditions (Azrin, 1956). Other studies, directly comparing FT and FI schedules involving independent electric shock presentation, have found patterns similar to those involving independent food presentation.
These studies report that the rates of responding were lower for schedules with independent response-punishment relationships than dependent response-punishment relationships (McKeamey, 1974 ;Morse & Kelleher, 1970). Finally, in a Sidman avoidance paradigm in which responding is maintained by response contingent shock postponement, the juxtaposition of shock delivered according to an Ff schedule produced an increase in responding (Sidman, Hermstein , & Conrad, 1957 Studies that involve varying contingencies between responding and reinforcement (Herrnstein & Hineline, 1966;Sidman, et al. , 1957) have frequently been designed to address completely different issues than their differential suppressive effects . An experiment by Hermstein and Hineline (1966) using aversive conditioning in a Sidman avoidance paradigm was originally conducted in order to counter the argument that avoidance behavior can be adequately explained using the two-factor theory of avoidance learning (Mowrer, 1950). Mowrer's theory postulate s two underlying processes , a classically conditioned fear response due to a signaled shock contiguity, and an operant response that is strengthened by contingent fear reduction. Hermstein and Hineline's study was an attempt to remove the conditions under which a aversive stimuli is predictably paired with any overt stimulus. In their experiment the consequence of the subject's responding was a switch from a high to a low frequency of electrical shock. The results showed that rats were able to learn such a relationship and thereby showed that negative reinforcement can take the form of reduction in shock-frequency. These experimental results are important as they show that the relationship between response and reinforcement need not necessarily depend on contingency (Hermstein, 1969). Because the brief shocks are unpredictable and reinforcement takes the form of decreasing the average number of shocks, occasional pairings of shock and responses continue to occur and therefore this experimental procedure may been used to study response-reinforcement contiguity.
One technique for the elimination of responding that uses reinforcement is DRO.
The suppressive results of this technique have been well documented in the literature (Leitenberg, Rawson, & Mulick, 1975;Mulick, Leitenberg, & Rawson, 1976;Pacitti & Smith, 1977;Uhl, 1973;Zeiler, 1971) . The ORO procedure is best defined by the temporal parameters described by Uhl and Garcia (1969 (Rieg, Smith, & Collins-Pucino, 1988;Rieg, Smith, Russo, & Vyse, 1987;Uhl, 1973;Uhl & Garcia, 1969). Recent research has shown that the response elimination effect of DRO and EXT is transient at best and should therefore be termed "suppressive " rather than "elimination" techniques (Rieg, et al., 1988;Rieg, et al., 1987). In order to maximize reinforcement in the DRO procedure the subject is required to do anything other that the previously reinforced response, in other words the animal is required ill2.l to make the targeted response (Zeiler, 1970). What is important to note is that DRO is a _ response dependent schedule of reinforcement which at the same time allows for a change in contiguity, beyond a certain minimum delay. Extinction, on the other hand, by removing the reinforcement altogether, involves both the removal of the contiguity and contingency between responding and reinforcement, and it is therefore theoretically difficult to determine which of these variables contributes to response suppression.
Another type of situation through which response elimination has been studied involves alternate response (Alt-R) schedules. These schedules are· arranged so· that reinforcement may be available contingent on an alternate response . Generally this takes the form of making reinforcement contingent on responding on another lever, while no longer reinforcing lever presses on the original bar. This situation in which the contingency is removed from the original response and applied to an alternative response has been shown to be more effective than EXT (Leitenberg, et al., 1975;Lowry & Lachter, 1977;Pacitti, et al., 1977) and more effective than DRO (Vys e, Rieg, & Smith, 1985). Comparisons among DRO, Alt-R and EXT show Alt-R to produce the most rapid response elimination effects (Lowry, et al., 1977;Mulick, et al., 1976). All three of these schedules involve the removal of contingencies between responding and reinforcement and in the case of EXT · and Alt-R (on the original lever ) the removal of reinforcement altogether .
Schedules which deliver reinforcers independent of responding either at some fixed or variable interval of time have been designated Fixed Time (FT) and Variable Time (VT) schedules in order to distinguish them from analogous interval schedules (ie. FI & VI) which employ responsedependent reinforcers. The result of the removal of the contingency between responding and reinforcement in these schedules is that they also reduce responding. In FT schedules, reinforcers are delivered after some period of time ( t) independent of whether the subject makes a response or not. The same is true for VT schedules with the difference that a range of intervals with an average time elapse between reinforcements is employed.
The differences between response independent time schedules, response dependent interval schedules and extinction schedules is depicted spatially in Figure 1, according to their location contingency space.
In 1948 Skinner showed that food delivered to experimentally naive animals independent of · responding increased the probability of some Once responding was established subjects were exposed to a FT 11 sec schedule . It was predicted that with the removal of a deperidency between responding and reinforcement the rates should increase at first and eventually subside . However, results showed that even after extensive training on a response independent schedule, responding was maintained at significant levels.
Herrnstein (1966) explains these results as a corollary to the effects of response-dependent reinforcement. This corollary is that behavior will increase in frequency if it is followed closely by reinforcement.
Adventitious conditioning occurs because an animal is making some response when reinforcement occurs then the probability of that response is increased. If the next reinforcer occurs again sufficiently soon after the last reinforcement then th~ probability that the animal will still be making that response is high and subsequent reinforcement will again strengthen that response's probability. However, if the interval to the next reinforcement is lengthened, and because there is no specified dependency between responding and reinforcement , then the probability that the animal will be reinforced for engaging in some other behavior is increased. This will necessarily result in the reinforcement and strengthening of beha viors other than the one being measured . The implication of this analysis is that the rate of the target response will decrease and a lower rate of responding is therefore observed with response-independent than with response-dependent schedules.
A further implication of this is that contiguity and not contingency is the important factor effecting conditioning , but that contingency works by assuring contiguity.
Two theoretical explanations have been used to explain responding observed under schedules of reinforcement.
The molecular view describes responding as controlled by discrete events occurring at any given time (Baum, 1973). This mode of explanation therefore centers on measurements that can be made at the time of reinforcement, such as the quality of a response (measuring the strength of a response) or the time interval between a response and reinforcement. This view holds that temporal contiguity between response and reinforcer is central to the reinforcement effect. The molar interpretation of behavior assumes that the relationship between response rates and reinforcement rates should be computed over long periods of time and not just between individual responses and subsequent reinforcers. This orientation assumes that measures such as average response rates "cannot be assessed at any moment . . . and that order in behavior often only appears at the molar level" (Baum, 1981). The molar view then asserts that animals can discriminate among different schedules based on features other than the one-to-one correlation between responding and reinforcement. Although the two approaches to explaining behavior are not mutually exclusive , and could even be complementary , individual researchers have come to prefer one with an almost total disregard for the other (de Villiers, 1977). One of the purposes of the present experiment was to collect both molecular data and molar data on schedules and then to determine which data better conform to what we know about that schedule.
Several other researchers have investigated response dependent and independent schedules at both a molar and molecular level (Lattal, 1981;Lattal, et al., 1971;Rachlin & Baum, 1972;Zeiler, 1968;Zeiler, 1977). Rachlin and Baum (1972) used a concurrent schedule of reinforcement in which two sources of reinforcement are available simultaneously, to compare ORO and VT schedules of reinforcement which they termed as delayed and undelayed sources of reinforcement. Using pigeons on different combinations of VT 2 sec and ORO 2 sec , concurrent schedules they found that both of these schedules resulted in equivalent levels of response suppression. While this is interesting and contrary to the common expectation that necessarily delayed reinforcement would cause more suppression than randomly delayed reinforcement, they further state that in the VT schedule, at least occasionally a response must have occurred immediately preceding a response-independent reinforcer . When this temporal contiguity occurred they argue that the rate of responding would increase just as though there was a response-dependent relationship . They go on to state that when more response -independent reinforcers were delivered, the rate of responding was lower , which is again contrary to what one would expect if response-independent reinforcers were adventitiously reinforcing key pecking through contiguity because more frequent reinforcers would more frequently reinforce the target response.
In their conclusions, Rachlin and Baum assume that response-independent (VT) schedules allow for contiguous relationships between responding and reinforcement which lead to the increased rate of responding . Due to the equivalence of the VT and ORO response rates, they suggest that their findings would necessarily suggest a molar explanation of the results as a molecular explanation based on temporal contiguity would lead to the prediction of lower response rates in the ORO than the VT condition .
However, Rachlin and Baum did not collect any explicit contiguity data but relied on the logical implication that ORO involve s longer contiguities and therefore the VT schedule's equivalent response rates necessarily implies it's average underlying contiguity. These data are presented as evidence that many variables specified by a contingency need to be analyzed in order to determine the extent to which response dependencies control the rate of responding. The present study attempted to determine if response rates are associated with longer delays between response and reinforcement with DRO schedules, and higher response rates are associated with shorter delay between response and reinforcement with FT and VT schedules.
The process of studying response-independent schedules is facilitated when a particular response is under the control of the experimental situation. The alternative is for the experimenter to wait for some behavior to emerge through chance pairing of response and reinforcer (Herrnstein , 1966;Skinner, 1948). For this reason the field of response dependent research has typically utilized responding dependent on reinforcement and then subsequently studied schedules that do not have a dependency attached to them . For this reason animals in the present experiment were trained to respond on a fixed interval schedule and then various schedules were employed that allowed the differen .tiation of the effects of dependency and contiguity between responding and reinforcement. A total of four different schedules with multiple parameters that were either response-dependent or response-independent and either contiguous or non-contiguous were compared for their suppressive effects. The response-dependent schedule with a delay ed contiguity was a DRO procedure with response-reinforcement intervals of 10, 20 and 30 seconds, and a reinforcement-reinforcement interval of 20 seconds for all three delay intervals. Fixed time and variable time schedules were also run with intervals of 10, 20, and 30 seconds . The final group was exposed to a standard extinction contingency.
Previous studies comparing response-reinforcement intervals using DRO have found varying effects (Rieg, et al., 1988;Rieg, Vyse, & Smith, 1986;Vyse, et al., 1985). Generally, these studies have found that when the response-reinforcement interval is longer than the reinforcementreinforcement interval it leads to response elimination effects (Rieg, et al. , 1986) . Specifically when reinforcement is contingent on not responding greater response elimination effects are observed with responsereinforcement intervals twice the reinforcement-reinforcement interval (Rieg, et al., 1987) than response-reinforcement intervals equal to or shorter than the reinforcement-reinforcement interval.
These studies further found that, when reinforcement is again made contingent on responding that the longer the response-reinforcement interval used in DRO the slower the recovery . Also, when DRO is compared to EXT for its suppressive effects, Q.!1J.y_ when the response-reinforcement interval is longer than the reinforcement-reinforcement interval was the recovery of the original response retarded (Rieg, et al., 1986). In an effort to replicate these findings in the present study the DRO groups' delay intervals were selected so that they wou.ld be shorter, equal, and longer than the interreinforcement interval. The primary reason for the particular delay intervals for the DRO treatment groups and t intervals for the FT and VT treatment groups used in the present study was that these values had previously been shown to be effective (Latta!, et al., 1971;Rieg, et al., 1988;Uhl, et al., 1969;Wilkie, 1972 A study replicating Skinner's study on superstitious behavior was conducted by Staddon and Simmelhag (1971). This study used both FI response independent reinforcement (termed FT in the present study) and VI response independent reinforcement (termed VT here) .
Their findings indicate that animals exposed to FT schedules began emitting the targeted response later into the reinforcement-reinforcement interval than did animals exposed to the VT schedule. The implication is that if the VT animals began responding sooner after reinforcement they would have higher response rates and therefore show less response suppression. It was therefore predicted that when each of the fixed time intervals is compared to the equivalent variable time intervals, that the animals exposed to the fixed time interval schedule would show greater suppression .
Very few of the studies on reinforcement based response elimination have undertaken an analysis of the relative permanence of the treatment effects when reinforcement is again made contingent on the targeted response .
The studies that have investigated reacquisition have found the effectiveness of these response elimination techniques transitory at best with complete recovery within 60 minutes (Pacitti, et al., 1977) and 15 minutes (Vyse, et al., 1985) of reexposure to the original schedule . Data from Vyse et al. (1985) shows that within the 15 minutes differential effect of treatment have been found and that these differences were dependent on whether the animals had experienced a shorter or longer DRO delay interval. Pacitti and Smith (1977) found greater resistance to the reacquisition of original lever response (a VI 30 sec schedule) with half of their animals not responding until 36 minutes into the session.
In the present study as a measure of the each condition's response elimination durability, reinforcement was again made contingent on responding during a 30 minute reacquisition phase. It was predicted based on previous studies (Pacitti, et al., 1977;Rieg, et al., 1988;Rieg, et al., 1987;Rieg, et al., 1986;Vyse, et al., 1985), that all animals would recover to levels of the last day of acquisition within the one session. Furthermore, the DRO animals were expected to resume responding faster than the EXT animals because during reacqui si tion, the absence of the reinforcers delivered during treatment for not responding would serve as a discriminative stimulus for the changed schedule (Rieg, et al., 1988). During the treatment phase the DRO animals received reinforcement for not making lever responses. When during the reacquisition phase reinforcement is no longer delivered according to the DRO schedule, the termination of that schedule is signalled through the absence of reinforcers.
For the EXT group, if the subjects do not respond during the reacquisition phase there will be no stimulus provided that a change in contingencies has occurred.
Finally , it was predicted that if the animals exposed to the independent schedules of reinforcement showed response suppression during the treatment phase, they too would recover responding relatively soon into the reacquisition session, and they should recover more quickly than the DRO condition which had a non-response dependency associated with it during treatment.

Subjects
Subjects were 90 experimentally naive Sprague-Dawley male rats obtained from Charles River Breeding Laboratory. The subjects were housed separately and given ad libitum food and water prior to the experiment.
During the experiment all subjects were maintained at 80% of their free feeding weight.
The weights of the animals prior to experimentation ranged between 250g and 375g with a mean weight of 280g.

Apparatus
The apparatuses consisted of two Coulbourn Instruments model #E 10- Interface. Software was written by the experimenter specific to this project. Bio Serve 45-mg precision "Dustless" food pellets were used as reinforcers, and standard Purina Rat Chow (Rat, Mouse Hamster 3000) was used to supplement each animal's diet.

Procedure
Four days prior to shaping each subject was weighed and food deprived. Subjects were randomly assigned to one of the two experimental chambers. Two food pellets were placed in the food cup and the animal remained in the chamber for 10 min with the house lights on. After the second day of food deprivation and until the end of the study, each subject was returned to his home cage m the colony and was fed enough to maintain him at 80% of his free feeding weight.
Shaping In this way the animals were reinforced every 10, 20, or 30 sec dependent on which group they were in, regardless of whether they made a response or not. The same was true for the Variable Time schedule with the exception that these animal were reinforced on the average every 10, 20, or 30 sec. The extinction group received a regular extinction procedure where reinforcement was no longer presented whether a response had occurred or not.

Reacguisition.
The reacquisition phase of the experiment was run on the day following the 15 th treatment session. This phase consisted of one 30 min session during which an FI 20 sec schedule of reinforcement was in effect for all subjects.

Subject Attrition
Over the course of the experiment two subjects were eliminated due to equipment failure and four subjects were rejected for not meeting the shaping criterion within 75 min. All of these subjects were replaced with other animals so that the data from 90 subjects were used for data analysis.

General Considerations
Over the course of the experiment a total of 989,428 responses were recorded and analyzed , and a total of 239,093 reinforcers were delivered . A detailed breakdown of responding and reinforcements delivered by group and phase is given in Tables 2 and 3. These are presented to provide a global overview of response frequencies and a preliminary interpretation of the effects the experimental manipulation had.
The data indicate very stable rates of responding during the acquisition phase, with differences in response rates during the treatment and reacquisition phases. The    following statistical analyses were undertaken to separate out what differences were s_ ignificant and where those differences laid. Associated with each response and each reinforcer is the exact time into the session that each response and reinforcer occurred and from these the various interval data were computed and analyzed.

Statistical Analyses
Acquisition. For the eight session acquisition phase the dependent variables collected were the number of responses, the number of reinforcements, and the time between responses (response-response (R-R) intervals) for each animal.  (Winer, 1971). After transformation to a common log scale violations in homogeneity persisted [Em ax (10,8) = 94.204, I! > .01). However, studies by Box (1953) indicate that ANOVA is robust for violations in its underlying assumptions. Furthermore, Table 5 shows that the greate st deviations in variances occurred during the early sessions of this phase . Becau se the purpose of the acquisition phase was to establish the equivalence of responding just prior to treatment, which was found to be the case, all further data was analyzed using this common log transformed data . The log transformed data are graphically represented in     SESSION for all ten groups over the eight sessions.
The lack of significant differences for the treatment group effect establishes the equivalence of the groups' responding during this phase.
Omega squared values were computed for this design in order to determine effect sizes. For the responses made during acquisition the values for the group x session interaction was .001, .019 for treatment groups, and .216 for sessions. This indicated that while the majority of the variance was due to the interaction effect, most of the variance accounted for was due to increased response rate over the eight sessions .
The data for the mean number of reinforcements delivered by group during the acquisition phase are shown in Table 6 and Figure 4. An Em ax test on these data showed violations of homogeneity, LE.max (10,8) = 2636.599, l2. > .01] and were therefore transformed using a common log scale (Winer, 1971). These transformed data are reported in Table 7 and Figure 5. A 10 x 7 (group x sessions) ANOV A was computed on the transformed data (the ~ summary Table can      contingent and closely contiguous schedule such as FI will deliver more reinforcers with an increase in responding. Response-response interval data for the acquisition phase are reported in Table 8 and illustrated in Figure 6. An Em ax test of homogeneity of variance was found to be significant for these data,  (Winer, 1971 ). The means and standard deviations for the transformed data are represented in Table 9. All further data were analyzed using this common log transformed data, as seen in  In summary, all three dependent measures collected during the acquisition phase showed no differences between groups. There were significant differences between sessions indicating that animals in all groups acquired the lever press response during this phase and it seems that by the end of this phase all animals were responding at or approaching asymptotic rates. These results are important in that they    The means and standard deviations for the lever responses of each group during the treatment phase are presented in     group . During sess ions nine, ten, and 14 the EXT group responded more that the DRO 20 group. The data in Figure 9 show an early gradual decrease in responding over sessions, and a leveling out of the rate of responding in most groups is evident by the end of this phase . Furthermore, the FT and VT groups showed a decrease in responding which was not as great as that of the DRO animals, with the EXT group responding at times above those of the DRO animals.
The means and standard deviations for the reinforcers delivered during the treatment phase are presented in Table 12 and Figure 10.
Because the EXT group, by definition did not receive any reinforcers during this phase they were excluded from this analysis. In this case, the Em ax test of homogeneity of variance was found to be significant     The response-response interval data for the treatment phase are conveyed in Table 14 Table 15 and their means are shown in Figure 13. A 10 x 15 (treatment group x session) ANOV A was performed on these data (see Appendix F. l).     group. These data , seen in Figure 13, show variable response-response intervals for the DRO and EXT groups , yet these intervals were almost consistently above those of the FT and VT treatment groups.
The means and standard deviations for the respon se-reinforcement intervals are reported in Table I 6       and FT 20 groups. The same is true for sessions eight through 14 with the exception that during session nine and 12 and 14 the VT 30 group was significantly higher than the FT 10 group, and during session 12 and 14 the VT 10 group had longer response-reinforcement intervals than the FT 10 group and the VT 30 group was also higher than the FT 30 group. During the last session of this phase the same order was again true except that the VT 30 group had longer response-reinforcement intervals than the FT 10, FT 30, and VT 10 groups. These data, depicted in Figure 15, then show that the the ORO groups had consistently longer response-reinforcement intervals than the other groups and that during the latter two thirds of the phase the VT 30 group had greater response-reinforcement intervals than the other five non-contingent treatment groups. Figure 15 supports the finding that the DRO 30 group had consistently longer intervals between a response and the next delivered reinforcer than the other groups. Also, the two other DRO groups had longer intervals between response and reinforcement than the other groups.
Subjects were compared for the number of responses per reinforcement-reinforcement interval. These data are reported in Table 18 and graphically represented in Figure 16. Again the data were transformed using a common log transformation due to violations of       The final ANOV A performed on data collected during the treatment phase was on the reinfor ce ment-reinforcement intervals for the DRO groups.
Only these three groups were included because for the the FT groups the time between reinforcers was fixed at their particular t value, and for the VT groups the time between reinforcers was equal to the particular variable time schedule the particular group was exposed to. The means and standard deviations for the se three group s is reported in Table   20 and the means are plotted in Figure 18. for the treatment group effect, and .5 15 for the sess ions effect.
Simple effects tests (see Appendix 1.2) showed significant differences between the three DRO treatment groups for the first two sess ions only.
Tukey (hsd ) follow-up tests for the se two signific ant simple effects tests indicated longer intervals between reinfor cem ents for the DRO 30 groups for both sess ions one and two. Figure 19 shows a marked decrease for all three DRO treatment groups during the first four sessio ns and the DRO 30 group was · considerably higher than the DRO 20 and DRO 10 group s for the first two sessions of thi s pha se .     Table 23 and Figure 21. This indicated that most of the accounted for variance was due to the differences between the groups and to the increase in responding over minutes.
Simple · effects test s were performed for each minute during this phase. As previously, for each test computed, the Satterthwaite method (Winer, 1971)    The same is true for the DRO 30 groups where by the end of the treatment phase all responding was occ urring at response-reinforcer interva ls greater tha n 36 seconds. It is important to remember that for all the animals, whe n the R-R interva ls increased to values greater than the sR -SR interval, two reinforcers may have occ urred witho ut an intervening response. This implies that as response rate decreased over the sessions the relative freq uency of interreinforcement intervals without a response increased while that of the shortest response-reinforcer interval oecreased .

Leve r Res ponse Pattern Anal ysis
The following is a presentat ion of the cumulative records collected throughout this experiment. The limited selection of representative records was necessary as a tota l of 2160 records were collected during the course of the expe riment.
In general the records were selected on the basis of seve ral criteria . Each record is considered representative of response patterns observed in the animals with in that treatment conditio n unless othe rwise noted. Furthermore, the response total of any one record did not deviate more than .5 standard deviations from the mean of that represented group.
In    4 look similar, except when the animal does not make the target response during acquisition the record has no slashes and during treatment the animal continues to be reinforced. Record 4 shows the typical extinction curve (Skinner, 1938) with an initial burst of responding early in the session and longer and longer periods without a response towards the end of the session. Records 5 through 7 show response and reinforcement patterns for the three VT sc heduled groups. Particularly record 5 shows again that over the course of the session periods may occur during which no lever response is made but reinforcements are delivered. The last thre e cumulative records (records 8 through 10) repr ese nt those of animals within the three DRO conditions. When comparing these three records it is evident that with higher respon se -reinforcement interval s fewer reinforcements are achieved. This is of course due to the longer delay intervals impo sed on the schedu le when the animal respond s. An interesting observation can be made when comparing records from the animal in the EXT and that of the one from the DRO 10 group (reco rds 4 and 8). The record s are very similar . That is, initial bur sts of responding are followed by longer periods of time with no lever pre ssing occurring. The distinction is that for the EXT animal no reinforcer is ever delivered and for the . DRO animal reinfor cers are delivered every 20 seconds when no lever press occurred. When comparing all the records it is clear that the time based sc hedules maintained respo nding at · higher levels than did the contingency based and non-reinforcement ba sed sc hedule s, and this occurred within the first sessi on of this phase. However, it appears that for the se animal s two distinct pallems emerged. The first, reflected in records 2, 3, and 6, show a stable state and continuous respon se pattern. The other three records (1, 4, and 5) show stable responding with periods during which few or no lever responses occur followed again by bursts of continuous lever responding.
The cumulative records for subjects on EXT (record 4) and DRO (records 8 through 10) show that responding was suppressed with no reinforcement delivery for the EXT groups and reinforcements being delivered every 20 secs for the DRO subjects. All three DRO records have breaks in the 20 sec reinforcement interval indicating that the subjects made some responses distributed throughout the session . An interesting differentiation between FT groups, VT groups and the DRO groups, is that for the DRO treatment groups generally only one or two responses occurred followed by long periods of other behaviors; whereas the time based treatment groups with irregular response patterns showed series of non-lever responding followed by runs or bursts of responding.
Figures 34a, 34b, and 34c show representative cumulative records for subjects from each of the ten treatment groups during the reacquisition phase. It is clear from these records that by the end of this one session all the treatment groups were responding at rates similar to those at the end of the acquisition phase (sec Figures 31 a, 31 b, and 3 lc). An interesting observation is that within five reinforcements all animals were responding at !evels which they maintained until the end of the session.
However, there are noticeable differences at the beginning of this session.
All subjects showed a lower slope, or slower rate, early in the session. This is particularly true for the EXT animals (record 4) where the first response did not occur until 17 minutes into the session. · This finding supports the statistically significant results of the ANOV A (Appendix J.1) showing lower rates for this group well into the session. The DRO animals (records 8 through 10) also showed suppressed responding early in the session with recovery earlier than that of the EXT group.  The primary focus of this study was to investigate the relationship between response and reinforcer and to determine whether contiguity or contingency between these maintains behavior.
The information gathered may also be used to address the topic of operant methodology. This may be accomplished by differentiating the results obtained when collecting and analyzing the data at molar and molecular levels. Finally, if the reaso n for basic research on lower animals is to help better predict human behavior then the results should provide possible implications for the applied setting.
Several statements can be made about the general findings from the data collected in this experiment. Analysis of the acquisition phase clearly esta blished equivalent rate s of responding for all the animals in the ten treatment groups prior to the experimental manipulation of interest. Th e results from the treatment phase showed that all of the schedules used in the present study lead to response suppression. However, a comparison between the contingent treatment groups (DRO) with non-contingent treatment groups (FT and VT) in terms of overall responding showed that providing for a contingency between not making the targeted response and reinforcement , greatly effects the subject s' responding (see Figure 9).
At the same time the mean interval between responses increased for all the groups with those of the DRO animals becoming variable (see Figur e 14).
Furthermore , the time between the last respon se and the next scheduled reinforcer increased for all of the animals, with the change in the contingent delay treatment groups showing less of an increase (see Figure   15). Finally, data from the reacquisition phase when reinforcement is again made contingent on responding, indicated that the EXT animals were the slowest to recover, followed by the DRO subjects, and finally the FT and VT animals which showed very little duration of suppression during the reacquisition phase (see Figure 21).
The results from the treatment phase do not entirely support findings from previous studies which had shown that the longer the responsereinforcement interval the greater the degree of response suppression during treatment (Rieg, et al., 1987;Rieg, et al., 1986). These papers had indicated that the sequence from shorter to longer delay intervals in the DRO condition also produced lesser to greater response elimination during treatment.
The data depicted in Figure 9 do not support this conclusion.
That is, the greatest amount of response suppression was shown in the DRO 20 group during the middle half of the treatment phase. This group had a response-reinforcement interval equal to the reinforcementreinforcement interval. Furthermore, previous results had indicated that for DRO schedules, when the response-reinforcement interval was shorter than the reinforcement-reinforcement interval, less response suppression was observed when compared to an EXT schedule. This was not true in the present study. However, the particular DRO procedure used in the present study differed slightly than the one defined by Uhl and Garcia (1969).
Previously, when the delay interval was shorter than the interreinforcement interval then responding up to the reinforcementreinforcement interval minus t~e response-reinforcement interval had no effect on setting up a delay for an animal. For example, in the Rieg et al. study (1986) when a subject in the DRO 2 group responded up to eight seconds after the last delivered reinforcer (with reinforcementreinforcement intervals of 10 sec) no delay of reinforcement was incurred.
Only during the last two seconds prior to the next scheduled reinforcement was a delay incurred. The net effect of this procedure was that the subject was allowed to continue to respond during these eight seconds without setting the occasion for the next reinforcer to be delayed. The differences between previous studies and the present study is that each response during ·the reinforcement-reinforcement interval had a contingency associated with it. That is, no matter when the response was made within the reinforcement-reinforcement interval, the delay interval was juxtaposed onto the inter-reinforcement interval.
In the Rieg et al. studies (1986; where the response-reinforcement interval was shorter than the reinforcement-reinforcement interval any response up to the delay value had no contingency associated with it. Therefore, the present procedure which made each response during the inter-reinforcement interval contingent on delaying reinforcement, resulted in greater suppression for all the DRO groups compared to the EXT groups. It was probably due to the fact that a greater proportion of responding was paired with delays that greater suppressive effects were observed in the DRO treatment groups than the EXT treatment groups. The implication of this differentiation between the procedure used here and that used previously was that when the contingency between response and reinforcement was consistent, ie. every response delayed the next scheduled reinforcer, greater response suppression was observed than when only partial pairing between response and reinforcement delay was in effect. Therefore, the particular findings of this experiment compared to earlier studies may be explained through a contingency explanation alone without the neces sity of contiguity arguments. The animals in all the FT and VT treatment groups decreased their target response rates slightly yet significantly. As stated above the sequencing effect of response suppression based on the length of the DRO interval was not as predicted for the DRO animals, Figure 9 shows that for the time based schedules the longer the period between reinforcers the less the respon se suppression. The question becomes , of course, what mechanism produces the lower response rates. That is, why does some "other" response besides lever pressing get started. An obvious explanation might be the adventitious reinforcement of what has been described as the stereotypical postreinforcement pause (Ferster, et al. , 1957). That is, subjects exposed to most temporal schedules of reinforcement exhibit behaviors such as grooming, investigating, and consummatory behavior just after reinforcement. Generally, it has been found that this postreinforcement pause is one-half to about two-thirds the average duration of the inter-reinforcement · interval (Schneider, 1969;Staddon & Simmelhag, 1971 ). These findings hold true for interreinforcement intervals between 16 and 512 seconds (Innis, 1981).
Furthermore, studies inve stigating the delay of reinforcement, clearly show that the longer the interval between response and reinforcement the less the conditioning (de Villiers, 1977;Rachlin, 1989). In the present study, if we assume that behaviors during the postreinforcement pause lasted up to seven, 14 and 2 I seconds for the three temporal parameters of the non-contingent treatment groups (ie . Ff and VT intervals of 10, 20, and 30 seco nd s), then ~he delay between non-lever responding and reinforcement is much shorter for the 10 second FT and VT groups. This argument would be true if the data had shown greater response suppressio~ for animals with shorter inter-reinforcement intervals than longer inter-reinforcement intervals. However, the analysis on response data do not show this to be true (see Appendix D.1 and subsequent follow-up tests).
The implication of this is that contiguity between non-lever responding and reinforcement may not account for the suppressive effects observed in animals exposed to non-contingent sc hedules of reinforcement.
Skinner's early demonstration of superstition (1948) was explained in terms of co nditionin g merely through temporal terms. He suggested that behaviors could be conditioned through temporal pairings alone, and this would happen without a contingency between response and reinforcement (1948, p168) . In the present study, animals with lower values of t and therefore with significantly more exposures to reinforcement (see Figure   11) and llil. associa ted contingency showe d the greatest amount of non target re sponding, ie. greater suppression of the target respon se.
According to Skinner 's early explanation then, animals with more pairing s between some response other than lever pressing and reinforcement, compared to animals with fewer non-conting e nt re spon se-reinforcement pairings, should be more likely to acquire the other response . The rea son of course being the greater number of pairings between non lever press respo nding and reinforcem e nt. One has to rem e mber that whil e significant, ther e was not a lot of suppre ssio n in any of the FT and VT treatment gro up s.
An implication of thi s idea that greater respo nse suppression for the 10 sec FT and VT trea tment groups was due to greater densitie s of reinforcement for the FT 10 and VT 10 groups is that if the animals in the FT 20, FT 30, VT 20, and VT 30 tre atment gro ups had been run for an additional week or two so that the tota l numb e r of reinforcers were equal to the FT and VT 10 sec treatment groups, they would have leve ls of response suppre ss ion equivalent to the FT 10 and VT 10 treat ment group s. One way to assess thi s idea would be to run the anim als each day so th at the y rece ived equal number s of reinforcers durin g eac h session. This was not do ne in the prese nt study, because each sess ion lasted 30 minute s, and therefore the total rei nfo rcers rece ived va ried as a function of the schedul e the anim als were on (see Table 12). It is possible though to look at the last da y of the treatment phase for the t = 30 sec FT and VT animal s and co mpare those response data with response data from the t = IO sec FT and VT animals when the number of reinforcements were equal.
At  Table 10 for the number of responses made per session). In both cases with a larger parameter of t, but equivalent numbers of reinforcers delivered, higher response rates were observed. The implication of this analysis is that the increased suppressive effects of shorter reinforcement-reinforcement intervals · is not due to an increase in the number of temporal pairings between non-lever responding and reinforcement.
That is the animals in the Fr and VT treatment groups were not learning a temporal contiguity between responses other than the target response and reinforcement. The results here are then empirical arguments against the notion of superstitious conditioning based on adventitious reinforcement and its emphasis on response-reinforcer temporal contiguity as the reinforcement effect.
An interesting finding of the present study is that all three ORO groups modified their responding in such a way so that by the third session they were receiving as many reinforcers as the FT 20 and VT 20 groups and were earning close to the maximum number of reinforcers possible during any one session. The implication of this observation is that behavioral maximization had taken place. Maximization theory postulates that an animal given a choice between two alternatives will choose among them so as to maximize reinforcement.
Furthermore, given only one targeted response, maximization occurs when the level of responding maximizes the probability of reinforcement (Catania, 1984). For the ORO animals, in order to maximize their rate of reinforcement they had to make any non-target response. This occurred within just three sessions in the present study.
Herrnstein (1961) demonstrated that when reinforcement is available on two separate schedules of reinforcement animals responded so that the relative response rates on the two schedules equalled the relative reinforcement frequency of the two alternatives . This relationship was quantified by Herrnstein m 1970, and has been shown to hold in different types of concurrent schedules across several species of animals (de Villiers, 1977).
In the research literature on schedules, one of the most debated issues is the relationship between matching and maximization (Commons, Mazur, Nevin, & Rachlin, 1987;de Villiers, 1977). Some authors have argued that maximization is fundamental in that matching will only occur when it conforms to maximization (Shimp, 1969) and others have postulated the opposite (Herrnstein, 1961;Herrnstein, 1970). Some authors have stated that matching can only be studied if two responses are measured and one can estimate the relative · frequencies of each (Catania, 1984). However, it is not impossible to conceive of animals matching based on the amount of time they spend on one schedule of reinforcement (Baum & Rachlin, 1969;de Villiers, 1977;Herrn _ stein, 1971). Baum and Rachlin (1969) studied matching which could only be measured in terms of the time allocated to each response, that of standing in one specific location . Their results indicated that the ratio of time spent in one specific location varied as a proportion of the reinforcement ratios.
The argument can therefore be made that with standard lever responding the time spent responding determines the number of reinforcements earned.
The data collected in the present experiment can then be said to conform to the matching law. It is clear that the animals in the DRO conditions responded so as to match their responding to the available reinforcement rate. Furthermore, by spending time emitting other responses than lever pressing these subjects were able to maximize their rates of reinforcement because responding on the lever actually decreased the density of reinforcement (to zero if an additional response occurred within the response-reinforcement interval).
The data comparing response-response intervals showed that the animals exposed to the ORO contingency had longer inter-response intervals than those exposed to the time based schedules.
In general ,-with the exception of three sessions the animals in the ORO 30 group had the longest times between responses (see figure 12 for absolute responseresponse intervals) . Furthermore, the transformed data depicted in Figure   13 show that the most dramatic increase in response-response times occurred within the first half of the treatment phase, especially for the ORO and EXT treatment groups. When one compares this observation with cumulative records collected for these animals (see Figures 32, Cumulative Records 8, 9, and 10), we see bursts of responding where the time allocation of responding is clearly distributed between "other" behaviors and lever pressing. These data suggest that the DRO animals were responding in such a way as to maximize their reinforcement rates.
The data comparing response-response intervals also showed that with the exception of the FT 10 group, all other non-contingent treatment groups showed only a slight increase in their times between responses, and that these intervals were relatively consistent throughout the session (see Figure 13). The question whether matching or maximizing was occurring or not is more difficult if not impossible to apply to non-contingent schedules.
That is, with these schedules, either responding or not responding yielded the same amount of reinforcement, and it is therefore difficult to assay if the animals are responding in order to maximize reinforcement. The question then results in a descriptive analysis of what the animals were doing when the schedule was in effect.
The group data show a great deal of variability during this phase. Some of the animals in each of the Ff and VT groups were . maintaining their response rates from the acquisition phase, hence shorter response-response intervals, while at the same time some of the animals were emitting non-lever response behaviors.
When one looks at the raw data for lever responding it is evident that some of the animals dramatically reduced their response rates and some increased their rates of responding. The raw data and cumulative records , then, indicate that the Ff and VT schedules produced two types of responding in the animals, one group that increased their response rates , and one that decreased their response rates.
When one contrasts the cumulative records of the animals whose response rates decreased over the course of the treatment phase with those of the DRO animals the dramatic drop in responding is evident for the DRO animals only (see Figure 8). The FT and VT animals decrease their responding much more gradually (see Figure 35, Cumulative Records 2 and 4) than the DRO animals (see Figure 32, Cumulative Records 8 through 10).
By the end of the treatment phase, almost half the animals in the FT and one third the animals in the VT treatment condition had cumulative records similar to those of Figure 35, Cumulative Record 1. These response patterns are not unlike those of the DRO treatment groups (see Figure 33, Cumulative Records 8 through 10). The only difference was that when the ORO animals made a response a delay to the next scheduled reinforcer was incurred. This is evidenced by a break in the downward slashes of the reinforcement pen . This then sugge sts that the differences in these two patterns of behavior are the result of the accidental contiguity of the "other" behavior in some of the animals and lever pressing in others.
One additional pattern of outcome warrants comment in light of maximization analysis and because of the fixed nature of the scheduled reinforcement delivery . The analys is comparing groups in terms of the number of responses made between two successive reinforcers is interesting.
The sequencing pattern here is identical to that of the total number of responses made by each group during the treatment phase .
That is, the longer the Ff and VT parameter the higher the number of responses per reinforcement-reinforcement interval.
For the ORO subjects the same is true early during the phase until response rates had decreased to such an extent that only very few responses were occurring between reinforcements (see Figure 16). The prediction based on the findings by others (Staddon, et al., 1971) that the VT treatment groups would show less suppres sion than FT treatment groups was supported by the data. Staddon and Simmelhag (19 7 1) report perseverance of "interim" activities for animal with fixed reinforcement-reinforcement intervals ver ses earlier resumption of "terminal" activitie s for animals scheduled with variable reinforcementreinforcement interval s . In their study, interim activities were defined as those related to adjunctive behavior generally occurring just after food delivery, and terminal activitie s were de scribed as discriminated operants occur ring toward the end of the reinfor cement-reinforcement interval and continuing until food delivery (generall y lever responding or key pecking).
In the present study, where only terminal activities were measured, the animals with fixed value s of t showed greater suppression of responding. Molecular analysis of cumulative records collected during the treatment phase support this finding. Figure 35, Cumulative Record 2, shows that a scalloped effect characteristic of FI responding (Ferster, et al., 1957) is evidenced by the FT 20 animal. Just after reinforcement , no responses occurred, and toward the end of the reinforcementreinforcement interval the probability of responding increased .
Furthermore, comparison of FT and VT groups in terms of their responsereinforcement intervals showed that the animals with fixed interreinforcement intervals, with more predictability as to when the next reinforcer would be delivered, had shorter response-reinforcement intervals . This finding raises the possibility that the animals might have been using temporal cues as the occurrence of the next reinforcer . And if the temporal relationship between response and reinforcement is more predictable , as is the case in the FT animals, then greater response suppression is to be expected in tho se treatment groups over the variable time treatment groups. This finding is indeed supported by the response suppression data (see Figure 9). Finally , if this type of temporal discrimination was indeed being learned by both the FT and VT animals then the fact some animals continued to respond throughout the treatment The comparison between contingent (ORO groups) and noncontingent (FT and VT groups) treatment groups in terms of overall responding supported the prediction by showing that providing for a contingency of not making th_ e targeted response and reinforcement greatly effects the subject's responding.
That is, the DRO groups showed dramatically greater suppression during the treatment phase than either the Ff or VT groups (see Figure 8 for absolute rates of responding during the treatment phase).
The particular findings of this study comparing the animals exposed to the contingent schedules with non-contingent schedules do not support those of Rachlin and Baum (1972).
In that study the data indicated that responding was equivalent in animals exposed either to a DRO contingency or VT schedule. However, others (Henton, et al., 1978;Zeiler, 1976) comparing DRO and VT using conjoint schedules showed that the DRO schedule reduced the probability of a lever press more than the VT schedule. The present study is the first to replicate these findings using a between subjects design and only with one schedule in effect at a time. Looking at the molar analysis of lever responding (see Appendix D.1) we can conclude that the removal of a response-reinforcer dependency maintains behavior and the change to a dependency involving not responding and reinforcement quickly reduces behavior .
The molar data that was collected to directly assess the responsereinforcement contiguity was the one comparing the responsereinforcement intervals for the different treatment conditions. The animals with no contingency showed that initially they were responding within seconds of the next reinforcer and the interval between responding and reinforcement continued to grow throughout . the treatment phase (see Figure 15). The results for animals with a specified contingency of not making th e target response and reinforcement showed that the interval from response to reinforcer was maintained . This finding is also supportive of a contiguity analysis. The animals in the Ff and VT conditions had short temporal pairings between response and reinforcement. That is, if they continued to respond, the interval between response and reinforcement could have been ve ry short as reinforcement was delivered independent of responding. Howev er, animals in the DRO condition had response-reinforcement intervals of at least 10 seconds for the ORO 10 treatment group and at least 20 seconds for the ORO 20 and at least 30 seconds for the ORO 30 treatment groups. Therefore, for the ORO animals the interval or contiguity between response and reinforcement was necessarily extended. For the ORO animals the behaviors that were most contiguous with reinforcement were necessarily "other" behaviors than lever pressing.
Because this experiment used a between groups design, did not wait for stable states to emerge, and collected predominantly molar dependent measures, the conventional studies comparing response-dependent and response-independent schedules may arguably not be the most adequate for comparison to the present results (Gamzu & Schwartz, 1973;Latta!, et al., 1971;Zeiler, 1977). However, there exists a body of literature that has focused on the transition of response-dependent to response-independent schedules (Catania & Keller, 1981;Hermstein, et al., 1966;Imam & Latta!, 1988;Lachter, 1971;Lattal, 1972;Zeiler & Solano, 1982). A criticism to studying just those transitions from response-produced to responseindependent reinforcement is that just after the transition, responding will continue to be maintained due to the contiguities between responses and reinforcers. That is, if contiguity is what maintains animals responding and the animals continue to respond on a response independent schedule they will experience sufficient adventitious pairings of response and reinforcer to maintain responding. The procedures used in the present study were designed to assess this argument. This was achieved by directly comparing time schedules with no contingency and no experimenter-specified contiguity, with ORO schedules with a contingency and a fixed minimum or artificial contiguity . The molar analysis of total response rates showed that this had a dramatic effect with greater response suppression seen in the ORO and EXT groups (see Figure   9). showed that the time between response and reinforcer remained relatively constant over this phase for the ORO treatment groups (see Figure 15). By definition of the parameters of the ORO schedule the response-reinforcer interval for the ORO treatment groups will necessarily be greater than the non-contingent time based treatment groups. What is evident in Figure 15 is that the interval from response to reinforcer increased for the Ff and VT groups through the treatment phase. Furthermore, the figure shows that the interval was very short for these time based treatment groups.
The molecular data collected throughout the treatment phase did not indicate the same pattern as the molar analysis. The responsereinforcement bin analysis did show that the interval between lever press and food delivery gets longer over the 15 session treatment phase (see  Figure 30 (data from the DRO 30 treatment group), where no minimal responsereinforcement intervals were recorded for the latter sessions and during sessions 10 and 15 most of the responding was occurring more than 40 seconds before the next scheduled reinforcer . Of course, by definition, no response-reinforcement interval of less than 30 seconds was possible in this group.
The differences between molar and molecular analysis findings invite further comment. One reason that has been proposed for the greater suppressive effects of the DRO over the Ff and VT schedules is that the contiguity between lever response and reinforcement for the DRO animals is exaggerated.
In addition, the only short temporal relations possible would reinforce the "other" behavior m the DRO animals. Thus the level of analysis focusing on contiguity can best be explained using a molecular analysis . The other reason that has been presented as an account for the greatly suppressive effects of the DRO animals involves a contingency analysis.
For the DRO animals, every response resets the interval to the next scheduled reinforcer. That is when the DRO animals decreased their mean rates of responding, they necessarily increased their rate of reinforcement.
The data then, collected on schedules with a delayed contiguity and specified contingency between responding and reinforcement, conform best to a molar analysis. This finding leaves the researcher in a quandary as to what type of data to collect, and once the data is obtained, what is the best way to analyze that data. Molar measures tend to focus on data collected over longer periods of time rather than smaller segments as the molecular orientation does . It seems that neither molar nor molecular approaches to understanding behavior are better, but that the advantage of one cannot be determined without an understanding about how a particular process operates (Zeiler, 1989). If different processes of behavior operate at different time spans then understanding these dictates the analysis of choice. That is, for the present experiment, invt?stigating contiguity would best be achieved using molecular response bin analysis, while determinants of contingency will be best ascertained using group means on response -reinforcer intervals.
The analyses from the acquisition phase show that all groups were responding at equivalent rates prior to exposure to the treatment condition .
This finding was necessary to establish in order to draw conclusions from the effectiveness of each treatment condition. Several researchers have argued that at least initially, the particular dependency the subject was exposed to would effect response patterns when that dependency was removed (Latta!, 1972;Rescorla & Skucy, 1969;Rieg, et al., 1987;Zeiler, 1968). That is, the baseline schedule during an acquisition phase will determine the degree to which responding will be effected during a treatment phase. Latta! (1972) presents the observation that with an FI schedule the response rate is much higher than a VI schedule of reinforcement.
The effectiveness of a subsequent non-contingent schedule in reducing responding, will be greater for those previously exposed to the VI schedule. The argument is that if animals are emitting behaviors other than lever responding prior to treatment then once response independent reinforcement is initiated they will have a higher probability of adventitious pairing of non-lever behaviors and reinforcement. The present study clearly established the equivalence of response rates prior to exposure to the experimental manipulations of interest.
In the present study when animals exposed to a schedule of reinforcement that is both contiguous and contingent (ie. FI schedule) and are subsequently exposed to response-independent schedules, with no contingency or contiguity (EXT schedule) or no contingency and a nonspecified contiguity (Time schedules), then the removal of a contingency between response and reinforcer may allow for the maintenance of responding . That is, the effectiveness of EXT and DRO over both Fr and VT as response elimination techniques may be due to the high response probability schedule used during the acquisition phase. If a schedule had been used with a lower rate of reinforcement or one that generates slower rates of responding the suppressive effects for the FT and VT groups might have been enhanced.
The analysis comparing lever responding between treatment groups during the reacquisition phase establishes the temporary effects of these response elimination procedures.
For all the groups responding at the end of the treatment phase had been depressed, but when reinforcement was again made contingent on responding, resurgence of responding was noticed.
Throughout the first seven minutes those animals that had been exposed to the non-contingent schedules tended to recover faster than those exposed to the contingent schedules. When responding began for individual animals their rates quickly became similar to those of the last day of the acquisition phase. This supports previous assertions that EXT and ORO procedures do not "eliminate" responding but rather "suppress" responding (Boe & Church, 1967;Rieg, et · al., 1986). Again, the point should be made that these techniques should not be labeled response elimination procedures but response suppression techniques.
It is interesting that the particular pattern of recovery observed when comparing the ORO and EXT treatment groups is dissimilar to those of previous studies (Rieg, et al., 1988;Rieg, et al., 1986). These studies had found that ORO animals whose response-reinforcement interval were shorter than the reinforcement-reinforcement interval recovered as fast or faster than EXT animals, and that when the response-reinforcement was greater than the reinforcement-reinforcement interval recovery was slower than EXT animals. Although this pattern was true for the first three minutes of this phase, this was not the case for minutes five through 18 (see Figure 21). The difference between the present study and tho se reported earlier is that the length of each session was twice those of the previous studies and the treatment phase consisted of an additional five sessions. A frequent criticism of our previous findings was that the treatment phase was too short, having been five or ten, 15 minute session s.
In the present study the session lengths were doubled and extended to fifteen days. For the ORO and EXT treatment groups differences were found for the first four of the five minute intervals . The data indicated that at least initially the EXT and ORO groups were responding at lower levels during reacquisition than during acquisit ion. After 15 minute s the se tre at ment groups were responding at the · sa me level that they had been prior to treatment. These analyses, in conjunction with the minute by minut e analysis de scribed above, have sev eral implications for the ongoing effectiveness of re spon se elimination procedures. One way to view the reacquisition phase is to interpret it as a test of the effectiveness of the treatment conditions. In this way, animals exposed to respon se independent reinforcement during treatment show no suppression of re sponding when the dependence bet ween response and reinforcer is again initiated . At the same time, behavior for animals whose depen dency was maintained, yet for which the temporality between targeted response and reinforcer was lengthened, show greatly reduced levels of responding when the targeted behavior was again made temporally contiguous with reinforcement.
In the present study not only were the DRO and EXT treatment groups responding less than the FT and VT groups during reacquisition, but they also responded less during the early part of the session compared to the acquisition session.
Previously, researchers have argued that the increased effectiveness of DRO when compared to EXT is due to the fact that subjects learn not to emit the target response and that the response alternate to responding is strengthened, and that during EXT response suppression without conditioning of other behaviors occurs (Uhl, et al., 1969;Zeiler, 1971 That is for the DRO animals an appreciable change had occurred. The phenomenon of "resurgence" will then explain their recovery of responding (Epstein, 1983;Epstein & Skinner, 1980). The observation is that when recently reinforced behavior is no longer reinforced, that behaviors previously reinforced will tend to occur. Therefore, the early recovery of the DRO animals may be attrib~table to the resurgence of the previously reinforced response. The same argument may explain the recovery of the FT and VT subjects. While their response rates were not as dramatically suppressed as those of the contingent groups, they too showed resurgence of lever responding. That is, for the animals that had decreased their lever responding and were emitting other behaviors, the changes in the schedule from · free reinforcement to contingent reinforcement tended to produce resurgence of lever responding.
There are several implications that this research project has for response suppression when used in applied settings. Most general is the importance basic research has for applied applications. It is inevitable that the plethora of research comparing treatment conditions will produce a series of different solutions. The que stio n then becomes if two different studies comparing the same treatment procedures yield different findings, which is the most valid, and more importantly, what are the causes for those particular differences . This argument has been raised before by Sidman (1960). Furthermore , is the issue of the particular parameters employed when comparing different treatment conditions (Va n Houten, 1987). For example, unless the optimal temporal parameter of a schedule is employed, conclusions about the efficacy of any one treatment over another should be regarded with skepticism.
There is no doubt that the operant chamber with its accurately controlled environment and response alternatives allows for a setting in which the relations between response and reinforcer can empirically be manipulated and studied. There is very little doubt that the experimental laboratory is a contrived setting. That is, the experimenter has the maximum control in setting up the occasion for a reinforcer, and also specifying the exact temporal relationship between response and reinforcer.
In the applied setting excessive behaviors generally involve a long chain of behaviors and the targeted response may not be as discrete as a switch closure due to a lever press. For example, O'Neil, White, King, and Carek (1979) decreased the probability of rumination using both EXT and DRO procedures. However, rumination involves a long chain of behaviors culminating in the ejection or reswallowing of food. As O'Neil et al. ( 1979) point out treating different parts of the behavioral chain will have differential effect. It is therefore difficult to assess the exact contingency between reinforcement and any one of the ongoing behaviors. Furthermore, because the behaviors are internalized it may be even more difficult to assess the temporal relationship between response and reinforcer. It is for these three reasons that experimental laboratory research on schedule parameters can provide the greatest benefit for applied researchers.
The effectiveness of extinction for the suppression of responding was clearly established in the treatment phase and especially during the reacquisition phase. _ At the same time, schedules which involve a contingency between not responding and reinforcement are also effective in reducing responding . A comparison between EXT and DRO showed only slight differences during treatment. While some studies comparing DRO and EXT have shown DRO to be more effective than EXT (O'Neil et al., 1979), others have found the opposite to be true (Redd, 1986), while still others have shown no differences between DRO and EXT (Heidorn & Jensen , 1984).
The comparison between contingent and non-contingent schedules from the present study also have implications for applied settings. In the present study providing for a contingency between not responding and reinforcement (DRO) was more effective in reducing unwanted behaviors than no contingency between response and reinforcer (FT and VT).
In the applied settings comparisons between contingent and non-contingent reinforcement generally involve the implementation of a contingent relationship where previously there had been none. That is, these are generally A-B type designs where because of ethical · considerations a reinstitution of pre-treatment conditions is never implemented. The results from the present study are indeed supportive of the findings (George & Hopkins, 1989;Johnson, McGlynn, & Topping, 1973). It seems clear that as a response suppression technique, ORO is clearly superior to both FT and VT schedules . At the same time, if exposure to the treatment phase is sufficiently long then EXT should the the treatment of choice .
In addition to conclusions about differences between particular schedule . over another this project has furnished considerable information about the underlying processes involved in learning. First is the compari son between the four types of schedules compared. The data showed that both ORO and EXT schedules were more effective at reducing responding than were either FT and VT schedules. As a measure of the durability of the response suppression procedures, results from the reacquisition phase indicated that EXT followed by DRO produced the slowe st recovery to pre -treatment levels .
Second is the comparison between the underlying processe s hypothe sized to be the contributing factors for reinforcement theory. Generally, it might be said that merely removing a contingency between an established response and its reinforcer will not greatly effect the behavior .
In the present study the response suppres sion was not as great for the two time based schedules as for ORO or EXT .
However, a second contributin g factor in addition to the contin ge ncy between not responding and reinforcement , it may be argued, for the effectiveness of DRO, was the lengthening of the interval between the targeted response and reinforcement. It must therefore be concluded that a combination of contingency between not responding and reinforcement concurrent with and longer response reinforcer contiguity will greatly enhance the effectiveness of applied response suppression techniques.
Finally, the differential effectiveness of the treatment conditions needs to be assessed in terms of the level of data collection and analysis. The present analyses showed different effects for molar and molecular interpretations especially for the response-reinforcer interval data . Some important implications that this study has for the applied literature is that both contingency and contiguity are important aspects of reinforcement and should be considered in determining treatment programs.
Furthermore, time based schedules are poor and at least unpredictable methods for reducing behavior. Also, DRO does not seem to suppress behavior as effectively as EXT if the treatment program is extended over long periods of time. At the same time, for short treatment programs, DRO may be more effective. Finally , all three techniques used here must be considered "suppression" techniques and not elimination techniques, because as treatment is discontinued and reinforcement is again made contingent and contiguous with responding the behavior will recover to its original levels very rapidly.
de Villiers, P. (1977). Choice in concurrent schedules and a quantitative fonnulation of the law of effect. In W.