THE DIFFERENTIAL REINFORCEMENTAOF ' ' LOW RATES AND STIMULUS CONTROL, ‘ Thesis for the Degree of Ph. D. MICHIGAN STATE UNIVERSITY THOMAS J. KRAMER 1970 W“ This is to certify that the thesis entitled The Differential Reinforcement of Low Rates and Stimulus Control presented by Thomas J. Kramer has been accepted towards fulfillment of the requirements for 2h 1 D 1 degree iwy ABSTRACT THE DIFFERENTIAL REINFORCEMENT OF LOW RATES AND STIMULUS CONTROL BY Thomas J. Kramer Research in the past several years has suggested that stimuli associated with some positive reinforcement schedules may possess some of the properties normally at- tributed to stimuli associated with extinction, such as an inhibitory generalization gradient around the training stimulus. One important question is whether this inhi- bition is the result of discrimination training or the result of the contingencies of the reinforcement schedule itself. The purpose of Experiment I was to investigate the effect of the presence or absence of discrimination training on the shape of the generalization gradient around a stimu- lus correlated with a differential-reinforcement-of—low-rate schedule (DRL), a schedule which requires the subject to pause a certain minimum time between responses in order to obtain reinforcement. Eighteen White Carneaux pigeons were trained to peck during a successive discrimination under a multiple DRL 8-sec - DRL 8-sec schedule for 20 days (Phase Thomas J. Kramer I). Each unreinforced response was followed by darkening the response key for 0.8 sec in order to provide feedback for each criterion response. A plain green key was corre- lated with the first component and a white vertical line on a green background with the second component. After 20 days the subjects were divided into three groups of six each (Phase II). During Phase II the schedule was changed to multiple DRL 8-sec - DRL 64-sec for one group, multiple DRL 8-sec - Ext for the second, and DRL 64-sec only for the third. The DRL 8-sec component was correlated with the plain green key, and either the DRL 64-sec or Ext component with the white vertical line on the green background. After 15 days a standard generalization test in extinction was administered to all subjects along the line-orientation dimension, the dimension correlated with the white vertical line. The results showed that the generalization gradient for the DRL 64-sec group was inhibitory in shape, while the gradients for the other two groups were not. These results indicate that standard discrimination training (1) is not always necessary for the production of inhibitory stimulus control, and (2) discrimination training does not always lead to inhibitory stimulus control around the stimulus which controls the lower response rate. The purpose of Experiment II was to assess the ef- fect of removal of the feedback for each unreinforced Thomas J. Kramer response on the generalization gradients and on the rate of responding in the presence of the DRL 8-sec correlated stimulus. The procedure was identical to Experiment I ex- cept that no feedback was provided for each unreinforced response. The results of Experiment II showed that none of the group gradients along the line-orientation dimension were inhibitory. Comparing the response rate data of Experiment II with that of Experiment I, the subjects main- tained on either the multiple DRL 8-sec - DRL 64-sec or the multiple DRL 8-sec - Ext schedule showed a higher rate of responding in Experiment II (no feedback) during the DRL 8-sec component than the comparable subjects in Experiment I, and also greater variability in response rates. In addition, the subjects in Experiment II showed an increase in the rate of responding (behavioral contrast) during Phase II while the subjects in Experiment I did not. Almost all the difference in the rate of responding between the feedback vs. no-feedback subjects was accounted for by the difference in the per cent of responses with interresponse times less than 0.8 sec. Dated“? Z‘, (220 Signed THE DIFFERENTIAL REINFORCEMENT OF LOW RATES AND STIMULUS CONTROL by .'!"' 1 Thomas J? Kramer A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Psychology 1970 DEDICATION To Nancy 2+2=5 ii ACKNOWLEDGMENTS Almost all thesis chairmen deserve the sincere thanks and appreciation of their students. In the case of Dr. Mark Rilling, however, I feel that something more is deserved. He has not only been an excellent advisor, but also a thoughtful and considerate person, both within and apart from the advisor-student relationship. Any pro- fessional success which I may have achieved so far is in large measure due to his guidance. He not only guided me in the right directions, but more importantly, knew where the guidance should end. I hope to treat any students who work with me in the same way. Thanks are also due to Dr. Glenn Hatton, who has often lent a willing ear and given sound advice, as well as Dr. M. Ray Denny and Dr. Robert Raisler, all of whom served on my committee. I am also grateful to Ralph Richards and Allen Vieth, not only for assisting with the running of my subjects on occasion, but also for helpful criticism of my research. iii TABLE OF CONTENTS Page LIST OF TABLES O O O O O O O O O O O 0 v LIST OF FIGURES . . . . . . . . . . . . vii INTRODUCTION 0 O O O O O O O O O O O 1 Some Characteristics of Discrimination Training . . . . . . . . . . . . 2 Differential Reinforcement of Low Rates and Stimulus Control . . . . . . . . 5 Behavioral Contrast . . . . . . . . . ll EXPERIMENT I . . . . . . . . . . . l3 MethOd O O O O O 0 O O O O O O 0 14 Results . . . . . . . . . . . . . 17 Discussion . . . . . . . . . . . . 38 EXPERIMENT II . . . . . . . . . . . 45 methOd I O O O O O O O O O O O O 45 Results . . . . . . . . . . . . . 46 Discussion . . . . . . . . . . . . 79 REFERENCES I O O O O O O I O O O O 0 I 85 APPENDICES A O O O O O O O O O O O O O O O 88 B O O O O I O O O O O O O O O O 89 C O O O O O O O O O O O I O O O 90 D O 0 O O O O O O O O O I O O O 91 iv Table LIST OF TABLES Reinforcements per 100 min for the Three 5-day Periods During Phase II (days 1 to 15) for the Multiple DRL 8-sec - DRL 64-sec Group and the DRL 64-sec Group in the Presence of $2, the DRL 64-sec Component . . . . . . . . Reinforcements per 100 min for the Five 5-day Periods (Days 16 to 20 of Phase I and Days 1 to 20 of Phase II) for the Multiple DRL 8-sec - DRL 64-sec Group and the Multiple DRL 8-sec - Ext Group During the DRL 8-sec Com- ponent (82) . . . . . . . . . Reinforcements per 100 min for the Three 5-day Periods During Phase II (Days 1 to 15) for the Multiple DRL 8-sec - DRL 64-sec Group and the DRL 64-sec Group in the Presence of 82, the DRL 64-sec Component . . . . . . . . Reinforcements per 100 min for the Five 5-day Periods (Days 16 to 20 of Phase I and Days 1 to 20 of Phase II) for the Multiple DRL 8-sec — DRL 64-sec Group and the Multiple DRL 8-sec - Ext Group During the DRL 8-sec Component (82) . The Total Number of Responses and the Per Cent of Each Groups Total Re- sponses at Each of the Five Wave- length Stimuli for the Feedback and No-feedback Stimuli . . . . . . . Summary Table for the Analysis of Vari- ance for the DRL 8-64 Groups, with and Without Feedback, and the DRL 8-Ext Groups, With and Without Feed— back, for the Generalization Gradients Along the Line-Orientation Dimension . V Page 37 39 65 67 71 88 Page Summary Table for the Analysis of Vari- ance for the DRL 8-64 Groups, With and Without Feedback, and the DRL 64 Groups, With and Without Feedback, for the Generalization Gradients Along the Line-Orientation Dimension . . . . . . . . . . . 89 Summary Table for the Analysis of Vari- ance for the DRL 64 Groups, With and Without Feedback, and the DRL 8-Ext Groups, With and Without Feedback, for the Generalization Gradients Along the Line-Orientation Dimension . . 90 Summary Table for the Analysis of Vari- ance for the DRL 8-64 Groups, With and Without Feedback, and the DRL 8—Ext Groups, With and Without Feed- back, for the Generalization Gradi- ents Along the Wavelength Dimension . . 91 vi Figure LIST OF FIGURES Page Individual Generalization Gradients, Expressed in Terms of Number of Re- sponses to Each of the Six General- ization Stimuli, After Exposure to the White Vertical Line Correlated with the DRL 64-sec Component During Discrimination Training in Phase II . . . 18 Individual Generalization Gradients, Expressed in Terms of Number of Re- sponses to Each of the Six General- ization Stimuli, After Exposure to the White Vertical Line Correlated with the Extinction Component During Discrimination Training in Phase II . . . . . . . . . . . . 20 Individual Generalization Gradients, Expressed in Terms of Number of Re- sponses to Each of the Six General- ization Stimuli, After Exposure to the White Vertical Line Correlated with the DRL 64-sec Contingency During Non-Differential Training in Phase II . . . . . . . . . . . 22 Group Generalization Gradients, Expressed in Terms of the Mean Number of Re- sponses per Group to Each of the Six Generalization Stimuli, in the Presence of the White Vertical Line. The Ap- propriate Component is Underlined for Each Group . . . . . . . . . . . 25 Group Generalization Gradients, Expressed in Terms of the Mean Number of Re- sponses per Group to Each of the Five Wavelength Stimuli, Around the 555 nm Stimulus which was Correlated with the DRL 8-sec Component for Both Groups During Discrimination Training in Phase II . . . . . . . . . . . . 29 vii Figure 11. Group Generalization Gradients, Ex- pressed in Terms of the Mean Number of Responses Per Group to Each of the Five Wavelength Stimuli, Around the 555 nm Stimulus Which was Corre- lated with the DRL 8-sec Component . 12. Daily Average Group Response Rates (Responses per Min) During Phase II for the Component Correlated with $1, the 555 nm Stimulus, and the Com- ponent Correlated with $2, the White Vertical Line. The Point at "Ses- sion 0" Represents a Group Average Response Rate for Each Particular Component During the Last Five Days of Phase I . . . . . . . . . 13. The Average Daily Response Rates for the Feedback (F) Groups and the No- Feedback (NF) Groups Combined During the DRL 8-sec Component for Days 6 to 20 of Phase I and Days 1 to 20 of Phase II. The Vertical Lines at Each Point Represent : 1 Standard Error of the Mean for Each Group . . . 14. The Average Daily Per Cent of the Total Responses with IRTs of 0.8 sec or Less for the Feedback (F) and the No- Feedback (NF) Groups Combined During the DRL 8-sec Component for Days 6 to 20 of Phase I and Days 1 to 20 of Phase II. The Vertical Lines at Each Point Represent i 1 Standard Error of the Mean for Each Group . . . ix Page 58 6O 73 77 INTRODUCTION Research in the past several years has suggested that stimuli associated with some positive reinforcement schedules may possess some of the properties normally at— tributed to stimuli associated with extinction. More specifically, schedules of positive reinforcement which maintain low rates of responding produce behavioral contrast and peak shift in a manner similar to extinction, when alternated with another schedule which maintains a higher rate of responding (Guttman, 1959; Terrace, 1968; Weisman, 1969). It has been suggested that these behavioral phe- nomena, peak shift and behavioral contrast, are the result of an inhibition of responding in the presence of the stimu— lus which maintains the lower response rate (Thomas & Williams, 1963; Terrace, 1966a; Weisman, 1969). The question then arises as to whether this inhibition is the result of discrimination training or is the result of the contingencies of reinforcement per se which control the be- havior. The primary purpose of Experiment I was to in- vestigate the effect of the presence or absence of dis- crimination training on the shape of the generalization gradient around a stimulus correlated with a differential- reinforcement-of-low-rate schedule. Experiment II was conducted to assess the effect of removal of a feedback stimulus for each peck on the generalization gradients and response rates. Some Characteristics of Discrimination Training Hanson (1959) trained pigeons to peck a response key illuminated from behind with a 550 nanometer (nm) light on a variable-interval schedule with a mean interreinforce- ment interval of 1 min (VI l-min). After five days of training a second stimulus, either 555, 560, 570, or 590 nm for four separate groups, alternated with 550 nm. Re- sponding in the presence of this second stimulus was never reinforced (extenction). A fifth group did not receive dis- crimination training, but was reinforced on the VI l-min 'schedule in the presence of 550 nm for an equal number of sessions. A11 generalization gradients obtained from the groups given discrimination training reached a peak at 540 nm. The gradient for the control group peaked at 550 nm, the training stimulus. This shift of the gradient peak from the training stimulus in a direction away from the stimulus correlated with extinction has been called "peak shift." Many experimenters have demonstrated the peak shift phenomenon after discrimination training on the wave- length continuum (Honig, Thomas, & Guttman, 1959; Honig, 1962; Pierrel & Sherman, 1960, 1962; Thomas, 1962; Thomas & Williams, 1963; Terrace, 1964, 1966c; Friedman & Guttman, 1965; Stevenson, 1966) and with the line-orientation di- mension (Bloomfield, 1967). Some investigators have suggested that the peak shift is the result of the formation of an underlying U- shaped inhibitory gradient around the extinction-correlated stimulus (Thomas & Williams, 1963; Terrace, 1966a; Weisman, 1969). An inhibitory gradient is a stimulus generalization gradient, usually obtained during extinction, in which re- sponding increases along the stimulus continuum as the distance from the stimulus that has previously been corre- lated with extinction increases (Catania, 1968). To di— rectly evaluate the possibility that peak shift is the result of an underlying inhibitory gradient the reinforced stimulus and the unreinforced stimulus must be on orthogonal dimensions. Jenkins and Harrison (1962) trained pigeons to discriminate a 1000 cps tone correlated with extinction from white noise correlated with VI reinforcement. They later obtained an inhibitory gradient around the 1000 cps tone. Honig, Boneau, Burnstein, and Pennypacker (1963) found that the post-discrimination generalization gradient around a black vertical line on a white background was in- hibitory if the stimulus was correlated with extinction, whereas the gradient was excitatory if the stimulus had been correlated with positive reinforcement. Terrace (1966b) found a similar result when the reinforced stimulus was on the wavelength dimension and the extinction- correlated stimulus was on the line-orientation dimension, provided that the subjects had responded in extinction. In all these studies the inhibitory gradients were relatively shallow when compared to the excitatory gradients. It does not appear, however, that discrimination training must occur with extinction in order to produce peak shift or inhibitory stimulus control. Guttman (1959) exposed pigeons to a multiple VI l-min VI 5-min schedule. In a multiple schedule two or more component schedules Operate in alternation, each in the presence of a different stimulus (Catania, 1968). In Guttman's experiment the key light during the VI 1-min component was illuminated with a 550 nm light and during the VI 5-min component with a 570 nm light. The two schedule components alternated every 5 min, two of each component per session for 12 days. Across the 12 days of acquisition, the response rates in the two components diverged. The rate in the VI 1—min component became progressively higher (up to 60 per min) while the response rate in the VI 5-min component stabilized at about 20 responses per min. The post-discrimination general- ization gradient was unimodal, reaching a peak at 540 nm, indicating a shift away from the stimulus correlated with the VI S-min component. Terrace (1968) replicated this finding using a similar procedure. Weisman (1969) exposed pigeons to a similar procedure except that the VI l-min component was correlated with a plain green key light while the VI 5-min component was correlated with a white vertical line on the green background. A post-discrimination gener- alization test along the line-orientation dimension yielded a U-shaped inhibitory generalization gradient centered around the VI S-min stimulus. In summarizing the results of his experiment Guttman stated that "a stimulus associated with relatively weak reinforcement can become functionally negative [p. 338]." He also pointed out that stimuli associated with extinction and relatively weak reinforcement schedules may both possess suppressive or inhibitory properties. "The obvious inference is that during training the behavior in the presence of this stimulus is under an active re- straint, and that the observed level of performance is the resultant of this discriminative restraint and the positive factor of reinforcement [p- 3391-" The results obtained by Terrace and Weisman would seem to make this conclusion more defensable. Differential Reinforcement of Low Rates and Stimulus Control Under a schedule which differentially reinforces low rates of responding (DRL), interresponse times (IRTs) that end after a predetermined time interval are reinforced. The IRTs shorter than this time reset the timing contingency. For example, if a subject is responding under a DRL lO-sec schedule, all IRTs of 10 sec or more are followed by rein- forcement and all IRTs less than 10 sec "reset" the con- tingency of waiting at least 10 sec to respond for rein- forcement. In many studies concerned with DRL, IRTs are recorded in class intervals, usually of 1, 2, or 3 sec widths. For example, all IRTs from 0 to 1 sec are recorded in the first category, all IRTs from 1 to 2 sec in the second category, etc. When the data are recorded in this way, a plot of the relative frequency of responses in the various class intervals usually produces a bimodal curve once stable performance is obtained. The first mode occurs at the first category, after which responding drops to near zero for the next category. As the IRT category in which the shortest reinforced IRT falls is approached, the rela- tive frequency of responses in each category begins to rise, reaching a maximum at or just before the first rein- forced category. After this the relative frequency de- creases to zero (Kramer, 1970). The large number of IRTs in the first category has been a subject of considerable interest to many investi- gators. These short IRTs have been called "bursts" of re- 8ponding by Sidman (1956), defined as any sequency of two or more responses in which no consecutive responses are separated by more than 2 sec. It is somewhat surprising that these bursts occur with such a high frequency since they are relatively far removed in time from reinforced IRTs. Blough (1963, 1966), using sensitive recording equipment, noted that pigeons reinforced for certain IRTs tended to emit responses with IRTs which clustered at around 0.4, 0.8, and 1.2 sec. To further investigate these short IRTs, Blough (1966) recorded their occurrence in 0.1 sec categories up to 2 sec. Although the details of this experiment are somewhat complicated, suffice it to say that while the independent variables of varying reinforcement probabilities and varying IRT category widths significantly affected IRTs greater than 0.8 sec, IRTs less than this were insensitive to these contingencies. In addition, while the behavior of all pigeons was quite similar under the various conditions, individual differences in the prob- ability of IRTs less than 0.7 sec were large. The data also indicated that response probabilities in the short IRT region showed the greatest session-to-session variability. Kramer (1970), in a review of the DRL contingency, noted that not all investigators have found a large number of bursts being emitted by their subjects on DRL schedules. Kelleher, Fry, and Cook (1959) found very few bursts with rats, although in this case they were providing an audible "click" for each bar depression as feedback. Noting that this was an unusual finding, they attempted to investigate further. After the rats had been exposed to a DRL 20-sec - LH 5-sec schedule for over 120 hrs, the click was removed for three animals. The effect of removing the auditory click was consistent; all animals increased the percentage of bursts, although the magnitude of the increase was small. Kramer (1968) recorded a large percentage of bursts for pigeons on both DRL 20-sec and DRL 30-sec. However, virtually no bursts occurred after a reinforced response. Each reinforced response was followed by immediate feedback; the key light went out and the food magazine was presented. Sidman (1956) also found a reduced number of bursts follow— ing reinforced responses. Here again food presentation provided immediate feedback. These results suggest that bursts of responding may be due to a lack of stimulus feedback for a criterion response. In any automated experiment, a certain minimum criterion must be exceeded in order for the response of the subject to be recorded as a response electrically. For example, the pigeon must exert at least some minimum amount of force on the response key, and the rat must move the bar at least some minimum distance, in order for a response to be recorded. On the other hand, these subjects may be making subcriterion responses that are not recorded electrically, but that are responses nonetheless to the subject. If subcriterion responses occasionally precede a criterion response that is reinforced, then these sub- criterion responses would also be strengthened through con- tiguity with reinforcement. In summarizing the data on bursts of responding Kramer (1970) suggested that manipu- lation of the kind and amount of stimulus feedback for each response might be a fruitful way of reducing the frequency of response bursts. Another way of investigating stimulus control is through generalization testing. Hearst, Koresko, and Poppen (1964) conducted a systematic study of the shape of the generalization gradients after responding under several reinforcement schedules. In Experiment I a group of pigeons was trained to respond under either a DRL 6-sec or DRL lO-sec schedule. A second group was trained under a VI l-min schedule. After 14 days a generalization test was conducted. The obtained gradients were quite different for the two schedules. For the VI group data, the gradient was sharply peaked at the training stimulus. In contrast, the group data for the DRL birds was quite flat in comparison, with the peak occurring at the training stimulus. Of the 24 birds given DRL training, however, only five had their peak at the training stimulus, whereas 18 of 20 VI gradients peaked at the training stimulus. The authors suggested that the relatively flat DRL gradients may have been the result of several factors, one of which was the following: "DRL is usually considered to be one of these 'semi-average' schedules, and therefore its negative properties might interact with its positive properties to affect the peak and slope of the generalization gradients. The fact that only 10 5 of the 24 DRL 83 had a gradient peak at the CS lends weight to the notion that DRL possesses inhibitory effects [p. 375]." One implication and extension of these results is that DRL schedules with longer time intervals may produce inhibitory generalization gradients. Weisman (1969), in Experiment II of the study pre- viously referred to, investigated the shape of the general- ization gradient after discrimination training under a multiple VI 1-min DRL schedule. Initially, pigeons were exposed to a multiple VI l-min VI l-min schedule in the presence of a green light and a white vertical line which alternated randomly. After 20 sessions reinforcement in the presence of the white vertical line was contingent on a DRL schedule. The value of the DRL component was adjusted daily so that the number of reinforcements per session was equal for both components. The DRL value was eventually raised to 16 or 20 sec. A generalization test along the line—orientation dimension, the dimension which had been correlated with the DRL component, yielded an inhibitory gradient for four out of four birds. The question still unanswered is whether the inhibitory gradient obtained by Weisman was a function of the DRL schedule itself, as the results of Hearst gt_gl. (1964) might suggest, or a function of discrimination training, as was implied by Guttman (1959). ll Behavioral Contrast Another behavioral phenomenon which appears to be highly correlated with peak shift and inhibitory gradients due to discrimination training is behavioral contrast (Terrace, 1966a,b,c, 1968; Weisman, 1969). Behavioral contrast has been defined as a change in the rate of re— sponding from the baseline level in the presence of one stimulus in a direction away from the rate of responding generated in the presence of another stimulus (Reynolds, 1961). The behavioral contrast literature has been ex- tensively reviewed by both Terrace (1966a) and Dunham (1968). Terrace (1968) has concluded that "a sufficient condition for both phenomena (behavioral contrast and peak shift) is a reduction in the rate of responding to one of two alternating discriminative stimuli [p. 737]." Weisman (1969) has obtained complementary data and has suggested that peak shift is an indicator of inhibitory stimulus control. If this is true, then the occurrence of behavioral contrast and inhibitory generalization gradients should also be highly correlated in the same way as behavioral contrast and peak shift appear to be. With respect to DRL and behavioral contrast, Reynolds and Catania (1961) and Reynolds and Limpo (1968) have demonstrated that behavioral contrast results when a DRL schedule is the constant component in a multiple schedule and the rate of responding in the other component 12 is reduced by either extinction (Reynolds & Catania, 1961) or by signaling the availability of reinforcement in the second component (Reynolds & Limpo, 1968). But the amount of behavioral contrast found by Reynolds and Catania was small. It appeared that the greatest increase occurred for IRTs less than 3 sec, and that the probability of occurrence of longer IRTs was little affected. Reynolds and Limpo found that behavioral contrast was slow in developing. Here again short IRTs appeared to be effected the most. EXPERIMENT I The main question under investigation in the first experiment can be stated simply as follows: is discrimi- nation training necessary for the development of inhibitory stimulus control, as measured by a generalization gradient, after training on a schedule of reinforcement which gener- ates low rates of responding? An extension of the results of Hearst gt_gl. (1964) suggests that a schedule which maintains a low rate of responding may, of itself, yield inhibitory generalization gradients. The results of Weisman (1969) have shown that inhibitory gradients do develop after discrimination training around the stimulus correlated with the schedule which maintained the lower response rate. Guttman's (1959) statement that a "stimulus associated with relatively weak reinforcement can become functionally negative" [p. 338] implies that discrimination training is necessary to generate inhibitory stimulus control. Of secondary interest was the effect of a feedback stimulus, delivered immediately after each unreinforced response, on the rate of responding, particularly during discrimination training. 13 14 Method Subjects The subjects were 18 naive White Carneaux female pigeons, ranging in age from six to 10 years, maintained at 80 per cent of their free-feeding weight. Apparatus The experimental environment was a standard operant conditioning chamber, 12 X 14 X 13 in. The front panel contained three Lehigh Valley Electronics pigeon keys, 1 in. in diameter, mounted 8 1/2 in. above the floor with a horizontal separation of 3 in. Behind the center key an Industrial Electronics Engineers in-line display unit was mounted. The display cell could illuminate the center key with six orientations of a 1/8 in. wide by 7/8 in. high white line (:90°, -60°, -30°, 0°, +30°, +60°, of departure from vertical) and five wavelengths (501, 538, 555, 576, and 606 nm peak wavelength). A force of approximately 15 g was required to close the key and record a response. A rectangular opening located below the center key permitted access to grain which was raised to the feeding position and illuminated for 2 1/2 sec for reinforcement. Two house lights, used only during shaping, were mounted on the front wall. A third house light (yellow) was mounted on the rear wall. Programming and recording equipment were located in an adjoining room. 15 Procedure All animals were given approximately one hr of general habituation to the chamber, followed on the second and third days by manual shaping of approach to the food magazine and the pecking response to the center key. During shaping the center key was illuminated with either the white vertical line on the green (555 nm) background or the plain green (555 nm) stimulus. The stimuli alternated every 2 min. Once at least 10 responses had been reinforced on a continuous reinforcement schedule in the presence of each stimulus, the first daily session started. During Phase I (acquisition training) the responses of all subjects were reinforced on a multiple DRL 8-sec - DRL 8-sec schedule. The white vertical line on the green background was associ- ated with one component and the plain green stimulus with the other. Under this schedule all IRTs of 8 sec or longer are reinforced, whereas all IRTs less than 8 sec reset the timing contingency. During reinforcement the response key was not illuminated. In addition, after each unreinforced response, the key light was blacked out for 0.8 sec to provide immediate stimulus feedback for each response which activated the microswitch. The time of 0.8 sec was se- lected from the data of Blough (1966) which suggested that interresponse times in this area were not sensitive to the manipulation of independent variables. No other illumina- tion was provided in the chamber. 16 The stimuli on the center key were presented ac- cording to one of six irregular orders (Gellerman series) for 4—min periods, and eight periods of each stimulus were given each day for a total of 16 periods. The 4-min peri- ods were separated by 10-sec periods during which the center key was out and the yellow light in the rear of the chamber was illuminated. Responding during this lO-sec period had no scheduled contingencies. Phase I lasted for 20 daily sessions. The subjects were run on the average of six days a week. During Phase II the 18 subjects were divided into three groups of six each. Two groups received discrimi- nation training on a multiple schedule; multiple DRL 8-sec - DRL 64-sec for one group and multiple DRL 8-sec - Extinction (Ext) for the other discrimination group. For both groups the plain green stimulus (555 nm) was correlated with the DRL 8-sec component, while the white vertical line on the green background was correlated with either the DRL 64-sec component or the Ext component. As before, the stimuli were presented for 16 periods of 4 min duration, eight of each stimulus component. The remaining group of six subjects received DRL 64-sec only, in the presence of the white vertical line on the green background. This group received only eight 4-min periods daily. Thus, the number of periods where the stimulus was the white vertical line on the green background was the same for all three groups. 17 After 15 sessions of Phase II for all groups, all subjects were given a brief warmup on the next day (2 4-min periods of each schedule component for the discrimination groups and 2 4-min components for the DRL 64-sec group), followed by a generalization test in extinction. The six line-orientation stimuli were presented in each of 12 randomized blocks for a total of 72 stimulus-on presen- tations, each separated by 10-sec periods when no stimulus was presented and the yellow light in the rear of the chamber was illuminated. Six test stimulus sequences were used, with a different sequence for each subject in each group. The number of key-pecks during each 30-sec stimulus presentation was recorded. The two groups given discrimination training during Phase II were then given five additional sessions as before in Phase II. On the following day generalization testing in extinction was administered along the wavelength di- mension using the same procedure as for the line-orientation dimension. The five wavelength stimuli were presented in each of 12 randomized blocks for a total of 60 30-sec pres- entations. The data were recorded as before. Results Generalization Gradients Figures 1, 2, and 3 show the individual generali- zation data, expressed as a function of the number of 18 Figure l.--Individual generalization gradients, ex- pressed in terms of number of responses to each of the six generalization stimuli, after exposure to the white vertical line correlated with the DRL 64-sec component during dis- crimination training in Phase II. NUMBER OF RESPONSES 19 Mult DRI. fir-QRL Q4 382 .. 947 low A 11111. 11111. 30347 .540 2 A 1. 1. 1111111 11111. 2450 _1169 .. 1. . 2. _ "0.70 0.003090 60300 6090 - + - + DEGREES FROM VERTICAL 20 Figure 2.--Individual generalization gradients, expressed in terms of number of responses to each of the six generalization stimuli, after exposure to the white vertical line correlated with the extinction component during discrimination training in Phase II. NUMBER OF RESPONSES .0700 21 Mult DRL 8- 5.25.1 + 70-090 9000 F "74 P p )- 306090 — + DEGREES FROM VERTICAL 22 Figure 3.--Individual generalization gradients, expressed in terms of number of responses to each of the six generalization stimuli, after exposure to the white vertical line correlated with the DRL 64-sec contingency during non-differential training in Phase II. NUM BER O F RESPONS ES 23 637 20 10 , 1 1 , 635 20 I. , 1 11 . 9 l0 b 0'. .0 70 . . + + DEG REES FROM VE RTIC Al. 24 responses to each of the six line-orientation stimuli, for the multiple DRL 8—sec - DRL 64-sec group, the multiple DRL 8-sec - Ext group, and the DRL 64-sec group, respectively. For the multiple DRL 8-sec - DRL 64-sec group (DRL 8-64 group), individual differences are striking. The individual gradients range from clearly excitatory (Birds 2450 and 947) to clearly inhibitory (Bird 1169). The individual gradients for the multiple DRL 8-sec - Ext group (DRL 8-Ext group), shown in Figure 2, show more uniformity. The gradients for the four birds with a level of responding above zero at all stimuli uniformly show a minimum number of responses at or near the training stimulus (0°, :30°), and a maximum number of responses to the peripheral stimuli (:90°, :60°). The individual gradients for the DRL 64-sec group (DRL 64 group), shown in Figure 3, indicate the most individual uniformity. With the exception of Bird 387, all gradients reach a minimum at or near the training stimulus (0°, 130°) and a maximum at the peripheral stimuli. Figure 4 shows the mean number of responses for the three groups at each of the line-orientation stimuli. A subjects by treatments (stimuli) repeated-measures AOV was performed for each group gradient. The statistical analysis of the absolute number of responses across the various stimuli revealed that there was no significant effect for the DRL 8-64 group (F < 1), no significant effect for the DRL 8-Ext group (F = 1.80, df 5/25, p > .10), but that the 25 Figure 4.--Group generalization gradients, expressed in terms of the mean number of responses per group to each of the six generalization stimuli, in the presence of the white vertical line. The appropriate component is under- lined for each group. MEAN RESPONSES 26 DRI. B-DRI. 64 o DRLB'EXT I URI. 64 A )— \ ’/ \ ,- OE! l I l I L a] 6 0 6O 90 + DEGREES FROM VERTICAL 27 effect of stimuli was significant for the DRL 64 group (F = 7.42, df 5/25, p < .001). This result, combined with inspection of the group and individual gradients, indicates that the gradient for the DRL 64 group is a U-shaped, in- hibitory gradient. A two-factor AOV with repeated measures on one factor (stimuli) was performed for comparisons between groups. Statistical comparison of the group gradients of the DRL 8-64 group and the DRL 8-Ext group indicated that there was no significant difference between groups (F = 1.13, df 1/10, p > .20), that the overall effect of stimuli was not significant (F < 1), and that the interaction of the groups and the stimuli was not significant (F = 1.26, df 5/50, p > .20). Statistical comparison of the DRL 8-64 group gradient with the DRL 64 group gradient indicated no significant difference between groups (F < 1) and no sig- nificant difference due to stimuli (F < 1). However, the interaction of the groups with the stimuli was significant (F = 3.57, df 5/50, p < .01). This result indicates that the two group gradients have a different shape. The overall conclusion is that, for the contingencies used in this ex— periment, discrimination training did not lead to inhibi- tory stimulus control while comparable training on the same schedule (DRL 64-sec) in isolation did yield inhibitory stimulus control. 28 Although the DRL 8-Ext group and the DRL 64 group differ procedurally by two factors, a comparison of the generalization gradients is still of some interest. The effect of the stimuli produced a statistically significant difference (F = 8.12, df 5/50, p < .001). However, neither the effect of groups (F = 2.36, df 1/10, p > .10) nor the interaction of groups with stimuli (F = 1.55, df 5/50, p > .20) yielded a significant difference. Both the DRL 8-64 group and the DRL 8-Ext group were given a second generalization test along the wave- length dimension, the dimension correlated with the DRL 8-sec component. The general tendency for animals in both grOUps was for an excitatory gradient reaching a peak at the training stimulus (555 nm), with the fewest number of responses at the peripheral stimuli (501 and 606 nm). Two birds in the DRL 8-64 group were exceptions to this trend. Birds 347 and 1169 emitted the highest number of responses to the 538 nm stimulus. Nevertheless, both of these gradi- ents were excitatory in shape. Figure 5 shows the mean number of responses for both groups at each of the five wavelength stimuli. Both group gradients are excitatory and symmetrical. Statistical analysis of the absolute number of test responses revealed that the effect of stimuli was significant for both the DRL 8-64 group (F = 6.47, df 4/20, p < .005) and for the DRL 8-Ext group (F = 16.84, df 4/20, p < .001). Statistical 29 Figure 5.--Group generalization gradients, expressed in terms of the mean number of responses per group to each of the five wavelength stimuli, around the 555 nm stimulus which was correlated with the DRL 8-sec component for both groups during discrimination training in Phase II. MEAN RESPONSES 30 sow emu-oat 64 _. IDRLB-EXT )— 1. ! ! | | J l 555 576 606 WAVElENGTH 31 comparison of the group gradients with each other indicated no significant differences between the groups (F = 1.54; df l/lO, p > .10) and no significant interaction (F < 1). Of course the overall effect of stimuli was highly signifi- cant (F = 21.10, df 4/40, p < .001). These results indicate that (l) generalization around a stimulus correlated with DRL 8-sec is excitatory after discrimination training, and (2) the generalization gradient around a stimulus correlated with DRL-8 sec is not differentially affected by whether the schedule in the other component is reinforced (DRL 64-sec) or not reinforced (Ext), as long as both maintain a lower response rate. Response Rates Figure 6 shows the mean daily response rates of both discrimination groups for the DRL 8-sec component (81) and the mean daily response rates of all groups for either the DRL 64-sec component (the DRL 8-64 group and the DRL 64 group) or Ext (the DRL 8-Ext group). This component is designated as $2. The points at Session 0 represent the group mean averages for the last five days (days 16 to 20) of Phase I (acquisition) training. These last five days of Phase I were chosen as the baseline rate of responding since inspection of the data indicated that the rate of responding had reached as asymptotic level and showed little day-to-day changes. Therefore, these last five days will be used as a baseline point from which to determine 32 .H mmmnm mo m>MU m>flm umma may mafiuso pcocomaoo amasofluumm some now mama omcommmu ommuo>m moonm m mucmmmummu =0 :onmom= um ucHom one .ocfia HMUHDH0> muflzz may .mm nuHB consamuuoo unocomeoo map How can .mSHDEHum Ea mmm may .Hm suHB Umpmaouuoo ucmcomfioo on» MOM HH ommnm mcfluso ACHE mom noncommouv mommy oncommmu msoum mmmum>m >HHMQII.m mhsmflh 33 mZO_mmmm ...1 1:11 1... I. I G. 19 \l. 1i...- U‘AxI/ltivlz p o. (1010,40,: a Oil-\O / I \‘\‘/ 1/ 1 4\1 C\ /C\4 I. /\Q/‘ ‘IC . 4 /, . . .0 b, m 2% w\ o I o 0 xx x o 10:10. . ... ”1w... ... . U .— < #0 duo . :3 -3: a = N— 0 v0 d¢0104¢0 o «m 5 NIW/SBSNOJSBU 34 whether the daily response rates changed during Phase II. It is quite clear that all three groups showed a reduction in the rate of responding in the presence of 82. Con- sidering days 1 through 15 of Phase II, one bird (#9) in the DRL 64 group exceeded the baseline rate on days 2, 5, and 6. One subject in the DRL 8-64 group exceeded this baseline rate on day 4. For all other birds every daily rate was below the average 5-day baseline rate. Statistical comparisons of the three groups with each other were carried out using the average rate of re- sponding per group in the presence of $2 for days 11 through 15 of Phase II. This comparison was made because the rates of responding were asymptotic by this time, and also because these days just preceded the line-orientation generalization test. The analyses indicated that the average rate of responding for both the DRL 8-64 group and the DRL 64 group differed significantly from the DRL 8-Ext group (t = 3.65, df 5, p < .02; t = 3.23, df 5, p < .05, respectively, two-tailed). However, the rate of responding for the DRL 8-64 group and the DRL 64 group were not sig- nificantly different (t = 0.37, df 5, p > .50). Therefore, the difference in the shape of the generalization gradients for these two groups cannot be attributed to a difference in asymptotic response rates. Inspection of the daily response rates during the DRL 8-sec component (81), shown in Figure 6, for the DRL 35 8-64 group and the DRL 8-Ext group during Phase II shows only small rate changes from day-to-day. For the DRL 8-64 group there is a slight rate increase for the first day compared to the previous 5-day baseline rate (mean change of 1.1 responses per minute) and for the first five days combined (mean change of 1.1 responses per minute). For succeeding days there was a very slight but gradual de- crease. For the DRL 8-Ext group there was a drop in the response rate for the first day of Phase II for all animals which averaged 2.5 responses per min, followed on the second day by an increase for all animals which averaged 2.5 re- sponses per min. For succeeding days the response rate de- clined slightly and gradually. Except for the first day of discrimination training the daily average response rates for both groups were quite similar. Between days 15 and 16 the generalization test along the line-orientation dimension was given. Note that the response rates on days 15 and 16 show little change. For statistical analysis the rates of responding were grouped into 5-day blocks, thus collapsing the data into five 5-day blocks (the last five days of Phase I and 20 days of Phase II). A two-factor AOV with repeated measures on one factor was performed using the response rates during the DRL 8-sec component for the DRL 8-64 and the DRL 8-Ext groups across the five 5-day blocks. The results showed no significant differences due to groups 36 (F < 1), 5-day blocks (F < 1), or the interaction of groups with treatments (F = 1.49, df 4/40, p > .10). A subjects by treatments repeated-measures analysis across the five 5-day blocks for the DRL 8-Ext group also showed no significant rate changes (F < 1). However, the same analysis for the DRL 8-64 group was significant (F = 2.98, df 4/20, p < .05). In order to determine which means differed from each other, a Newman-Keuls test was performed (Winer, p. 114). The results indicated that the only significant difference occurred between the last five days of Phase I and the first five days of Phase II (p < .05). The mean change was an increase in the rate of responding of 1.1 responses per min. Reinforcement Rates Table 1 shows the 5—day average number of rein— forcements per 100 min for days 1 through 15 for the groups that responded on the DRL 64-sec component in the presence of 52. Both individual and group averages are shown. For the DRL 8-64 group the reinforcement rate increased over the three 5-day blocks, although some individual exceptions occurred. For the DRL 64 group, the mean average rein- forcement rate increased to its maximum value during days 6 to 10 and remained at about that level for days 11 to 15. Note that both groups have equal reinforcement frequencies for the 11-15 day block. 37 TABLE l.--Reinforcements per 100 min for the three 5-day periods during Phase II (days 1 to 15) for the multiple DRL 8-sec - DRL 64-sec group and the DRL 64-sec group in the presence of $2, the DRL 64-sec component. Reinforcements Per 100 Minutes Subjects Days 1-5 Days 6-10 Days 11-15 Multiple DRL 8-sec - DRL 64-sec 382 3.4 2.5 4.4 2,450 3.1 3.1 5.6 1,169 2.5 2.5 4.5 540 3.7 8.1 6.2 347 1.8 5.6 8.1 947 4.4 5.0 6.8 Mean 3.2 4.5 5.9 DRL 64-sec 387 2.5 10.0 10.0 9 1.2 3.1 1.9 637 5.0 9.3 9.3 542 5.6 7.6 6.2 635 1.2 3.7 6.2 97 5.6 2.5 1.9 Mean 3.5 6.0 5.9 Total Mean (A) 0 u U1 0 DJ U1 \9 38 A two-factor AOV with repeated measures on one factor (three 5-day blocks) was carried out for comparisons between groups. The blocks effect was significant (F = 6.39, df 2/20, p < .001). There was no significant effect due to groups (F < l) or the interaction of groups with blocks (F < 1). Table 2 shows the 5-day average number of rein- forcements per 100 min for the five 5-day blocks (days 16 to 20 of Phase I and days 1 to 20 of Phase II) for the com- ponent correlated with $1, the green (555 nm) stimulus, for the DRL 8-64 group and the DRL 8-Ext group. Again, both group averages and individual averages are shown as well as the combined average for both groups. The overall trend (combined group average) is for an initial decrease fol- lowed by a monotonically increasing rise in reinforcement frequency. This is also characteristic of the group aver- ages for the DRL 8-64 group, while the DRL 8-Ext group did not show the initial drop. Note that within either group there are some animals whose reinforcement rate varies from the group average. Statistical comparison of the groups across the 5-day blocks indicated no significant differences due to groups (F < 1), blocks (F < 1), or the interaction of groups with blocks (F < 1). Discussion The data from this experiment lead to several con- clusions. First of all, on the basis of the generalization 39 TABLE 2.--Reinforcements per 100 min for the five 5—day periods (days 16 to 20 of Phase I and days 1 to 20 of Phase II) for the multiple DRL 8-sec - DRL 64-sec group and the multiple DRL 8-sec - Ext group during the DRL 8-sec component (82). Reinforcements Per 100 Minutes Phase I Phase II Phase II Phase II Phase II Subjects Days Days Days Days Days 16-20 1-5 6-10 11-15 16-20 Multiple DRL 8-sec - DRL 64-sec 382 164.2 208.3 205.9 213.7 268.5 2,450 418.1 180.9 245.9 295.7 322.7 1,169 125.3 123.9 141.7 217.9 128.5 540 240.9 212.1 251.1 251.9 252.6 347 306.4 203.2 208.4 173.2 220.1 947 121.0 135.2 157.3 148.1 135.0 Mean 229.3 177.2 201.7 216.8 22.12 Multiple DRL 8-sec - Ext 193 382.9 297.0 351.4 334.3 377.4 592 145.9 187.1 158.2 281.5 307.8 564 182.0 214.0 230.6 119.9 133.1 17 154.8 253.5 235.7 90.0 115.7 1,174 195.6 233.2 334.2 421.5 400.8 450 82.0 57.7 67.3 92.9 89.2 Mean 190.5 207.1 229.5 223.4 237.3 Total Mean 210.0 192.2 215.6 220.0 229.3 40 data of the DRL 64 group, it appears that discrimination training is not a necessary prerequisite for the production of inhibitory stimulus control. The data indicate that some positive reinforcement schedules, at least long DRL schedules, can produce inhibitory stimulus control. These animals did, however, show a reduction in the rate of re- sponding from a previously established rate of responding. Secondly, a reduction in the rate of responding during discrimination training does not always lead to in- hibitory stimulus control. This is most evident from the data of the DRL 8-64 group, where only one out of six sub- jects showed a clear inhibitory gradient. This was the case even though all six birds clearly showed a rate re- duction and discrimination between the two schedules. The fact that the DRL 64 group produced an in- hibitory generalization gradient while the DRL 8-64 group did not can only be attributed to the fact that the DRL 64 subjects were not given discrimination training. The groups did not differ with respect to asymptotic response rates on the DRL 64-sec component, they were not signifi- cantly different with respect to reinforcement rate, and they did not differ with respect to the total number of re- sponses emitted during the generalization test. The fact that the DRL 8-64 group did not show clear inhibitory stimulus control around the stimulus correlated with the DRL 64-sec component is somewhat 41 paradoxical since the DRL 64 group did. The literature suggests that discrimination training should heighten in— hibitory stimulus control when compared to a non- discrimination procedure. Just the opposite was found in this experiment. One possible explanation suggests itself from the generalization data of the DRL 8-Ext group. While - several investigators (Honig gt_al,, 1963; Terrace, 1964; Farthing & Hearst, 1968) have shown clear inhibitory gradi- ents around a stimulus correlated with extinction after VI - Ext training, the extinction gradients for the DRL 8-Ext group were equivocal with respect to shape. Two animals failed to response enough to show any gradient. All animals emitted a low number of responses during the gener- alization test. The same number of presentations of six line-orientation stimuli was used in this experiment as was used by Farthing and Hearst in a generalization test after 16 sessions of VI l-min - Ext training. They obtained clear inhibitory stimulus control around the stimulus corre- lated with extinction (a 1/8 in. black line on a white field). But their six subjects emitted three times as many total responses as the DRL 8-Ext subjects in the present experiment (1535 vs 452). This suggests that possibly the alternation of the DRL 8-sec schedule with either the DRL 64-sec schedule or extinction created a ceiling effect. The "maintained" rate of responding in this experiment was ap- proximately nine responses per min while the VI 1-min 42 schedule used by Farthing and Hearst maintained an average rate of responding of 70 responses per min. This does not negate any conclusions drawn with respect to the DRL 64 group. The animals in this group did not have any "refer- ence" rate of responding, at least for 15 days, and thus there was no other conditioned low rate. One problem with this explanation is that the total number of responses for the DRL 8-64 group and the DRL 64 group did not differ significantly during generalization testing, although the total was slightly higher for the DRL 64 group. A more hypothetical explanation may be that in- hibitory stimulus control around a given stimulus indicates that the particular stimulus is aversive. Further, a stimulus is only aversive in a relative sense as Guttman (1959) suggested. Applying this explanation to the present data, the DRL 64-sec schedule and extinction were not par- ticularly aversive when contrasted with a DRL 8-sec schedule, a schedule which maintains a rather low rate of responding when compared to most other reinforcement schedules. For the DRL 64 group there was no other con- ditioned low rate stimulus in the animals' immediate past history. Therefore, any aversiveness conditioned to the DRL 64-sec schedule would not be dissipated by reference to another low rate schedule. A logical extension of this hy- pothesis would be that increasing the rate of responding in the presence of 81 would lead to greater inhibitory stimulus 43 control in 52, provided that $2 rates are approximately equal. The generalization gradients around the stimulus correlated with the DRL 8-sec component were clearly ex- citatory. This is in contrast to the results of Hearst et a1. (1964) who found that the generalization gradients around a DRL 6- or lO-sec stimulus were essentially flat. The present gradients were taken after discrimination training and after a prior generalization test, two Opera- tions which tend to sharpen stimulus control. The signifi- cant point is that a DRL schedule can lead to excitatory stimulus control. As previously mentioned, behavioral contrast refers to a change in the rate of responding in the constant com- ponent of a multiple schedule in a direction away from the change in the rate of responding in the other component (Reynolds, 1961a). Terrace (1968) concluded that "reducing the rate of responding to a discriminative stimulus by ex- tinction, by changing from a VI 1-min to a VI 5-min schedule of reinforcement, by a DRL schedule, or by electric shock punishment are functionally equivalent in that in each case, (behavioral) contrast and the peak shift result [p. 737, my parentheses]." The results of the present experiment indi- cate that some of these operations do not always produce behavioral contrast. For the DRL 8-Ext group behavioral contrast in the DRL 8-sec component did not occur, even 44 though there was a clear and immediate reduction in the response rate in the extinction component. For the DRL 8-64 group a significant rate increase was found for the first five days of discrimination training compared to the five baseline days. But this rate increase averaged only 1.1 responses per min and did not persist past the first five days of Phase II as would normally be the case. Reynolds and Limpo (1968) found that behavioral contrast developed gradually over 15 days and was greatest for days 11 through 15, when the constant component was a DRL 35-sec schedule. Considering the low magnitude and the brevity of the rate change for the DRL 8-64 group, the rate increase of this group during the DRL 8-sec component is uncharacter- istic of the normal contrast phenomenon. The general lack of behavioral contrast in this experiment suggests that a reduction in the rate of responding in one component does not always lead to an increased rate of responding in the constant component. EXPERIMENT II The procedure used in the first experiment was somewhat unusual in that each unreinforced response was followed by darkening the response key for 0.8 sec (feed- back). This was done because past experiments had indi- cated that (1) short IRTs (less than 0.8 sec) were not under stimulus control (Blough, 1963, 1966), and (2) short IRTs accounted for the major part of the rate increase during discrimination training when a DRL schedule was the constant component (Reynolds & Limpo, 1968; Reynolds & Catania, 1961). Experiment II was undertaken (l) to determine the effect of this feedback on the rate of responding, particu- larly during discrimination training, and (2) to determine the possible effect of feedback on the generalization gradients. Method Subjects The subjects were 18 naive White Carneaux female pigeons, ranging in age from six to 10 years, maintained at 80 per cent of their free-feeding weight. 45 46 Apparatus and Procedure All aspects of the apparatus and procedure of this experiment were identical to the first with one exception. After each unreinforced peck the key light remained illumi- nated. The key light was darkened only during reinforcement and during the 10 sec between 4-min periods. To briefly summarize the procedure, all animals were given 20 days of training on a multiple DRL 8-sec — DRL 8-sec schedule (Phase I). Then the birds were divided into three groups of six each (Phase II). One group re- ceived 15 days of multiple DRL 8-sec - DRL 64-sec training, the second group 15 days of multiple DRL 8-sec - Ext train- ing, and the third 15 days of DRL 64-sec training. The DRL 64-sec schedule and extinction were correlated with the white vertical line on the green background. After the 15th day of Phase II all animals were given a generalization test along the line-orientation dimension as in Experiment I. The two discrimination groups were then given five ad- ditional days of Phase II training, followed on the next day by a generalization test along the wavelength dimension. Results Generalization Gradients Figures 7, 8, and 9 show the individual generali- zation data, expressed as a function of the number of re- sponses to each of the six line-orientation stimuli, for 47 Figure 7.--Individual generalization gradients, expressed in terms of number of responses to each of the six generalization stimuli, after exposure to the white vertical line correlated with the DRL 64-sec component during discrimination training in Phase II. NUMBER OF RESPONSES 48 Mult DRI. B-DRL 64 30 201 I7 .2618 60 0 — + - + ossaess FROM vsancu 9O 49 Figure 8.--Individual generalization gradients, expressed in terms of number of responses to each of the six generalization stimuli, after exposure to the white vertical line correlated with the extinction component during discrimination training in Phase II. NUMBER OF RESPONSES 50 Mult DRL B-EXT 275 ._ 242 ~ - 4. - W 29 P 757 1 l 1 L J 1 L J l I J 833 , 245 )- ? “-1-?- . obo-o WWO — + - + DEGREES FROM VERTICAL 51 Figure 9.--Individua1 generalization gradients, expressed in terms of number of responses to each of the six generalization stimuli, after eXposure to the white vertical line correlated with the DRL 64-sec contingency during non-differential training in Phase II. NUMBER OF RESPONSES 52 DR]. 64 3. - 2. 298 r- 23 '01- , 1 1 1 1 1 1 3 658 2 l0 , 1 1 1 1 1 .1 no 5. 4 0 2.1. ID ".6030 0 I060" 60 30 — + - + DEGREES FROM VERTICAL 53 the multiple DRL 8-sec - DRL 64-sec group, the multiple DRL 8-sec - Ext group, and the DRL 64—sec group, re- spectively. For the multiple DRL 8-sec - DRL 64-sec group (DRL 8-64 group), two birds show excitatory gradients (Birds 109 and 2618) while the other four birds show either flat (Bird 17) or inhibitory gradients. The individual gradients for the multiple DRL 8-sec - Ext group (DRL 8-Ext group) are similar to the gradients of the corresponding group in Experiment I. Again, two birds (275 and 242) emitted too few responses for any conclusions to be drawn about a gradient. The gradient for one bird is excitatory (Bird 29), while the remaining gradients are inhibitory. The individual gradients for the DRL 64-sec group (DRL 64 group) show the most variability. Two of the gradients are excitatory (Brids 185 and 3821), two are flat (Birds 298 and 23), one is inhibitory (Bird 658), and the sixth is nonsymmetrical. Figure 10 shows the mean number of responses for the three groups at each of the line-orientation stimuli. It is clear by inspection that neither the DRL 8-64 nor the DRL 64 group gradients are inhibitory. The DRL 8-Ext group gradient is flat with a very slight inflection for the :90° stimulus. A subjects by treatments (stimuli) repeated- measures AOV was performed for each group gradient. The statistical analysis of the absolute number of responses across various stimuli revealed no significant effect for 54 Figure lO.--Group generalization gradients, ex- pressed in terms of the mean number of responses per group to each of the six generalization stimuli, in the presence of the white vertical line. The appropriate component is underlined for each group. MEAN RESPONSES 55 M‘ 40" 301* A r- \A 201. /1\‘/ \ A R /- 10+. \\ /"“’~l’/ — V""""" 0 6 3 3 90 DEGREES FROM VERTICAL either the DRL 8-64 group (F < l), the DRL 8—Ext group (F = 1.45, df 5/25, p > .20), or the DRL 64 group (F < l). A two-factor AOV with repeated measures on one factor (stimuli) was performed for comparisons between groups. Statistical comparison of the DRL 8-64 group and the DRL 8-Ext group indicated that there was a significant effect due to groups (F = 9.16, df 1/10, p < .025), but no significant effect due to stimuli (F < 1) or to the inter- action of groups and stimuli (F < 1). Statistical compari— son of the DRL 8-64 group gradient with the DRL 64 group gradient indicated no significant differences due to either groups, stimuli, or the interaction of groups and stimuli (all F values less than 1). Comparisons between the DRL 8-Ext group and the DRL 64 group showed no significant difference due to groups (F = 4.43, df 1/10, .05 < p < .10), no significant effect due to stimuli (F < l), and no sig- nificant interaction (F < 1). Therefore, the only signifi— cant difference found in the generalization data was that the DRL 8-Ext group emitted significantly fewer responses during the generalization test than did the animals in the DRL 8—64 group. Both the DRL 8-64 group and the DRL 8-Ext group were given a second generalization test along the wavelength dimension, the dimension correlated with the DRL 8-sec com- ponent. Again, the general tendency was for an excitatory gradient reaching a peak at the training stimulus (555 nm), 57 with the fewest number of responses at the peripheral stimuli (501 and 606 nm). Two birds in the DRL 8-64 group and one bird in the DRL 8-Ext group showed a slight devi— ation from this pattern with a maximum number of responses at either 538 or 576 nm. Figure 11 shows the mean number of responses for both groups at each of the five wavelength stimuli. Both group gradients are excitatory and symmetrical. Statistical analysis of the absolute number of responses during general- ization revealed that the effect of stimuli was significant for both the DRL 8-64 group (F = 9.79, df 4/20, p < .001) and the DRL 8-Ext group (F = 15.69, df 4/20, p < .001). Statistical comparison of the group gradients with each other indicated no significant difference due to groups (F < 1) and no significant interaction (F < 1). Of course the overall effect of stimuli was highly significant (F = 23.73, df 4/40, p < .001). Response Rates Figure 12 shows the mean daily response rates of both discrimination groups for the DRL 8-sec component (31) and the mean daily response rates of all groups for either the DRL 64—sec component (the DRL 8-64 group and the DRL 64 group) or Ext (the DRL 8-Ext group). This component is designated as $2. The points at Session 0 represent the group mean average for the last five days (days 16 to 20) of Phase I (acquisition) training. These last five days of 58 Figure ll.--Group generalization gradients expressed in terms of the mean number of responses per group to each of the five wavelength stimuli, around the 555 nm stimulus which was correlated with the DRL 8-sec component for both groups during discrimination training in Phase II. MEAN RESPONSES 59 093]. 8-DRL 64 301' - 0m - E x t _ 71 I \ 701— ’ \ / b l \ I \ 60'- , \ / \ — / \ / \ so I L l— J \ .. / \ I \ / \ 40H— / \ / \ ’ \ 30L- / \ I \ 1. / \ J \ 201— I. 'r l I l | SOI 538 555 576 606 WAVELENGTH 60 .H mmmsmmhemmmp m>flm ummH on» mcfiuso ucmcomEoo HMHSOHDHMQ some now open oncommou mmmuo>m msoum m mucmmonmou =0 coflmmmm= um ucflom one .mcHH H00HDH0> muflnz onu .mm spas poumHmHuoo neocomaoo ecu Ugo .mdaseflum E: mmm on» .Hm AuHB woumamumoo unocomEoo msu How HH ommnm mcfluso AGHE mom momcommmuv moumu oncommmu msoum ommuw>m maflmott.ma mnnmam 61 mZO.mmwm ON 0— n.— O— m p O . _ . 1 . _ :o t./ It“ I. I, .I, \I, .1— ( ( #08, 1a (1!... x 1m 4 \It/tl/l v I‘ L I L 4 4/\ / .10; t = 2.49, df 5, p > .05, respectively, two-tailed). Inspection of the daily response rates during the DRL 8—sec component (81), shown in Figure 12, for the DRL 8-64 group and the DRL 8-Ext group during Phase II shows an increase in the rate of responding. With the exception of the first day for the DRL 8-64 group, every daily average response rate for both groups during Phase II is higher than the baseline rate. In general the rates of responding tended to increase across the 15 days of Phase II with the largest change occurring during the first five days, al— though there was considerable day-to-day variability. For days 11 through 15 the average increase from the baseline was 2.0 responses per minute for DRL 8-64 group and 2.6 responses for the DRL 8-Ext group. For statistical analysis the rates of responding were grouped into 5—day blocks, thus collapsing the data into five blocks (the last five days of Phase I and 20 days of Phase II). A two-factor AOV with repeated measures on one factor was performed using the response rates during the DRL 8-sec component for the DRL 8-64 and the DRL 8-Ext grOUpS across the five S-day blocks. The results showed no significant effect due to groups (F < 1) and no significant interaction between groups and 5-day blocks (F < 1). How- ever, the effect of the 5-day blocks was significant 64 (F = 4.27, df 4/40, p < .01). In order to determine which 5-day means differed from each other, a Newman-Keuls test was performed. The results indicated that each 5-day mean during Phase II was significantly higher than the baseline (p < .01 for each comparison). In addition, the mean for days 16 through 20 was found to be significantly higher than the mean for days 1 through 5 of Phase II. All other comparisons were not significant (p < .05). This result indicates that, for this experiment, discrimination training in general led to behavioral contrast; i.e., an increase in the rate of responding in the constant component. Reinforcement Rates Table 3 shows the 5-day average number of rein- forcements per 100 min for days 1 through 15 for the groups that responded on the DRL 64-sec component in the presence of 82. Both individual and group averages are shown. For the DRL 8-64 group the reinforcement rate increased to its maximum value during the second 5-day block, and then showed a slight decrease for days 11 through 15. For the DRL 64 group the mean average reinforcement rate decreased during days 6 through 10 when compared to days 1 through 5. The reinforcement rate then increased to its maximum for days 11 through 15. For both groups there are large in- dividual differences in magnitude and in pattern across the three 5-day blocks. 65 TABLE 3.--Reinforcements per 100 min for the three S—day periods during Phase II (days 1 to 15) for the multiple DRL 8-sec - DRL 64-sec group and the DRL 64-sec group in the presence of 82, the DRL 64-sec component. Reinforcements Per 100 Minutes Subjects Days 1-5 Days 6-10 Days 11-15 Multiple DRL 8-sec - DRL 64-sec 508 6.2 8.1 7.9 216 5.6 5.0 3.0 109 1.9 2.5 1.9 17 4.3 9.3 13.7 2,618 5.6 12.1 8.7 19 7.0 6.2 5.6 Mean 5 l 7.2 6.8 DRL 64-sec 23 13.1 8.1 11.2 185 5.0 5.0 5.0 3,821 3.7 3.7 6.2 110 3.1 6.2 6.8 298 5.0 5.7 7.0 658 4.3 3.7 9.3 Mean 5.7 5.4 7.6 Total Mean 5.4 6.3 7.2 66 A two-factor AOV with repeated measures on one factor (three 5—day blocks) was carried out for comparisons between groups. There was no significant effect due to groups (F < 1), blocks of days (F = 2.23, df 2/20, p > .10), or the interaction of blocks with groups (F = 1.44, df 2/20, p > .20). Table 4 shows the 5-day average number of rein- forcements per 100 min for the five 5-day blocks (days 16 to 20 of Phase I and days 1 to 20 of Phase II) for the DRL 8-sec component correlated with 81 for the DRL 8-64 group and the DRL 8-Ext group. The reinforcement rates for the DRL 8-64 group were monotonically decreasing across the five blocks of days, while the reinforcement rate for the DRL 8-Ext group decreased and then increased. This was also the case for the combined mean reinforcement rate. Again, within either group, there were some animals whose rein- forcement rate varied from the group average. Statistical comparison of the groups across the 5—day blocks again indicated no significant differences due to either groups (F = 1), blocks (F = 1.11, df 4/40, p > .20), or the interaction of groups with blocks (F==2.53, df 4/40, p > .05). Thus, even though there was a signifi- cant increase in the rate of responding during the DRL 8-sec component for the discrimination groups in this experiment there were no significant changes in the reinforcement rate, although there was a slight drop initially. The largest decrease in reinforcement rate occurred between the 67 TABLE 4.--Reinforcements per 100 min for the five 5-day periods (days 16 to 20 of Phase I and days 1 to 20 of Phase II) for the multiple DRL 8-sec - DRL 64-sec group and the multiple DRL 8-sec - Ext group during the DRL 8-sec component (82). Reinforcements Per 100 Minutes Phase I Phase II Phase II Phase II Phase II Subjects Days Days Days Days Days 16-20 1-5 6-10 11-15 16-20 Multiple DRL 8-sec - DRL 64-sec 508 129.7 117.5 92.8 136.5 118.1 216 226.8 226.3 201.9 163.0 78.4 109 376.6 352.8 284.2 266.3 288.3 17 297.8 59.6 111.8 85.5 70.4 2,618 292.0 234.6 267.7 233.6 212.6 19 284.9 349.4 332.8 367.5 336.8 Mean 251.3 223.4 215.2 208.7 184.1 Multiple DRL 8-sec - Ext 275 316.8 338.6 243.1 256.2 361.9 245 304.5 306.7 240.4 309.4 374.7 242 173.4 125.9 88.1 121.8 148.4 833 274.5 239.3 227.9 241.7 287.4 757 216.6 211.0 335.2 310.9 345.7 29 107.6 126.0 88.9 61.6 44.8 Mean 232.2 224.6 203.9 216.9 260.5 Total Mean 241.8 224.0 210.0 212.8 222.3 68 baseline days and days 6 through 10. For 10 out of 12 animals the average reinforcement rate was lower for days 6 through 10 when compared to the five baseline days. Feedback and Generalization Gradients The overall effect of feedback can best be seen by comparing the subjects in Experiment [(feedback) with those in Experiment II (no feedback). In addition, the overall effect of the schedule contingencies in 52 were of im- portance. Therefore, a three-factor AOV with repeated measures on one factor (stimuli) was carried out for each of the three possible comparisons for the generalization gradients along the line-orientation dimension. The re- sults of comparing the DRL 8-64 groups with the DRL 8—Ext groups in both experiments are shown in Appendix A. This analysis indicated that the only significant difference was due to the schedule contingency (p < .01). This means that the DRL 8-Ext groups emitted fewer responses during the generalization test than the DRL 8-64 groups. There was no significant effect due to feedback or stimuli, and no sig- nificant interactions. The results of the statistical comparison of the DRL 8-64 groups with the DRL 64 groups in both experiments are shown in Appendix B. In view of the past analyses it is not surprising that the results yielded no significant effects due to feedback, schedule contingencies, stimuli, and no significant interactions. 69 The DRL 64 groups were also compared with the DRL 8-Ext groups in both experiments. This statistical com- parison (Appendix C) indicated a significant effect due to the schedule contingency (p < .05). This result indicates that the DRL 8-Ext groups emitted significantly fewer re- sponses during the generalization test than the DRL 64 groups. The overall effect of feedback and of stimuli was not significant. Of the four interactions, only the Feed- back X Stimuli interaction was significant (p < .05). This result, combined with inspection of the data and previous results, indicates that the gradient for these two feed— back groups is inhibitory and of different shape than the combined gradient for the no-feedback groups. Further analysis indicated that this interaction was primarily due to the difference in the gradients of the DRL 64 groups. Statistical comparison of the DRL 64 feedback gradient with the DRL 64 no-feedback gradient showed no group effect (F < 1), no treatment effect (F < 1), but a significant groups and treatments interaction (F = 2.66, df 5/50, p < .05). On the other hand, the same comparison for the two DRL 8-Ext groups yielded a significant treatments effect (F = 2.89, df 5/50, p < .05), but no significant group (F < l) or interaction effect (F < 1). The generalization gradients along the wavelength dimension were also analyzed in a similar manner. A 2 X 2 X 5 (repeated measures) AOV indicated that there was 70 a significant effect due to feedback (p < .005, Appendix D). This indicates that the feedback groups (Experiment I) emitted fewer responses during the generalization test along the wavelength dimension than did the no-feedback groups. As would be expected there was a significant ef- fect due to stimule (p < .001), indicating an excitatory generalization gradient. The only other significant effect was due to the interaction of feedback with stimuli (p < .05). At first glance this result suggests that the combined gradients for the feedback vs. no-feedback groups have a different shape, but this is a matter of interpre— tation. The statistical analysis measures the mean differ— ence in the number of responses across the five wavelength stimuli. The analysis indicates that at least two of these means are different. Psychologically, the important datum is the per cent of each groups total responses emitted at each point. Table 5 shows both the total number of re- sponses and the per cent of each groups total responses for the feedback and no-feedback groups, as well as the differ- ence, across the five wavelength stimuli. It is clear that the difference in the number of responses differs across the five wavelength stimuli, but the percentages do not. Therefore, the shape of the two gradients do not differ. In summary, the feedback variable generally had little effect on the gradients along the line-orientation dimension. Only the combined effect of the DRL 64-sec 71 TABLE 5.--The total number of responses and the per cent of each groups total responses at each of the five wavelength stimuli for the feedback and no-feedback stimuli. Number of Responses 501 nm 538 nm 555 nm 576 nm 606 nm Feedback 172 407 591 428 214 No—Feedback 297 655 889 660 247 Difference 125 248 298 232 37 Per Cent of Responses Feedback 09 22 33 24 12 No-Feedback 11 24 32 24 09 Difference 02 02 01 00 03 72 contingency in isolation and feedback produced a reliable single-group inhibitory gradient. Feedback led to fewer responses during the generalization test along the wave- length dimension but no difference in the shape of the gradients. With respect to schedule contingencies, the Ext component for the DRL 8-Ext groups led to fewer responses during generalization testing than for either the DRL 8-64 group or the DRL 64 group. Feedback and 81 Response Rates The determination of the effect of feedback on the rate of responding in the DRL 8-sec component (81) was a primary reason for Experiment II. Since the response rates during the DRL 8‘sec component for the DRL 8-64 groups and for the DRL 8-Ext groups did not differ from each other in Experiment I or Experiment II, the data were collapsed into two groups: feedback (F) and no feedback (NF). Figure 13 shows the daily mean response rates for days 6 through 20 of Phase I and days 1 through 20 of Phase II for the feedback subjects and the no-feedback subjects. The vertical lines indicate :1 standard error of the mean for each point. It is obvious that the no-feedback group main- tained a higher rate of responding, since every daily rate is higher for this group than for the feedback group, and the daily pairs of standard errors never overlap. It is also apparent that the response rate variability is greater for the no-feedback group. For every daily comparison but 73 an: .msonm nooo new cooE onu mo Houuo wumocoum a H ucomoumon pawom sumo um mocfia amoeuuo> one .HH ommnm mo om CD A memo paw H omofim mo om cu 0 meme How ucocomeoo oomtm me on» mcwuse mocHnEoo mmooum flmzv Moonooomtoc osu Ugo masonm xoonooom onu How mouou omcommon eaflop ommuo>o onett.ma ousmem mZO.mmmm o o m— m! J. _ OL— M LOW m. o— 74 T 2mm<1m .f .mm<1m J- NIW/SJSNOJSSB 75 one the standard error was greater for the no-feedback group. Of particular interest was the effect of discrimi— nation training (Phase II) on the rate of responding in the 81 component. The curve for the feedback group shows a slight rise on day two of discrimination training from the baseline level, followed by a general slow and slight de- crease in the rate. The curve for the no-feedback group shows a sharper increase in the rate of responding at the onset of discrimination training, followed by a slower but general increase in the rate of reSponding. Between days 15 and 16 of Phase II, the line-orientation generalization test was administered. It can be seen from Figure 13 that this generalization test was followed on the next day by a sharp increase in the rate of responding for the no-feedback subjects but little change for the feedback group. Statistical comparisons between the feedback and no-feedback groups were not carried out due to the differ- ence in variability. But previous analyses showed that the feedback subjects showed no significant response rate changes (Experiment I) while the no-feedback subjects showed a significantly higher rate of responding for each of the four 5-day periods of Phase II compared to the last five days of Phase I. The conclusion is that the no- feedback subjects showed behavioral contrast while the feedback subjects did not. 76 Figure 14 shows the per cent of the total number of responses which had IRTs of 0.8 sec or less for the feed- back and no-feedback groups for days 6 through 20 of Phase I and days 1 through 20 of Phase II. Daily standard errors for each group are also shown. The difference in the per cent of bursts is quite clear. Also, the difference in variability is large and stable. After day 10 of Phase I each daily standard error for the feedback group is smaller than for the no-feedback group. For Phase II the standard error of the no—feedback group is about five times that of the feedback group. There is little change in the per cent of 0.8 sec IRTs for the feedback group. After day 10 of Phase I the range of values was from a maximum of 3.3 per cent to a minimum of 1.8 per cent. In contrast, the per cent of 0.8 sec IRTs shows much more day-to-day variability for the no- feedback group. The per cent shows an irregular decrease during Phase I to a minimum at day 20. This is followed by an increase during Phase II. During Phase II the per cent of 0.8 sec IRTs varied between 30 and 36 per cent. This percentage is very similar to the percentage increase in the rate of responding of the no-feedback group over the feedback group. The data suggest that the majority of the difference in the rate of responding of the feedback and no-feedback groups is accounted for by the per cent of 0.8 sec IRTs. 77 I .msoum sumo How cmoE onu mo Honuo mummcmum H + ucomoumou ucflom sumo um mocfla Hmofluuo> one .HH ommnm wo om ou H memo mom H ommnm mo cm on m memo How unocomeoo oomtm qmo one mcfluso mocHQEoo mesoum Amzv xomamoomtoa one mam Amv xomnmoom onu How mmoa no oom m.o mo memH Suez momcommou Hmuou on» mo ucoo mom eaflmm ommuo>m onett.va ousmflm loam}..— 0 78 mZO_m—mum Sm— m. figgo to— “.z I a . Law On m :mm41m 1T .mm.10 Error 20718.50 20 1035.88 Within 55 6704.33 120 C (Stimuli) 225.14 5 45.03 <1 -- AC 90.34 5 18.07 <1 -- BC 369.38 5 73.88 1.26 >.20 ABC 134.30 5 26.86 <1 -- Error 5885.17 100 58.85 88 APPENDIX B TABLE B-l.--Summary table for the analysis of variance for the DRL 8-64 groups, with and without feedback, and the DRL 64 groups, with and without feedback, gradients along the line-orientation dimension. for the generalization Source of Variance SS df MS F p Between 55 30398.31 23 A (Feedback) 305.39 1 305.39 <1 -- B (Schedule) 106.78 1 106.78 <1 -- AB 2241.30 1 2241.30 2.16 >.lO Error 27744.84 20 1038.24 Within 85 10812.33 120 C (Stimuli) 116.97 5 23.39 <1 -- AC 493.61 5 98.72 1.09 >.20 BC 297.39 5 59.48 <1 -- ABC 818.53 5 163.71 1.80 >.10 Error 9085.83 100 90.86 89 APPENDIX C TABLE C—1.--Summary table for the analysis of variance for the DRL 64 groups, with and without feedback, and the DRL 8-Ext groups, with and without feedback, for the generali- zation gradients along the line-orientation dimension. Source of Variance SS df MS F p Between 55 31083.23 23 A (Feedback) 950.70 1 950.70 <1 - B (Schedule) 7140.25 1 7140.25 6.21 .05 AB 1.00 1 1.00 <1 - Error 22991.28 20 1149.56 Within 55 6816.33 120 C (Stimuli) 300.39 5 60.80 1.16 20 AC 633.47 5 126.69 2.41 .05 BC 110.92 5 22.18 <1 - ABC 520.16 5 104.03 1.98 .05 Error 5251.39 100 52.51 9O APPENDIX D TABLE D-l.--Summary table for the analysis of variance for the DRL 8-64 groups, with and without feedback, and the DRL 8-Ext groups, with and without feedback, for the generali- zation gradients along the wavelength dimension. Source of Variance SS df MS F p Between 85 22040.80 23 A (Feedback) 7300.80 1 7300.80 10.27 <.005 B (Schedule) 128.13 1 128.13 < 1 ~- AB 388.80 1 388.80 < 1 -- Error 14223.07 20 711.15 Within 85 50373.20 96 C (Stimuli) 32322.91 4 8080.73 42.98 <.001 AC 1901.11 4 475.28 2.53 <.05 BC 535.95 4 133.99 < 1 ABC 574.29 4 143.57 < 1 Error 15038.93 80 187.99 91 IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII llIlllllfllllijlfllw|1|W|jtflifl|lfllfllflllfllHI)