INDUCED MOTIVATION. April Melissa Becker, B.S. Thesis Prepared for Degree of MASTER OF SCIENCE UNIVERSITY OF NORTH TEXAS. August 2011 APPROVED:

Transcription

1 INDUCED MOTIVATION April Melissa Becker, B.S. Thesis Prepared for Degree of MASTER OF SCIENCE UNIVERSITY OF NORTH TEXAS August 2011 APPROVED: Jesús Rosales-Ruíz, Major Professor Shahla Alai-Rosales, Committee Member Jonathan Pinkston, Committee Member Richard Smith, Chair of the Department of Behavior Analysis Thomas Evenson, Dean of the College of Public Affairs and Community Service James D. Meernik, Acting Dean of the Toulouse Graduate School

2 Becker, April Melissa. Induced motivation. Master of Science (Behavioral Analysis), August 2011, 54 pp., 1 table, 20 figures, references, 29 titles. In the avian training community, a procedure has been utilized to maintain food reinforcer efficacy at high body weights. Elements of this procedure include limited holds and closed economies. To test this procedure, a baseline performance of keypecking on an FR 15 schedule at 80% ad lib weight for two pigeons was established. By imposing limited holds and a closed economy, rates of responding were increased compared to baseline, even while the pigeons were over 90% of their ad-lib body weights.

3 Copyright 2011 by April Becker ii

4 ACKNOWLEDGEMENTS Without the continual and flexible support from my advisor, Jesus Rosales-Ruiz, I would not have been able to undertake this experiment. He provided permission to explore the topic that I wished and extended that support to me through the two years of necessarily slow work on apparatus construction and procedural experimentation before the meat of data collect could even begin. Not only did he give me permission to follow my own reinforcers, he did everything in his power to provide me with the practical opportunity to do so and, once the opportunity came to fruition, encouraged me to behave creatively with respect to my goals. The project would also not have been possible without the help of my brother Damon Becker, whose talent as an innovative engineer (capable of working within my budget) enabled me to construct my apparatus. His infinite patience and willingness to spend hours on the phone while I constructed, asked questions, or troubleshot problems provided me with an entirely separate education. I would also like to thank Dave Palmer, Allen Neuringer, and Manish Vaidya, who personally advised and supported me in the construction of my apparatus. Any quality in this study directly stems from their encouragement and teaching. I also want to thank Dr. Neuringer for generously lending me a manufactured pellet dispenser, without which the project would have been much more difficult. Of course, the study owes its existence to Steve Martin, Cassie Malina, Phung Luu, Barbara Heidenrich, and Sid Price, whose professional achievements and talent inspired it and whose continual advice made it possible. I owe enormous thanks to the supporters and donors behind the Society for the Advancement of Behavior Analysis, who provided the grant which funded this experiment. I owe perhaps the most to my family and friends who provided continual encouragement, especially to my parents, my siblings, and Daniele Ortu, whose supporting enthusiasm never waned for a second. iii

5 TABLE OF CONTENTS Page ACKNOWLEDGEMENTS... iii LIST OF TABLES...v LIST OF FIGURES... vi INTRODUCTION...1 METHOD...8 RESULTS...15 DISCUSSION...23 REFERENCES...51 iv

6 LIST OF TABLES Table 1: Hourly Deprivation...30 Page v

7 LIST OF FIGURES Page Figure 1: Overall rate vs. pre-session weight, Subject Figure 2: PRP vs. pre-session weight, Subject Figure 3: IRI vs. pre-session weight, Subject Figure 4: Overall rate vs. pre-session weight baseline comparisons, Subject Figure 5: PRP vs. pre-session weight baseline comparisons, Subject Figure 6: IRI vs. pre-session weight baseline comparisons, Subject Figure 7: Overall rate vs. pre-session weight scatter plot, Subject Figure 8: PRP vs. pre-session weight scatter plot, Subject Figure 9: IRI vs. pre-session weight scatter plot, Subject Figure 10: Overall rate vs. pre-session weight, Subject Figure 11: PRP vs. pre-session weight, Subject Figure 12: IRI vs. pre-session weight, Subject Figure 13: Overall rate vs. pre-session weight scatter plot, Subject Figure 14: PRP vs. pre-session weight scatter plot, Subject Figure 15: IRI vs. pre-session weight scatter plot, Subject Figure 16: Final baseline sessions IRI probabilities...46 Figure 17: Timeouts, non-utilized opportunities, and reinforcers per minute, Subject Figure 18: Timeouts, non-utilized opportunities, and reinforcers per minute, Subject Figure 19: Graphs using first 20 minutes of experimental sessions, Subject Figure 20: Graphs using first 20 minutes of experimental sessions, Subject vi

8 INTRODUCTION Avian trainers have utilized a practice dubbed psychological appetite for many years in order to maintain the effectiveness of food as a reinforcer even when birds are heavy relative to their free-feed weight. The complex procedure was developed fairly recently - within the past several decades. The trainers incentives to develop such a procedure included issues related to the physiology of birds in particular. Birds do not store fat in the same way as mammals, and when they are maintained at very low body weights, it can adversely affect their health. This point becomes more relevant for birds who are housed outdoors overnight in cold climates. Without sufficient energy reserves, these birds can easily die from the cold. In addition, birds who are performing physically demanding routines in shows and demonstrations need sufficient caloric intake to maintain rigorous daily exertion and the high muscle mass required for it. Such a bird s performance and health may be affected if they are constantly energy restricted. For a trainer, the ability to effectively maintain fluent, non-hesitant behavior holds importance along with these considerations. In a free-flight bird show, the animals work under choice conditions; the opportunity to fly away and contact many other reinforcers elsewhere remains constantly available. Thus, trainers have a unique interest in keeping their birds heavy without diminishing the reinforcing power of food; it is possible that this environment has contingency shaped them into accomplishing such ends. Trainers developed their procedure under the rationale that a wild animal competing for limited food resources will be more likely to eat slightly more than they physically need, should the opportunity present itself. This behavior pattern should be adaptive, since extra food consumed under such conditions may enable the animal to survive longer through food shortages. Several avian trainers have described a version of this procedure explicitly, stating 1

9 that it increases the hunger of birds even at high body weights relative to ad lib (Malina & Martin 2003). Though the procedure has never been tested under controlled conditions, trainers have a great deal of non-controlled data for at least dozens of animals (including weight records and training logs) that indicate that the procedure might produce behavior with low latencies and high topographical magnitudes even when the birds are close to or over 100% of their ad-lib weights. Even taking for granted that these observations are reliable and that the phenomenon directly proceeds from the operations that the trainers describe, the complexity of the procedure presents innumerable variables as candidates for involvement. Additionally, because deprivation has been most frequently investigated in the laboratory under free operant conditions, the hunger that the trainers describe under their discrete trial conditions may or may not be synonymous with the effects of food deprivation as classically studied. Description of Procedure At the beginning of the procedure, the animal starts at lib weight. The animal is introduced to the training situation and his/her normal daily ration (the amount eaten daily under free-feeding conditions) is adjusted according to the trainers assessment of hunger in the session. Usually, this involves food restriction for the day. The animal continues to be restricted and their weight drops rapidly until the trainers determine that he/she is acting hungry in session. The criteria by which the animal is judged to be sufficiently hungry may vary dynamically, and so may the amount of weight initially dropped. Training in a new environment, training a new behavior, training with a novice animal, or training with new trainers might inspire larger weight drops than otherwise. On the other hand, some trainers will often drop weight only after all other manipulable variables that might affect the animals performance have been adjusted or made constant, so the weight drop might be slight. Once the animal is deemed hungry, their 2

10 daily food ration is increased to maintain their weight from day to day. After shaping is completed or the training environment stabilizes or once they ve been deemed hungry at the same weight for several days, their diet will be adjusted so that they can gradually gain weight. Weight gain is purposefully much slower than weight loss. If response latencies, body language, or other cues cause the trainer to describe a loss of motivation, food may again be restricted and weight may again drop. The idealized result of this procedure is a pattern of weight loss and gain (Malina & Martin, 2003). Weight gains, though more gradual than losses, eventually sum to the point that the animal works at weights close to or above ad lib. Trainers conduct the procedure by controlling two things: the amount of food an animal gets during the day and the contingencies under which they receive the food. They control these things in a dynamic way, responding to multiple conditions in order to make decisions. The three major stimuli involved in controlling those decisions are. the behavior of the bird during training sessions or performance opportunities, the nature of the training environment or task relative to the bird s history, and the current weight and weight history of the bird. The behavior of the bird can be evaluated according to latency, magnitude, topography, etc. The training environment or task may affect decisions by virtue of being novel or complex. Additionally, the bird may or may not have a history of training in strange environments, may or may not be experienced at learning new behaviors, or there may be other contingencies in the environment (such as an aversive object) to consider. The recent weight history of the bird (gain or loss, magnitude of change) might also weigh into the decision, as may more distant history such as the bird s historical performance under different levels or patterns of weight regulation. A few tangential points about the environments created by these trainers deserve mention. First, the animals involved almost never receive food outside their training 3

11 contingencies. Small amounts of food are used to maintain husbandry and management behaviors (such as shifting from holding to stage environments and back), but those rations are purposefully as small as possible. Thus, the situation can be conceived of as a largely closed economy, because no supplemental food is provided outside the contingency (Bauman, 1991). Additionally, the trainers always employ a strict limited hold. The limited hold may be as long as 5-10 seconds during conditions of shaping or early training; but once the response is shaped, it is rarely as much as 5 seconds. The discrete-trial nature of the conditions as well as the short limited hold may be necessary to produce the observed effects. Deprivation Behavior analysts often describe deprivation (and, more broadly, establishing operations) as antecedent states affecting reinforcer efficacy. The efficacy of a reinforcer typically is inferred by the rate, the accuracy, or another dimension of responses maintained by the reinforcer in question. Many basic experiments control for deprivation by normalizing the pre-session weight of the animal and describing the weight as a fraction of ad-lib weight, which is the individual s weight under free-feeding conditions. Experiments have investigated deprivation by altering body weight via post-session feedings and, under several weight values, testing some dimension (see above) of responses maintained by food or water (i.e. Holtz & Azrin, 1963; Herrnstein & Loveland, 1974). Others have altered the periods of pre-session hourly deprivation or the removal of food for a certain number of hours before the session (i.e. Clark, 1958; Belke, Pierce & Jenson, 2004). Still others have examined the effects that the restriction of one reinforcer may have on behavior maintained by a different reinforcer (i.e. Belke, Pierce & Jenson 2004, Premack & Premack 1963) or the 4

12 effects that antecedent manipulations or events (such as exposure to an activity or to demands) might have on responses (i.e. O Reilly et al., 2008; Carbone et al., 2007). The dimensions of behavior affected by deprivation and, more broadly, motivating operations vary. Many basic studies examine the effects of deprivation on responses per minute and/or inter-response interval (IRI) and post reinforcement pause (PRP) in a free-operant scenario where those dimensions are not prescribed by the contingency (i.e. Sidman & Stebbins, 1954; Schull, 2004; DeMarse, Kileen & Baker, 1999; Belke, Pierce & Jensen, 2004.). Other measures shown to be sensitive to deprivation include progressive ratio responding, the shape of extinction curves (Skinner, 1938), and avoidance behavior (Leander, 1973). In addition, different dimensions of a response are not equally affected by deprivation operations. For example, PRP and IRI are not equally affected by deprivation under VI schedules (Shull, 2004). In sum, the literature may lead scientists to take deprivation and establishing operations as broad categories of antecedent operations that share the same behavioral process. In the case of food deprivation, pre-session body weight is used as an index of this deprivation state. Yet, despite the common terminology used to refer to the operations involved, the data don t establish the existence of any single-dimensional deprivation state that controls the efficacy of a reinforcer. It establishes that individual manipulations that control body weight also predictably affect responding under certain conditions. Yet not just that manipulation, but others including hourly food restriction, other MOs, schedules of reinforcement, and manipulation of stimulus control elements can affect the same dimensions of a reoccurring response. The interrelations between these manipulations have not yet been thoroughly characterized, nor has it been established that they don t operate by means of common underlying behavioral processes, nor has it been established that any subset affects a motivational state, while others do not. 5

13 Though some work has been done to make that separation (Michael 1982), the effects of operations in both categories still share common effects on dependent measures, and no other measurement of motivation is forthcoming apart from those common effects. Further research in deprivation should ideally either demonstrate the known behavioral processes that underlie deprivation effects or else uncover evidence of a separate process. If the trainers procedure described here effectively produces high rates of responding at high body weights, it could be a step in the direction of elucidating the mechanism of deprivation operations. If the conditions of the procedure break the relationship normally seen between body weight and response measures, it could aid in the discovery of why that relationship normally holds to begin with. Translation to the Operant Chamber The current investigation sought to explore the possibility of replicating the described effects of the avian trainers procedure in the operant chamber so that it could be analyzed with regard to classically described deprivation. Two possible initial strategies might be employed toward these ends. One is to bring the variables involved in the procedure into the operant chamber one at a time and test their individual effects. The other is to try to transfer the complex procedure and adjust its elements until weight gain is achieved without major decrements in particular response measures. After such a transfer, the different variables involved in the procedure might be tested individually against that baseline. This investigation takes the second approach and constitutes only the first step of the progression. It attempts to generate the complex procedure in the operant chamber so that weight gain can occur without major decreases in various measures of response rate. 6

14 Several considerations contribute our choice of this strategy. First, as Sidman (1960) points out, the first step in investigation is often to establish that a particular phenomenon is possible; The introduction of a new control technique may result in the demonstration of a previously unobserved, unmeasured, or uncontrolled type of behavior. Very often, however, experiments are carried out for the specific purpose of demonstrating a particular behavioral effect. Second, the possibility that the avian trainers field observations resulted from the combined or conditional effects of the various aspects of their procedure meant that testing individual variables without the possibly necessary conditions provided by the others might have led us to miss the effect. Third, if the effect was not caused by the described procedure at all, testing individual elements would hardly be expected to bear any relevance to our purpose and would have taken more time and resources away from other research. 7

15 METHOD Subjects Subjects for this experiment were two pigeons captured humanely from the north side of Dallas. The subjects were under veterinary care for the duration of the experiment, and institutional (IACUC) approval was granted for all procedures. The birds re-release sites are identical to those in which they were captured. U.S. Fish and Wildlife issued a letter of permission for the trapping, the experimental process, and the re-release of the birds. The sex of the subjects was unknown (they will both be referred to as he ). Apparatus All experiments were conducted in an operant chamber 12 x 12 x 15.5 with a single key 8.5 from the floor. A pellet dispenser delivered food to a cup that was accessible through an opening 2 from the floor, directly under the key. The edge of the opening was set 1 above the floor of the cup, and an infrared beam was set 1.5 from floor of the cup and directed parallel to the edge of the opening. This beam, when broken, was used to measure when the bird s head was in the cup. Measurement Keypecks were measured by the closure of a switch behind the key. Headpokes were measured by the breaking of the infrared beam in the food cup. Weight was measured to the nearest gram with a kitchen gram scale on which the birds stood. A number of different dependent variables were derived from these raw measurements and their relation to programmed events such as the keylight (on or off) and the dispenser (pellet delivery). IRI (inter-response interval). IRI measurements were the time from the onset of one keypeck to the onset of the next, excluding the intervals in which both a pellet delivery and a 8

16 headpoke occurred and the intervals in which the onset of the keylight occurred (ergo, all darkened-key intervals were counted unless it was the last one of a dark key period). IRI measurements during the dark period that were longer than 5 seconds were later eliminated from the data, because Subject 1 acquired a superstitious dark-peck response that had IRIs significantly longer than those to the lighted key. All IRI measurements in a session were used to calculate an mean and a median for that session. PRP (post-reinforcement pause). PRP measures were defined relative to the first headpoke after pellet delivery. The PRP was the time between the offset of this headpoke and the next keypeck, even if other headpokes occurred in the interim. All PRP measurements in a session were used to calculate an mean and a median for that session. Overall rate. The overall rate for a session was determined by taking the total number of keypecks that occurred to the lighted key and dividing it by the entire amount of time (in minutes) that the keylight was on during that session. Reinforcers per minute. This measure was determined by taking the total number of pellets delivered in a session and dividing it by the entire amount of time (in minutes) that the keylight was on during the session. Missed opportunities per minute. We counted the number of times in each experimental session that the keylight lighted and was then not pecked for 10 seconds (after which it darkened again) and divided this number by the total session length to derive the Missed Opportunities per Minute measure. Data analysis. Session analyses took into account all data from baseline sessions, but only the first part of experimental sessions. Experimental data was considered up to the point at which the key had been lit for a cumulative of 20 minutes in that session or until the end of the 9

17 session if the key was lit for a cumulative of less than 20 minutes (which was rarely the case). We analyzed the data this way to control for the variability of session length that might otherwise have affected session means, and to make the data easily comparable to baseline sessions. We also measured experimental sessions for the first 20 minutes of each session regardless of keylight status. These measurements are presented separately in Figures 19 and 20. In baseline sessions, all measures were the same since the keylight did not turn off. Preliminary Procedure Acquisition of subjects and dietary changes. After their capture, each subject was immediately driven to an apartment in Denton, TX where they were given unlimited food and water and allowed to calm down in a covered, dark room for a few days. They were provided with any food they would eat at first, mixed with Nutriblend Gold TM pigeon pellet and a mix of rapeseed, millet, and canary seed. They were weighed daily using a scale inside their cage, as was their food. Each bird visited the veterinarian once within their first week of capture. Otherwise, they never left the apartment. Once the birds were eating the seed and pellet reliably, other food was phased out. Once the birds were eating pellet reliably, the seed was scaled back to 5g per day. Each bird was then taught to eat TestDiet TM brand pellets by providing unlimited TestDiet and limiting their Nutriblend Gold TM until they would eat the TestDiet TM without hesitation. After this period, they were again provided unlimited Nutriblend Gold TM and 5g of seed per day. They never again were provided TestDiet TM in the home cage, though it was all that was available in the experimental chamber. All supplemental feeding in the home cage was Nutriblend Gold TM and seed. The birds had unlimited access to water in their home cage, and grit was provided once per week. After the experiment had been underway for some time, grit became always available in the home cage. 10

18 Establishment of ad-lib weights. The subjects weights were allowed to adjust for at least two weeks after all diet changes were completed or until the weight stabilized. Stability was assessed visually; the weight measurements had to be showing no trend for at least a week to be considered stable. The mean of all stable weight measurements was then used as 100% ad lib weight. For Subject 1, ad lib was 344g, for Subject 2 it was 328g. Shaping sessions. Once ad-lib weight was established, the birds food bowls were removed from their home cages about 12 hours before their first training session. In these sessions, they were shaped to peck a lightened key, not to peck a darkened key, and their schedule was adjusted to FR 15. They were sometimes provided with food between shaping sessions and sometimes were not. Shaping, discrimination training, and schedule adjustment took approximately three to five sessions per bird. Sessions varied in length. During shaping, Subject 1 was always 322g (93.6% ad lib) or heavier. Subject 2 was always 311 (94.8%) or heavier. Data Collection Baseline. After initial training, the birds entered baseline. They were provided with 20 minutes in the chamber roughly every 12 hours. The keylight darkened after the 20 minutes had elapsed. Subject 1 was fed in the home cage after the session at the discretion of the experimenter (if she thought that weight loss was too rapid, or more than approximately 4g per day) until he reached 80% of baseline. Subject 2 was not fed after sessions until he reached 80% of baseline. Subject 1 reached 80% in 12 days (starting from 96.8%). Subject 2 reached 80% after 16 days (starting from 94.8%). Once the birds weighed less than 80% at the beginning of their sessions, they were fed supplemental food after sessions if their post-session weight was not 80% or above. The amount 11

19 of supplemental food was equal to the difference between 80% and the post-session weight. The birds remained in this condition until all rate measures had reached stability for at least 10 consecutive sessions. Measures were: post-reinforcement pause (session mean and session median), inter-response interval (session mean and session median), and overall session rate (pecks/second). Each one of these 5 measures was considered stable if the current session s measurement lay within ±25% of the mean of the most recent 10 sessions. An additional stability criterion was a visual inspection of each of the data paths to assure that they were not consistently trending in any direction. The last few sessions of Subject 2 s baseline were unstable, but he entered the next condition anyway because he had been stable for more than 15 sessions prior to that and because the trend of the last sessions was in the opposite direction from the effect expected during the experimental condition. Experimental phase. After baseline, the birds were again allowed unlimited access to food in their home cages. They were allowed this access for two full weeks before sessions were re-initiated. At this point, their economy was switched from open to closed so that no supplemental feeding occurred in home cages. The birds each experienced a different transition phase into the experimental condition, as explained below. Contingencies. During the experimental condition, the FR 15 schedule was re-imposed. If the bird did not peck for 10 seconds after consuming a reinforcer or 0.6 seconds after a previous peck, the keylight was darkened for 20 seconds. If the keylight lighted and the pigeon did not make the first peck within 10 seconds, it would darken for another 20. Pecks to the dark key delayed its illumination for another 20 seconds from the time of that peck (this addition was 12

20 made October in response to Subject 1 s acquisition of superstitious slow pecking to the darkened key). Procedural differences between subjects. Experimental conditions were slightly different between subjects. Subject 1 experienced several different conditions as the experimenters searched for one that would produce weight gain without rate decrement. The data shown here include only the experimental period after the final procedure had been established. Subject 2 was shaped into the final procedure (which was the end product of exploration with Subject 1) by starting under CRF with the limited holds imposed and rapidly increasing to FR 15. FR15 was imposed for the first time on the 38 th session of the experimental period for Subject 2. Session length. Subject 1 s sessions before 12/28/10 varied from approximately 30 to 80 minutes, and the length was determined mostly by his weight. In general, he was given enough time in the chamber to increase his weight slightly from session to session or, if he repeatedly failed to respond to the key when it lighted, his session was terminated early. On 12/28/10, his session length was standardized to 45 min for losing weight and 1 hour for gaining weight. Subject 2 s session lengths were varied during the shaping period and a little after, but standardized to 1 hour to lose weight and 1.5 or 2 hours to gain weight. Hourly deprivation. The experimenter tried to make all sessions roughly 12 hours apart, though some exceptions created a range (see table 1 for means, medians, and standard deviations). There were two reasons for this variation. First, only one person was available to run birds, and scheduling did not always permit consistency. Second, the birds did not always gain weight effectively by increasing their time in the chamber. Thus, to test other methods of weight gain, they were provided an extra session in the day on a few occasions. This method was tested especially with Subject 2 early in his experimental condition, which is the reason for 13

21 his high standard deviation. This strategy was discontinued by the time the session lengths were standardized. Analysis of the hourly deprivation values showed a slight correlation with the DV s (shorter deprivation, in general, with slower performance), however hourly deprivation ranges were not very dissimilar between baseline and experimental phases, and generally erred on the longer side in baseline. As the DV s of interest here showed faster performance during the experimental phase, any confounding influence from this variable should have been in the opposite direction of the observed effects. Weight changes. During experimental phases, the birds weights were increased and decreased gradually. Decreases were more sudden than increases, both to model the applied procedure and because it was easier to drop each bird s weight in a closed economy than to increase it. Return to baseline. After Subject 1 had increased and decreased weights between 80% and 100% ad lib several times, he were returned to baseline (the timing contingencies were removed, sessions were 20 minutes) until his weight decreased again to just above 80%. 14

22 RESULTS Subject 1 Figures 1 through 3 show overall rate, PRP (post-reinforcement pause), and IRI (interresponse interval) session means and medians against pre-session weight for Subject 1. The blue box in each graph shows the 80% - 100% ad-lib range for pre-session weight. The red box in Figure 1 shows the range of ± 25% of the mean rate among the last 10 stable baseline sessions. The red and green boxes in Figures 2 and 3 show the same ± 25% range for mean and median PRP and IRI measures, respectively. The graphs include the last 10 stable sessions (ending on 2/16/10) of baseline, the experimental period from 7/27/10 and later, and the return to baseline through 4/17/11. The 5 months between the end of baseline and the beginning of the experimental period constituted the inductive phase of the experiment; they were spent spent refining the procedure and adjusting equipment. Thus, the data from those sessions are omitted here. The experimental period on all graphs is broken into the variable length condition, the 45 min condition, and the hour condition. During the variable length period, sessions ranged from approximately 20 minutes to 1.5 hours, more often adhering to a 45 min to 1.25 hour range. The 45-min period and 1-hour periods were consistent to within approximately 10 minutes. Figure 1 shows the overall rate of keypecking against pre-session weight. Rates quickly rose above baseline levels during the experimental phase. At first they rose and fell in accordance with weight (note that this trend is counterintuitive). They then stayed somewhat steady but high during weight gains and losses with the exception of a small spike roughly correlated with a period of weight loss around session 200. During the 45 minute condition, the subject lost weight as expected and rates rose gradually. They continued to increase constantly during the 1hour sessions through one period of weight gain and one period of weight loss. They 15

23 then declined steadily with increasing weight. During return to baseline, weight decreased steadily while rates declined to baseline levels and below. Overall, the baseline and return to baseline periods saw rates consistently lower than during the experimental period, even when low-weight baseline sessions are compared to high-weight experimental sessions. The experimental period was marked by occasions in which rate increased along with weight and other periods in which it increased as weight decreased. Figure 2 shows PRP session means and medians against pre-session weight. The experimental period began with a dramatic drop in PRP values corresponding with weight increase. During a subsequent decline in weight, PRP values began to increase slightly, then rose steadily as the bird gained weight again. A subsequent weight loss corresponded with a drop in PRP, then steady weight gain with a slight PRP rise then a dramatic PRP decrease. The 45-min condition saw highly correlated drops in weight and PRP, and the 1-hour condition saw steady PRP values lower than any in the experiment despite two weight gains and one loss. About midway through the second weight gain, PRP again started to increase, reaching a plateau during a weight loss. Return to baseline saw lowering weights and drastically increasing PRP values back into baseline ranges. Figure 3 shows IRI values against pre-session weight. At the beginning of the experimental period, IRI medians increased and stayed relatively constant through the variablesession length period. The means rose and fell in a tight negative correlation with pre-session weight. As weight was gained, mean but not median IRI values decreased. As weight was lost, mean but not median IRI values increased. The exception to this rule started around session 200, when a corresponding drop then slow raise in both values continued through the remainder of the variable session length condition. Otherwise, the negative correlation continued through the 45-16

24 minute and hour conditions, except that during the last ~40 sessions of the 1-hour condition IRI measures became more disperse and median values increased from their previously steady state. During return to baseline, IRI measures increased drastically and continued with the variability that had begun in the final part of the 1-hour condition rather than dropping back to baseline values. Figures 1 through 3 do not show all baseline sessions. Figures 4 through 6 directly compare the entire baseline with return to baseline for Subject 1. Figure 4 shows that overall rate was generally higher when the bird was lighter and lower when he was heavier both during baseline and return to baseline. Figure 5 shows that PRP values became steadily smaller during the first baseline period until they reached stability. In the return to baseline, they started out low but transitioned to levels comparable to the final portions of baseline, despite the progressive decreases in weight. Figure 6 shows that IRI measures during return to baseline did not recover baseline values. In fact, median values continued to stay at levels comparable to the experimental period and means remained variable in a range well above the steady state that characterized the final part of baseline. Figure 7 shows the general correlations between overall rate and pre-session weight. Although the correlation coefficient is quite low (r 2 = 0.051), there is a slight trend during the experimental period for increased weights to correspond to slower rates. Baseline shows the opposite trend when considered in its entirety; however, the last 10 steady-state baseline data points (black) fit a line roughly parallel to the experimental one. Notably, though the stable baseline correlation coefficient is larger than that of the experimental period, it remains small (r 2 = 0.228). Also notably, the baseline data points fit a similarly sloped line when compared to the 17

25 experimental period, but very different values; baseline data show a slower rate at comparable weights as compared to the experimental period. Figure 8 shows general correlations for the mean PRP measurement. Both baseline and experimental conditions produced trends that generally saw longer PRPs at higher weights, as would be expected. However, while the low correlation coefficient hints at wide variation (r 2 = 0.273), it hides the pattern of that variation. As we saw in Figure 2, some periods of the experiment were characterized by a seemingly robust disconnect between pre-session weights and PRP values. Values from two of those periods are labeled on the correlation graph and on Figure 2 as periods 1 and 2 of disconnect. The lines fit to these data clusters differ markedly from the overall experimental correlation. Both the slope of the line and the correlation coefficient change drastically for these clusters, indicating that within these periods, the relationship between PRP and weight was broken. Figure 8 also shows a wide separation between baseline and experimental PRPs. Experimental PRPs are much smaller at much larger weight values as compared to baseline. Figure 9 shows correlations for IRI. As was discerned from Figure 3, experimental periods showed a slight negative correlation between pre-session weight and IRI as opposed to the more intuitive positive one seen in baseline. Subject 2 Figures 10 through 12 show overall rate, PRP, and IRI session means and medians against pre-session weight for Subject 2. The graphs are structured the same as with Subject 1. The experimental condition consisted of a short period of variable session length followed by a 1-hour condition and a hour condition. Subject 2 had no temporal break between his experimental and baseline periods; however, there was one break during the 1-hour session 18

26 condition. The bird had lost so much weight so quickly that the experimenter fed him 10g of food per day in lieu of sessions so that he would gain some weight and not drop below 80% of ad lib. All sessions were measured for Subject 2 in the same way as Subject 1. Figure 10 shows the overall rate against pre-session weight for Subject 2. After a long, stable baseline, rate quickly increased as weight decreased. After the feedup days, rate again dropped, but rose again quickly as weight dropped a second time. Rates stayed high during a subsequent weight gain during the hour condition; however, the trend is not without exception during this period as there are several low rate values that occur at heavy weights. During another weight gain and loss, rates became highly variable, often staying high at high weights and other times dropping to baseline levels or below. Figure 11 shows PRP against pre-session weight for Subject 2. During baseline, PRP values increased and still showed some tendency to trend upward at the end of baseline. The experimental conditions immediately produced very short PRPs. As weight dropped into the 1- hour condition, this low PRP became even more stable. After the 10g feeding days, PRP increased a little but decreased quickly again. After this, during two weight gains and one weight loss, PRP remained well below baseline levels but tended to increase slightly during two separate weight peaks. Figure 12 shows IRI against pre-session weight for Subject 2. Unlike with Subject 1, IRI median and mean values immediately decreased during experimental conditions, though not drastically enough to take them out of baseline range. After this, IRI did not show the same close negative correlation with pre-session weight that Subject 1 had shown. During the last portions of the 1.5 to 2 hour sessions, IRI measures tracked weight in such a manner that sometimes the trends were going in opposite directions and sometimes they were going the same 19

27 direction. Notably, median IRI measures varied similarly to mean measures for Subject 2, where they had been fairly constant for Subject 1. Figure 13 shows correlations between overall weight and rate for Subject 2. Trends were similar to Subject 1; correlation between rate and weight was slight but negative for both data sets. Baseline data showed much lower rates at the same weight as compared with experimental data. Figure 14 compares PRP and weight for Subject 2. As before, PRPs were much shorter than during baseline, even at very high weights. Data points from the section of the PRP data that looked disconnected from weight trends (marked disconnected Period 3 in Figure 11) did not fit a line with a shallower slope compared to the entire experimental period as had occurred for Subject 1. Instead, the slope was the same but the correlation coefficient was again much lower (r 2 = , from r 2 = ). Figure 15 compares IRI with weight. As with Subject 1, the slope of the line fit to the experimental data is negative, whereas in baseline it s positive, and in stable baseline it is almost flat. Thus, though the IRI measure did not visibly track weight to the same degree as it did for Subject 1, the line indicates a slight inverse relationship. As with Subject 1, no dependent variables correlated tightly with pre-session weight. No calculated correlation coefficient for either Subject was greater than.5 other than the baseline IRI/weight comparison for Subject 2 (r 2 = ). Most others were less than.3, many less than.1. Control Calculations It could be argued that observed differences in dependent variables between baseline and experimental periods might be attributed to the signaling involved in the limited hold. When one 20

28 long IRI occurs, the light extinguishes. If the next IRI after a long IRI is more likely also to be long, then a signal that prevents such a response should artificially create an apparent increase in rate measures. With this in mind, the last session of baseline data for each Subject were analyzed to see if a long IRI consistently preceded another long IRI. Figure 16 shows that long IRIs were not likely to be followed by another long IRI, so it is unlikely that the data can be explained in this way. Timeouts per minute and non-utilized opportunities per minute were included in the analysis because of the limited hold. The birds could conceivably have generated high rates of responding at all body weights yet at the same time varied their performance along these dimensions. For example, a single peck at the lighted key followed by a delay long enough to cause a timeout, if not followed by another peck until the key re-lighted, would produce no IRI or PRP measurement but would constitute an event that could change in frequency along with body weight. Similarly, not pecking at a re-lighted key until it turned off again would produce no PRP or IRI measurement, though it would decrease the overall rate. Events like this, if affected by pre-session weight, would more directly affect the measures mentioned above than our primary DVs, and for this reason these measures were plotted (Figures 17 and 18). The relationship between timeouts per minute and body weight is discernable but not consistent, and we noted a slow increase in timeouts per minute over the entire period. Timeouts per minute for Subject one appeared to become more related to body weight during the 1-hour session lengths and for Subject 2 during the hour sessions. Missed opportunities per minute did often spike, almost always just before or during a weight loss, however not all weight losses included such a spike, no absolute value of body weight always correlated with a spike, and occasionally a 21

29 spike would occur during weight gain. Thus, the relationship between either timeouts or nonutilized opportunities and body weight was not consistent. Reinforcers per minute were also plotted against weight as another measure of rate (signifying a completed ratio). Figures 17 and 18 show that reinforcers per minute were not increased during experimental sessions as were other rate measures. Trends in this measure predicted weight changes especially during the constant-length sessions as expected. All the analyses thus far use the data from the entire 20 minute baseline and baseline reversal sessions along with the first part of experimental sessions up until the keylight had been lit for a cumulate of 20 minutes. Entire experimental sessions were not considered because insession effects of passing time and food consumption could systematically skew results. While our analysis does compare equal time periods in the presence of the Sd (the lighted key), it does not compare completely equal total time periods. All experimental data were thus also analyzed using only the first 20 minutes of the experimental session. Results of this analysis can be seen in Figures 19 and 20, and they do not differ substantially from the data considered in our primary analysis. Subject 1 achieved 94% of his ad-lib weight while still keypecking at overall rates well above baseline and PRP well below baseline, with IRI means close to baseline, IRI medians not much higher than baseline, and PRP values all substantially below baseline. Subject 2 reached 96% of his ad-lib weight while still keypecking at overall rates well above baseline and IRI/PRP measures below baseline. 22

30 DISCUSSION Summary The data show that the experimental conditions were sufficient to produce faster responding at high body weights than was possible at low body weights under typical laboratory conditions. Subject 1 achieved 94% of ad-lib during the experimental conditions without dependent variable measures dropping to baseline ranges. Subject 2 achieved the same at 96%. For Subjects 1 and 2, PRP (post-reinforcement pause) and overall measures were shorter than those of baseline during experimental conditions. IRI (inter-response interval) measurements were also shorter, even at high body weights for Subject 2. For Subject 1 the mean IRI was similar to baseline and the median IRI longer than baseline. However, a return to baseline did not restore these short median IRIs for Subject 1. Thus, the experimental conditions were sufficient to produce the apparently hungry behavior that the animal trainers claim to produce even at high body weights. Within experimental conditions, pre-session body weights did covary slightly as expected with PRP and overall rate, however the correlation with IRI was negative, all correlations were slight (less than.5 in all but one case), and the PRP correlation disappeared entirely during certain periods, but not during others. Values from the periods of PRP disconnection, when considered apart from the rest of the values in the correlation, showed a decreased correlation coefficient for both subjects and, for Subject 1, also a completely different slope (Figures 8 and 14). The reasons for this inconsistency remain unknown. Such noise in the data might indicate lack of control over one or more unknown critical variables. It also could represent the variation typically seen during transition phases. As the experimental conditions dealt with dynamic rather than steady states of pre-session weight, this seems even more likely. If this is 23

31 the case and deprivation measures must be at constant levels for some period of time before they become predictive, or if changes in pre-session weight do not affect responding until some threshold is reached, it indicates that the immediate influences of weight manipulation on rate are not captured best by the weight measures themselves. It s also possible that some features of the experimental conditions created an insensitivity of rate measures to pre-session body weights during this experiment, but that these features might not have been precisely applied or were interacting with other factors to produce mixed results. IRI showed a slight inverse relationship to pre-session body weight for both subjects during experimental conditions. This result is notable for several reasons; it is counterintuitive (one would expect deprived animals to have shorter rather than longer IRIs), the same relationship in the baseline condition is reversed or weaker, and IRI has been previously shown to be somewhat insensitive to body weight (Shull, 2004). The latter point was established using VI schedules and open economies, so the trend from this study might either have been too slight to be shown in that data or VI conditions may not have been sufficient to produce it. The fast response rates achieved here at high body weights carry implications for practical application. In animal training, where food remains the most often utilized reinforcer, such carefully designed conditions may achieve robust training results while requiring little weight manipulation. Bird trainers purportedly already utilize the procedure to this advantage. In fact, the cyclic rise and fall in body weight that preceded global weight gain seen in our data (even when it was not programmed via session length manipulations) reflects the pattern of weight loss and gain that the trainers produce. However, the disconnection between weight and response rate appears to have been achieved only during short periods, and only slightly and briefly in Subject 2. Our data neither demonstrate nor disprove that such disconnection can be 24

32 reliably produced nor maintained indefinitely at very high weights using this procedure. If further experimentation with this schedule uncovers a method to achieve such results, it may inform the trainers procedures and make them more efficient. However, such a disconnection may also be disadvantageous in application if it corresponds with a disconnection of all the classic effects of deprivation and satiation from body weight. Trainers describe some advantages in the behavior of somewhat sated animals; namely, sated animals tend to vary their responses over a wider range as compared to very deprived animals (Carlton, 1962; McSweeney, 1974). When a trainer is shaping or when practical considerations make variable behavior desirable, the elimination of this effect may constitute a problem. One possible advantageous result of the procedure modeled here may be that food reinforcers can produce rapid behavior at high weights while variability remains somewhat increased along with body weights. However, as variation of responses was not measured here, these data can t speak to that possibility. If these results are general to non-food-related deprivation, it may constitute a practical strategy for teachers. Responding that is somewhat insensitive to deprivation operations, or which remains relatively robust during satiation, might prove practical since otherwise uncontrollable fluctuations in access to reinforcers may interfere with performance. Additionally, such a procedural option may inform a method to make therapy less restrictive. In a therapeutic situation, applying limited holds may constitute a less restrictive approach to effective treatment than applying a deprivation procedure, and this study suggests that such a manipulation could be more effective anyway. Such a schedule could provide a greater Freedom of individual movement and access to preferred activities in a least restrictive environment without decreasing the efficacy of an intervention (Van Houten et al., 1988) 25

33 Implications The higher rate of response during experimental conditions at high weights compared to lower response rates during baseline conditions at low rates can be explained by the differing contingencies between the two conditions. Namely, if behavior during PRP and IRI intervals is part of the entire motion cycle of a keypeck response, the keylight darkening may have negatively punished all motion cycles with non-key-contacting components longer than 10 or 0.6 seconds, respectively. It may also have negatively reinforced components considerably shorter than those cutoffs. If this is the case, neither the closed economy nor the weight fluctuations should be necessary to produce the effect. Thus, the next research step should be to compare a baseline performance to one where the limited holds are imposed without involving weight fluctuations and while keeping the economy open. An alternate experiment would test the economy itself by opening and closing it under constant-weight conditions without the limited holds. Whatever occurs during typical food deprivation operations to produce effects on behavior, this study shows that such effects can be modified by one or both of our IVs (closed economy and limited holds). This can be seen in the differences between PRP, IRI, and overall rate measures during the experimental conditions as compared to baseline. The literature indicates that the conditions of this experiment are not the only ones under which deprivation operations may show contextual or counterintuitive effects. Conrad, Sidman and Herrnstein (1958), for example, found that under DRL schedules, satiation over a long (10 hour) session produced shorter rather than longer IRT s (inter-response times, identical to IRIs). Thus, under these conditions a satiated animal actually responds more quickly than a deprived one. Interestingly, Holtz & Azrin found the opposite effect on IRT when using pre-session body 26

34 weight as an index of satiation with the same schedule (Holtz & Azrin 1963). Extraneous contingencies and restrictions may also produce changes in the effects of reinforcement. Premack and Premack (1963) showed that having recently introduced or withheld a running wheel from the environment could affect response rates, even given the same deprivation state, the same immediate antecedent state, and the same reinforcement contingency. The effects of deprivation also depend on schedules of reinforcement (i.e. Carlton 1961), and can also depend on history. For example, Fischer and Fantino (1965) found that in chained schedules, low body weights had an adverse affect on performance in the initial chain, but only in the beginning of the experiment. Once the birds had experienced the chain for a longer period of time, the initial link became insensitive to changes in deprivation measures. These contextual, counterintuitive, and inconsistent effects of deprivation imply the need for further research and perhaps corresponding shifts in conceptualization. We typically consider motivating operations as antecedent manipulations that hold direct influence on responding that can combine with the direct effects other variables in order to produce a predictable result. Yet such a conception requires us to make exceptions or opposite predictions for contexts such as those pointed out here, or else to create a new category for behavior insensitive to deprivation. A more parsimonious account might explain not just the effects of classic deprivation, but also these counterintuitive effects and insensitivities seen under varying contexts. The elucidation of the behavioral processes underlying hunger would better enable a researcher to ask the question of whether such processes are readily generalizable or not to nonfood-related motivating operations. From a practical standpoint, the problem boils down not to a question of conceptual categorization of operations that affect the same DV, but to the ability 27

35 of a behavior analyst to control behavior to a particular degree using particular reinforcers. A useful investigation will lead to a technology in which reinforcers might be more easily created and the environments that elicit their associated behavior made more or less powerful in doing so. The experimental condition of this study may constitute a useful tool in this effort, and further investigation into the component parts of the study may lead to better control of behavior. Summary As an exploratory study, this investigation can t elucidate the processes underlying deprivation, nor is it aimed at building on the deprivation literature directly. Rather, it investigates a new behavioral phenomenon that professional animal trainers claim to have discovered and utilized. Thus, these data are purposed to provide questions ripe for systematic analysis: Science, it is argued, proceeds by manipulating variables in a systematic fashion, and by unifying the results of such manipulation within a conceptual framework. The simple demonstration of a behavioral effect is held to be only the prelude to systematic investigation. (Sidman 1960, p. 23) The data presented here demonstrate that, for more than one organism, placing limited holds and a closed economy onto an FR schedule makes it possible to keep response rates high even at high body weights and may help produce a disconnect in the relation between the two. Still unknown are the necessary and sufficient conditions for this to occur, the relative importance of the variables, the reasons for the inconsistency between PRP and pre-session weight within the experimental condition, and the reasons for a slight but consistent inverse relationship between IRI and pre-session weight. The study asks many more questions than it answers, and since any experiment worth its salt will raise more questions than it answers 28

36 (Sidman 1960, p. 8), we hope that the data may contribute to a fruitful exploration of the conceptualization of deprivation in the future. 29

37 Table 1 Hourly Deprivation HH:MM:SS Subject 1 Baseline Experimental Return to Baseline Mean 14:05:16 12:50:32 11:55:02 Median 12:22:30 12:25:00 12:03:53 St. Dev. 5:20:49 3:31:20 1:58:55 Subject 2 Mean 13:39:52 13:13:31 Median 13:16:27 12:12:40 St. Dev. 2:13:37 7:17:56 30

38 Figure 1. Overall rate vs. pre-session weight, Subject 1. 31

39 Figure 2. PRP vs. pre-session weight, Subject 1. 32

40 Figure 3. IRI vs. pre-session weight, Subject 1. 33

41 Figure 4. Overall rate vs. pre-session weight baseline comparisons, Subject 1. 34

42 Figure 5. PRP vs. pre-session weight baseline comparisons, Subject 1. 35

43 Figure 6. IRI vs. pre-session weight baseline comparisons, Subject 1. 36

44 Figure 7. Overall rate vs. pre-session weight scatter plot, Subject 1. 37

45 Figure 8. PRP vs. pre-session weight scatter plot, Subject 1. 38

46 Figure 9. IRI vs. pre-session weight scatter plot, Subject 1. 39

47 Figure 10. Overall rate vs. pre-session weight, Subject 2. 40

48 Figure 11. PRP vs. pre-session weight, Subject 2. 41

49 Figure 12. IRI vs. pre-session weight, Subject 2. 42

50 Figure 13. Overall rate vs. pre-session weight scatter plot, Subject 2. 43

51 Figure 14. PRP vs. pre-session weight scatter plot, Subject 2. 44

52 Figure 15. IRI vs. pre-session weight scatter plot, Subject 2. 45

53 Figure 16. Final baseline sessions IRI probabilities. 46

54 Figure 17. Timeouts, non-utilized opportunities, and reinforcers per minute, Subject 1. 47

55 Figure 18. Timeouts, non-utilized opportunities, and reinforcers per minute, Subject 2. 48

56 Figure 19. Graphs using first 20 minutes of experimental sessions, Subject 1. 49