THE EEFECT OF A WDELMG TECHE‘UGLUE 0N 'E’HE LEARNENG Q? FQUR TASm BY RETARDATES That‘s gm! {‘E‘m Degree oi; M. A. lfiCEEGAN STATE UBSWEBSETY Thomas Edward Evans £968 1111111111111111111111” ' 31293 01073 2117 T . bn f . r42 “TENN . .—,1 "a“. ABSTRACT THE EFFECT OF A MODELING TECHNIQUE ON THE LEARNING OF FOUR TASKS BY RETARDATES Thomas Edward Evans The present study investigated the effectiveness of a parti- cular method, modeling, in training severely and profoundly retarded children on four different tasks. §s were ten institutionalized severely and profoundly retarded individuals, matched on C.A. and I.Q. and randomly assigned to a modeling or a no-modeling condition. The four tasks were (1) a prOperty concept attainment task, (2) a relational concept attainment task, (3) a word identification task, and (4) a Spelling task. Training took place on a teaching machine (MUDRAFA). The prOperty concept attainment task provided clear support for the facilitatory effects of modeling. The word identification and relational concept attainment tasks also provided some support. The spelling task showed no differential effects of modeling. Possible reasons for interaction of task difficulty with amount of facilitation were offered. The complex nature of compound cues in the word identification and relational concept attainment tasks would necessitate a longer modeling period than the preperty concept task to provide an equal amount of facilitation, while the present study THE EFFECT OF A MODELING TECHNIQUE ON THE LEARNING OF FOUR TASKS BY RETARDATES by Thomas Edward Evans A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Psychology 1968 ACKNOWLEDGMENTS Special appreciation is due Dr. M. Ray Denny, chairman, for his guidance and encouragement, as well as helpful criticism which aided immeasurably in the completion of this study. Special appreciation is also due Dr. Lester Hyman, a committee member, for the many hours spent in helping the author present the study in a coherent fashion. I am particularly grateful as well to the third committee member, Dr. Uleman, for the confidence he placed in my work. The assistance of Bryan Zartman in the actual running of the study is deserving of my sincere gratitude. To Dr. Marion J. Skowronski, M.D., Howell State Hospital, go the author's special thanks for his c00peration and enthusiasm which helped to make the study possible. ii Chapter I. II. III. IV. V. INTRODUCTION METHOD . . . RESULTS . . DISCUSSION . SUMMARY . . REFERENCES . . . . . TABLE OF CONTENTS iii Page 21 32 39 41 LIST OF TABLES Page I.Q., M.A., C.A., and cause of deficiency for indiVidual ES 0 O O O O O I O O O O O O O O O O O O 7 Trials-to-criterion scores for_§s on four learning tasks: P.C.A., R.C.A., W.I.D., and SOP. O O O O O O O O O O O O O O O O O O O O O O O 22 Mean trials-to-criterion for modeling and no-modeling groups under the four stages of training, from ERRORrIRRELEVANT and ERRORrRELEVANT data . . . . . . . . . . . . . . . . . . 24 Trials-to-criterion for individual §s on three learning tasks: P.C.A., R.C.A., and WOIODO O O O O O O O O O O O O O O O O O O O O O O 25 Statistical summary: Paired-difference t—test for three discrimination tasks . . . . . . . . . 27 Mean trials-to-criterion for §s with C.A. over/under 25 and I.Q. over/under 18 from ERROR—IRRELEVANT and from ERRORrRELEVANT data . . . . . . . . . . . . . . . . . . . . . . . . . 31 iv Figure 1. LIST OF FIGURES MUDRAFA with P.C.A. stimuli (Straight—Crooked) MUDRAFA with R.C.A. stimuli (Over-Under) MUDRAFA with W.I.D. stimuli (Rough) . . . . . MUDRAFA with S.P. stimuli (Rough) . . . . . . Trials-to-criterion scores of individual_§s across four learning tasks: Preperty concept attainment, spelling, word identification, and relational concept attainment . . . . . . Per cent correct for each quartile of performance previous to criterial run for_§s who reached criterion on a particular task (prOperty concept attainment was not included since 4 of 5 modeling Ss had no pre-criterial errors, Table 1). . . . . . . . . . . . . . . Per cent correct for each quartile of performance previous to the cut-off point for Se who did not reach criterion on a particular task . . . Page 10 ll 12 13 23 28 29 INTRODUCTION In recent years psychologists interested in learning have investigated imitative behavior in children, i.e., the effects of modeling on children's learning. Theories of selective attention, information processing, and vicarious reinforcement (to mention a few) have brought forth a wide variety of research on modeling. This study attempts to explore the effects of modeling on the learning of several tasks by severely and profoundly retarded children. Several studies demonstrated that modeling of the correct response has a facilitatory effect on learning. Kanfer and Marston (1963) suggested that one learning variable, vicarious re- inforcement (V.R.) may account in part for the facilitation of modeling on learning. As described by Kanfer and Marston, V.R. is an inferred construct. The assumption is that gs observing a model are responding vicariously and are similarly reinforced vicariously when the model is reinforced. -§s observing a model are also gain- ing information about the potential reinforcing value of cues, but Kanfer and Marston place less emphasis on the informational input value of modeling. Kanfer and Marston presented a verbal task to three groups of Ss. The V.R. group observed another S performing a task correctly and receiving social reinforcement prior to their own participation in the task. A second group observed a.§'per- forming the task correctly without reinforcement. The control group did not observe a_§ performing the task prior to their participation in the task. The performance of the V.R. group was significantly better than the performance of the control group. The performance of the non-reinforcement imitating group fell between the performance of the V.R. and control groups. These results support the theoretical contention of Kanfer and Marston that modeling influences learning through two more basic social learning variables, vicarious responding and vicarious re- inforcement. For Kanfer and Marston's position it is necessary that the model be another person. Other theorists such as Dollard and Miller (1941) consider vicarious responding to be a part of observational learning. This implies that the "social" model is not intrinsically necessary except as a mechanical device for demonstrating the correct response. Dollard and Miller suggested that learning by observation may direct the_§'s attention toward relevant environmental cues. Fletcher (1966) investigated infor— mational input in greater detail. He posited that implicit responses occurred during observational learning. In a discrimi- nation task gs were given 3, 6, or 9 trials of observation prior to active participation. During these observational trials a light would flash beside the correct alternative. The light flash was to have elicited an implicit response in the observing SS. Fletcher found that the number of light flashes was related to subsequent performance on trial and error discrimination trials, supporting his predictions. Gormezano and Grant (1958) conducted a study in which the con- cept to be attained was imitation, i.e., the §S had to learn to imitate the behavior of the model. The model was performing a two- choice discrimination task. They found that increasing the ratio of reinforcement of the cues in the discrimination task to the model's behavior increased the difficulty of learning to imitate. These results imply that a modeled response may direct the imitators attention towards various discriminable cues. Gormezano and Grant concluded that a "selective attention" interpretation (e.g., Miller and Dollard, 1941) of conceptual learning is appro- priate in explaining their findings. McDavid (1962) designed a similar study. _§s were presented with a three-choice discrimination task, with two salient cues (color and position). The model made choices consistently (100%), partially (67%), or randomly (33%) associated with a color cue. The cue which was defined by the experimenter as relevant was the response of the model, i.e.,_§ was being trained to imitate. Therefore, the reinforcement ratio of the color cue varied. A selective attention position would predict that the 67% group would learn slower than the other groups, since S's attention is inappropriately directed toward an irrele- vant discriminable cue. The 33% group would suffer no interference, and thus learn faster, since color is associated with modeling (and hence reinforced) randomly. The 100% group would learn fastest of the three groups due to the redundant nature of the stimulus situation. The results fully supported the predictions of the selective attention position. The evidence clearly demonstrates a facilitatory effect of modeling on learning. The purpose of the present research is to discover if the facilitory effects of modeling will still be Operating with Se in the severely and profoundly retarded range of intelligence (I.Q.‘s from 10-32). A review of attempts to identify the behavioral deficits of the retarded might help the reader to understand why modeling may be a particularly appropriate technique in training these children. Benoit (1957) pointed out the importance of attention in learning and observed that the highly distractable retarded_§s might suffer from an inability to maintain attention. House and Zeaman (1961) also attribute various learning deficiencies to an attention deficit, which in turn they relate to trace-perseveration error tendencies. This refers to a failure of retarded Ss to inhibit responses made on prior trials. Denny (1964) suggested that the inhibition deficit is more basic than the attention deficit. The S, not being able to inhibit responding to interfering stimuli, is constantly being distracted, so the task stimuli are not attended to sufficiently for learning to occur. Finally, Luria and Vinogradovia (1959) note that verbal control of behavior fails to develop in the severely retarded due to their failure to inhibit responses to external stimuli. One approach to the analysis of learning is quite eXplicit in its implications for training mentally retarded individuals. This position, known as "Elicitation Theory" (Denny and Adelman, 1955, Denny, 1966), focuses upon the eliciting power of a stimulus. Since retardates tend to perseverate and do not benefit from negative cues as much as normals (Eimas, 1965; House and Zeaman, 1958; Kaufman, 1959; House, Orlando, and Zeaman, 1957), a maximally designed learning situation would provide external prevention of inappropriate responding while maintaining appropriate responding. Yascolt (1965) used an apparatus known as the Multiple Dif- ferential Response and Feedback Apparatus (MUDRAFA) designed with these principles in mind (Denny, 1966). Yascolt had a great deal of success in teaching mentally retarded Ss with this apparatus. The present study uses a similar, though more sophisticated, apparatus. In review, retardates are theoretically unable to maintain attention (Benoit, 1957), while modeling theoretically facilitates learning by a selective attention mechanism (McDavid, 1962; Zeaman and House, 1963). Retardates suffer from an inability to inhibit errors (Denny, 1964), but modeling may provide a method for con- sistently eliciting the appropriate response, thus effectively preventing errors. A modeling technique is logically suited to facilitate learning in the retarded. The present study provided a test of the strength of this method, since S are from the severely and profoundly retarded range. The prediction is that, to the extent that modeling provides for elicitation of the correct approach response in the presence of a positive relevant cue and prevents the elicitation of approach to negative and irrelevant cues, the probability of any_§ learning will be enhanced. METHOD SUBJECTS. Fourteen institutionalized mentally retarded child- ren and adults, ranging in age from 14 to 40 years, and in recorded I.Q. from 10 to 32, were selected from Howell State Hospital for this experiment. One § continued to emit stereotyped verbal behavior and refused to Operate the apparatus, while another simply refused to "go to school" because it was "...too much work”. Two other_§s were eliminated due to extreme spasticity. The remaining ten §s were divided into pairs matched on the basis of C.A. and I.Q. Seven Ss were confined to wheelchairs due to various physical deformities and spasticity. (See Table 1 for I.Q., C.A., M.A., and type of retardation.) PRE-TEST. Ten §s were tested prior to training to determine whether they would be eligible for training on various concepts. Each §_was assigned to be trained on one set of property concepts, i.e., concepts which designate an abstract character of many stimuli, such as a color, shape, or texture. Each S was also assigned to be trained on one set of relational concepts, i.e., concepts which designate an abstract relationship between stimuli, such as positional relationships or similarity relationships. The concepts which were tested for were ROUGH-SMOOTH, and STRAIGHT- CROOKED for property abstractions, and OVER—UNDER, BETWEEN-BESIDE, Table l. I.Q., M.A., C.A., and cause of deficiency for individual Ss. R.G.-#e1 T.B.-#e2 C.R.-#e3 G.E.-#e4 H.G.-#e5 D.E.—#c1 12 14 32 13 26-30 10 3 yr. 2371‘. ,lO'mO. Ayro ,8m00 zyro , 81110. 3er ’lOmOO— 1949-51110. 3yr.,6mo. 6 yr. 40 yr. 14 yr. 16 yr. 36 yr. 24 yr. 32 yr. Congenital Congenital Mongolism Familial Mechanical Birth brain de- cerebral locomotor injury at injury fect with deficit ataxia birth Encephalogy hydroceph- Cerebral palsied Spastic Congenital alus Unknown prenatal quadri— spastic internus influence plagia Infantile paralysis (continued) 7a Table l. I.Q., M.A., C.A., and cause of deficiency for individual §s (continued). 17 29-33 18 1964-26 I.Q. 27 1966—19 2yr.,2m0. 5 yr Byro ,411'10.‘ 4 yr. M.Ao 4yr.,4mo. 13 27 yr. 26 20 C.A. Mongolism Prenatal Mongolism Symtomatic Cause of influence epilepsy deficiency Spastic due to quadri— brain plagia disorder Birth injury and SAME-DIFFERENT for relational abstractions. For each concept the_§s were given four chances to demonstrate that they understood the spoken word. For prOperty abstraction, S merely picked out the stimuli with the correct prOperty. For relational abstraction the S pointed to the stimulus which was gg§£_another or held one stimulus ggg£_another, etc., or identified a stimulus which was the gag; as a standard. Four-out-of-four correct responses for any concept precluded training on that con- cept. Three-out-of-four correct responses indicated another testing session was necessary for adequate determination of the S's under- standing. Less than three correct responses for any concept qualified_§ for training on that concept. .§3 were similarly tested on recognition of the printed word. None of the SS had previously learned to read any of the words being trained on, and hence, none failed to qualify for the word recognition training. FEE-TRAINING. All_§s were taught to imitate E'by the follow- ing procedure: _§ said "Do this", and raised his right hand. If_§ did not imitate the response, g raised_§'s hand, said ”Very good", and gave S a miniature marshmallow. After S imitated this response adequately, §_modeled a second response for_§. .E said ”Do this", and slapped his knee, physically guiding_§'s response when neces- sary as before. When S was imitating this second response con- sistently, §_alternated the first and second responses. This training continued, with a new response added by §_when adequate performance on the alternation of responses already learned was attained. At criterion, § could successfully imitate 25 successive different responses at the command of "Do this". No §_failed to complete pre-training, and thus, none were eliminated on this basis. APPARATUS. A Npltiple Differential Response And Feedback .Apparatus (MUDRAFA) was used in this study. It consisted of a circular face with a cross-like Opening in the center. {ounted in the center by counter—balanced springs, was a bar which could be moved to any position within the Opening. Each of the four slots forming the cross-like opening could be closed without visibly changing the appearance of the opening. A light was mounted in the bar, and was controlled by an on—Off foot switch. Figs. 1, 2, 3, and 4 illustrate various stimulus configurations placed on MUDRAFA for various training procedures. Two machines were present during sessions with the modeling group, one for S and one for E. Only one machine was present during sessions with the non-modeling group. PROCEDURE. General training procedure: The specific procedure in training varied for each concept. In the concept attainment tasks, stimuli were placed on MUDRAFA by E, and S responded appro— priately to the verbal cue, e.g., "Go to the straight one", "Push the bar over the ball". Since the concepts were dichotomous pairs, the concept attainment stage of training proceeded by presenting the verbal cue for one concept for the first 24-trial session, and the verbal cue for the Opposite concept for the second 24-trial session. Twelve trials on each of the concepts were run during the third session. On the fourth day the verbal cues for each concept were alternated every third trial; during the remainder of the training the verbal cues were randomly presented. The word Figure l MUDRAFA with P.C.A. stimuli (Straight-Crooked) 10 Figure 2 MUDRAFA with R.C.A. stimuli (Over-Under) 11 Circle Cut Out 4 Figure 3 MUDRAFA with W.I.D. stimuli (Rough) 12 Figure 4 MUDRAFA with S.P. stimuli (Rough) 13 FROUGII 14 identification and spelling tasks were similar in procedure, except the stimulus materials were linguistic, i.g., "GO to the word straight", "Spell straight". For the word identification and the spelling tasks, there was no alternation. Since the concepts taught were dichotomous, and therefore the cue words were related tO each other in meaning, it could be argued that simultaneous training of the dichotomous concepts should facilitate learning through transfer. There was no logical reason for maintaining simultaneous training for word identification or spelling. That is, the reading and spelling of the second word was not trained simultaneously with the first. After S had been run to either criterion or to the termination point on the first word, training on the second word began. The slot, position, and order cues were randomized by §.in such a way as to prevent slot or position fixations, i.e., ir- relevant cues were rewarded only at chance level. Errors were prevented by closing all but the correct slot for the first ten trials, and opening just one alternative slot for the second ten trials. From the twentieth trial on, all slots were opened. How~ ever, if_§ committed three consecutive errors, or four errors in five trials at any time during training, he was immediately returned to the earliest stage with only the correct slot Open to repeat the same progression of slot Openings just mentioned. No slots were closed during the spelling task, since the spelling pro- cedure provided enough prompting bY.§ to assure correct responding, as will be described later. Inter—trial-interval was not held 15 constant because the amount of time required to change stimuli between trials varied depending on the task. Specific Procedures STAGE ONE: Property concept attainment (PCA) Teaching Rough-Smooth: A two-choice discrimination task. _E placed two stimuli on the MUDRAFA, each at the end of a slot. One stimulus was abrasively rough to the touch, the other quite smooth. I Three different pairs of stimuli, having in common this rough-smooth variation were used. (Two pieces of sandpaper of a rough grade were used, one with a smooth, visually non—discriminable plastic covering. A I Also, pairs of pipe cleaners were used as stimulus objects. One of the two in each case would have abrasive particles glued to the cleaner, making it very rough to the touch.) §_was directed to touch the two items on MUDRAFA, and to push the bar to the rough- smooth one. Teaching Straight-Crooked: A two-choice discrimination task. §_placed the stimuli at the ends of two slots, as above. The stimuli were distinctly straight or crooked. A pair of identically colored pipe cleaners, one straight and the other crooked, was the first set of stimuli. Straight and crooked lines drawn on 5"x7" cards composed the second set of stimuli (see Fig. l). §_was directed to push the bar to the straight—crooked one. STAGE TWO: Relational concept attainment (RCA) Teaching Over-Under: A two-choice discrimination task. (An initial four—choice discrimination task proved too difficult, so all .§$ were switched to the two-choice task.) §_p1aced one red circular 16 cut-out on a plastic rod which was fastened in the middle of the widest slot on MUDRAFA. The bar could be pushed on either side of the plastic rod. With this slot in a horizontal position, the bar could be pushed under or over the red disc. .E directed_§ to push the bar under-over the red disc. Since responses were to either half of one slot, prevention of errors by Efs hand (rather than mechanical blocking) was necessary (see Fig. 2). Teaching Between-Beside: A four-choice discrimination task. §_placed eight red paper discs on MUDRAFA, with two discs for each of the slots. The bar could be pushed either beside, under, over, or between the discs. E instructed_§ to push the bar between— beside the red discs. Teaching Same-Different: A four-choice discrimination task. .E placed eight paper cut-outs on MUDRAFA, two at the end of each slot. The cut-outs were triangles, circles, and squares. When teaching "different", E would place three identical pairs and one dissimilar pair on MUDRAFA. When teaching "same", §_wou1d place three dissimilar pairs and one identical pair on MUDRAFA. A single command of "Go to the ones which are different-same" was given by _§. Form cues were randomized. STAGE THREE: Letter identification There were no differential conditions during this stage of training, since it was simply devised to facilitate learning in STAGE FOUR. All S were trained in a four-choice discrimination task on MUDRAFA to push the bar to the correct letters upon the verbal command, e.g., "Go to 0". Six consecutive correct responses 17 to each letter were required before the §_qualified for the word identification task. STAGE FOUR: Word identification (WID) The whole word method: A four-choice discrimination task. §_placed four cards with words or pseudo-words on MUDRAFA, one at the end of each slot. §s were given the verbal command "GO to the word rough”. For the first four days of training on rough, the three incorrect alternatives contained the same letters as the correct word (see Fig. 3). For the above example, the second set of alternatives were STRAIGHT, CROOKED, and SMOOTH. The letters of formal similarity in the new set of alternatives were also in quite similar positions as the letters in the correct word. If criterion was still not reached after six more days, the task on the final two days was changed to a two-choice discrimination task. The incorrect alternative in the final task was the antonym of the correct alternative. STAGE FIVE: Spelling (SP) A serial learning task. ‘E placed a printed sample of the word to be spelled on the lower right corner of MUDRAFA. The individual letters of the word were randomly placed along the slots (see Fig. 4). Initially, E said, "Spell rough", pointed to the first letter in the word and said, "Push the bar to R". If §_did not respond correctly, E took S's finger and pointed it to the first letter in the word, then to the same letter along the slot (a forced matching). Then E again said "Push the bar to R". (Six Ss required this prompt- ing and four Ss did not.) §_pointed to the next letter in the word, 18 and said, ”Push the bar to 0”, again forcing, as above, the response of pointing and matching when necessary. This procedure was continued until the_§ had responded to each letter in the proper sequence. Completion of the sequence constituted a trial. Twelve trials were run per session. A gradual fading of externally supportive cues was undertaken. Only if errors or non—responding indicated lack of ability to respond differentially were the prompting cues of pointing and matching provided. When the_§ could respond to the verbal command with self-controlled usage of the pointing and matching cues, the physical guidance of E_had been completely faded out. _§s would show partial differential responding, for example, by going properly to the first letter or two. The E73 guidance was then required to complete the sequence. The E's guidance was gradually removed as the_§ could complete a progressively larger portion of the sequence without the Efs guidance. The individual spoken letters, e.g., "Go to R”, were next to be eliminated. When the_§ seemed to comprehend the usage of the standard (printed word), it was no longer necessary to speak the letters individually as cues, although §_continued to speak the letters after each correct response. The S at this stage guided his responses by following the left-right sequence of the written.word, i.e., pointing and matching without the overt point- ing reSponse. The final fading of externally supportive cues was simply the elimination of the standard. Some Ss required a gradual fading of the standard, while a few could go directly from 'standard' to 'no standard' without a drop in performance. The gradual fading 19 Of the standard was accomplished by partial exposure of each of the letters in the sequence. _§ sould match the partially exposed letter in the standard with the letter on the board. The degree of exposure was gradually lessened until no exposure of the next letter was necessary for correct anticipation. The standard could then be eliminated, and the_§ would push the bar to the letters in prOper sequence with the command "Spell rough". Experimental conditions. “Ss were divided into two groups, with each S in one group matched approximately on C.A. and M.A. with an _S in the other group. In the modeling group, all training included modeling of the correct response by the E, with the_§ saying "Do this, go to______9, then pushing the bar on his machine to the apprOpriate stimulus, followed by S's imitation of the correct approach response. For the non-modeling group, §_would say only "Go to _____", without demonstrating the correct response. Both groups were given social reinforcement as well as candy on an 8:1 fixed ratio schedule. There were two MUDRAFAs during sessions with the modeling group, one for the E, and one for the_§. When the E demonstrated the correct response, the conditioned reinforcer (a bright light in the bar) flashed on, as it did when_§s responded correctly. The modeling was phased, so that, for each task, the §_was given 100% modeling for the first 12 responses, 66% for the next 12 responses, and 33% from then on. When any_§ demonstrated proficiency on the task during early trials, modeling was discontinued to allow criterion to be reached. Operationally, proficiency was identified 20 by three consecutive, correct non-modeled responses. A free- responding situation was instigated when S demonstrated proficiency until S either made an error or attained criterion. If §_made an error, modeling was reinstated at whatever phase was in Operation proceeding the non-modeling test trials. The trial on which the modeling occurred (for phases other than the 100% phase) was systematically randomized. RESULTS Trials to criterion were counted in two ways. Since S was allowed to respond until a correct response was made, a trial could be defined as terminating in a correct response (ERROR-IRRELEVANT). Each incorrect response was also recorded, so that the occurrence of a response (correct or incorrect) could be considered a trial (ERROR—RELEVANT). For individual data on trials-to-criterion for each task, the error-relevant data is presented in Table 2 and Fig. 5, and the error—irrelevant data in Table 4. The data for group means for both error-relevant and error-irrelevant analysis is in Table 3. Because of the skewness of the trials-to-criterion data a square root transformation was applied. Such skewness is a common characteristic of the numbers of trials-to-criterion. A t—test for related measures (paired difference test) was used to test for the significance of the difference between the modeling and no~modeling conditions for each task. For the error-irrelevant data, the effects of modeling on PCA were significant (£.= 2.88, ELE. = 4, p, <:.025), while the effects of modeling on the other tasks, though in the expected direction, were not significant. If only_§s who learned are considered, the effects of modeling on WID approached statistical significance (£_= 2.64, d.f. = 2, p. <1.l). The RCA 21 22 Aemsaaucoov com. «mm. mus. mmm. 1 1 «as .:.o .m.o mo Awm. was. mas. ans. 1 1 Nmm .=.o .o.a so oom. oom. ooh. cow. + 1 ms .a.m .H.u mo coo. mmo. «mm. was. 1 1 mmq .=.o .a.e No «as. «as. can. mas. + 1 «on .:.o .m.m Ho namacamuum men. omw. mmo. mac. + No mma .:.o .o.m mm uamucoo + as o .:.o .m.u am Hmaoaumama ooo.a 0mm. 0mm. ooo.H + ma Ha .a.m .m.o mm mew. es“. ANA. sea. 1 am ssm .p.o .m.e mm «mm. owe. Ham. Ham. 1 oma SAN .m.m .u.m Hm sow. cow. mma. coo. + 1 oo .m.m .m.u we see. Boa. owe. Goo. + 1 men .o.m .o.a so + 1 sm .m.m .H.o mo mam. ems. com. owe. + 1 mm .m.m .<.e No was. Non. Goo. 0mm. 1 1 Gas .m.m .m.m Ho unwaaamuum + ma 9 .m.m .o.m mm umooaou mmm. com. mmm. see. + NH om .m.m .m.u s” Annmaoum + s o .m.m .m.o mm + ma 0 .o.m .m.a mm + am 0 .m.m .u.e an ape cum cam uma uuameos uamoaoo noofinsm o\m xmmu moawuuonw ca uoouuoo N m1v emameoz ¥+ Iaoz mHmHuH * waflaumoa .WW msouw Hmunoaawomxo How ooumwucouommwo madame meado081noa mam voaoooa nous .moafiuwmov a“ uoouuoo uamo won ousw mouhamdm nonuu5m mum 9H3 mam .«om .avafi Mom mouoom nowHouHMOIOulmHmwuH .N manna 22a .cumOH uoc owo n 1 .oonumoa n + « + ss .o.m .s.o oo 1 oms .o.m .o.s so + o .o.m .s.o mo + Nos .o.m .<.o No 1 oms .o.o .m.o so + s .m.m .o.m om ocsssmam + om .o.o .o.o so + as .m.o .m.o so + oa .o.m .m.a No 1 oms .m.o .o.m so amo. .oma. Noax1wosa. ooh. .soo. was. woso. + sos .mos .m.o .o.o mo 1 oss .oom .o.o .o.s so moo. .ooo. oom. .moa. oom. .omo. has. .oss. + sh .oss .o.o .s.o mo coo. .mso. «on. .mma. mos. .omo. Nss. .Noo. + osN .Nss .o.m .<.e No moo. .aoo. ass. .oso. oos. .Noo. sas. .Nso. 1+ oso .mmm .o.o .m.o so aosooo ooo. ooo.s ooo.s ooo. + s .o os .o .o.m .o.o mo 1sosocmos oao. .oso. cos. .ooo. oso. .mao. moo. .muo. 1 NNs.st omm .AAN .o.m .o.o so once coo. .ooa. ooo.s.ooa. ooo. .omo. ooo.s.ook. + N .mm o .ms .o.o .m.o mm oos. .ooa. ooo. .omo. smo. .ooo.s omm. .ooa. + No .os sh .sm .o.m .o.a No oso. .ooa. mas. .oka. oam. .omn. Boo. .sso. 1 so .No omu .mam .o.o .o.m so ans oum oam oos m1 oosoooz oosoooa oamoaoo oomnoso o\o ammo moafiuumau as uoouwoo N * icy 1ooz wowsumoq ososuo o .Aooossuooov mw.aoouw Hmuoosfluonxo wow moumfiuaouommwo msmsuu ooaoooalnoo mom oosooofi spas .mossusooo as uoouuoo uamo Hoe cuss mommamao nonuusm mum QHZ vow .fiosfi you mouoom nowuouHHOIOulmHosuH .N Odome Figure 5 Trials-to—criterion scores of individual Ss across four learning tasks: PrOperty concept attainment, spelling, word identification, and relational concept attainment. 23 54o~+ 520'“— g rror relevant) Number of Inch. '0 Critenon T 60—— 40—4 .— 204~ ‘5??? § 5? NO Modeling-SS o._—- Model 1nq Si __.‘ V ———O 1L 24 Table 3. Mean trials-to-criterion for modeling and no~modeling groups under the four stages of training. Two calcula- tions are given for each stage; one from ERRORrIRRELEVANT (E.I.) data and one from ERRORFRELEVANT (E.R.) data. Stage of Training, Concept Attainment Reading, Spelling_ Word Serial Property Relational Identification Learning a: _ .— 3 E010 - X g 86 E010 - X 3 41 .11 5 _ _ 58 EORO - X 172 EORO - X = 41 '6' ..1 .1 'U ——————— -1 ——————————————— F- —————— -1 >23 .1 11.1. - E = 145 12.1. i 238 11.1. - T 145.5 12.1. - SE = 80 0:0 mzz _. __ __ 8 32.11. - x = 192 E.R. x 351 E.R. - x 263 E.R. - x - 80 .3 .,.| vo m ._ __ g :3. E01. "' X 45 E010 - X a 48 U a dim -; " 1:0 E.R.-}x 52 ER.-X=48 -s1s '3’. ——————— —-1 ———————————————— L- —————— .—1 'o __ __ __ A __ .3. g E.I. x =44.6 E.I. x 173 13.1. - x 125 13.1. - x .1. 60 3'33. - SE = 22 E.R. i 212 E.R. — E 174 E.R. - If s 60 25 so + mom m nuOOEmlswsom ohm I «0% m ww¥oouolu£wfimuum ms + mos m cocoam1aw=om om + «as o noooso1cosom mum 1+ Hoe m Hoodsluo>o sesame o + mas o noooao1ao=om 1sosoamos cum 1 cos m nuooamlnwsom ouoz ms + mus m aoooam1oo=om om + «as m oosoouo1ooosmuom okm 1 sms o nooo5m1noaom cum I was m womanluo>o cum I sue m HOthIHObo sm + moo m unOHOMMAQIoaom cum 1 Nos m woodbluo>o Hum + HO§ m woonpluo>o unoenfimuuo omH + mow m woonDIHO>o umoooou sm + «ma m wownplno>o Honosumaom mm + moo m ucOHOMMHQIoSmm cum 1 Nos m woonaluo>o cum 1 sea m oowmomlaoozuom 00 + mow m nuooamlnwoom mum + «0* m :uooamlswsom mm + mas o cocoao1ow=om ‘ oo + mas o omsoouo1oawsouom ohm 1 sue m sooosm1aw=om oamaesmuoo om + mo* m nuooamlnwsom unooaoo ms + so» m suooamlswoom huHOQOHm ss + moo o aoooao1swsoe mm + mas o omsoono1ozossuoo on + sms m aoooao1oosom mHmHuH a I I cumoa no: man pummosm uaooaou + 1 coupons A.xmou wnaaaoam ecu so sumo Honow>sonfi you N osome oomv .mumu Hz<>msmmmHlmommm aoum meoo mouowfim ona .cOHuMOHMHucooH ouos one .uaoeawmuuo umoocoo Hm:o«umsou .usoedamuum unoonoo huuoaoum .mxmmu mafiawooa wows» so mm.Ho=oH>Hoos How mouoom sowuouHuOIOulmHmauH .q mHan 26 data was moderately supportive, though non-significant, while SP data showed no effects. For the error-relevant data, the modeling effects were significant (E.= 3.818, g§_= 4, p_1<.02) for PCA, borderline for WID (£_= 2.449, df_= 4,.2 <:.05) and moderately sup- portive for RCA, with SP again showing no effects. (See Table 5 for statistical summary.) Vincentized learning curves were plotted in quartiles, with per cent correct on non-modeled (test) trials on the vertical axis. Individual trials-to-criterion for each task determined the division points. The error-relevant data was used, since per cent correct curves must recognize errors as trials. Fig. 6 illustrates the learning curves for Group E and Group C.§$ who learned on the RCA and WID tasks, as well as the two combined. The PCA data was not represented because four-of—five Group E.§§ made no errors on non-modeled trials during this task. That is, errorless discrimi- nation learning took place for four-of-five_§s in the modeling group during PCA training. The spelling data was not included because it was irrelevant to the purpose of the per cent correct analysis. Fig. 7 shows similar learning curves for non-learners. Each of these curves represents only one or two Se. The most prominent feature of Fig. 6 (learners) is that modeled Ss display a distinct advantage during early trials (first quartile), often losing this advantage during the following quartile, but in all cases holding a similar or greater advantage at the last quartile. (The WID curve is probably an artifact, since two of three Group E Ss who learned WID did so in very few trials, and hence a 50 per 27 Om. ssoo nova ms 1 so moss 1 m uum>oaom A%Hno muoauooav Aww Hamv Aww Hamv uowum oovm a 1 oo ”moo 1 o ~.vm1s. ms 1 oo ”moss 1 o so.vMs ms 1 oo mosoo 1 o Amado muonumoav Amw.asmv Amw.aamv unm>osouuH s.vm “N 1 so mso.~ 1 o NVmVs. ms 1 oo moons 1 o omovm ms 1 so mo: 1 o Hobo nosumOHmwunooH ouoz unoonoo Hooowumaom umooooo auuomoum .mxmmu noaumnflewuomfio woman you umoutw monouomwfianooufimm 1 muweeom Hmosumfiuoum .m manoe Figure 6 Per cent correct for each quartile of performance previous to criterial run for_§s who reached criterion on a particular task (prOperty concept attainment was not included since 4 of 5 modeling Ss had no pre-criterial errors, Table l). 28 VIP! ken! Lvor' an. I Modal 1 ng S; (E) v No Mods-ling 53 (C) RCA 8006 WID A 7000'" . lD WID 60% ID M81 RCA 50% o °RCA 40% 20% (90 l s: l 2 nd 3 rd 4'}, Quortiles Figure 7 Per cent correct for each quartile of performance previous to the cut-off point for Ss who did not reach criterion on a particular task. 29 ' Lorrec Per Cent 10. 7o 50 3: l0 Modelong 511 (E) No Modelhg 51 (C) °——_-—“-—O RCA ' O (\./ \\ l.—-__ \\ . ’/ —9§8IAL \ I \ / \ . ’/ - 'WID \\ , - 3 1" A0101“ ‘ r‘LV —-""—6RCA lst 2nd 3rd 4th Quartiles 30 cent correct measure for these_§s might represent one error in two responses, with two responses being a quartile. Table 2 contains the information [indirectly] of how many trials are included for each quartile for a particular S. Both Figs. 6 and 7 demonstrate the initial superiority which modeled Ss have in the first quartile. In Table 6, Se were divided according to two criteria, I.Q. scores and C.A. .§§ with I.Q. scores of over or under 18 and C.A. of over or under 25 years of age were grouped in a 2x2 matrix, and the mean of trials-to-criterion scores for each cell were computed. The choice of division points was mainly determined by the particular §S in the sample, and the points are not intended to be definitive. I.Q. exerted the expected effect, with higher I.Q._§s doing better in all cases than the lower I.Q. Ss. The effect of C.A. was less clear. With higher I.Q. Ss, higher C.A. was correlated positively with learning. With lower I.Q. Ss, higher C.A. was correlated with very poor learning. Ss with I.Q. scores under 18 and C.A. under 25 years performed much more like S3 of the higher I.Q. groups than_§s of the lower I.Q.-higher C.A. group. There appeared to be an inter- action between C.A. and I.Q. which gravely lowered the probability of learning in older, lower I.Q. Ss. There were two Ss each for three divisions (over 18, over 25; over 18, under 25; and under 18, under 25) and four_§s in one division (under 18, over 25). See Table l for individual I.Q., M.A. and cause of deficiency information. 31 sos 1 m 1 .s.m 8s 1 M 1 so a... Hmmlwmléé HmHIMIéémm. 1 Honosuoaom 1. Hgosumaom m .4 ~s1m1.s.m 31m1.s.o o1m1.s.m sm1m1.s.m In. ~s1m1.m.m 31m1.m.o am1m1.s.m o1m1.m.m mo1m1.m.m «m1m1.s.m81 «m1x1.m.m om1x1.m.m ..0 1 Nuuonoum .1 Nuuonoum cum I x 1 .H.m mam 1. mm 1 .H.m _S Nss1w1.m.m Non-M1.m.mms 1 samosoosom 1 smaosoosmm w. m oo1m1.s.u ao1m1.s.m ossumsso so~1m1.s.o Hm omuwml .m.m moalMI .m.m menwml .H.m wNHIMI .m.m «Helm... .m.m omHIMI .H.m8"1 omlxléé womnxudfi .O .1 Nuuomoum .1 Nuuomonm . mawadomm :OHOMOHMHHGOVH ucogfiwuflw wafidaoam coausfimwudooH uaoaafimuuw v.33 umooaou 9.83 umoudou wdaawmuu .45 gum Madison“. mo madam mm Home: .<.o 5......» mm mm 996 .<.o sud; mm .A.m.mv sumo Hz<>msmm 1mommm now one was A.H.mV sumo Hz<>mAMMMHLmommm scum one mno>aw own maosumanoamo 03H. .wH wooo=\uo>o .o.H one mu wooaa\uo>o .<.o no.3 m1m1 pom doauouHHOIOuumHflsuu new: .0 3an DISCUSSION The contention that modeling is a useful technique in training children of the severely and profoundly retarded range has received only qualified support. Data from the property concept attainment task supported modeling. Data from the word identification and relational concept attainment tasks each gave borderline support; the word identification task data provided the stronger support, particularly with the error-relevant scale of trials-to-criterion. When non-learners were eliminated the evidence for the facilitation of modeling approached statistical significance for the word identi- fication task, with error—irrelevant data, but even including non- learners the error—relevant data analysis was significant. Hence the error-relevant scores provided better support for modeling, which is consistent with the idea that modeling facilitation may be partially due to error elimination. The spelling task was clearly not affected by modeling. The spelling task was essentially a serial learning task, while the others could be considered discrim- ination tasks. That modeling would facilitate discrimination tasks and not a serial learning task might be inferred from a selective attention position. A close look at the learning curves will be necessary to explain the differential effects of modeling on the three discrim- 32 33 ination tasks. It was mentioned above that modeling was expected to facilitate learning to the extent that the required response (an approach response) was elicited in the presence of positive relevant cues and prevented from occurring in the presence of negative or irrelevant cues. In every case, Group E.§s demonstrated an initial superiority with fewer errors in early trials. This was due, of course, to the fact that the_§ was merely imitating Efs responses on many of these early trials. Nevertheless, the S was making an approach response to the apprOpriate cues, and building up an S-R habit. Whether the S-R habit will be effective in the solution of the discrimination task is a problem of attention. House and Zeaman (1963) have shown that moderately retarded Ss can utilize positive and negative compounds of stimuli in forming discriminations. That is, the functional stimulus in an S-R habit might be a combination of different dimensional cues, such as position, sequence, form, color, etc., and these compounds Of stimuli can serve as negative cues (as in an avoidance S-R habit) or positive cues (as in an approach S-R habit). Studies mentioned earlier (Gormezano and Grant, 1958; McDavid, 1962) have suggested that modeling facilitates learning through a selective attention mechanism. Consider an alternative, though not necessarily conflicting, explanation. If the modeling condition provides for more approach responses to positive, relevant cues, and lessons the likelihood of approach responses to irrelevant or negative cues, the probability of sampling an effective stimulus compound 34 (positive) would be enhanced. An understanding of the formation of an effective stimulus compound could be facilitated by referring to the particular cues of the stimulus compound. Those combinations of cues which would enable the S to reSpond differentially to the total stimulus array are effective stimulus compounds. Discrimination problems vary in complexity. The number of different cues which must be sampled simultaneously to form an effective compound varies correspondingly. Another way of saying this is to refer to the amount of information required of a stimulus compound for it to be effective, i.e., the amount of information required for a stimulus compound to be effect- ive varies with the complexity of the task. In a complex discrimi— nation task, several dimensions of information, i.e., differential cues, may be necessary to form an effective stimulus compound. Several combinations of these necessary cues may be sampled as insufficient or ineffective stimulus compounds. Because these necessary cues are correlated with reinforcement, i.e., partially relevant to the task, the use of ineffective stimulus compounds may result in a fairly high ratio of reinforcement; thus the_§ is dis- couraged from attempting to form new compounds and discovering an effective stimulus compound, i.e., one correlated perfectly with reinforcement. Thus, the more complex the task, the smaller the probability of sampling an effective compound, since there are more possible inef- fective compounds. Predictions from this analysis would be in harmony with predictions from a selective attention prediction. 35 Both the relational concept attainment task and the word identifica- tion task could be solved using stimulus compounds. The word identi- fication would be the easier of the two theoretically because there is a finite, predictable number of possible ineffective compounds, which could be delineated by considering combinations of letters, positions and sequences. The number Of possible ineffective stimulus compounds for relational concept attainment with the procedure in this study is less specifiable, although a different training pro- cedure might allow for the number of compounds to be estimated. The number of possible compounds for the property concept attainment task is much smaller than for either the word identification task or the relational concept attainment task. The more complex discrimination tasks would require a longer modeling period than the easier prOperty concept attainment task to insure that effective stimulus compounds have had a chance to form habit strengths with the apprOpriate response. The present study employed a constant modeling period for all tasks, so the interaction of task difficulty and modeling would be expected. A simple example might make this analysis clear. If a S was asked to find the word DOG amongst the alternatives OGD, GDO, and DGO in a free responding situation, and the S chose OGD, E would say to the_§ "No, that is wrong". Consequently, the S could formulate a negative compound cue of 06, having sampled only the first two letters and their relative sequence. On subsequent trials, the probability of a correct choice is smaller, because a conflicting negative habit has begun to deve10p. If this response occurred after 36 several correct responses to DOG, the probability of sampling CD as the negative compound cue is much higher, since DC has already been sampled as a positive compound cue. A rational procedure would provide for several correct responses during the initial stage of training before allowing free responding and errors; the errors would then have a greater chance of providing effective negative compound cues. This is essentially what the modeling condition did for Group E.§s on the prOperty concept attainment task, and to a lesser extent on the word identification and relational concept attainment tasks. Modeling failed to facilitate in learning to spell. This is contradictory to the expectations of the effect of vicarious rein- forcement. Since the nature of the reinforcer is different from any used in the V.R. studies of the past, and is really more like Fletcher's implicit response condition (1965), this negative evidence is not particularly devastating to the V.R. position. The parameters which the present study has attempted to relate to modeling, that is, the elicitation of the approach response in the presence of positive relevant stimuli, were not effected by modeling in the spelling task. Appropriate responding was assured by the spelling procedure itself; consequently modeling should not have produced an effect. From this perspective, the spelling task could be viewed as a control task. Of course, if modeling facilitates through some inferred process of vicarious participation, the modeling condition should have facili— tated the spelling task, but it did not. This study leaves many important questions about what modeling can do unanswered. The assertion that merely providing more modeling 37 trials will increase the probability of effective stimulus sampling needs empirical validation. A study which simultaneously varied the number of modeling trials, as well as stimulus complexity (as defined by the number of possible ineffective stimulus compounds) might be a good place to start. Possibly more than an increase in modeling trials will be necessary to insure effective stimulus sampling. If §5 continue to respond to an ineffective stimulus compound during modeling, they will not benefit from modeling sufficiently to solve the problem. That this might occur is suggested by the observation that retardates tend to perseverate more than normals. Isolating an effective com- pound or components of a compound would be the next logical attempt to overcome the learning deficiencies of the retarded_§s. A method of isolating an effective compound can be illustrated by referring to the earlier example. If_§ could not learn to discriminate DOG, a new set of stimuli would be presented, with two letters and their posi— tions as the effective cues. For example, D-G as the correct alternative, and DG—, -DG, and G—D as the incorrect alternatives would be a problem which would require fewer dimensions Of information to form an effective stimulus compound. After the stimulus compound of D-G had gained positive strength and the other compound had gained negative strength, the original task of discriminating DOG from the original alternatives would be tried. If necessary, the construction of effective compounds could begin with only one letter and its position on the card as the effective cues, e.g., D—-, -D-, —-D, etc., with the progression to the above D—G stage and the original task 38 following. A progression Of specificity can be seen, in which the stimulus elements are manipulated to allow fewer ineffective stimulus com— pounds. The implication is that if the S does not learn to correctly respond to stimuli at one level of complexity, the stimuli are reduced in complexity until the S can respond differentially. The incidental finding of an apparent interaction of C.A. and I.Q. is not very meaningful due to the small number of Ss involved and the lack of equality in the number of §s in the C.A./I.Q. cells. The very large differences between the low I.Q. groups as a function of age are thought provoking, however, and suggest that a more com— prehensive study with a large number of Ss might be worthwhile. SUMMARY The present study investigated the effectiveness of a particular method, modeling, in training severely and profoundly retarded child- ren on four different tasks. Ss were ten institutionalized severely and profoundly retarded individuals, matched on C.A. and I.Q. and randomly assigned to a modeling or a non-modeling condition. The four tasks were (1) a property concept attainment task, (2) a re- lational concept attainment task, (3) a word identification task, and (4) a spelling task. Training took place on a teaching machine (MUDRAFA) specifically designed to overcome posited inhibition and attention deficits of the retarded. The property concept attainment task provided clear support for the facilitatory effects of modeling. The word identification task and relational concept attainment task also provided support, though of borderline statistical significance. The spelling task showed no differential effects of modeling. Possible reasons for the differential effects of modeling across tasks, i.e., interaction with task difficulty, were Offered. The complex nature of compound cues in the word identification and relational concept tasks might necessitate a longer modeling period than the prOperty concept task to provide an equal amount of facili- tation, while the present study employed modeling periods of equal length for all tasks. 39 40 A study in which the number of modeling trials and the complex— ity of cues were varied at several levels was suggested to test this explanation. Other techniques which might help to overcome the inability of retarded Ss to c0pe with complex stimulus compounds were tentatively offered. A general approach to training the retarded was implied, which stressed the necessity to (1) break down stimulus complexity to a level that is discriminable to the_§ (as indicated by the_§'s ability to respond differentially and (2) to provide external elicitors, e.g., modeling, powerful enough to assure that the correct response will occur consistently in early training so that effective stimulus sampling (the development of effective stimulus compounds) and the prevention of the development of inapprOpriate S-R habits will be more probable. REFERENCES Benoit, E. P. Relevance of Hebb's theory of the organization of behavior to educational research on the mentally retarded. .éEf.£3 Ment. Defic., 1957, 61, 497-506. Denny, M. R. A theoretical analysis and its application to training the mentally retarded. In Ellis, N. (Ed.), International Review of Mental Retardation, Vol. II. New York, Academic Press, 1966. Denny, M. R. Research in learning and performance. In Stevens, N. and Heber, R. (Eds.), Mental Retardation: A Review of Research. University of Chicago Press, 1964. Denny, M. R. and Aldman, H. M. Elicitation theory: An analysis of two typical learning situations. Psychol. Rev., l955,_62, 290—296. Eimas, P. Stimulus compounding in the discrimination learning of kindergarten children. _J. exp._§. Psychol., l965,_2, 178-185. Fletcher, Harold J. Discrimination learning by retardates as a function number of implicit response trials. Psychon. Sci., 1966, 4, 158, 159. House, B. J. and Zeaman, D. Role of positive and negative cues in the discrimination learning of mental defectives. Percept. mot. Skills, 1957, 1) 73-79. House, B. J. and Zeaman, D. Effects of practice on the delayed response of retardates. J, Comp. Physiol. Psychol., l96l,_§4, 223—229. House, B. J. and Zeaman, D. Visual discrimination learning in imbeciles. Amer. g. ment. Defic., 1958, 63, 447-452. Kanfer, F. H. and Marston, A. R. Human reinforcement: Vicarious and direct. g, exp. Psychol., 1963, 65, 292-296. Luria, A. R. and Vinogradovia, O. S. An objective investigation of the dynamics of semantic systems. Brit. J, Psychol., 1959,_§Q, 41 42 McDavid, J. R. Effects of ambiguity of environmental cues upon learning to imitate. J, abnorm. soc. Psychol., 1962,.65, 381-386. Miller, M. E. and Dollard, J. Social Learning and Imitation. New Haven, Yale University Press, 1941. Yascolt, M. Analysis of a method of training the mentally retarded. Unpublished M.A. thesis, Michigan State University, 1965. Zeaman, D. and House, B. J. The role of attention in retardates discrimination learning. In Ellis, N. R. (Ed.), Handbook of Mental Deficiency: .Psychological theory and research. New York: McGraweHill, 1963. NM 161953 11111111111111"