OBSERVATIONAL LEARNING AND CHILDREN WITH AUTISM:DISCRIMINATION TRAINING OF KNOWN AND UNKNOWN STIMULI

JAIME A. DEQUINZIO, BRIDGET A. TAYLOR AND BRITTANY J. TOMASI

ALPINE LEARNING GROUP

We extended past observational learning research by incorporating stimuli already known to par-ticipants into training. We used a multiple-baseline design across three participants to determinethe effects of discrimination training on the discrimination of consequences applied to modeledresponses using both known and unknown pictures. During baseline, participants were exposedto modeled correct and incorrect picture labels and were observed to imitate modeled responsesthat were incorrect and followed by negative feedback. During discrimination training, wetaught participants to label known pictures regardless of observed responses and consequences.With unknown pictures, we taught participants to imitate correct and reinforced modeledresponses, and to say, “I don’t know,” when modeled responses were incorrect and received neg-ative feedback. Test sessions measured responding to known and unknown pictures and showedacquisition over baseline levels. Generalization to pictures not associated with training was vari-able. Implications for teaching observational learning to children with autism are discussed.Key words: autism, discrimination training, observational learning

Observational learning was defined by Cata-nia (1998) as the learning of new responses thatemerges following observation of modeledresponses and their consequences. When mod-eled responses are followed by reinforcement,the observer is more likely to engage in thatresponse later in time. When modeledresponses are followed by extinction or punish-ment, the observer is less likely to engage inthose modeled responses later in time. Masiaand Chase (1997) offered a behavior analysis ofobservational learning, wherein they hypothe-sized that observational learning is influencedby generalized imitation, stimulus generaliza-tion, and acquisition of relevant conditionaldiscriminations. For example, a math problemand the response of the model is a compoundstimulus. The teacher’s feedback (either positiveor negative) following the modeled responsechanges the function of the compound

stimulus; the model’s response becomes a dis-criminative stimulus or s-delta for imitationconditional upon the teacher’s feedback. Thus,the observer must respond conditionally to thepresence and absence of reinforced and pun-ished responses. Masia and Chase also arguedthat emitting the response later in time mightbe a function of the presence of contextual stim-uli, or contextual control by additional instruc-tions provided to the model, and that imitationwill result in reinforcement. Perhaps the use ofrules and rule-governed behavior is a goodexample of the role of contextual control duringobservational learning (e.g., “When he gets itright, say what he says”), but more research isrequired to evaluate this interpretation.Given the complexity of observational learn-

ing, it is not surprising that early researchersdemonstrated that children with autism do notreadily learn by observing others (Lovaas, Koegel,Simmons, & Long, 1973; Varni, Lovaas, Koe-gel, & Everett, 1979). This has led to a develop-ing body of research evaluating procedures forteaching observational learning responses to chil-dren with autism (e.g., Greer, Dudek-Singer, &Gautreaux, 2006; MacDonald & Ahearn, 2015;

We would like to thank the teachers, students, andfamilies of the Alpine Learning Group who participated inthis study.

Address correspondence to: Jaime A. DeQuinzio,Alpine Learning Group, 777 Paramus Road, Paramus NJ,07652. E-mail: jdequinzio@alpinelearninggroup.org

doi: 10.1002/jaba.481

JOURNAL OF APPLIED BEHAVIOR ANALYSIS 2018, 51, 802–818 NUMBER 4 (FALL)

© 2018 Society for the Experimental Analysis of Behavior

802

Pereira-Delgado & Greer, 2009; Rehfeldt, Lati-more, & Stromer, 2003; Taylor & DeQuinzio,2012). Targeting observational learning deficitshas important educational, social, and economicimplications; if children with autism are able tolearn by observing others, it may lead toincreased social and educational opportunitiesand may reduce reliance on costly one-to-oneinstruction. Although current research is promis-ing in that it indicates that some children withautism may acquire observational learningresponses and may learn new information duringobservational learning contexts, more research isneeded to continue to evaluate variables thatmight influence learning through the observationof others.Emerging research has indicated that some

responses, if taught to children with autism,may facilitate learning during observationallearning contexts. For example, in a study byTaylor, DeQuinzio, and Stine (2012), compo-nent skills (i.e., attending and imitation)hypothesized to be associated with observationallearning were taught to three children withautism. Participants observed peers engaged ininstructional interactions during which peersemitted correct responses to sight wordsunknown to participants. Participants weretaught to attend to instructional stimuli and tomonitor the correct responses of their peers byproducing an imitative response and a matchingresponse indicating the participant had observedthe instructional stimuli. In the condition inwhich participants were taught to monitor theresponses of peers, participants learned newsight words more quickly than in the conditionin which they were not required to monitor theresponses of peers. In addition, once monitoringresponses were acquired in the training context,performance in reading the sight words in thenontraining condition increased, suggesting thatgeneralization of the monitoring responsesoccurred. These results suggested that learningto monitor the responses of a model facilitatedobservational learning. In this study, however,

participants observed only correct responses thatwere followed by reinforcement. They were notrequired to discriminate between reinforced(correct) responses and nonreinforced (incor-rect) responses.Expanding on these results, DeQuinzio and

Taylor (2015) incorporated modeled correctand incorrect responses, as well as teacher-delivered consequences that either did or didnot reinforce the model’s response. In thisstudy, the authors taught children with autismto discriminate between the reinforced and non-reinforced responses of the model. Acquisitionof unknown picture labels was then measuredduring test sessions. During baseline, partici-pants were exposed to modeled correct andincorrect responses and the respective conse-quences to each response, but participants werenot taught to discriminate the consequences ofmodeled responses. During discriminationtraining, participants were taught to imitate thereinforced responses of the model and to say, “Idon’t know,” when the model’s responses werenot reinforced. Training consisted of stating arule to the participants (i.e., “Say what she sayswhen she gets it right,” or “When she gets itwrong, say, ‘I don’t know’”) and reinforcingaccurate discriminations. All participantslearned the discrimination between reinforcedand nonreinforced trials during the training ses-sions. Test sessions conducted 1 and 10 minafter sessions throughout baseline, discrimina-tion training, and generalization showed thatthe participants maintained the discrimination.More specifically, the participants said, “I don’tknow,” in response to unknown picture labelsemitted by the model and not reinforced by theinstructor during training. By contrast, they saidthe correct picture label to unknown picturelabels emitted by the model and reinforced bythe instructor. Generalization to stimuli notassociated with training was variable across thefour participants. This study illustrated theimportance of systematic instruction for teach-ing learners with autism to make conditional

803OBSERVATIONAL LEARNING

discriminations during observational learningsessions.Although these studies elucidated many

important considerations when teaching chil-dren with autism to learn through observation,there were two critical issues associated witheach. First, the previous studies did not assess ifthe observed consequences were functional forthe participants. Although we assumed that theobserved consequences would have functionedas reinforcers and punishers for the participants,these consequences were not evaluated empiri-cally prior to observational learning sessions.Second, participants were exposed to mod-

eled responses that were initially unknown tothe participants. By learning to engage in com-ponent responses of observational learning(e.g., imitation, attention to instructional stim-uli, and discrimination of consequences) partici-pants acquired other responses (e.g., sightwords) that were not previously in their reper-toires. Still unclear from these studies is theeffect of incorporating stimuli already known toparticipants (e.g., stimuli that evoke correct pic-ture labels) into observational learning. Wouldparticipants continue to say the correct label ifmodeled responses were incorrect and followedby negative feedback, or would they imitate theincorrect response? Further, if participants withautism did indeed make this error, could weteach them to respond to known and unknownstimuli in the context of observational learning?Thus, the purpose of the current study was

to extend Taylor et al. (2012) and DeQuinzioand Taylor (2015) by ensuring that theobserved consequences provided to the modelswere functional for the participants and byincluding both unknown and known stimuliduring observational learning sessions.

METHOD

Participants and SettingThe participants were three children with

autism who attended a behaviorally based

school where the primary teaching was basedon applied behavior analysis. All participantshad experience using token economies and hada history of learning simple discriminations aswell as match-to-sample conditional discrimina-tions. All demonstrated generalized imitativerepertoires. All participants in this study hadsmall-group instruction for a portion of theschool day, in which a variety of academic sub-jects were taught such as science, social studies,and reading. All participants had learned to say,“I don’t know,” in the presence of unknownstimuli as part of prior instructional program-ming. Participants were selected to participatein this study because they had either observa-tional learning goals or group-instruction goalsthat were specified within their IndividualizedEducation Program. Sessions took place inrooms other than the participants’ classrooms,such as other classrooms, offices, and meetingrooms. The sessions were all conducted byteachers familiar with the participants andtrained in both applied behavior analysis andthe procedures for this study.Billy was 7 years old at the time of the study.

His age-equivalent score on the Peabody PictureVocabulary Test-Fourth Edition (PPVT-IV)(Dunn & Dunn, 2007) was 6 years, 3 months.His age equivalent score on the ExpressiveVocabulary Test (EVT-II) (Williams, 2007) was5 years, 11 months. His instructional programstargeted skills such as participating in groupinstruction, imitating behavior of a group, lis-tening to and following directions related toworksheets, and describing events. Billy’s otherinstructional programs at the time of the studyincluded demonstrating a variety of functionalresponses in the community, tying shoes, andcommenting to an individual about his ownplay. In addition, Billy had been taught to par-ticipate in dyad instruction, use the phrase “Idon’t know,” and selectively imitate a model.Hank was 12 years old at the time of the

study. His age-equivalent score on the PPVT-IV was 7 years, 10 months. His age equivalent

JAIME A. DEQUINZIO et al.804

score on the EVT-II was 5 years, 8 months.Hank’s instructional programming focused onexpressive language, including describing topics,initiating conversations, and asking and answer-ing social questions. His instructional programsalso included prevocational skills, self-care skillssuch as brushing teeth, and community skillssuch as completing a schedule of prevocationaltasks and ordering food from a fast food restau-rant. Additionally, Hank had been taught touse the phrase “I don’t know,” to use “yes” and“no” functionally, and to participate in bothdyad and group instruction.Rich was 12 years old at the time of the

study. His age-equivalent score on the PPVT-IV was 4 years, 3 months. His age equivalentscore on the EVT-II was 6 years. His instruc-tional programs included answering generalknowledge questions, listening to and followingdirections related to worksheets, asking socialquestions, and following instructions with adelay. In addition, Rich’s instructional pro-grams included prevocational skills, self-careskills, domestic skills, and community skillssuch as replenishing supplies, operating a wash-ing machine and dryer, and demonstrating avariety of functional responses in the commu-nity. Rich had been taught to imitate vocalmodels, label pictures, use the phrase. “I don’tknow,” and participate in both dyad and groupinstruction.Teachers were trained in the clinical applica-

tion of applied behavior analysis and had workedat the school for 2 to 5 years. All teachers weretrained by the first author to implement theobservational learning protocol. Adult modelswere varied depending on schedules and avail-ability, and consisted of adult staff members atthe school including other teachers, instructionalaides, supervisors, volunteers, and interns.

Experimental DesignA concurrent multiple-baseline-across-

participants design was used to evaluate the

effects of discrimination training on correctresponding during the four trial types describedbelow. The third baseline was delayed by fivesessions for the third participant.

Dependent MeasuresThe main dependent measure was summa-

rized as the percentage of correct responses dur-ing test sessions conducted 10 min after eachbaseline, discrimination training, and generali-zation session, as well as during discriminationtraining after the model. There were four cor-rect responses and trial types that included(a) correctly labeling pictures in the presence ofstimuli that were known by the participant,when modeled responses were correct and rein-forced; (b) correctly labeling pictures in thepresence of stimuli that were known by theparticipant, when modeled responses wereincorrect and followed by negative feedback;(c) correctly labeling pictures in the presence ofunknown stimuli, when modeled responseswere correct and reinforced; and (d) saying, “Idon’t know,” in the presence of unknown stim-uli when modeled responses were incorrect andfollowed by negative feedback (see Table 1).We also summarized data from the test ses-

sions as a) the mean percentage of responsescorrect in the presence of known stimuli onboth correct, reinforced trials and incorrect,negative feedback trials, and b) the mean per-centage of responses correct in the presence ofunknown stimuli on both correct, reinforcedtrials and incorrect, negative feedback trials.

Interobserver Agreement and TreatmentIntegrityInterobserver agreement (IOA) was calcu-

lated for responses during test sessions con-ducted after baseline, discrimination training,and generalization. An agreement was countedif both the teacher and a second observer inde-pendently scored a response as correct or incor-rect in the same trial. IOA was calculated on a

805OBSERVATIONAL LEARNING

trial-by-trial basis by dividing the number ofagreements by the number of agreements plusdisagreements and converting the result to apercentage. For Billy, IOA was collected for25% of test sessions and for all conditions ofthe concurrent-operant assessment. The meanIOA was 94% (range, 94%-100%) duringbaseline, 100% during discrimination training,and 100% during generalization. Interobserveragreement for the concurrent-operant assess-ment for Billy was 95%. For Hank, IOA wascollected for 30% of test sessions and was100% during baseline, discrimination training,and generalization. For Rich, IOA was collectedfor 21% of the test sessions and was 100% dur-ing baseline, discrimination training, andgeneralization.Treatment integrity (TI) data were collected

on the accurate implementation of the indepen-dent variable during baseline, discriminationtraining, generalization, and test sessions. Anindependent observer used a checklist contain-ing all of the procedures described above(i.e., environmental arrangement, accuratedelivery of feedback based on trial and sessiontype, use of error correction, schedules of rein-forcement, and presentation of stimuli during

test sessions). Data on TI were collected during14%, 11%, and 13% of sessions for Billy,Hank, and Rich, respectively, and TIwas 100%.

MaterialsA set of 12 pictures was used as training

stimuli during discrimination training sessions,including six known items (e.g., toaster, fork)and six unknown items (e.g., a pizza cutter, agarlic press, a pastry scraper) identified duringthe pretest. For each participant, we also chose12 additional pictures of six known items(e.g., knife, cup, plate, microwave) and sixunknown items (e.g., corkscrew, can opener,funnel, sharpening rod) to use during generali-zation sessions. Generalization sessions wereconducted approximately every three trainingsessions. For both training and generalizationsets, three of the known and three of theunknown stimuli were further divided intoreinforced and negative feedback categoriesusing random assignment. Pictures were pre-sented on 7.6 cm x 12.7 cm index cards. Noneof the pictures chosen were used at other timesduring the participants’ regular instructionalprograms.

Table 1Procedural Components Across Each Stimulus Category

StimulusModel’sResponse Consequence Correct Response for Participant Error Correction Rule

Known toParticipant Correct Reinforcement (Positive

feedback +edible)Says correct response usedduring pretest

You know it.She got it right.Say what you know.

Incorrect Negative feedback Says correct response used during pretestDoes not imitate model

You know it.She got it wrong.Say what you know.

Unknown toParticipant Correct Reinforcement

(PositiveFeedback+edible)

Imitates model You don’t know it. Shegot it right.

Say what she said.

Incorrect Negativefeedback

Says “I don’tknow”

You don’t know it. Shegot it wrong.

Say, “I don’t know.”

JAIME A. DEQUINZIO et al.806

Individualized token systems were used ineach session to reinforce on-task behavior. Jellybeans and M&Ms® were used as reinforcersand were identified using a single-operant rein-forcer assessment described below.

Single-Operant Assessment of Tangible ItemsTo ensure that the stimuli provided to the

model contingent on correct responding func-tioned as reinforcers for participant behavior,we conducted a single-operant reinforcer assess-ment similar to the procedures used by Smaby,MacDonald, Ahearn, and Dube (2007). Wecompared the rate of responding (i.e., placingbeads in a small plastic bin) across reinforce-ment and extinction phases, in which eachresponse produced a putative reinforcer or noconsequences, respectively. Reinforcement andextinction phases were alternated in an ABABreversal design. Overall, for all participants,responding was higher during conditions inwhich candy was provided and lower duringextinction conditions (data available uponrequest).

Concurrent-Operant Assessment of VerbalFeedbackTo determine relative preference for the

feedback statements (e.g., “That’s right,”“That’s wrong”) provided to the model, weconducted a concurrent-operant reinforcerassessment similar to the procedures describedby Hanley, Piazza, Fisher, Contrucci, andMaglieri (1997). We measured the number ofresponses allocated to two simultaneously avail-able response options during consecutive 2-minsessions. Response 1 was placing items in a bin60 cm to the left of the participant, andresponse 2 was placing items in a bin 60 cm tothe right of the participant. During the firstphase, responding on the left produced negativefeedback (e.g., “That’s wrong.”) and respond-ing on the right produced positive feedback(e.g., “That’s right!”). Negative feedback

statements were variations of the statement“That’s wrong,” including, but not limited to,“I’m sorry that’s wrong,” “No, that’s wrong,”and “You are wrong.” Positive feedback state-ments were variations of the statement “That’sright!” including, but not limited to, “Terrific,that’s right!” “Yes, that’s right!,” and “You areright!” These statements were later used duringobservational learning. During the secondphase of the assessment, the contingency wasswitched; responding on the left produced posi-tive feedback, and responding on the right pro-duced negative feedback. These two phaseswere alternated with a baseline comparisoncondition during which no feedback was pro-vided for either response within an ABABreversal design. For all participants, respondingwas higher when positive feedback was pro-vided as opposed to when negative feedbackwas provided (Figure 1).

PretestTo determine which pictures were known

and unknown, we conducted a pretest. We pre-sented 30-35 pictures randomly, three timeseach, and asked, “What’s this?” If the partici-pant failed to label the picture on all three tri-als, we considered it unknown. If theparticipant labeled the picture correctly on allthree trials, we considered it known. If the par-ticipant responded correctly on some trials andincorrectly on others, we removed it from thestimulus pool and excluded it from the study.We did not provide any prompts or reinforce-ment contingent upon participant responses.Pictures were assigned to the discriminationtraining and generalization conditions asdescribed in the materials section above.

BaselineWe conducted baseline sessions to evaluate

responding on test sessions prior to exposing theparticipants to modeled responses that were fol-lowed by reinforcement and negative feedback,

807OBSERVATIONAL LEARNING

in the presence of both known and unknownstimuli. There were six known/reinforced trials,six known/negative feedback trials, sixunknown/reinforced trials, and six unknown/negative feedback trials all presented in randomorder throughout the session, totaling 24 trials(Table 1).The participant and the adult model sat next

to each other at a table, and the teacher satacross from them. An adult model was chosen

rather than a peer model to better control forthe arrangement of correct and incorrectresponses. The edible reinforcer identified inthe single-operant assessment was placed on thetable in view of both the participant and theadult model. The teacher held either a knownor unknown picture between the two, ensuredthat the participant was attending to the picture(Taylor et al., 2012), and said, “What is this?”to the adult model. If the participant did not

Sessions

Res

po

nse

per

Min

on

Lef

t an

d R

igh

t

Billy

Hank

Rich

Baseline Positive FB Right

Negative FB Left

Positive FB Left

Negative FB Right Baseline

Right Left

Figure 1. Response per minute on the left and right of the concurrent-operant assessment of feedback during base-line, when positive feedback (FB) was provided for responding on the right and negative feedback was provided forresponding on the left, and when positive feedback was provided for responding on the left and negative feedback wasprovided for responding on the right.

JAIME A. DEQUINZIO et al.808

look when the teacher held up the stimulus,she provided the instruction “Look” or “Every-one look” prior to asking, “What’s this?” Theadult model provided a preplanned correct orincorrect response, and the teacher provided aprogrammed consequence depending on thetrial type (Table 1).There were two trial types that differed with

respect to programmed consequences for themodel: reinforced trials and negative feedbacktrials. The order of reinforced and negativefeedback trials was randomized. During rein-forced trials, the adult model answered cor-rectly, and the teacher provided the positivefeedback statement (e.g., “That’s right!” “Yes,that’s right!” “You got it right!”) and an ediblereinforcer (e.g., jelly beans). The teacher neverused the picture label in the feedback statementso that learning of new picture labels could beattributed to observation of the modeledresponse rather than to the teacher’s label ofthe picture. During negative feedback trials, theadult model answered incorrectly, and theteacher provided negative feedback, (e.g., “I’msorry, that’s wrong,” “No that’s wrong,” “Youare wrong”) to the model and removed the edi-ble reinforcer from the table. Again, the teachernever used the correct label in the feedbackstatement. A fixed-time 30-s schedule of tokendelivery was used with the participants to pro-mote sitting quietly with hands down through-out all baseline, discrimination training, andgeneralization sessions to decrease the likeli-hood of the occurrence of any problem behav-ior. After 15 tokens were earned, they wereexchanged for one edible reinforcer.

Discrimination TrainingTraining was conducted to teach participants

to discriminate the reinforced responses of theadult model from the responses that were fol-lowed by negative feedback in the presence ofboth known and unknown stimuli. One sessionwas conducted 3 days per week. The

arrangement and presentation of trials, as wellas the teacher’s interaction with the adultmodel, was similar to that of baseline, with oneexception. After providing the consequence tothe adult model, the teacher immediately pre-sented the picture to the participant and said,“What is this?” Correct responses producedpraise and an edible. Incorrect responses pro-duced an error-correction procedure includinga rule and additional practice opportunity. Thepraise statements and error-correction rules dif-fered and were specific to each of the four trialtypes (described below; Table 1). During errorcorrection, the teacher provided the error-correction rule, asked the adult model to labelthe picture again, provided the consequence,and gave the participant the chance to practicethe response described in the rule statement. Ifthe participant answered correctly on this error-correction trial, general praise was provided,but the edible reinforcer was not provided. Ifthe participant still did not answer correctly,the trial was terminated, and the teacher movedon to the next trial. (Table 1).Known/reinforced trials. If the participant said

the correct answer when the teacher held upthe card and asked, “What is it?” the teacherprovided praise to the participant (i.e., “Youare right! Good saying what you know!”) anddelivered the edible reinforcer. If the partici-pant answered incorrectly or did not answer atall within 3 s, the teacher initiated the error-correction procedure by stating the rule, “Youknow it. She got it right. Say what you know.”Known/negative feedback trials. If the partici-

pant said the correct answer, the teacher pro-vided praise (i.e., “You are right! Good sayingwhat you know when she got it wrong!”) anddelivered the edible reinforcer. If the partici-pant answered incorrectly, imitated the incor-rect response from the model, or did notanswer at all within 3 s, the teacher initiatedthe error-correction procedure by stating therule, “You know it. She got it wrong. Say whatyou know.”

809OBSERVATIONAL LEARNING

Unknown/reinforced trials. If the participantsaid the correct answer, the teacher providedpraise (i.e., “Good saying what she said whenshe got it right!”) and delivered the edible rein-forcer. If the participant answered incorrectlyor did not answer at all within 3 s, the teacherinitiated the error-correction procedure by stat-ing the rule, “You don’t know it. She got itright. Say what she said.”Unknown/negative feedback trials. If the par-

ticipant said the correct answer (i.e., “I don’tknow”), the teacher provided praise(i.e., “Good saying ‘I don’t know’ when shegets it wrong!”) and delivered the edible rein-forcer. If the participant answered incorrectlyor did not answer at all within 3 s, the teacherinitiated the error-correction procedure by stat-ing the rule, “You don’t know it. She got itwrong. Say, ‘I don’t know.’”

GeneralizationTo determine the extent to which the dis-

crimination learned during the discriminationtraining sessions would occur in the presence ofstimuli not associated with discrimination train-ing, we conducted generalization sessions. Gen-eralization sessions were conductedapproximately every three to five training ses-sions, using a different set of pictures. Generali-zation sessions were, procedurally, the same asbaseline and consisted of the same trial typespresented during training.

Test SessionsTest sessions were conducted 10 min after

baseline and discrimination training sessions.During this 10-min interval, participantsreturned to regularly scheduled programs. Dur-ing the test sessions, the model was not present.The teacher presented the pictures used in theprior baseline, discrimination training, or gen-eralization session. The pictures were presentedin a random order two times each. During eachtrial, the teacher sat across from the participant,

held a picture in view, and asked, “What’sthis?” The teacher then recorded the partici-pant’s response without providing reinforce-ment or prompts and moved on to the nexttrial. A fixed-time 30-s schedule of token deliv-ery was used to promote sitting quietly withhands down throughout the session. The mas-tery criterion was 90% of total responses cor-rect for two consecutive sessions.

Procedural Modifications for RichOne stimulus originally identified in the pre-

test as unknown (i.e., ice cream scoop) andanother stimulus originally identified as known(i.e., grill) were removed from the training seton sessions 18 and 23, respectively. Duringtraining, Rich started to label the ice creamscoop correctly even though the correctresponse was never modeled (i.e., ice creamscoop was assigned to the unknown/negativefeedback set in which the model never providedthe correct label). The picture of the grill wasidentified as known in a pretest. However, overtime, he stopped providing the correct responseand instead said nothing at all on these trials.We decided to remove both stimuli from theset to prevent misinterpretation of the results.

RESULTS

Figure 2 displays the percentage of correctresponses to training stimuli (top panels) andgeneralization stimuli (bottom panels) duringtest sessions on known/reinforced trials, known/negative feedback trials, unknown/reinforcedtrials and unknown/negative feedback trials forBilly, Hank, and Rich. With the systematicintroduction of discrimination training, weobserved increases in the total level of correctresponding to training stimuli (top panels) forall participants. For Billy, baseline respondingfor known/reinforced trials was high butdecreased slightly. On known/negative feedbacktrials, Billy’s responding was moderate andincreased to 100%. Responding on unknown/

JAIME A. DEQUINZIO et al.810

reinforced trials was initially low and increased to100%, and responding on unknown/negativefeedback trials remained near zero with a slightincrease on session 4. With the introduction ofdiscrimination training, Billy’s respondingincreased to 100% for all trial types. For Hank,baseline responding on known/reinforced trialswas at 100%; on known/negative feedback trials,his responding was moderate and variable.Responding on unknown/reinforced trials gradu-ally increased to 100%, and responding onunknown/negative feedback trials remained atzero. With the introduction of discriminationtraining, Hank’s responding increased to 100%for all trial types. For Rich, baseline responding

on known/reinforced trials was at 100% anddecreased slightly near the end of baseline.Responding on known/negative feedback trialswas moderate and variable. Responding onunknown/reinforced trials remained at zero, andresponding on unknown/negative feedback trialswas initially low and increased at the end of base-line. With the introduction of discriminationtraining, Rich’s responding increased to 100%for known/reinforced trials, unknown/negativefeedback trials, and known/negative feedback tri-als, and increased but remained under criterionlevels for unknown/reinforced trials.Responding to generalization stimuli (lower

panels) was more variable across the three

Figure 2. Percentage of responses correct to training stimuli (top panels) and generalization stimuli (bottom panels)during test sessions on known/reinforced trials (solid squares), known/negative feedback trials (open squares), unknown/reinforced trials (solid circles) and unknown/negative feedback trials (open circles) for Billy, Hank, and Rich.

811OBSERVATIONAL LEARNING

participants. For Billy, baseline responding togeneralization stimuli on known/reinforced tri-als and known/negative feedback trials was100%. Responding on unknown/reinforced tri-als increased slightly, and responding onunknown/negative feedback trials remained atzero. With the introduction of discriminationtraining, Billy’s responding to generalizationstimuli eventually increased to 100% for theunknown/negative feedback trials and remainedat 67% for the unknown/reinforced trials. ForHank, baseline responding to generalizationstimuli on known/reinforced and known/nega-tive feedback trials was high. Responding onunknown/reinforced trials remained moderateat 67%. With the introduction of discrimina-tion training, Hank’s responding to unknown/negative feedback generalization stimuliincreased to 100%, and responding tounknown/reinforced generalization stimuliremained at 67%. For Rich, baseline respond-ing to generalization stimuli on known/rein-forced and known/negative feedback trialsdecreased during baseline. Responding onunknown/reinforced trials remained at zero,and responding on unknown/negative feedbacktrials was initially low and increased at the endof baseline. With the introduction of discrimi-nation training, Rich’s responding to generali-zation stimuli persisted at 100% for unknown/negative feedback and known/reinforced trialsbut remained at zero for unknown/reinforcedtrials and remained relatively the same forknown/negative feedback trials.Figures 3 and 4 display the mean percentage

of correct responses to training stimuli for alltrial types. For all participants, pretest measuresrepresent responding to known stimuli prior toexposure to the model. Responding on test ses-sions during baseline and discrimination train-ing represent responding following exposure tothe model. Figure 3 shows the mean percentageof correct responses to known training stimuliduring test sessions for both reinforced trials(left panel) and negative feedback trials (right

panel). During baseline, when modeledresponses were correct and reinforced (leftpanel), correct responding (i.e., imitation ofcorrect responses that were reinforced)remained high for Hank and Rich. For Billy,correct responding decreased from pretest mea-sures during baseline. With the implementationof discrimination training, the mean percentageof correct responses increased for Billy. Duringbaseline, when modeled responses were incor-rect and followed by negative feedback, correctresponding (i.e., stating the correct picturelabel) for all participants dropped from pretestmeasures. When discrimination training wasimplemented, correct responding increased forall participants. Figure 4 shows the mean per-centage of correct responses to training stimulion test sessions in the presence of unknownstimuli on both reinforced trials (left panel) andnegative feedback trials (right panel). Partici-pants did not label any pictures correctly dur-ing pretest measures. During baseline, whenmodeled responses were correct and followedby reinforcement (left panel), correct respond-ing (i.e., imitation of modeled responses thatwere correct and reinforced) increased for Billyand Hank, but not for Rich. On these trials,correct responding increased for all participantsonce discrimination training was implemented.During baseline, when modeled responses wereincorrect and followed by negative feedback(right panel), correct responding (i.e., saying, “Idon’t know”) increased for Billy and Rich, butnot for Hank. Correct responding increased forall participants on these trials when we imple-mented discrimination training.

DISCUSSION

One outcome of this study replicates thoseof DeQuinzio and Taylor (2015) in that chil-dren with autism learned to discriminate thecontingencies applied to modeled responseswhen stimuli were unknown. That is, theylearned new picture labels when modeled

JAIME A. DEQUINZIO et al.812

responses were accurate and reinforced, and tosay, “I don’t know” when modeled responseswere inaccurate and followed by negative feed-back. We extended the research on observationallearning by incorporating known stimuli intodiscrimination training sessions. Inconsistent

responding by participants to the known stimuliwhen the modeled response was incorrect and fol-lowed by negative feedback suggests that incorporat-ing known stimuli into these sessions is a beneficialcomponent to aid in acquisition of the conditionaldiscriminations involved in observational learning.

Mea

n P

erce

nta

ge

of

Corr

ect

Res

ponse

s to

Know

n S

tim

uli

Modeled Response Correct

Reinforced

PT

Modeled Response Incorrect

Negative Feedback

BL DT

PT BL DT PT BL DT

PT BL DT PT BL DT

PT BL DT

Billy

Rich

Hank

Figure 3. The mean percentage of correct responses to training stimuli on test sessions during pretest (PT), baseline(BL), and discrimination training (DT) in the presence of known stimuli on both correct reinforced trials (left panel)and incorrect negative feedback trials (right panel).

813OBSERVATIONAL LEARNING

Furthermore, our results support those of Pereira-Delgado and Greer (2009) who incorporatedknown stimuli into observational learning training

so that participants could learn to label the correctand incorrect respones of a model by choosinggreen and red blocks, respectively.

Mea

n P

erce

nta

ge

of

Corr

ect

Res

ponse

s to

Unkno

wn S

tim

uli

Modeled Response Correct

Reinforced

PT BL

Modeled Response Incorrect

Negative Feedback

DT

PT BL DT PT BL DT

PT BL DT PT BL DT

PT BL DT

Billy

Rich

Hank

Figure 4. The mean percentage of correct responses to training stimuli on test sessions during pretest (PT), baseline(BL), and discrimination training (DT) in the presence of unknown stimuli on both correct reinforced trials (left panel)and incorrect negative feedback trials (right panel).

JAIME A. DEQUINZIO et al.814

As Figure 2 illustrates, and consistent withpast research (DeQuinzio & Taylor, 2015),some participants in the current study acquireda few of the unknown stimuli in baseline bybeing exposed to the model’s correct and rein-forced responses. Also, because participants hadlearned to say, “I don’t know,” in prior instruc-tional programming, they occasionally used thisresponse in baseline. Therefore, responding onthese trials was scored as correct. However, oncediscrimination training occurred and partici-pants were taught the rules (e.g., “You don’tknow it. She got it wrong. Say, ‘I don’t know’”),accuracy during test sessions increased.Some of the participants’ responding was

inconsistent. As Figure 3 (left panel) illustrates,on the pretests participants always or almostalways (Billy) responded accurately to knownstimuli. During baseline, however, respondingvaried across participants. Correct respondingmaintained for Rich and Hank and decreasedfor Billy. This occurred even though all partici-pants were exposed to a model saying a correctand reinforced response that matched theirresponse during the pretest. For some trials,Billy added a suffix to his response (e.g., heresponded correctly with the label “garlic press”during the pretest but changed it to “garlicpressing” during baseline). It was unclear whythis shift occurred. It may have been related tobeing exposed to all trial types in baseline,which included correct and incorrect responsesof the model and both negative feedback andreinforcing consequences to modeled responses.Billy may also have combined words fromother picture labels with “ing” (e.g., “rollingpin” and “sharpening rod”) that were targets inthis set. Correct responding increased for Billyonce discrimination training was implemented.As Figure 3 (right panel) illustrates, when

participants were exposed to the model sayingan incorrect response that was followed by neg-ative feedback during baseline, correct respond-ing decreased for Hank and Rich. We observedparticipants imitating the inaccurate responses

of the model even though these were knownstimuli as was demonstrated during the pretest.This indicated that participants were not dis-criminating the contingencies applied to mod-eled responses and perhaps were relying onimitative repertoires with stronger histories ofreinforcement. Once discrimination trainingoccurred (i.e., learning the rule, “You knowit. She got it wrong. Say what you know”), cor-rect responding increased on these trials and weobserved less imitation of incorrect responses.Similar to DeQuinzio and Taylor (2015), we

included unknown stimuli and analyzedresponding in the presence of modeledresponses that were either reinforced or fol-lowed by negative feedback. As Figure 4 (leftpanel) illustrates, participants did not respondcorrectly to any of the stimuli during pretestsessions. When baseline was initiated, Billy andHank learned several new picture labelsthrough exposure to the model’s correct andreinforced responses. Rich, on the other hand,did not. The differences among participants inacquiring new labels by being exposed to theaccurate responses of the model may be due tothe absence of direct contingencies for imita-tion during baseline. Once discriminationtraining was implemented and contingenciesfor imitation were in place, all three partici-pants showed an increase in correct picturelabels during test sessions, with Billy and Hankreaching criterion levels. Rich, however, didnot learn all the target stimuli.When modeled responses were followed by

negative feedback (Figure 4, right panel), Billyand Rich responded correctly on some trials inbaseline by saying, “I don’t know;” a responsethat had a history of reinforcement in their reg-ular academic programming. Although this wasalso true for Hank, he did not use the “I don’tknow” response in baseline. Additionally, onsome trials, all three participants occasionallyimitated the incorrect responses that were mod-eled and followed by negative feedback, indicat-ing a weak discrimination of consequences

815OBSERVATIONAL LEARNING

applied to modeled responses. With the intro-duction of discrimination training, however, allthree participants showed increases in the use of“I don’t know,” indicating the effectiveness ofdiscrimination training. Billy did not reach crite-rion levels of responding; on some trials, insteadof saying, “I don’t know,” Billy continued toimitate responses of the model that were incor-rect and followed by negative feedback. We areunclear as to why he did this, but we observedthat Billy imitated one of the negative feedbackstatements used by the teacher (e.g., “I’m sorry,that’s wrong”) while laughing and looking at theteacher during some of the trials. These resultsare consistent with those of Ingvarsson and Hol-lobaugh (2010) and Ingvarsson, Tiger, Hanley,and Stephenson (2007) who found that youngchildren with developmental disabilities dis-played inconsistencies demonstrating generaliza-tion of “I don’t know” responses or failed todiscriminate between known and unknownstimuli following the training of this response.In addition to the use of known stimuli, we

expanded on past research by ensuring conse-quences for the modeled responses were empiri-cally derived prior to use during observationallearning. Specifically, we evaluated the reinfor-cing value of the edibles provided to the modelusing a single operant reinforcer assessment.We also determined relative preferences for thepositive and negative feedback statements usinga concurrent operant assessment. We did not,however, conduct an avoidance assessment todetermine whether the removal of the edibleand the negative feedback statement functionedas punishers. It is important to note thatDeQuinzio and Taylor (2015) did not conductempirical analyses of feedback statements ortangible items provided for correct responding,yet participants learned the target discrimina-tions. Thus, we are unsure of the possibleimpact of these stimuli on observational learn-ing. Future research could compare the effectsof empirically and nonempirically derived con-sequences to determine this.

Past research has demonstrated that childrenwith autism can learn to respond to rules con-taining “if/then” statements, describing theantecedent and behavior (Tarbox, Zuckerman,Bishop, Olive, & O’Hora, 2011). Similarly, inthis study, we applied the use of rules toobservational learning. The teacher used spo-ken rules describing the antecedent (e.g., “Youknow it. She got it wrong”) and behavior(“Say what you know”) as part of an error-correction procedure. It is unclear if these rulesenhanced discrimination in the observationallearning context. Future studies might com-pare the use of discrimination training withthese rules to training without these rules. Forexample, a multiple baseline design can beused in which baseline includes a discrimina-tion training procedure that does not use rules,such as differential reinforcement. The inter-vention could include differential reinforce-ment plus the rules. One could thendetermine if observational learning increases oris more efficient once the rules are introduced.In addition, because rules were used as part ofan error-correction procedure in our study, itwould be interesting to examine the effects ofminimizing errors in an errorless-teaching for-mat. It could be the case that allowing errorsduring conditional discrimination trainingnegatively affected acquisition in this study.Employing errorless teaching procedures in aredesigned study could reveal whether permit-ting errors during the discrimination trainingin the current study negatively affectedacquisition.In this study, accurate responses during test

sessions were not reinforced. We did this fortwo reasons. First, we wanted to ensure thatresponding was under the control of theinstructional stimuli used during observationallearning and that an operant reinforcementcontingency did not influence respondingduring tests. Second, we wanted to moreclosely simulate naturally-occurring observa-tional learning, in which the learner observes

JAIME A. DEQUINZIO et al.816

a reinforcement contingency for modeledresponses and does not directly contact thereinforcement contingency for his or her ownbehavior (Catania, 1998). Future researchcould compare the effects of reinforcementand extinction during tests of observationallearning.There are several other variables that might

have affected responding. For example, wetaught four conditional discriminations simulta-neously using mixed trials. This required manytrials (i.e., 24) so that all trial types (i.e., known/reinforced, known/negative feedback, unknown/reinforced, unknown/negative feedback) couldbe presented an equal number of times. We alsoassigned three target stimuli to each trial typeand required that each stimulus be presented atleast two times. This combination of variablesmay have interfered with acquisition of the dis-crimination. It is possible that limiting the num-ber of stimuli would have facilitated thediscrimination and resulted in more consistentresponding across participants. In addition,while the many trial types were necessary to eval-uate observational learning variables, they werenonetheless cumbersome to implement. It maybe that variations in instructional procedures,such as video-taped instructors and models, maybe more efficient to implement as they wouldnot require two adults (Spriggs, Gast, &Knight (2016).Observational learning research with people

with autism is still in its infancy. Even so, thisstudy makes an important contribution to thefoundational research in the field. We incorpo-rated known stimuli and used explicit rules toenhance discrimination during observationallearning and demonstrated that children withautism can learn complex conditional discrimi-nations through observational learning. Suc-ceeding research should continue to isolatevariables that influence observational learningand strive toward the creation of a reliableteaching technology for improving this impor-tant repertoire.

REFERENCES

Catania, A. C. (1998). Learning (interim 4th ed.). Corn-wall-on-Hudson, NY: Sloan.

DeQuinzio, J. A., & Taylor, B. A. (2015). Teaching chil-dren with autism to discriminate the reinforced andnonreinforced responses of others: Implications forobservational learning. Journal of Applied BehaviorAnalysis, 48, 38-51. https://doi.org/10.1002/jaba.192

Dunn, L. M., & Dunn, D. M. (2007). Peabody picturevocabulary test (4th ed.). Bloomington, MN: Pearson.

Greer, R. D., Dudek-Singer, J., & Gautreaux, G. (2006).Observational learning. International Journal ofPsychology, 41, 486-499. https://doi.org/10.1080/00207590500492435

Hanley, G. P., Piazza, C. P., Fisher, W. W.,Contrucci, S. A., & Maglieri, K. A. (1997). Evalua-tion of client preference for function-based treatmentpackages. Journal of Applied Behavior Analysis, 30,459-473. https://doi.org/10.1901/jaba.1997.30-459

Ingvarsson, E. T., & Hollobaugh, T. (2010). Acquisitionof intraverbal behavior: Teaching children withautism to mand for answers to questions. Journal ofApplied Behavior Analysis, 43(1), 1-17. https://doi.org/10.1901/jaba.2010.43-1

Ingvarsson, E. T., Tiger, J. H., Hanley, G. P., &Stephenson, K. M. (2007). An evaluation of intraver-bal training to generate socially appropriate responsesto novel questions. Journal of Applied BehaviorAnalysis, 40, 411-429.

Lovaas, O. I., Koegel, R., Simmons, J. Q., & Long, J. S.(1973). Some generalization and follow-up measureson autistic children in behavior therapy. Journal ofApplied Behavior Analysis, 6, 131-166. https://doi.org/10.1901/jaba.1973.6-131

MacDonald, J., & Ahearn, W. H. (2015). Teachingobservational learning to children with autism. Jour-nal of Applied Behavior Analysis, 48, 800-816. https://doi.org/10.1002/jaba.257

Masia, C. L., & Chase, P. N. (1997). Vicarious learningrevisited: A contemporary behavior analytic interpre-tation. Journal of Behavior Therapy and ExperimentalPsychiatry, 28, 41-51. https://doi.org/10.1016/S0005-7916(96)00042-0

Pereira-Delgado, J. A., & Greer, R. D. (2009). The effectsof peer monitoring training on the emergence of thecapability to learn by observing instruction receivedby peers. The Psychological Record, 59, 407-434.

Rehfeldt, R. A., Latimore, D., & Stromer, R. (2003).Observational learning and the formation of classes ofreading skills by individuals with autism and otherdevelopmental disabilities. Research in DevelopmentalDisabilities, 24, 333-358. https://doi.org/10.1016/S0891-4222(03)00059-3

Smaby, K., MacDonald, R. P. F., Ahearn, W. H., &Dube, W. V. (2007). Assessment protocol for identi-fying preferred social consequences. Behavioral

817OBSERVATIONAL LEARNING

Interventions, 22, 311-318. https://doi.org/10.1002/bin.242

Spriggs, A. D., Gast, D. L., & Knight, V. F. (2016).Video modeling and observational learning to teachgaming access to students with ASD. Journal ofAutism and Developmental Disorders, 46, 2845-58.doi: https://doi.org/10.1007/s10803-016-2824-3.

Tarbox, J., Zuckerman, C. K., Bishop, M. R.,Olive, M. L., & O’Hora, D. P. (2011). Rule-Governed Behavior: Teaching a preliminary reper-toire of rule-following to children with autism. TheAnalysis of Verbal Behavior, 27(1), 125-139.

Taylor, B. A., & DeQuinzio, J. A. (2012). Observationallearning and children with autism. BehaviorModification, 36, 341-360. https://doi.org/10.1177/0145445512443981

Taylor, B. A., DeQuinzio, J. A., & Stine, J. (2012).Increasing observational learning of children withautism: A preliminary analysis. Journal of AppliedBehavior Analysis, 45, 815-820. https://doi.org/10.1901/jaba.2012.45-815

Varni, J. W., Lovaas, O. I., Koegel, R. L., &Everett, N. L. (1979). An analysis of observationallearning in autistic and normal children. Journal ofAbnormal Child Psychology, 7, 31-43. https://doi.org/10.1007/BF00924508

Williams, K. T. (2007). Expressive vocabulary test (2nded.). Bloomington, MN: Pearson.

Received March 15, 2017Final acceptance April 24, 2018Action Editor, Tiffany Kodak

JAIME A. DEQUINZIO et al.818