Multiple exemplar training involves teaching target words as

613

Socially important behavior can be changed deliberately. The preceding chapters describe basic principles of behavior and how practitioners can use behavior change tactics derived from those principles to increase appropriate behaviors, achieve desired stimulus controls, teach new behaviors, and decrease problem be- haviors. Although achieving initial behavior changes often requires procedures that are intrusive or costly, or for a variety of other reasons cannot or should not be continued indefinitely, it is almost always important that the newly wrought behavior changes continue. Similarly, in many instances the intervention needed to produce new patterns of responding cannot be implemented in all of the envi- ronments in which the new behavior would benefit the learner. Nor is it possible in certain skill areas to teach directly all of the specific forms of the target behav- ior the learner may need. Practitioners face no more challenging or important task than that of designing, implementing, and evaluating interventions that produce behavior changes that continue after the intervention is terminated, appear in rele- vant settings and stimulus situations other than those in which the intervention was conducted, and/or spread to other related behaviors that were not taught di- rectly. Chapter 28 defines the major types of generalized behavior change and de- scribes the strategies and tactics applied behavior analysts use to achieve them.

P A R T 1 2

Promoting Generalized Behavior Change

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

614

C H A P T E R 2 8

Generalization and Maintenance of Behavior Change

Key Terms

behavior trap contrived contingency contrived mediating stimulus general case analysis generalization generalization across subjects generalization probe

generalization setting indiscriminable contingency instructional setting lag reinforcement schedule multiple exemplar training naturally existing contingency

programming common stimuli response generalization response maintenance setting/situation generalization teaching sufficient examples teaching loosely

Behavior Analyst Certification Board® BCBA® & BCABA® Behavior Analyst Task List©,Third Edition

Content Area 3: Principles, Processes, and Concepts

3-12 Define and provide examples of generalization and discrimination.

9-28 Use behavior change procedures to promote stimulus and response generalization.

9-29 Use behavior change procedures to promote maintenance.

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 615

Sherry’s teacher implemented an intervention that helped Sherry to complete each part of multiple-part, in- school assignments before submitting them and begin- ning another activity. Now, three weeks after the program ended, most of the work Sherry submits as “finished” is incomplete and her stick-with-a-task-until- it’s-finished behavior is as poor as it was before the in- tervention began.

Ricardo has just begun his first competitive job working as a copy machine operator in a downtown business of- fice. In spite of his long history of distractibility and poor endurance, Ricardo had learned to work indepen- dently for several hours at a time in the copy room at the vocational training center. His employer, however, is complaining that Ricardo frequently stops working after a few minutes to seek attention from others. Ricardo may soon lose his job.

Brian is a 10-year-old boy diagnosed with autism. In an effort to meet an objective on his individualized educa- tion program that targets functional language and com- munication skills, Brian’s teacher taught him to say, “Hello, how are you?” as a greeting. Now, whenever Brian meets anyone, he invariably responds with, “Hello, how are you?” Brian’s parents are concerned that their son’s language seems stilted and parrot-like.

Each of these three situations illustrates a com- mon type of teaching failure insofar as the most socially significant behavior changes are those

that last over time, are used by the learner in all relevant settings and situations, and are accompanied by changes in other relevant responses. The student who learns to count money and make change in the classroom today must be able to count and make change at the conve- nience store tomorrow and at the supermarket next month. The beginning writer who has been taught to write a few good sentences in school must be able to write many more meaningful sentences when writing notes or letters to family or friends. To perform below this stan- dard is more than just regrettable; it is a clear indication that the initial instruction was not entirely successful.

In the first scenario, the mere passage of time re- sulted in Sherry losing her ability to complete assign- ments. A change of scenery threw Ricardo off his game; the excellent work habits he had acquired at the voca- tional training center disappeared completely when he arrived at the community job site. Although Brian used his new greeting skill, its restricted form was not serving him well in the real world. In a very real sense, the in- struction they received failed all three of these people.

Applied behavior analysts face no more challeng- ing or important task than that of designing, imple- menting, and evaluating interventions that produce generalized outcomes. This chapter defines the major

types of generalized behavior change and describes the strategies and tactics researchers and practitioners use most often to promote them.

Generalized Behavior Change: Definitions and Key Concepts When Baer, Wolf, and Risley (1968) described the emerg- ing field of applied behavior analysis, they included gen- erality of behavior change as one of the discipline’s seven defining characteristics.

A behavior change may be said to have generality if it proves durable over time, if it appears in a wide variety of possible environments, or if it spreads to a wide vari- ety of related behaviors. (p. 96)

In their seminal review paper, “An Implicit Technol- ogy of Generalization,” Stokes and Baer (1977) also stressed those three facets of generalized behavior change—across time, settings, and behaviors—when they defined generalization as

the occurrence of relevant behavior under different, non- training conditions (i.e., across subjects, settings, peo- ple, behaviors, and/or time) without the scheduling of the same events in those conditions. Thus, generaliza- tion may be claimed when no extratraining manipula- tions are needed for extratraining changes; or may be claimed when some extra manipulations are necessary, but their cost is clearly less than that of the direct inter- vention. Generalization will not be claimed when simi- lar events are necessary for similar effects across conditions. (p. 350)

Stokes and Baer’s pragmatic orientation toward gen- eralized behavior change has proven useful for applied behavior analysis. They stated simply that if a trained behavior occurs at other times or in other places without it having to be retrained completely at those times or in those places, or if functionally related behaviors occur that were not taught directly, then generalized behavior change has occurred. The following sections provide def- initions and examples of the three basic forms of gener- alized behavior change: response maintenance, setting/ situation generalization, and response generalization. Box 28.1, “Perspectives on the Sometimes Confusing and Misleading Terminology of Generalization,” dis- cusses the many and varied terms applied behavior ana- lysts use to describe these outcomes.

Response Maintenance

Response maintenance refers to the extent to which a learner continues to perform the target behavior after a portion or all of the intervention responsible for the

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

616 Part 12 Promoting Generalized Behavior Change

*Response maintenance can be measured under extinction conditions, in which case the relative frequency of continued responding is de- scribed correctly in terms of resistance to extinction. However, using resistance to extinction to describe response maintenance in most applied situations is incorrect because reinforcement typically follows some oc- currences of the target behavior in the post-treatment environment.

Applied behavior analysts have used many terms to de- scribe behavior changes that appear as adjuncts or by- products of direct intervention. Unfortunately, the overlapping and multiple meanings of some terms can lead to confusion and misunderstanding. For example, maintenance, the most frequently used term for behavior changes that persist after an intervention has been with- drawn or terminated, is also the most common name for a condition in which treatment has been discontinued or partially withdrawn. Applied behavior analysts should distinguish between response maintenance as a measure of behavior (i.e., a dependent variable) and maintenance as the name for an environmental condition (i.e., an in- dependent variable). Other terms found in the behavior analysis literature for continued responding after pro- grammed contingencies are no longer in effect include durability, behavioral persistence, and (incorrectly) re- sistance to extinction.*

Terms used in the applied behavior analysis literature for behavior changes that occur in nontraining settings or stimulus conditions include stimulus generalization, set- ting generalization, transfer of training, or simply, gen- eralization. It is technically incorrect to use stimulus generalization to refer to the generalized behavior change achieved by many applied interventions. Stimulus gen- eralization refers to the phenomenon in which a response that has been reinforced in the presence of a given stim- ulus occurs with an increased frequency in the presence of different but similar stimuli under extinction condi- tions (Guttman & Kalish, 1956; see Chapter 17). Stimulus generalization is a technical term referring to a specific behavioral process, and its use should be restricted to those instances (Cuvo, 2003; Johnston, 1979).

Terms such as collateral or side effects, response variability, induction, and concomitant behavior change are often used to indicate the occurrence of behaviors that have not been trained directly. To further complicate matters, generalization is often used as a catchall term to refer to all three types of generalized behavior change.

Johnston (1979) discussed some problems caused by using generalization (the term for a specific behavioral process) to describe any desirable behavior change in a generalization setting.

This kind of usage is misleading in that it suggests that a single phenomenon is at work when actually a number of different phenomena need to be described, explained, and controlled. . . . Carefully designing procedures to opti- mize the contributions of stimulus and response general- ization would hardly exhaust our repertoire of tactics for getting the subject to behave in a desirable way in non- instructional settings. Our successes will be more frequent when we realize that maximizing behavioral influence in such settings requires careful consideration of all behav- ioral principles and processes. (pp. 1–2)

Inconsistent use of the “terminology of generaliza- tion” can lead researchers and practitioners to incorrect assumptions and conclusions regarding the principles and processes responsible for the presence or absence of generalized outcomes. Nevertheless, applied behavior analysts will probably continue to use generalization as a dual-purpose term, referring sometimes to types of be- havior change and sometimes to behavioral processes that can bring such changes about. Stokes and Baer (1977) clearly indicated their awareness of the differ- ences in definitions.

The notion of generalization developed here is an essen- tially pragmatic one; it does not closely follow the tradi- tional conceptualizations (Keller & Schoenfeld, 1950; Skinner, 1953). In many ways, this discussion will sidestep much of the controversy concerning terminology. (p. 350)

While discussing the use of naturally existing con- tingencies of reinforcement to maintain and extend pro- grammed behavior changes, Baer (1999) explained his preference for using the term generalization:

It is the best of the techniques described here and, inter- estingly, it does not deserve the textbook definition of “generalization.” It is a reinforcement technique, and the textbook definition of generalization refers to unreinforced behavior changes resulting from other directly reinforced behavior changes. . . . [But] we are dealing with the prag- matic use of the word generalization, not the textbook meaning. We reinforce each other for using the word prag- matically, and it serves us well enough so far, so we shall probably maintain this imprecise usage. (p. 30, emphasis in original)

In an effort to promote the precise use of the tech- nical terminology of behavior analysis and as a reminder that the phenomena of interest are usually products of multiple behavior principles and procedures, we use terms for generalized behavior change that focus on the type of behavior change rather than the principles or processes that bring it about.

Box 28.1 Perspectives on the Sometimes Confusing

and Misleading Terminology of Generalization

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 617

behavior’s initial appearance in the learner’s repertoire has been terminated. For example:

• Sayaka was having difficulty identifying the low- est common denominator (LCD) when adding and subtracting fractions. Her teacher had Sayaka write the steps for finding the LCD on an index card and told her to refer to the card when needed. Sayaka began using the LCD cue card, and the ac- curacy of her math assignments improved. After using the cue card for a week, Sayaka said she no longer needed it and returned it to her teacher. The next day Sayaka correctly computed the LCD for every problem on a quiz on adding and subtracting fractions.

• On Loraine’s first day on the job with a residential landscaping company, a coworker taught her how to use a long-handled tool to extract dandelions, root and all. Without further instruction, Loraine continues to use the tool correctly a month later.

• When he was in the seventh grade, one of Derek’s teachers taught him how to write down his assign- ments and keep materials for each class in sepa- rate folders. As a college sophomore, Derek continues to apply those organizational skills to his academic work.

These examples illustrate the relative nature of gen- eralized behavior change. Response maintenance was ev- ident in Sayaka’s performance on a math quiz one day after the cue card intervention ended and also in Derek’s continued use of the organizational skills he had learned years earlier. How long a newly learned behavior needs to be maintained depends on the importance of that be- havior in the person’s life. If covertly reciting a telephone number three times after hearing it enables a person to remember the number long enough to dial it correctly when he locates a telephone a few minutes later, suffi- cient response maintenance has been achieved. Other be- haviors, such as self-care and social skills, must be maintained in a person’s repertoire for a lifetime.

Setting/Situation Generalization

Setting/situation generalization occurs when a target be- havior is emitted in the presence of stimulus conditions other than those in which it was trained directly. We de- fine setting/situation generalization as the extent to which a learner emits the target behavior in a setting or stimulus situation that is different from the instructional setting. For example:

• While waiting for his new motorized wheelchair to arrive from the factory, Chaz used a computer sim- ulation program and a joystick to learn how to op-

erate his soon-to-arrive chair. When the new chair arrived, Chaz grabbed the joystick and immedi- ately began zipping up and down the hall and spin- ning perfectly executed donuts.

• Loraine had been taught to pull weeds from flower- beds and mulched areas. Although she had never been instructed to do so, Loraine has begun remov- ing dandelions and other large weeds from lawns as she crosses on her way to the flowerbeds.

• After Brandy’s teacher taught her to read 10 differ- ent C-V-C-E words (e.g., bike, cute, made), Brandy could read C-V-C-E words for which she had not received any instruction (e.g., cake, bite, mute).

A study by van den Pol and colleagues (1981) pro- vides an excellent example of setting/situation general- ization. They taught three young adults with multiple disabilities to eat independently in fast-food restaurants. All three students had previously eaten in restaurants but could not order or pay for a meal without assistance. The researchers began by constructing a task analysis of the steps required to order, pay for, and eat a meal appropri- ately in a fast-food restaurant. Instruction took place in the students’ classroom and consisted of role-playing each of the steps during simulated customer–cashier interac- tions and responding to questions about photographic slides showing customers at a fast-food restaurant per- forming the various steps in the sequence. The 22 steps in the task analysis were divided into four major compo- nents: locating, ordering, paying, and eating and exiting. After a student had mastered the steps in each compo- nent in the classroom, he was given “a randomly deter- mined number of bills equaling two to five dollars and instructed to go eat lunch” at a local restaurant (p. 64). Observers stationed inside the restaurant recorded each student’s performance of each step in the task analysis. The results of these generalization probes, which were also conducted before training (baseline) and after train- ing (follow-up) are shown in Figure 28.1. In addition to assessing the degree of generalization from the class- room, which was based on the specific McDonald’s restaurant used for most of the probes, the researchers conducted follow-up probes in a Burger King restaurant (also a measure of maintenance).

This study is indicative of the pragmatic approach to assessing and promoting generalized behavior change used by most applied behavior analysts. The setting in which generalized responding is desired can contain one or more components of the intervention that was imple- mented in the instructional environment, but not all of the components. If the complete intervention program is required to produce behavior change in a novel environ- ment, then no setting/situation generalization can be claimed. However, if some component(s) of the training

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

618 Part 12 Promoting Generalized Behavior Change

100

50

0

1.0 2.0 3.0 4.0

Baseline Training Follow-Up

Covert Novel Restaurant

Overt Novel Restaurant

Covert 1-Year Follow-Up

100

50

0

1.0 2.0 3.0 4.0

100

50

0

2.0 3.0 4.0

5 10 15 20 25 30 35 Probe Sessions

P er

ce n

t C

or re

ct R

es p

on se

s

1.0

Student 1

Student 2

Student 3

Figure 28.1 Percentage of steps necessary to order a meal at a fast-food restaurant correctly performed by three students with disabilities before, during, and after instruction in the classroom. During follow-up, the closed triangles represent probes conducted at a Burger King restaurant using typical observation proce- dures, open triangles represent Burger King probes during which students did not know they were being observed, and open circles represent covert probes conducted in a different McDonald’s 1 year after training. From “Teaching the Handicapped to Eat in Public Places: Acquisition, Generalization and Maintenance of Restaurant Skills” by R. A. van den Pol, B. A. Iwata, M. T. Ivanic, T. J. Page, N. A. Neef, and F. P. Whitley, 1981, Journal of Applied Behavior Analysis, 14, p. 66. Copyright 1981 by the Society for the Experimental Analysis of Behavior, Inc. Reprinted by permission.

program results in meaningful behavior change in a gen- eralization setting, then setting/situation generalization can be claimed, provided it can be shown that the com- ponent(s) used in the generalization setting was insuffi- cient to produce the behavior change alone in the training environment.

For example, van den Pol and colleagues taught Stu- dent 3, who was deaf, how to use a prosthetic ordering device in the classroom. The device, a plastic laminated sheet of cardboard with a wax pencil, had preprinted ques- tions (e.g., “How much is . . . ?”), generic item names (e.g., large hamburger), and spaces where the cashier could write responses. Simply giving the student some money and the prosthetic ordering card would not have enabled him to order, purchase, and eat a meal indepen- dently. However, after classroom instruction that included guided practice, role playing, social reinforcement (“Good job! You remembered to ask for your change” [p. 64]),

1Because the majority of the examples in this chapter are school based, we have used the language of education. For our purposes here, instruction can be a synonym for treatment, intervention, or therapy, and instructional setting can be a synonym for clinical setting or therapy setting.

corrective feedback, and review sessions with the pros- thetic ordering card produced the desired behaviors in the instructional setting, Student 3 was able to order, pay for, and eat meals in a restaurant aided only by the card.

Distinguishing Between Instructional and Generalization Settings

We use instructional setting to denote the total envi- ronment where instruction occurs, including any aspects of the environment, planned or unplanned, that may in- fluence the learner’s acquisition and generalization of the target behavior.1 Planned elements are the stimuli and

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 619

Figure 28.2 Examples of an instructional setting and a generalization setting for six target behaviors.

Instructional Setting 1. Raising hand when special education

teacher asks a question in the resource room.

2. Practicing conversational skills with speech therapist at school.

3. Passing basketball during a team scrimmage on home court.

4. Answering addition problems in vertical format at desk at school.

5. Solving word problems with no distracter numbers on homework assignment.

6. Operating package sealer at community job site in presence of supervisor.

Generalization Setting 1. Raising hand when general education

teacher asks a question in the regular classroom.

2. Talking with peers in town. 3. Passing basketball during a game on the

opponent’s court. 4. Answering addition problems in

horizontal format at desk at school. 5. Solving word problems with distracter

numbers on homework assignment. 6. Operating package sealer at community

job site in absence of supervisor.

events the teacher has programmed in an effort to achieve initial behavior change and promote generalization. Planned elements of an instructional setting for a math lesson, for example, would include the specific math problems to be presented during the lesson and the for- mat and sequencing of those problems. Unplanned as- pects of the instructional setting are elements the teacher is not aware of or has not considered that might affect the acquisition and generalization of the target behavior. For example, the phrase, how much in a word problem may acquire stimulus control over a student’s use of ad- dition, even when the correct solution to the problem re- quires a different arithmetic operation. Or, perhaps a student always uses subtraction for the first problem on each page of word problems because a subtraction prob- lem has always been presented first during instruction.

A generalization setting is any place or stimulus situation that differs in some meaningful way from the in- structional setting and in which performance of the tar- get behavior is desired. There are multiple generalization settings for many important target behaviors. The student who learns to solve addition and subtraction word prob- lems in the classroom should be able to solve similar problems at home, at the store, and on the ball diamond with his friends.

Examples of instructional and generalization settings for six target behaviors are shown in Figure 28.2. When a person uses a skill in an environment physically re- moved from the setting where he learned it—as with Behaviors 1 through 3 in Figure 28.2—it is easy to un- derstand that event as an example of generalization across settings. However, many important generalized outcomes occur across more subtle differences between the in- structional setting and generalization setting. It is a mis- take to think that a generalization setting must be somewhere different from the place where instruction is

provided. Students often receive instruction in the same place where they will need to maintain and generalize what they have learned. In other words, the instructional setting and generalization setting can, and often do, share the same physical location (as with Behaviors 4 through 6 in Figure 28.2).

Distinguishing between Setting/Situation Generalization and Response Maintenance

Because any measure of setting/situation generalization is conducted after some instruction has taken place, it might be argued that setting/situation generalization and response maintenance are the same, or are inseparable phenomena at least. Most measures of setting/situation generalization do provide information on response main- tenance, and vice versa. For example, the post-training generalization probes conducted by van den Pol and col- leagues (1981) at the Burger King restaurant and at the second McDonald’s provided data on setting/situation generalization (i.e., to novel restaurants) and on response maintenance of up to 1 year. However, a functional dis- tinction exists between setting/situation generalization and response maintenance, with each outcome presenting a somewhat different set of challenges for programming and ensuring enduring behavior change. When a behavior change produced in the classroom or clinic is not observed in the generalization environment, a lack of setting/ situation generalization is evident. When a behavior change produced in the classroom or clinic has occurred at least once in the generalization setting and then ceases to occur, a lack of response maintenance is evident.

An experiment by Koegel and Rincover (1977) illus- trated the functional difference between setting/situation generalization and response maintenance. Participants

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

620 Part 12 Promoting Generalized Behavior Change

100

50

0

Child 1

Instructional Setting

Generalization Setting

100

50

0 100

50

0

Child 2

Child 3

50 100 150 200 250 300 350 400 450 Trials

P er

ce n

t C

or re

ct T

ri al

s

Figure 28.3 Correct responding by three children on alternating blocks of 10 trials in the instructional setting and 10 trials in the generalization setting. From “Research on the Differences Between Generalization and Maintenance in Extra-Therapy Responding” by R. L. Koegel and A. Rincover, 1977, Journal of Applied Behavior Analysis, 10, p. 4. Copyright 1977 by the Society for the Experimental Analysis of Behavior, Inc. Reprinted by permission.

were three young boys with autism; each was mute, echolalic, or displayed no appropriate contextual speech. One-to-one instructional sessions were conducted in a small room with the trainer and child seated across from each other at a table. Each child was taught a series of imitative responses (e.g., the trainer said, “Touch your [nose, ear]” or “Do this” and [raised his arm, clapped his hands]). Each 40-minute session consisted of blocks of 10 training trials in the instructional setting alternated with blocks of 10 trials conducted by an unfamiliar adult stand- ing outside, surrounded by trees. All correct responses in the instructional setting were followed by candy and so- cial praise. During the generalization trials the children received the same instructions and model prompts as in the classroom, but no reinforcement or other conse- quences were provided for correct responses in the gen- eralization setting.

Figure 28.3 shows the percentage of trials in which each child responded correctly in the instructional set- ting and in the generalization setting. All three children learned to respond to the imitative models in the instruc- tional setting. All three children showed 0% correct re- sponding in the generalization setting at the end of the experiment, but for different reasons. Child 1 and Child 3 began emitting correct responses in the generalization

setting as their performances improved in the instruc- tional setting, but their generalized responding was not maintained (most likely the result of the extinction con- ditions in effect in the generalization setting). The imita- tive responding acquired by Child 2 in the instructional setting never generalized to the outside setting. There- fore, the 0% correct responding at the experiment’s con- clusion represents a lack of response maintenance for Child 1 and Child 3, but for Child 2 it represents a fail- ure of setting generalization.

Response Generalization

We define response generalization as the extent to which a learner emits untrained responses that are functionally equivalent to the trained target behavior. In other words, in response generalization forms of behavior for which no programmed contingencies have been applied appear as a function of the contingencies that have been applied to other responses. For example:

• Traci wanted to earn some extra money by helping her older brother with his lawn mowing business. Her brother taught Traci to walk the mower up and down parallel rows that moved progressively from

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 621

one side of a lawn to the other. Traci discovered that she could mow some lawns just as quickly by first cutting around the perimeter of the lawn and then walking the mower in concentric patterns in- ward toward the center of the lawn.

• Loraine was taught to remove weeds with a long weed-removal tool. Although she has never been taught or asked to do so, sometimes Loraine removes weeds with a hand trowel or with her bare hands.

• Michael’s mother taught him how to take phone messages by using the pencil and notepaper next to the phone to write the caller’s name, phone num- ber, and message. One day, Michael’s mother came home and saw her son’s tape recorder next to the phone. She pushed the play button and heard Michael’s voice say, “Grandma called. She wants to know what you’d like her to cook for dinner Wednesday. Mr. Stone called. His number is 555- 1234, and he said the insurance payment is due.”

The study by Goetz and Baer (1973) described in Chapter 8 of the block building by three preschool girls provides a good example of response generalization. Dur- ing baseline the teacher sat by each girl as she played with the blocks, watching closely but quietly, and displaying neither enthusiasm nor criticism for any particular use of the blocks. During the next phase of the experiment, each time the child placed or rearranged the blocks to create a new form that had not appeared previously in that ses- sion’s constructions, the teacher commented with enthu- siasm and interest (e.g., “Oh, that’s very nice—that’s different!”). Another phase followed in which each re- peated construction of a given form within the session was praised (e.g., “How nice—another arch!”). The study ended with a phase in which descriptive praise was again contingent on the construction of different block forms. All three children constructed more new forms with the blocks when form diversity was reinforced than they did under baseline or under the reinforcement-for-the-same- forms condition (see Figure 8.7).

Even though specific responses produced reinforce- ment (i.e., the actual block forms that preceded each in- stance of teacher praise), other responses sharing that functional characteristic (i.e., being different from block forms constructed previously by the child) increased in frequency as a function of the teacher’s praise. As a re- sult, during reinforcement for different forms, the children constructed new forms with the blocks even though each new form itself had never before appeared and therefore could not have been reinforced previously. Reinforcing a few members of the response class of new forms in- creased the frequency of other members of the same re- sponse class.

Generalized Behavior Change: A Relative and Intermixed Concept

As the examples presented previously show, generalized behavior change is a relative concept. We might think of it as existing along a continuum. At one end of the con- tinuum are interventions that might produce a great deal of generalized behavior change; that is, after all compo- nents of an intervention have been terminated, the learner may emit the newly acquired target behavior, as well as several functionally related behaviors not observed pre- viously in his repertoire, at every appropriate opportu- nity in all relevant settings, and he may do so indefinitely. At the other end of the continuum of generalized out- comes are interventions that yield only a small amount of generalized behavior change—the learner uses the new skill only in a limited range of nontraining settings and situations, and only after some contrived response prompts or consequences are applied.

We have presented each of the three primary forms of generalized behavior change individually to isolate its defining features, but they often overlap and occur in combination. Although it is possible to obtain response maintenance without generalization across settings/ situations or behaviors (i.e., the target behavior continues to occur in the same setting in which it was trained after the training contingencies have been terminated), any meaningful measure of setting generalization will entail some degree of response maintenance. And it is common for all three forms of generalized behavior change to be represented in the same instance. For example, during a relatively quiet shift at the widget factory on Monday, Joyce’s supervisor taught her to obtain assistance by call- ing out, “Ms. Johnson, I need some help.” Later that week (response maintenance) when it was very noisy on the factory floor (setting/situation generalization), Joyce sig- naled her supervisor by waving her hand back and forth (response generalization).

Generalized Behavior Change Is Not Always Desirable

It is hard to imagine any behavior that is important enough to target for systematic instruction for which re- sponse maintenance would be undesirable. However, un- wanted setting/situation generalization and response generalization occur often, and practitioners should de- sign intervention plans to prevent or minimize such unwanted outcomes. Undesirable setting/situation gen- eralization takes two common forms: overgeneralization and faulty stimulus control.

Overgeneralization, a nontechnical but effectively descriptive term, refers to an outcome in which the

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

622 Part 12 Promoting Generalized Behavior Change

behavior has come under the control of a stimulus class that is too broad. That is, the learner emits the target be- havior in the presence of stimuli that, although similar in some way to the instructional examples or situation, are inappropriate occasions for the behavior. For example, a student learns to spell division, mission, and fusion with the grapheme, –sion. When asked to spell fraction, the student writes f-r-a-c-s-i-o-n.

With faulty stimulus control, the target behavior comes under the restricted control of an irrelevant an- tecedent stimulus. For example, after learning to solve word problems such as, “Natalie has 3 books. Amy has 5 books. How many books do they have in total?” by adding the numerals in the problem, the student adds the numerals in any problem that includes the words “in total” (e.g., “Corinne has 3 candies. Amanda and Corinne have 8 candies in total. How many candies does Amanda have?”).2

Undesired response generalization occurs when any of a learner’s untrained but functionally equivalent re- sponses results in poor performance or undesirable out- comes. For example, although Jack’s supervisor at the widget factory taught him to operate the drill press with two hands because that is the safest method, sometimes Jack operates the press with one hand. One-handed re- sponses are functionally equivalent to two-handed re- sponses because both topographies cause the drill press to stamp out a widget, but one-handed responses com- promise Jack’s health and the factory’s safety record. Or, perhaps some of her brother’s customers do not like how their lawns look after Traci has mowed them in concen- tric rectangles.

Other Types of Generalized Outcomes

Other types of generalized outcomes that do not fit eas- ily into categories of response maintenance, setting/situ- ation generalization, and response generalization have been reported in the behavior analysis literature. For ex- ample, complex members of a person’s repertoire some- times appear quickly with little or no apparent direct conditioning, such as the stimulus equivalence relations described in Chapter 17 (Sidman, 1994). Another type of such rapid learning that appears to be a generalized outcome of other events has been called contingency ad- duction, a process whereby a behavior that was initially selected and shaped under one set of conditions is re- cruited by a different set of contingencies and takes on a

2Examples of faulty stimulus control caused by flaws in the design of in- structional materials, and suggestions for detecting and correcting those flaws, can be found in J. S. Vargas (1984).

new function in a person’s repertoire (Adronis, 1983; Johnson & Layng, 1992).

Sometimes an intervention applied to one or more people results in behavior changes in other people who were not directly treated by the contingencies. Genera- lization across subjects refers to changes in the behav- ior of people not directly treated by an intervention as a function of treatment contingencies applied to other peo- ple. This phenomenon, which has been described with a variety of related or synonymous terms—vicarious rein- forcement (Bandura, 1971; Kazdin, 1973), ripple effect (Kounin, 1970), and spillover effect (Strain, Shores, & Kerr, 1976)—provides another dimension for assessing the generalization of treatment effects. For example, Fan- tuzzo and Clement (1981) examined the degree to which behavior changes would generalize from one child who received teacher-administered or self-administered token reinforcement during a math activity to a peer seated next to the child.

Drabman, Hammer, and Rosenbaum (1979) combined four basic types of generalized treatment effects— (a) across time (i.e., response maintenance), (b) across set- tings (i.e., setting/situation generalization), (c) across be- haviors (i.e., response generalization), and (d) across subjects—into a conceptual framework they called the generalization map. By viewing each type of generalized outcome as dichotomous (i.e., either present or absent) and by combining all possible permutations of the four cate- gories, Drabman and colleagues arrived at 16 categories of generalized behavior change ranging from maintenance (Class 1) to subject-behavior-setting-time generalization (Class 16). Class 1 generalization is evident if the target be- havior of the target subject(s) continues in the treatment setting after any “experiment-controlled contingencies” have been discontinued. Class 16 generalization, which Drabman and colleagues (1979) called the “ultimate form” of generalization, is evidenced by “a change in a nontar- get subject’s nontarget behavior which endures in a dif- ferent setting after the contingencies have been withdrawn in the treatment setting” (p. 213).

Although Drabman and colleagues recognized that “with any heuristic technique the classifications may prove arbitrary” (p. 204), they provided objectively stated rules for determining whether a given behavioral event fits the requirements of each of their 16 classifications. Regardless of whether generalized behavior change con- sists of such distinctly separate and wide-ranging phe- nomena as detailed by Drabman and colleagues, their generalization map provided an objective framework by which the extended effects of behavioral interventions can be described and communicated. For example, Stevenson and Fantuzzo (1984) measured 15 of the 16 generalization map categories in a study of the effects of teaching a fifth-grade boy to use self-management

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 623

techniques. They not only measured the effects of the in- tervention on the target behavior (math performance) in the instructional setting (school), but they also assessed effects on the student’s math behavior at home, disruptive behavior at home and at school, both behaviors for a non- treated peer in both settings, and maintenance of all of the above.

Planning for Generalized Behavior Change

In general, generalization should be programmed, rather than expected or lamented.

—Baer, Wolf, and Risley (1968, p. 97)

In their review of 270 published studies relevant to generalized behavior change, Stokes and Baer (1977) concluded that practitioners should always “assume that generalization does not occur except through some form of programming . . . and act as if there were no such an- imal as ‘free’ generalization—as if generalization never occurs ‘naturally,’ but always requires programming” (p. 365). Of course, generalization of some type and de- gree does usually occur, whether or not it is planned. Such unplanned and unprogrammed generalization may be suf- ficient, but often it is not, particularly for many learners served by applied behavior analysts (e.g., children and adults with learning problems and developmental dis- abilities). And if left unchecked, unplanned generalized outcomes may be undesirable outcomes.

Achieving optimal generalized outcomes requires thoughtful, systematic planning. This planning begins with two major steps: (1) selecting target behaviors that will meet natural contingencies of reinforcement, and (2) specifying all desired variations of the target behav- ior and the settings/situations in which it should (and should not) occur after instruction has ended.

Selecting Target Behaviors That Will Meet Naturally Existing Contingencies of Reinforcement

The everyday environment is full of steady, dependable, hardworking sources of reinforcement for almost all of the behaviors that seem natural to us. That is why they seem natural to us.

—Donald M. Baer (1999, p. 15)

Numerous criteria have been suggested for determining whether a proposed teaching objective is relevant or functional for the learner. For example, the age- appropriateness of a skill and the degree to which it rep- resents normalization are often cited as important criteria for choosing target behaviors for students with

disabilities (e.g., Snell & Brown, 2006). Each of these criteria was discussed in Chapter 3, along with numer- ous other issues that should be considered when select- ing and prioritizing target behaviors. In the end, however, there is just one ultimate criterion of func- tionality: A behavior is functional only to the extent that it produces reinforcement for the learner. This criterion holds no matter how important the behavior may be to the person’s health or welfare, or no matter how much teachers, family, friends, or the learner himself consid- ers the behavior to be desirable. To repeat: A behavior is not functional if it does not produce reinforcement for the learner. Said another way: Behaviors that are not followed by reinforcers on at least some occasions will not be maintained.

Ayllon and Azrin (1968) recognized this fundamen- tal truth when they recommended that practitioners fol- low the relevance-of-behavior rule when selecting target behaviors. The rule: Choose only those behaviors to change that will produce reinforcers in the postinterven- tion environment. Baer (1999) believed so strongly in the importance of this criterion that he recommended that practitioners heed a similar rule:

A good rule is to not make any deliberate behavior changes that will not meet natural communities of rein- forcement. Breaking this rule commits you to maintain and extend the behavior changes that you want, by your- self, indefinitely. If you break this rule, do so knowingly. Be sure that you are willing and able to do what will be necessary. (p. 16, emphasis in original)

Programming for the generalization and maintenance of any behavior for which a natural contingency of rein- forcement exists, no matter the specific tactics employed, consists of getting the learner to emit the behavior in the generalization setting just often enough to contact the oc- curring contingencies of reinforcement. Generalization and maintenance of the behavior from that point forward, while not assured, is a very good bet. For example, after receiving some basic instruction on how to operate the steering wheel, gas pedal, and brakes on a car, the natu- rally existing reinforcement and punishment contingen- cies involving moving automobiles and the road will select and maintain effective steering, acceleration, and braking. Very few drivers need booster training sessions on the basic operation of the steering wheel, gas pedal, and brakes.

We define a naturally existing contingency as any contingency of reinforcement (or punishment) that oper- ates independent of the behavior analyst’s or practi- tioner’s efforts. This is a pragmatic, functional conception of a naturally existing contingency defined by the ab- sence of the behavior analyst’s efforts. Naturally exist- ing contingencies include contingencies that operate

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

624 Part 12 Promoting Generalized Behavior Change

without social mediation (e.g., walking fast on an icy sidewalk is often punished by a slip and fall) and socially mediated contingencies contrived and implemented by other people in the generalization setting. From the per- spective of a special educator who is teaching a set of targeted social and academic skills to students for whom the general education classroom represents the general- ization setting, a token economy operated by the general education classroom teacher is an example of the latter type of naturally existing contingency.3 Even though the token economy was contrived by the teacher in the gen- eral education classroom, it is a naturally existing con- tingency because it already operates in the generalization setting.

We define a contrived contingency as any contin- gency of reinforcement (or punishment) designed and implemented by a behavior analyst or practitioner to achieve the acquisition, maintenance, and/or generaliza- tion of a targeted behavior change. From the perspective of the teacher in the general education classroom who designed and implemented it, the token economy in the previous example is a contrived contingency.

In reality, practitioners are often charged with the difficult task of teaching important skills for which there are no dependable naturally existing contingencies of re- inforcement. In such cases, practitioners should realize and plan for the fact that the generalization and mainte- nance of target behaviors will have to be supported, per- haps indefinitely, with contrived contingencies.

Specifying All Desired Variations of the Behavior and the Settings/ Situations Where It Should (and Should Not) Occur

This stage of planning for generalized outcomes includes identifying all the desired behavior changes that need to be made and all the environments and stimulus condi- tions in which the learner should emit the target behav- ior(s) after direct training has ceased (Baer, 1999). For some target behaviors, the most important stimulus con- trol for each response variation is clearly defined (e.g., reading C-V-C-E words) and restricted in number (e.g., solving multiplication facts). For many important target behaviors, however, the learner is likely to encounter a multitude of settings and stimulus conditions where the behavior, in a wide variety of response forms, is desired. Only by considering these possibilities prior to instruction can the behavior analyst design an intervention with the best chance of preparing the learner for them.

In one sense, this component of planning for gener- alized outcomes is similar to preparing a student for a fu-

3Token economies are described in Chapter 26.

ture test without knowing the content or the format of all of the questions that will be on the test. The stimulus con- ditions and contingencies of reinforcement that exist in the generalization setting(s) will provide that test to the learner. Planning involves trying to determine what the final exam will cover (type and form of questions), whether there will be any trick questions (e.g., confus- ing stimuli that might evoke the target response when it should not occur), and whether the learner will need to use his new knowledge or skill in different ways (re- sponse generalization).

List All the Behaviors That Need to Be Changed

A list should be made of all the forms of the target be- havior that need to be changed. This is not an easy task, but a necessary one to obtain a complete picture of the teaching task ahead. For example, if the target behavior is teaching Brian, the young boy with autism, to greet people, he should learn a variety of greetings in addition to “Hello, how are you?” Brian may also need many other behaviors to initiate and participate in conversa- tions, such as responding to questions, taking turns, stay- ing on topic, and so forth. He may also need to be taught when and with whom to introduce himself. Only by hav- ing a complete list of all the desired forms of the behav- ior can the practitioner make meaningful decisions about which behaviors to teach directly and which to leave to generalization.

The practitioner should determine whether and to what extent response generalization is desirable for all of the behavior changes listed, and then, make a priori- tized list of the variations of the target behavior he would like to see as generalized outcomes.

List All the Settings and Situations in Which the Target Behavior Should Occur

A list should be made of all the desired settings and sit- uations in which the learner will emit the target behavior if optimal generalization is achieved. Will Brian need to introduce himself and talk with children his own age, to adults, to males and females? Will he need to talk with others at home, at school, in the lunchroom, on the play- ground? Will he be confronted with situations that may appear to be appropriate opportunities to converse but are not (e.g., an unknown adult approaches and offers candy) and for which an alternative response is needed (e.g., walking away, seeking out a known adult). (This kind of analysis often adds additional behaviors to the list of skills to be taught.)

When all of the possible situations and settings have been identified, they should be prioritized according to

IS B

N 1-256-93044-X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

Chapter 28 Generalization and Maintenance of Behavior Change 625

their importance and the client’s likelihood of encoun- tering them. Further analysis of the prioritized environ- ments should then be conducted. What discriminative stimuli usually set the occasion for the target behavior in these various settings and situations? What schedules of reinforcement for the target behavior are typical in these nontraining environments? What kinds of reinforcers are likely to be contingent on the emission of the target be- havior in each of the settings? Only when she has an- swered all of these questions, if not by objective observation then at least by considered estimation, can the behavior analyst begin to have a full picture of the teaching task ahead.

Is the Pre-Intervention Planning Worth It?

Obtaining all of the information just described requires considerable time and effort. Given limited resources, why not design an intervention and immediately begin trying to change the target behavior? It is true that many behavior changes do show generalization, even though the extension of the trained behavior across time, set- tings, and other behaviors was unplanned and unpro- grammed. When target behaviors have been chosen that are truly functional for the subject and when those be- haviors have been brought to a high level of proficiency under discriminative stimuli relevant to generalization settings, the chances of generalization are good. But what constitutes a high level of proficiency for certain behav- iors in various settings? What are all of the relevant dis- criminative stimuli in all of the relevant settings? What are all the relevant settings?

Without a systematic plan, a practitioner will usu- ally be ignorant of the answers to these vital questions. Few behaviors that are important enough to target have such limited needs for generalized outcomes that the an- swers to such questions are obvious. Just a cursory con- sideration of the behaviors, settings, and people related to a child introducing himself revealed numerous factors that may need to be incorporated into an instructional plan. A more thorough analysis would produce many more. In fact, a complete analysis will invariably reveal more behaviors to be taught, to one person or another, than time or resources would ever allow. And Brian—the 10-year-old who is learning to greet people and introduce himself—in all likelihood needs to learn many other skills also, such as self-help, academic, and recreation and leisure skills, to name just a few. Why then create all the lists in the first place when everything cannot be taught anyway? Why not just train and hope?4

4Teaching a new behavior without developing and implementing a plan to facilitate its maintenance and generalization is done so often that Stokes and Baer 1977) called it the “train and hope” approach to generalization.

Baer (1999) described six possible benefits of list- ing all the forms of behavior change and all the situations in which these behavior changes should occur.

1. You now see the full scope of the problem ahead of you, and thus see the corresponding scope that your teaching program needs to have.

2. If you teach less than the full scope of the problem, you do so by choice rather than by forgetting that some forms of the behavior could be important, or that there were some other situations in which the behavior change should or should not occur.

3. If less than a complete teaching program results in less than a complete set of behavior changes, you will not be surprised.

4. You can decide to teach less than there is to learn, perhaps because that is all that is practical or pos- sible for you to do.

5. You can decide what is most important to teach. You can also decide to teach the behavior in a way that encourages the indirect development of some of the other forms of the desired behavior, as well as the indirect occurrence of the behavior in some other desired situations, that you will not or cannot teach directly.

6. But if you choose the option discussed in number 5 above, rather than the complete program implicit in number 1, you will do so knowing that the de- sired outcome would have been more certain had you taught every desirable behavior change di- rectly. The best that you can do is to encourage the behavior changes that you do not cause directly. So, you will have chosen the option in number 5 either of necessity or else as a well-considered gamble after a thoughtful consideration of possibil- ities, costs, and benefits. (pp. 10–11)

After determining which behaviors to teach directly and in which situations and settings to teach those be- haviors, the behavior analyst is ready to consider strate- gies and tactics for achieving generalization to untrained behaviors and settings.

Strategies and Tactics for Promoting Generalized Behavior Change Various authors have described conceptual schemes and taxonomies of methods for promoting generalized be- havior change (e.g., Egel, 1982; Horner, Dunlap, & Koegel, 1988; Osnes & Lieblein, 2003; Stokes & Baer,

IS B

N 1

-2 56

-9 30

44 -X

Applied Behavior Analysis, Second Edition, by John O. Cooper, Timothy E. Heron, and William L. Heward. Published by Merrill Prentice Hall. Copyright © 2007 by Pearson Education, Inc.

626 Part 12 Promoting Generalized Behavior Change

Figure 28.4 Strategies and tactics for promoting generalized behavior change.

Teach the Full-Range of Relevant Stimulus Conditions and Response Requirements 1. Teach sufficient stimulus examples 2. Teach sufficient response examples

Make the Instructional Setting Similar to the Generalization Setting 3. Program common stimuli 4. Teach loosely

Maximize Contact with Reinforcement in the Generalization Setting 5. Teach the target behavior to levels of performance required by naturally existing

contingencies of reinforcement 6. Program indiscriminable contingencies 7. Set behavior traps 8. Ask people in the generalization setting to reinforce the target behavior 9. Teach the learner to recruit reinforcement

Mediate Generalization 10. Contrive a mediating stimulus 11. Teach self-management skills

Train to Generalize 12. Reinforce response variability 13. Instruct the learner to generalize

1977; Stokes & Osnes, 1989). The conceptual scheme presented here is informed by the work of those authors and others, and by our own experiences in designing, implementing, and evaluating procedures for promot- ing generalized outcomes and in teaching practitioners to use them. Although numerous methods and tech- niques have been demonstrated and given a variety of names, most tactics that effectively promote general- ized behavior change can be classified under five strate- gic approaches:

• Teach the full range of relevant stimulus conditions and response requirements.

• Make the instructional setting similar to the gener- alization setting.

• Maximize the target behavior’s contact with rein- forcement in the generalization setting.

• Mediate generalization.

• Train to generalize.

In the following sections we describe and provide examples of 13 tactics applied behavior analysts have used to accomplish these five strategies (see Figure 28.4). Although each tactic is described individually, most efforts to promote generalized behavior change entail a combination of these tactics (e.g., Ducharme & Holborn, 1997; Grossi, Kimball, & Heward, 1994; Hughes, Harmer, Killina, & Niarhos, 1995; Ninness, Fuerst, & Rutherford, 1991; Trask-Tyler, Grossi, & Heward, 1994).

Teach the Full Range of Relevant Stimulus Conditions and Response Requirements

The most common mistake that teachers make, when they want to establish a generalized behavior change, is to teach one good example of it and expect the student to generalize from that example.

— Donald M. Baer (1999, p. 15)

To be most useful, most important behaviors must be per- formed in various ways across a wide range of stimulus conditions. Consider a person skilled in reading, math, conversing with others, and cooking. That person can read thousands of different words; add, subtract, multiply, and divide any combination of numbers; make a multi- tude of relevant and appropriate comments when talking with others; and measure, combine, and prepare numer- ous ingredients in hundreds of recipes. Helping learners achieve such wide-ranging performances presents an enormous challenge to the practitioner.

One approach to this challenge would be to teach every desired form of a target behavior in every setting/ situation in which the learner may need that behavior in the future. Although this approach would eliminate the need to program for response generalization and setting/ situation generalization (response maintenance would re- main the only problem), it is seldom possible and never practical. A teacher cannot provide direct instruction on every printed word a student may encounter in the fu- ture, or teach a student every measuring, pouring, stir- ring, and sautéing movement needed to make every dish