Week 4 Fisher Chapter 15 Article PDF

# CHAPTER 15 ## Building Complex Repertoires from Discrete Behaviors ### Establishing Stimulus Control, Behavioral Chains, and Strategic Behavior George H. Noell, Nathan A. Call, Scott P. Ardoin, and Sarah J. Miller Complexity is a cardinal feature of human behavior. Humans routinely exhibit behav...

# CHAPTER 15 ## Building Complex Repertoires from Discrete Behaviors ### Establishing Stimulus Control, Behavioral Chains, and Strategic Behavior George H. Noell, Nathan A. Call, Scott P. Ardoin, and Sarah J. Miller Complexity is a cardinal feature of human behavior. Humans routinely exhibit behavior that is varied; is subtly discriminated; requires long-term planning; is maintained by delayed, ambiguous contingencies; is initiated by verbal rules; and is recursive. Human behavior's complexity, subtle capacity of adaptation, and incorporation of rich language largely accounts for our evolutionary dominance over the earth (Ehrlich & Ehrlich, 2008). This wondrous capacity has also created enormous challenges for individual human beings, as societies have developed increasingly complex behavioral requirements for successful adaptation. As the complexity of behavioral expectations increases, the demand for efficient teaching increases. Cognitive-constructivist approaches to teaching have achieved dominance in many areas, because they explicitly emphasize the complex, varied nature of the target material and its accompanying behavior (e.g., Haywood, 2004; Martens & Daly, 1999; Richardson, 2003). This approach to teaching makes sense to many consumers and is sufficient for many learners. By contrast, behavior-analytic teaching procedures have commonly emphasized an elemental approach, in which the behavior analyst reduces complex behavior to small teachable units. For some consumers, this approach initially appears reductive and far removed from the goal of teaching subtle, complex, and adaptive behavior. When behavior analysts have sought to establish more complex behavior, they have typically achieved it by combining simpler behaviors to form a more complex chain, or by elaborating on a simpler behavior to produce a more complex behavior (e.g., Sauttfr, LeBlanc, Jay, Goldsmith, & Carr, 2011). Due to different conceptual bases and emphasis on different outcomes (e.g., understanding vs. behaving), cognitivist and behavior-analytic approaches are commonly viewed as conflicting or competing approaches, but the two approaches have much in common. For example, they both identify some of the same procedures as effective, such as practice with feedback, but provide different explanations for how and why the procedures are effective (Carroll, Kodak, & Adolf, 2016, vs. Trapman, van Gelderen, van Steensel, van Schoo-ten, & Hulstijn, 2014). Both approaches seek to help individuals develop complex behavioral repertoires that include flexible and generalized responding to diverse stimuli. <start_of_image> Areas of complexity can be divided into five main areas: - subordinate and superordinate skills - sequencing - promoting variability in responding - ambiguity in natural criteria - establishing self-management skills A strength of the systematic, elemental building approach that behavior-analytic teaching adopts is that simplified behaviors can be taught to individuals who have difficulty acquiring new or complex behavior (Luiselli & Hurley, 2005). By contrast, some have criticized behavior-analytic approaches for failing to capture the symbolic meaning or underlying structure of complex behaviors and for fostering dependence on instructors (Hickey, Moore, & Pellegrino, 2001; Kroesbergen, Van Luit, & Maas, 2004). This criticism of behavior-analytic teaching appears quite reasonable when we examine studies for which the goal was to establish a specific response (e.g., Swain, Lane, & Gast, 2015). It appears less tenable in the context of broader and more systematic behavior-analytic approaches to teaching, such as direct instruction, that clearly emphasize meaning, structure, and behaviors that represent understanding (Liem & Martin, 2013). ## Complexity: Flexible, Diverse Behavior Behavioral complexity is difficult to define, because it exists as a relative comparison of behaviors within and across individuals. Driving to the corner market to buy a gallon of milk can be insurmountably complex for some individuals or quite simple for others. Behaviors also vary in complexity. For example, talking to a friend who is present is less complex than e-mailing the friend. Behaviors that were complex will become simple as an individual develops greater skill in a domain. Decoding a single word can be a complex process that includes many discriminations (Snow, Burns, & Griffin, 1998). Later in the process of becoming literate, reading that same word can become a simple behavior in which the reader perceives the word as a single stimulus. A common goal of teaching is to help learners master simple behaviors that will become elements of more complex behaviors. Complexity emerges in all parts of the antecedent-behavior-consequence (A-B-C) chain. For example, a behavior is more complex when its antecedent is ambiguous (Harding, Wacker, Cooper, Millard, & Jensen-Kovalan, 1994); when a delay occurs between the onset of the discriminative stimulus and the target behavior; or when the same stimulus is discriminative for different behavioral repertoires, depending on the context (Hughes & Barnes-Holmes, 2014). Complex stimuli whose functional properties change across contexts are common aspects of social interactions, academic activities, and vocational activities. Behaviors themselves are the most intuitive source of complexity. No generally accepted definition of behavioral complexity exists in the behavior-analytic literature. We suggest that the following five dimensions are important to consider in establishing new behavior: - subordinate and superordinate skills - sequencing - promoting variability in responding - ambiguity in natural criteria - establishing self-management skills Behaviors themselves are the most intuitive source of complexity. No generally accepted definition of behavioral complexity exists in the behavior-analytic literature. We suggest that the following five dimensions are important to consider in establishing new behavior: (1) subordinate and superordinate skills, (2) sequencing, (3) promoting variability in responding, (4) ambiguity in natural criteria, and (5) establishing self-management skills. Most complex behaviors include or require several subordinate or prerequisite skills. For example, fluent decoding skills are precursors to text search and reading comprehension skills (National Reading Panel, 2000). Another considerable challenge is identifying the required proficiency level for prerequisite behaviors before teaching the target skill. That a student must be able to complete addition and subtraction operations correctly before learning to balance a checkbook may be obvious. However, defining which operations and what accuracy and fluency levels to require before teaching the target skill may pose a considerable challenge (see Kelley, Reitman, & Noell, 2002, for a discussion of accuracy and fluency criteria in subordinate skills). A second source of complexity arises from behavioral chains, a series of behaviors that occur sequentially and produce a consistent end state (Cooper, Heron, & Heward, 2007). Sequences of a different order of behaviors or variations within the chain may produce the same end state. For example, many successful variations of the hand-washing chain are possible, dependent on personal preferences and environmental context. Each behavior in an established chain produces the conditioned reinforcer that serves as the discriminative stimulus for the next behavior in the chain (Cooper et al., 2007). The requirement that competence includes varied behavior, which is necessary for response generalization, creates a third source of complexity. For example, initiating play with peers requires variety across occasions and available materials (Ledbetter-Cho et al., 2015). Variable behavior that maintains contact with reinforcement is a common feature of social, vocational, and academic behaviors. For example, the person who tells the same joke over and over is not likely to receive continued reinforcement for joke telling, even if the joke initially occasioned laughter. Response generalization and behavioral flexibility are very challenging for some learners. Individuals who require many trials with carefully controlled antecedent and consequent stimuli to learn also have difficulty learning to respond to variability in natural contexts, and this is one of the central features of some developmental disabilities (Reitzel et al., 2013). A fourth source of complexity is evident for behaviors that are so varied that correct and incorrect responses are difficult to define. These behaviors are so common in human interaction that we might describe them as normative. For example, a response to a simple greeting may have a nearly infinite number of acceptable responses. A fifth source of complexity arises when behavior requires substantial planning, progress monitoring, and plan revision, also known as executive control (Mahy, Moses, & Kliegel, 2014) or self-management (Gureasko-Moore, DuPaul, & White, 2006). Executive control or self-management in this context refers to organizing and evaluating responses necessary to complete complex behaviors. Although the term executive control typically describes internal unobservable processes, planning, progress monitoring, and plan revision can be observable behaviors. The consequences of behavior can create an additional source of complexity in establishing and maintaining behaviors. Delayed consequences, thin reinforcement schedules, and small-magnitude, cumulative consequences are extraordinarily common in human endeavor (Malott, 1989) and are frequently problematic in establishing and sustaining behavior. Natural consequences can be sufficiently delayed, on such thin schedules, and so ambiguous in their presentation that they are insufficient to teach behavior or maintain existing behavior (Malott, 1989). Behavior-analytic teaching often focuses on specific responses, using tightly controlled procedures that can produce inflexible, tightly controlled responding. We should conceptualize this outcome, however, as a beginning rather than an end. This approach arose in part from demonstrations that individuals regarded as "unteachable" or "disabled" could learn far more and far faster than anyone thought was possible. The striking success in using principles of applied behavior analysis (ABA) led to ABA's successful application to typically developing individuals (Daly, Persampieri, McCurdy, & Gortmaker, 2005; Koscinski & Hoy, 1993). Moving from discrete teachable behaviors to elaborate flexible repertoires is a fundamental goal of teaching represented in the ABA literature (e.g., Reid, Lienemann, & Hagaman, 2013). In the balance of this chapter, we focus on critical issues for establishing new behaviors and the elaboration of those behaviors to produce more flexible, adaptive repertoires. We first discuss selected issues in the assessment of behaviors and individuals before teaching begins. We then discuss many specific procedures behavior analysts might use to establish new behavior. We present the procedures as they might arise in practice as an analyst moves from establishing an initial response, to elaborating on that response, to creating a more complex repertoire. Thus these sections progress from shaping and prompting, to chaining, to strategy instruction, and finally to generalization. ## Assessing Behaviors an Individuals Any program to establish new behavior should begin with an assessment of the individual's current skills, behaviors, goals, and preferences. We can conceptualize establishing new behavior as the answers to three questions. First, what do we expect the learner to do? Second, what does the learner know how to do? Third, what procedures can we use to build on what the learner does now, so that he or she can meet the new expectations? Although these questions are intuitive, complexity arises in the details. For example, most individuals will need to learn many new behaviors that likely overlap in function and topography. We often must prioritize the behaviors that we will teach. Space limitations preclude an extensive consideration of prioritizing strategies in this chapter. Generally, we should target those behaviors that have the broadest possible adaptive importance and those that are prerequisites of these broad and important behaviors. Researchers have used the term behavioral cusps to describe behaviors that make broad contributions to an individual's adaptive success (Rosales-Ruiz & Baer, 1997). Obvious examples include spoken language in social contexts, and reading in educational and vocational contexts. Instructional planning should begin by identifying criteria for competence that indicate when we should terminate instruction or shift to new targets. The end point might be age-appropriate oral-language skills or reading text and correctly answering comprehension questions with an intermediate stage to prepare the student to learn additional material. For example, teaching a child with autism spectrum disorder (ASD) to articulate targeted words in response to appropriate antecedent stimuli is less likely to be an end than a stage in the process of building oral-language skills. ## Shaping Shaping is an instructional approach that is particularly important for learners who have a low probability of exhibiting the target behavior even with prompting, but engage in some related behavior that we can use to begin instruction. Shaping involves increasing the probability of a behavior's occurrence through the gradual transformation of some property of responding. Differential reinforcement of successive approximations of a targeted operant class produces this transformation. Shaping across topographies modifies the topography of a response, and researchers have demonstrated it in several classic studies (Horner, 1971; Isaacs et al., 1960; Skinner, 1938). For example, Isaacs et al. (1960) shaped the eye movements of an individual diagnosed with comatose schizophrenia into lip movements, then speech sounds, and eventually recognizable words. Shaping within topography modifies the rate, magnitude, or some other property of the target operant. Researchers have used this type of shaping to increase the arm extension of an athlete during a critical step in pole vaulting (Rea & Williams, 2002) and to increase the duration individuals held their breath before measuring exhaled carbon monoxide levels during treatment for smoking cessation (Scott, Scott, & Goldwater, 1997). Learners typically emit a distribution of behaviors relevant to the targeted response dimension. Shaping uses extinction and reinforcement to shift this distribution, so that the proportion of responses that contain the desired response property increases (Galbicka, 1994). Continued differential reinforcement of responses above a criterion value produces differentiation in which an increasing proportion of behavior is at or nearer to the target behavior. Factors affecting the probability of successful shaping include properties of the initial response and the way we establish the criterion for reinforcement. The response targeted early in the process must occur at a sufficient level or rate to permit initial reinforcement. It must also approximate the target behavior, so that we can differentially reinforce the relevant response dimension. Finally, the initial response must have enough variability that we can provide differential reinforcement for responses that exceed an established criterion, thus shaping the response toward the target. Note that increased variability is a predictable side effect of both reinforcement (Skinner, 1938) and extinction (Lerman & Iwata, 1996), so initial responses that seem relatively invariant may be amenable to shaping. Determining which responses to reinforce and which responses to extinguish can be challenging. If the criterion for reinforcement is too low, we will reinforce a large proportion of responses and shaping will proceed slowly. By contrast, if the criterion is too high, we will reinforce a small proportion of responses, and responding may be extinguished. Galbicka (1994) recommended that instructors use percentile schedules to empirically determine the criterion for reinforcement during shaping. All responses that exceed a preestablished rank-ordered response from a sample of the previous responses produce reinforcement (e.g., the third highest from the last 10). Researchers have successfully used this approach to shape behaviors such as eye contact (Hall, Maynes, & Reiss, 2009) and academic-task engagement (Athens, Vollmer, & St. Peter Pipkin, 2007; Clark, Schmidt, Mezhoudi, & Kahng, 2016). The advantage of a percentile schedule is that it constantly updates the distribution of responses it uses to establish the criterion for reinforcement as responding varies, which keeps the proportion of reinforced responses constant. To date, researchers have conducted few studies to determine the optimal rank order and number of responses to sample to establish a percentile schedule. In a notable exception, Athens et al. (2007) found that percentile schedule procedures were more effective when they based the criterion for reinforcement on a larger number of observations. However, using more observations has the potential to decrease efficiency, because we must postpone reinforcement until the learner has emitted a sufficiently large number of responses to establish the criterion. ## Prompting When a response is not in a learner's repertoire or is not under appropriate stimulus control, prompting may be necessary to evoke the response so that we can reinforce it. Prompts are antecedent stimuli that increase the probability of a desired response. Prompting can help a stimulus become discriminative by increasing opportunities to provide differential reinforcement in its presence (Alberto & Troutman, 1986; Cooper et al., 2007; Demchak, 1990; Miltenberger, 2001). Researchers have used prompts to teach communication (Matson, Sevin, Fridley, & Love, 1990; Williams, Donley, & Keller, 2000), academic skills (Noell, Connell, & Duhon, 2006; Stevens, Blackhurst, & Slaton, 1991), leisure skills (DiCarlo, Reid, & Stricklin, 2003; Schleien, Wehman, & Kiernan, 1981; Oppenheim-Leaf, Leaf, & Call, 2012), social skills (Krantz & McClannahan, 1993; Garcia-Albea, Reeve, Brothers, & Reeve, 2014), self-help skills (Pierce & Schreibman, 1994; Taylor, Hughes, Richard, Hoch, & Coello, 2004), and vocational skills (Bennett, Ramasamy, & Honsberger, 2013). Behavior analysts have categorized prompts as stimulus and response prompts (Schoen, 1986; Wolery & Gast, 1984). Stimulus prompts are those in which we alter or present some property of the criterion stimulus (Etzel & LeBlanc, 1979). For example, a teacher uses a stimulus prompt when he or she places two pictures in front of the learner, and the correct picture is larger than the incorrect one. By contrast, response prompts involve teacher behavior to evoke the desired learner behavior. For example, a teacher uses a response prompt when he or she points to the correct picture after the learner did not respond when the teacher said, "Point to [correct picture]." A script is a stimulus prompt that we can use to facilitate complex behavior, especially conversational skills. Subtlety is a distinct advantage of prompting social behaviors such as conversations with a script, because more contrived prompts (such as vocal instructions) might be off-putting to conversational partners. We can fade the script length from a complete text to a single word or a symbol (e.g., Krantz & McClannahan, 1993), or eliminate the script entirely (e.g., Garcia-Albea et al., 2014). Researchers have used scripts as a caregiver-mediated intervention for children with ASD to promote verbal interactions during play (Reagon & Higbee, 2009). However, scripts are limited to learners who can read, and script fading can be lengthy with some learners. Response prompts exist along a continuum of intrusiveness in the amount of assistance required to evoke the desired behavior from the least intrusive verbal prompts, to the moderately intrusive gestural or model prompts, to the more intrusive physical prompts (Cooper et al., 2007; Miltenberger, 2001). We can deliver most prompts at different intrusiveness levels, such as a partial verbal prompt instead of a complete instruction or a physical prompt, to guide the learner to perform the first portion instead of the entire behavior. Modeling, in which the antecedent stimulus is topographically identical to the target behavior, is a prompt that can be especially effective for teaching complex behaviors (Bandura, Ross, & Ross, 1963). Video modeling and prompting are methods to demonstrate target behavior without requiring an instructor to model the behavior each time the learner receives the prompt. The prompt in video modeling is a video of an individual completing the behavior, which provides the learner with a visual overview of the steps in the behavior sequence. The prompt in video prompting is a clip of an individual completing one step of a behavior sequence and a video clip of the next step after the learner completes the previous step (Domire & Wolfe, 2014). Researchers have used both methods to teach complex chains of behavior, including social skills and daily living tasks (Ayres & Langone, 2005). There is some evidence that video prompting may facilitate acquisition of chained responses more effectively than video modeling (Cannella-Malone et al., 2006). Several factors can influence the effectiveness of modeling, such as the learner's observing the modeled behavior produce reinforcement, as well as the similarity between the model and the learner or between discriminative stimuli presented to the model and the learner (Bandura et al., 1963). Video self-modeling, in which the learner is also the model in the video, is one way to maximize similarity between model and learner (Buggey & Ogle, 2012). Although modeling is effective, two of its limitations are that the learner must have generalized imitation skills and attend to the model during instruction. Further research is needed to identify the components that influence the effectiveness of video modeling and video prompting, such as evaluating the effects of model type and of recording perspective (e.g., Domire & Wolfe, 2014). ## Prompt Fading Prompt dependence occurs when the prompt overshadows the criterion stimulus to such an extent that it never takes on discriminative properties in the absence of the prompt. Thus we must transfer stimulus control from the prompt to the criterion stimulus. Fading is a method of gradually removing a prompt so that the behavior eventually comes under control of the criterion stimulus in the absence of prompts. For example, Wichnick-Gillis, Vener, and Poulson (2019) used textual prompts embedded within instructional stimuli (i.e., scripts) to teach three children with ASD to engage in social interactions. During generalization, the scripts were gradually faded by removing one word at a time from the end of a given script until all words were faded. ### Fading Stimulus Prompts There are two primary methods of fading stimulus prompts: stimulus shaping and stimulus fading. In stimulus shaping, we alter the property of the criterion stimulus that is critical to the intended discrimination, so that the learner can initially make the discrimination. For example, when a chef is teaching a sous-chef to make subtle discriminations of saltiness, initial training may include samples with very distinct differences in salt content. Once the learner is reliably making the discrimination, we gradually diminish the altered stimulus property until the stimulus is representative of the criterion stimulus (Etzel & LeBlanc, 1979). Thus the difference in the amount of salt in the two samples during stimulus shaping may become smaller until the sous-chef can detect even subtle differences. In stimulus fading, we alter a property of the criterion stimulus other than the dimension critical for the discrimination (Etzel & LeBlanc, 1979; Wichnick-Gillis et al., 2019). For example, we can alter position prompts or the size of a target letter when teaching letter identification. In this case, neither position nor size is the critical dimension for discriminating the target letter from other letters. Rather, the form of the letter is the critical property. During stimulus fading, we bring the size or position of the target letter closer to the position or size of the alternative letter until the learner can discriminate the letter by its form. A review of studies comparing stimulus shaping and stimulus fading found that both were effective instructional approaches (Ault, Wolery, Doyle, & Gast, 1989). However, stimulus shaping appears to be more effective than stimulus fading, perhaps because stimulus fading requires the learner to shift the discrimination from an irrelevant stimulus dimension to the relevant dimension. Making such a shift may be difficult for some learners-especially those who selectively attend to certain dimensions of stimuli, such as some individuals with ASD (Wolery & Gast, 1984). ### Fading Response Prompts The five main fading procedures researchers have studied for transferring stimulus control from a response prompt to a criterion stimulus are least-to-most, graduated guidance, most-to-least, time-delay, and simultaneous prompting. Least-to-most prompting, or the system of least prompts, is adaptable to teaching behavior chains such as a series of motor responses (e.g., folding laundry) and discrete behavior such as object labeling. During least-to-most prompting, the instructor presents the criterion stimulus so the learner can emit the correct response independently. If the learner does not emit the correct response, the instructor presents increasingly intrusive prompts until the learner emits the target behavior. For example, an instructor may use least-to-most prompting to teach a learner to correctly identify sight words by first presenting the criterion stimulus, a flash card. The instructor provide a more intrusive prompt, such as the first syllable, if the learner does not make the target response before an interval expired (e.g., 5 seconds), and an even more intrusive prompt, the word, if the learner does not make the target response after the previous prompt. Wolery and Gast (1984) suggested that as instructors, we should present the criterion stimulus at each prompt level and use a constant response interval after each prompt. We should also consider whether successive prompts increase the probability of the target behavior. If not, we should use the least intrusive prompt that is likely to occasion the behavior. An advantage of least-to-most prompt fading is that the learner can emit the correct behavior without prompts. It also may be easier to implement than other strategies, because more intrusive prompts become unnecessary as the individual learns to emit the target behavior independently (Billingsley & Romer, 1983). During graduated guidance, the instructor gradually eliminates the controlling prompt by only presenting the level of prompt necessary to evoke the target behavior. A controlling prompt is one that consistently results in the learner's exhibiting the target behavior (Wolery et al., 1992). We can use least-to-most prompting to identify controlling prompts, which often include physical guidance. For example, parents taught yoga poses to their children by gradually fading the amount of physical guidance from a firm hold to shadowing their children (Gruber & Poulson, 2015). We can use graduated guidance to fade physical guidance and to transfer stimulus control to other types of controlling prompts, such as verbal prompts (Schoen, 1986). An advantage of graduated guidance is that the learner can be as independent as possible, because the instructor only provides the minimum amount of guidance necessary. A disadvantage is that fading is not systematic. Fading relies on subjective judgments about the required prompting level (Wolery & Gast, 1984), which the instructor must make rapidly, based on the learner's responses. This can affect implementation integrity and can be difficult to evaluate in the absence of systematic research. Most-to-least fading begins with the delivery of a controlling prompt, and the amount of assistance necessary for the individual to complete the behavior correctly varies across trials instead of within a trial. The intrusiveness of the prompt decreases or increases on subsequent trials, based on whether the learner meets a mastery or failure criterion, respectively, for the current prompt level. Note that if the first prompting level is a controlling prompt, the learner should always meet the mastery criterion, because the response should always occur following this most intrusive prompt. Graduated guidance and most-to-least prompt fading are well suited for teaching chained motor responses and for learners who require many response-reinforcer pairings to achieve independence (Wolery & Gast, 1984). Libby, Weiss, Bancroft, and Ahearn (2008) modified most-to-least prompt fading by inserting a delay between controlling prompts to allow learners an opportunity for independence. This time-delay fading procedure produced mastery almost as rapidly as least-to-most prompting, but with fewer errors. The two types of time-delay fading, constant and progressive (O'Neill, McDowell, & Leslie, 2018; Snell & Gast, 1981), begin with O-second-delay trials in which the criterion stimulus and prompt occur simultaneously. We typically use more O-second-delay trials for difficult tasks or lower-functioning learners. In progressive time-delay fading, the time between the criterion stimulus presentation and the prompt gradually increases after each trial, several trials, or each instructional session (Heckaman, Alber, Hooper, & Heward, 1998). With constant time-delay fading, the instructor delays the prompt for a specified time after presentation of the criterion stimulus, and this latency remains fixed during instructional sessions (Snell & Gast, 1981). Some advantages of time-delay fading include its low error rate and its simplicity. Constant time-delay fading is especially simple to use, which may produce better treatment integrity (Wolery et al., 1992). Simplicity may be especially important when caregivers or supervisees implement interventions. For example, DiPipi-Hoy and Jitendra (2004) showed that parents could implement constant time-delay procedures with good fidelity. During simultaneous prompting, the instructor delivers the controlling prompt immediately after presenting the demand (Morse & Schuster, 2004). That is, trials have a 0-second delay, and the instructor does not introduce a delay. The instructor conducts an assessment before instructional sessions to determine whether the learner can emit the target response without the prompt. Researchers have used simultaneous prompting to teach discrete behaviors (Tekin-Iftar, Acar, & Kurt, 2003), chained tasks (Parrott, Schuster, Collins, & Gassaway, 2000), and vocational tasks (Fetko, Schuster, Harley, & Collins, 1999). The advantages of simultaneous prompting are low error rate and simplicity, but it may be less sensitive to detecting the moment at which mastery occurs. However, instructors can conduct more trials per time than they can with prompt-delay procedures, and simultaneous prompting produces more rapid mastery than other prompting strategies do (Akmanoglu, Kurt, & Kapan, 2015; Schuster, Griffen, & Wolery, 1992; Swain et al., 2015). Studies comparing prompt-fading methods have produced conflicting results. Limitations of these comparative studies is that most participants displayed generalized imitation, attended well, waited for teacher assistance, and demonstrated clear preferences for potential reinforcers. Wolery and Gast (1984) suggested that constant time-delay may be more efficient for students who exhibit these behaviors, but that other prompt-fading strategies may be more appropriate for individuals who lack these skills. In general, research has shown that the prompt-fading methods described above can effectively transfer stimulus control to a criterion stimulus for at least some tasks and participants. We cannot draw further conclusions about prompt fading beyond the idiosyncratic variables evaluated, such as participant characteristics, tasks, and prompting variations. ## Chaining Each step or component of a behavior chain has its own conditioned reinforcers and discriminative stimuli (Kelleher, 1966; Skinner, 1938). That is, the consequence following completion of each component of the behavior chain may function as both a conditioned reinforcer for the previous behavior and a discriminative stimulus for the next one. We typically complete a task analysis for the chain of target behaviors and teach the chain by using forward chaining, backward chaining, or total-task presentation. During backward chaining and forward chaining, the instructor teaches one component of the behavior chain at a time. The instructor teaches each additional component as the learner meets the mastery criterion for the previous components. The difference is that in backward chaining, the instructor teaches the behavior chain in reverse order, starting with the last component; in forward chaining, behaviors are taught in the order they occur in the chain, starting with the first component. In total- or whole-task chaining, the learner performs the entire behavior chain on every instructional trial. During reverse chaining, the instructor physically guides the learner to perform the behavior chain's components until the last one, which the learner performs independently (Sternberg & Adams, 1982). The instructor teaches components in reverse order by physically guiding progressively fewer components (e.g., last two, last three) as the learner masters each component. In backward chaining with leap ahead (Spooner, Spooner, & Ulicny, 1986), the instructor does not teach every component directly, to increase time efficiency. Rather, the instructor conducts ongoing assessment to determine whether the learner can perform some components without training. A potential advantage of backward chaining is that the learner produces natural reinforcement by completing each component of the behavior chain, whereas completing initial components during forward-chaining produces conditioned reinforcement (Spooner et al., 1986). However, the natural consequences produced in backward chaining may not function as reinforcement for some individuals. The advantage of total-task presentation is that the learner has increased opportunities for conditioned reinforcement by practicing every component in the behavior chain on every trial. However, total-task presentation may be less time-efficient. Forward chaining may be easiest to use, because the instructor teaches components in the order in which they occur in the chain, and this may produce the best long-term performance (Watters, 1992). Smith (1999) showed that fewer errors occurred at the beginning of the chain during forward chaining and at the end of the chain during backward chaining. Thus an instructor should use forward chaining if a learner is unlikely to complete the chain after an error. Direct comparisons of the different chaining methods have shown mixed results (Ash & Holding, 1990; Hur & Osborne, 1993; Slocum & Tiger, 2011; Spooner & Spooner, 1984; Spooner, Weber, & Spooner, 1983; Watters, 1992; Wightman & Sistrunk, 1987). As with other methods described in this chapter, acquisition of skills via different chaining methods is likely to be idiosyncratic across populations and highly influenced by the target skill. ## Promoting Response Generalization and Variety We direct readers to Chapters 5 (Spradlin, Simon, & Fisher) and 6 (Podlesnik, Jimenez-Gomez, & Kelley) of this volume for discussions of generalization and methods useful for its facilitation, including descriptions of stimulus control, equivalence classes, and recombinative generalization. We review some issues about generalization relevant to establishing complex behaviors. For example, multiple-exemplar training, in which the instructor prompts and reinforces responding to several members of a stimulus class, can produce generalization and promote spontaneous or varied responding (Stokes & Baer, 1977). For example, Krantz and McClannahan (1993) used scripts with varied content that prompted comments about activities to teach children with ASD to initiate social interactions. Scripted comments increased, and spontaneous, unscripted comments also increased for several participants. Such variability in responding can be important, because it often determines whether a response will produce natural reinforcers. For example, peers may perceive a student who always asks for help using a single phrase in the same tone of voice as odd, and may ignore or shun the student. McClannahan and Krantz (2005) used scripts to teach several mand frames (i.e., I want, I need, or I would like) to participants and observed increases in spontaneous, novel mand frames. Similarly, Betz, Higbee, Kelley, Sellers, and Pollard (2011) taught varied responses to individuals with ASD by using stimulus prompts, each associated with a unique color. They faded the stimulus prompts and used the color prompts, and faded the color prompts for two of three participants who continued to emit the varied responses. We can promote response variability by manipulating the consequences of responding. Extinction-induced variability is one example in which extinction of a previously reinforced response increases the likelihood that various other responses will occur (Goetz & Baer, 1973; Sullivan et al., 2020). Extinction-induced variability has the unique benefit of eliminating instructional time for behaviors that are in the learner's repertoire but occur infrequently. For example, Valentino, Shillingsburg, Call, Burton, and Bowen (2011) implemented extinction for the signed mands of children with limited vocalizations, and observed increases in vocalizations. The chances for extinction of the original target behavior and emergence of problem behavior are disadvantages of extinction-induced variability. Another method to produce response variability is a lag schedule of reinforcement. The instructor reinforces a response in a lag schedule if it differs from a designated number of previously emitted responses (Falcomata et al., 2018; Page & Neuringer, 1985). For example, if an individual had previously emitted the response hi followed by hello, neither of these greetings would be eligible for reinforcement as the third response on a Lag 2 schedule. Only a novel greeting (e.g., good morning) would produce reinforcement, but a novel response or hi for the fourth greeting would produce reinforcement, because neither was one of the previous two responses. Researchers have used lag schedules to establish variety for verbal behavior (Esch, Esch, & Love, 2009; Falcomata et al., 2018; Lee, McComas, & Jawor, 2002; Silbaugh, Falcomata, & Ferguson, 2018; Wiskow, Matter, & Donaldson, 2018) and building block structures (Napolitano, Smith, Zarcone, Goodkin, & McAdam, 2010), and to maintain variable responding after discontinuation of the lag schedule (Heldt & Schlinger, 2012). A disadvantage is that patterned responding may emerge. Using variable-lag schedules may mitigate this problem (Lee et al., 2002). ## Strategic Instruction Strategies are more complex sequences of behavior that include assessment, planning, execution, and evaluation of a course of action. Importantly, a strategy is a behavioral process by which an individual chooses, orders, and evaluates behavior toward solving diverse problems, rather than a fixed behavioral sequence. We can distinguish skills and strategies by the unique roles that they play in learning and achievement. Alexander and Murphy (1999) describe skills as procedural knowledge that students develop, which enables them to perform tasks effectively with speed and accuracy. Students who achieve automaticity or fluency of skills can attend to more complex task dimensions (e.g., comprehending text after mastering decoding and sight words). For example, Wagner et al. (2011) reported that fluency in writing individual letters contributed to the quality and complexity of first- and fourth-grade students' writing. One critical goal of effective education is to provide students with sufficient opportunities to practice fundamental skills (e.g., decoding, computing basic math facts), so that they can develop adequate fluency to use those basic skills in a strategic manner (Ardoin & Daly, 2007). As they do with basic