Guest Blog by: Lynn Schumacher, M.S., BCBA, Kennedy Cloe, M.S., BCBA, and Lesley Shawler, Ph.D., BCBA
“Generalization has been, and doubtless will remain, a fundamental concern of applied behavior analysis” (Stokes & Baer, 1977, p. 350).
Lynn Schumacher, M.S., BCBA
Lynn Schumacher is currently a Ph.D student in the Behavior Analysis and Therapy Program. She has a Bachelor of Science degree in elementary and special education and a Master of Science in Applied Behavior Analysis. She completed both degrees in Emmitsburg, Maryland, at Mount Saint Mary’s University. She is also a BCBA supervisor at the Center for Autism Spectrum Disorder. She spent three years working in an Early Intensive Behavioral Interventions (EIBI) company, working in both home-based and clinic-based settings with children aged 2 years old to 10 years old. The focus of her master’s thesis was the feasibility of skill assessment testing to inform social preference assessments for children with Autism. Her main research area is preference assessments and social interactions. She is also currently working on a generalization project to assess the effects of Multiple Exemplar Training.
Kennedy Cloe, MS, BCBA
Kennedy Cloe is currently a BCBA for the Menta Education Group, a therapeutic day school for children with emotional and behavioral disorders and disabilities. She graduated cum laude with her bachelor’s degree in behavior analysis and therapy from Southern Illinois University – Carbondale (SIUC). She also received her master’s degree in behavior analysis from SIUC, with her master’s thesis focusing on effective and efficient teaching strategies to foster generative language with children with autism. Her current research interests include verbal behavior and the assessment and treatment of challenging behavior.
Lesley Shawler, Ph.D., BCBA
Lesley Shawler, Ph.D., BCBA, is an Assistant Professor in the Behavior Analysis and Therapy program at Southern Illinois University, Carbondale, where she teaches and supervises master’s and doctoral students pursuing their BCBA. Dr. Shawler’s main research interests include exploring strategies to promote generalization of treatment effects beyond controlled settings, both within the assessment and treatment of challenging behavior and when teaching verbal behavior to individuals with autism and other neurodevelopmental disabilities and within caregiver training.
Generalization is said to occur when an individual emits behavior that was learned in one context in a different, untrained context (Stokes & Baer, 1977). It may take place across settings, individuals, time of day, teaching materials, and even related behavior. Specifically, if an individual learns to emit a response in the presence of a specific set of contingencies, or stimulus class, the individual may then emit the response in the presence of similar but untrained stimuli (i.e., stimulus generalization) (Stokes & Osnes, 1989). For example, when a child learns to greet their mother by stating, “hello”, and the child then begins greeting their teacher and peers with, “hello,” without direct teaching. Conversely, response generalization occurs when an untrained but functionally equivalent behavior occurs in the presence of a trained stimulus (Skinner, 1953). For example, after learning to request an item by stating, “Can I have that toy?”, a child may later say, “I want that toy”, or “toy, please” to request an item without direct teaching.
Individuals with autism spectrum disorder (ASD) often have difficulties generalizing learned skills to untrained contexts without explicit instruction. Because generalization is essential to successful treatment outcomes (Cooper et al., 2019), it should be embedded in treatment planning from the outset. Swan et al. (2016) recommend that generalization should be a central treatment goal, rather than an afterthought addressed only once treatment effects are observed. This raises an important question: how should practitioners plan or program for generalization? In 1977, Stokes and Baer reviewed the extant literature and provided a framework for practitioners to program for generalization when treatment planning for individuals with developmental disabilities. Table 1 summarizes these strategies, providing definitions and corresponding examples from research.
Table 1
| Generalization Technique | Type | Definition | Examples of Research |
| Train and Hope | Passive | After effective behavior change has been produced, any instances of generative responding are noted. Generalization is not actively pursued; rather, it is simply documented. | Redd and Birnbrauers (1996) taught cooperative play to children with specific adults, but responding did not generalize to novel adults after teaching occurred. Kifer et al. (1974) taught negotiation techniques in one situation, and these skills generalized to novel situations without direct teaching. |
| Sequential Modification | Passive (could become active) | When effective behavior change is achieved in one setting, it is then assessed in other settings. If generalization does not occur, then the original intervention is systematically applied to each non-generalized setting. | Meichenbaum et al. (1968) found that when monetary reinforcers were provided during the afternoon session for on-task behavior, responding generalized to novel contexts, such as the morning session, only after the direct implementation of the intervention. |
| Introduction to Natural Maintaining Contingencies | Active | The transfer of behavioral control from arbitrary reinforcers to naturally maintaining stable environmental contingencies (the transfer of reinforcing control). This is one of the most dependable methods for programming and teaching generalization. | Baer and Wolf (1967) taught a group of preschoolers greeting responses. The greeting responses began under the control of the teacher’s instructions, but as the naturally maintaining contingencies of peer interactions (i.e., reinforcement) began to emerge, they generalized to the peers in the classroom. |
| Train Sufficient Exemplars (Multiple Exemplar Training; MET) | Active | The use of more than one example to teach a stimulus or a concept. This can also include teaching multiple responses (Holth, 2017). For example, teaching multiple examples of different foods can lead to the formation of the category of food. Generalization to untaught stimulus conditions and responses is programmed through teaching a sufficient number of exemplars rather than all possible exemplars. | Baer et al. (1968) showed that teaching several examples of imitation promoted accurate responding across novel, untaught imitation responses. |
| Train Loosely | Active | This strategy uses the negation of discrimination techniques, so there is little to no control over the stimuli or the response. Multiple stimuli are allowed to evoke a variety of functionally appropriate responses within the setting. | Schroeder and Baer (1972) compared (1) serial training with restricted vocal skills (teaching one response at a time with multiple stimuli), and (2) concurrent training with a broader range of stimuli when teaching echoic responses. The second method produced generalized vocal imitative responding. |
| Use Indiscriminable Stimuli | Active | In the everyday environment, behavior occurs in the presence of specific discriminative stimuli. This strategy uses intermittent reinforcement to create resistance to extinction via the partial reinforcement effect (Nevin, 2012), thereby promoting generalization across settings. | Schwarz and Hawkins (1970) reinforced appropriate classroom behavior in math class, which then generalized to other classes, such as spelling, without direct reinforcement. |
| Program Common Stimuli | Active | Incorporating stimuli found in the individual’s everyday learning environment into a new teaching environment. These stimuli acquire discriminative properties in the instructional context and later evoke behavior in their everyday environment, promoting generalization. | Walker and Buckley (1972) used the same academic materials in an experimental remedial classroom and a control classroom. Students learned the academic and social behavior to emit in the remedial classroom before transitioning to the control classroom. Students demonstrated generalized academic and social behavior when common stimuli were programmed. |
| Mediate Generalization | Active | Teaching an individual to engage in behavior that promotes self-mediated responding across multiple untrained or novel contexts. This strategy may be most effective for individuals with advanced verbal repertoires. | Broden et al. (1971) used self-recording and teacher praise for an 8th-grade girl’s studying behavior. Increased levels of studying generalized across time using the self-recording procedure. Self-recording is an example of mediating generalization because the individual can then use self-recording with other untaught areas. For example, self-recording could be used with other similar problems, which could result in generalization. |
| Train to Generalize | Active | Actively teaching generalization throughout the intervention. If generalization is considered a response itself, then a reinforcement contingency can be applied to it like any other operant. That is, when generalization is observed, reinforcement contingencies can be used to increase its future likelihood. | Goetz and Baer (1973) taught three preschool children to generalize block-building responses. Participants received descriptive social reinforcers for every different block form created, but not for repeating prior structures, resulting in novel and generative responding. |
Stokes and Baer (1977) described these strategies based on existing methods involved in programming and assessing for the occurrence of generalization. Passive methods simply assess whether generalization has occurred after treatment results are observed, while active methods facilitate the programming of generalization to promote the likelihood of generalization. The seven active strategies, specifically, are important because they allow for generalization to be programmed during treatment development, increasing the likelihood of generalized and effective behavior change (Baer et al., 1968). This emphasis also aligns with our ethics code of providing effective treatment (Behavior Analyst Certification Board, 2024, Section 2.01). Teaching only under tight or controlled contexts is unlikely to effectively change behavior for the long term. Moreover, it can be incredibly time- and resource-consuming to teach all stimulus examples in the everyday environment; therefore, it is vital to program for generalization as early as possible. One common method, MET, teaches different examples of the same concept, which can promote the likelihood of generalization across a stimulus class. However, despite its utility, there are still unknowns about this technique and how to increase the likelihood that generalization will occur. As such, we will now explore some areas within MET in need of further investigation.
MET
MET is a common strategy used to program for generalization when teaching individuals certain skills, such as labeling (tacting) stimuli in their environment. MET involves presenting more than one exemplar of a stimulus to teach class formation. Stimulus exemplars can have a wide range of topographies, including visual, audio, abstract, and three-dimensional. For example, to teach the concept of a “pig”, one could use varying images or objects of pigs (e.g., cartoon, figurine, abstract) (Holth, 2017). Teaching some verbal operants, such as echoics or textual behavior, often results in the emergence of expansive generalized repertoires. For example, once a learner contacts reinforcement for consistently echoing certain sounds (i.e., generalized echoic repertoire), novel sounds or words can be taught, and the individual can echo these sounds without explicit teaching. C
MET can be used to teach both stimulus and response generalization. Stimulus generalization may occur if multiple exemplars of the same stimulus are taught, such that the response topography remains consistent while the stimuli presented vary. For example, as shown in Figure 1, stimulus generalization occurs when the learner can emit the response “pig” in the presence of all four pigs after being taught “pig” with two or three pig exemplars. Response generalization, by contrast, occurs when the stimulus remains constant while the response topography varies. For example, after learning to tact “pig” or “piglet” when a pig is present, the learner may then tact “little piggy” in the presence of the same or similar pig without direct teaching.
Schnell et al. (2018) evaluated two procedures to assess for stimulus generalization based on two different MET procedural variations with three children with ASD. The procedural variations consisted of serial (S-MET) and concurrent MET (C-MET) conditions. In the S-MET condition, one exemplar was randomly selected as the teaching target, while the remaining three exemplars were reserved for generalization testing. In the C-MET condition, three exemplars were randomly selected to serve as teaching targets, and only one exemplar was reserved for generalization testing. Both MET variations were effective at promoting generalization of the tact responses, demonstrating that even with procedural variations, MET can be effective at promoting generalization. However, S-MET was more efficient than C-MET in terms of teaching time. Notably, Wunderlich et al. (2014) reported contrasting findings, showing that C-MET was more effective and efficient at teaching tacts that lead to stimulus generalization, as compared to S-MET. These mixed findings highlight the need for further research on the topic.
Holth (2017) reported some strengths and limitations of MET. A major strength of MET is that there are certain skills in which this teaching strategy is essential. These include abstraction and concept learning, identity matching, relations between stimuli, rule following, describing past events, and answering WH questions. These skills, when taught using MET, have the potential to generalize far beyond the learned exemplars and act as pivotal behaviors. However, one limitation of MET is that it may not be sufficient by itself in producing generalized performance. For example, Sprague and Horner (1984) taught developmentally delayed teenagers to use vending machines in which three different strategies were used: single case, which involved a single example, MET, and general case, which included a variety of examples with different important features. The most effective strategy was the general case, which incorporated a wide range of variation in the stimuli presented and responses reinforced. These findings suggest that when there is a wide variation in the examples that are taught and reinforced, generalization occurs at a higher rate (Sprague & Horner, 1984). Still, limited research exists on the parameters for selecting exemplars (e.g., what qualifies as a representative example of a stimulus class) or the number of exemplars that is considered “sufficient” to reliably promote generalization.
Stimulus Features within MET
Some research has evaluated different stimulus features that may be pivotal to achieving generalization. Hupp (1986) compared the use of three or five “good” exemplars when teaching tacts to six individuals with disabilities. Although slightly higher rates of generalization were observed with five examples compared to three examples, the difference was not statistically significant. Hupp incorporated what they considered “good” exemplars during tact teaching; however, they did not provide an operational definition of what qualifies as a “good” example. Instead, stimulus examples were presented to neurotypical adults to rate on a scale from 1 being good and 7 being poor, as adapted from Rosch (1973). Rosch proposed that stimulus categories are defined by distributed attributes, with some examples representing typical members–or prototypes–of the class. Exemplars that closely resembled the prototype were considered to have salient features, whereas those that diverged substantially were viewed as less salient.
Building on this work, Hupp and Mervis (1982) found that generalization was more likely when the “best” exemplars were included as opposed to including a diversity of exemplars. Additionally, using three “best” exemplars (i.e., MET) was superior to using one “best” exemplar to increase the likelihood of generalization. These findings suggest that the inclusion of relevant or “must have” features may be a critical determinant of generalization, beyond the number of exemplars alone (Halbur et al., 2021).
Other research has highlighted the importance of incorporating non-exemplars to strengthen discrimination and facilitate generalization. Johnson and Bulla (2021), for example, asserted that including non-exemplars helps learners identify the defining features of a target stimulus. For example, when teaching the tact pig, explicitly contrasting it with a nonexample such as a warthog, which shares some but not all features, can sharpen discrimination and promote generalization (Figure 2). For further discussion of general-case teaching procedures, see Hickey et al. (2024).
Recently, Williams et al. (2025) distinguished between “must have” and “can have” features of a stimulus. “Must-have” features are those shared by stimuli in the class, while the “can-have” features are those features that are not universally present across all exemplars. Figure 3 shows an example of “must-have” features for a pig would be a round snout, short legs, hooves, and a curly, short tail; while “can-have” features would be the color pink, having mud on it, and having pointy ears. Hupp and Mervis (1982) similarly asserted that the most accurate generalization depends on the representativeness of the exemplars—maximizing within-category similarity while minimizing between-category similarity.
Many of the studies mentioned above do not share guidelines on how different exemplars were selected. For example, Sprague and Horner (1984) reported using one example, multiple examples, and general-case examples, but did not define the different features of the vending machines that were included in the teaching context. The lack of clear selection guidelines in much of the MET literature poses a barrier for practitioners, who must decide which stimulus features are critical for inclusion in teaching. Encouragingly, preliminary guidelines have started to emerge to guide practitioners on making such decisions (see Johnson & Bulla, 2021; Hickey et al., 2024). Such guidelines are especially important when teaching children with ASD due to the potential for stimulus overselectivity. Stimulus overselectivity occurs when a child attends to selected stimuli or features in the environment and does not attend to other relevant stimuli (Bradshaw, 2013). For example, they may attend to something in the background of an image (mud), rather than the central image itself (pig). Overselectivity during teaching might be one explanation as to why individuals with ASD fail to generalize newly learned skills across novel and untrained settings (Bradshaw, 2013).
MET vs. MEI
Before closing, it is crucial to note the difference between MET and multiple exemplar instruction (MEI) for clarity and consistency of our behavioral technologies. LaFrance and Tarbox (2019) eloquently elucidated the difference between these two procedures. MET is the presentation of several different examples of the same stimulus, while MEI refers to rapidly and randomly rotating instructions targeting different verbal operants of the same response across a series of consecutive trials. MEI involves teaching a response across multiple verbal operants such as tacts, intraverbals, echoics, and so forth. In MEI, the same response is emitted across different verbal operants such that the response “pig” could be a tact, a mand, an echoic, or an intraverbal response. In other words, the function of the response is what varies (LaFrance & Tarbox, 2019).
Suggestions for Practitioners Implementing MET
To increase the effectiveness of MET in promoting generalization across responses, we recommend the following:
- Select a minimum of 3 exemplars with ‘must have’ features representative of the stimulus class being taught.
- Identify and ensure that all “must-have” features are present and visible.
- Vary the “can-have” features while keeping the “must-have” features the same. Also, vary the discriminative stimuli that should evoke responding (Song et al., 2021).
- Identify functional response classes and teach several response options.
- Assess for generalization and adjust the stimuli used to teach or conduct further analyses to determine faulty responding (Song et al., 2021).
Final Thoughts
Planning for generalization should be an integral part of treatment development. Including active strategies from the beginning can help increase the likelihood of successful generalization outcomes. Ultimately, interventions are only truly effective when responses extend beyond the teaching context and occur across everyday situations and environments (Baer et al., 1968).
References
Baer, D. M., Wolf, M. M., & Risley, T. R. (1968). Some Current Dimensions of Applied Behavior Analysis. Journal of Applied Behavior Analysis, 1(1), 91–97. https://doi.org/10.1901/jaba.1968.1-91
Baer, D. M., & Wolf, M. M. (1967, August). The entry into natural communities of reinforcement (ERIC No. ED020533) [Paper presentation]. Annual Meeting of the American Psychological Association, Washington, DC. ERIC.
Behavior Analyst Certification Board. (2024, January 11). Ethics codes. Behavior Analyst Certification Board. https://www.bacb.com/ethics-information/ethics-codes/
Bradshaw, J. (2013). Stimulus overselectivity. Encyclopedia of Autism Spectrum Disorders, 3007–3008. https://doi.org/10.1007/978-1-4419-1698-3_137
Broden, M., Bruce, C., Mitchell, M. A., Carter, V., & Hall, R. V. (1970). Effects of teacher attention on attending behavior of two boys at adjacent desks. Journal of Applied Behavior Analysis, 3(3), 205–211. https://doi.org/10.1901/jaba.1970.3-205
Cooper, J. O., Heron, T. E., & Heward, W. L. (2019). Chapter 30. In Applied Behavior Analysis (3rd ed., pp. 713–755). Pearson Education.
Erhard, P., Falcomata, T. S., Oshinski, M., & Sekula, A. (2022). The effects of multiple-exemplar training on generalization of social skills with adolescents and young adults with autism: A systematic review. Review Journal of Autism and Developmental Disorders, 11(1), 66–85. https://doi.org/10.1007/s40489-022-00328-4
Goetz, E. M., & Baer, D. M. (1973). Social control of form diversity and the emergence of new forms in children’s blockbuilding. Journal of Applied Behavior Analysis, 6(2), 209–217. https://doi.org/10.1901/jaba.1973.6-209
Greer, R. D., Stolfi, L., Chavez-Brown, M., & Rivera-Valdes, C. (2005). The emergence of the listener to speaker component of naming in children as a function of multiple exemplar instruction. The Analysis of Verbal Behavior, 21(1), 123–134. https://doi.org/10.1007/bf03393014
Guevremont, D. C., Osnes, P. G., & Stokes, T. F. (1988). Preschoolers’ goal setting with contracting to facilitate maintenance. Behavior Modification, 12(3), 404–423. https://doi.org/10.1177/01454455880123006
Halbur, M. E., Caldwell, R. K., & Kodak, T. (2021). Stimulus control research and practice: Considerations of stimulus disparity and salience for discrimination training. Behavior Analysis in Practice, 14(1), 272–282. https://doi.org/10.1007/s40617-020-00509-9
Hickey, C. R., Reeve, S. A., Reeve, K. F., Deshais, M. A., & Vladescu, J. C. (2024). Using general‐case procedures outside of autism intervention: A systematic review. Behavioral Interventions, 39(3). https://doi.org/10.1002/bin.2000
Holth, P. (2017). Multiple exemplar training: Some strengths and Limitations. The Behavior Analyst, 40(1), 225–241. https://doi.org/10.1007/s40614-017-0083-z
Hupp, S. C. (1986). Use of multiple exemplars in object concept training: How many are sufficient? Analysis and Intervention in Developmental Disabilities, 6(4), 305–317. https://doi.org/10.1016/s0270-4684(86)80011-8
Hupp, S. C., & Mervis, C. B. (1982). Acquisition of basic object categories by severely handicapped children. Child Development, 53(3), 760. https://doi.org/10.2307/1129390
Johnson, K., & Bulla, A. J. (2021). Creating the components for teaching concepts. Behavior Analysis in Practice, 14(3), 785–792. https://doi.org/10.1007/s40617-021-00626-z
Kazdin, A. E. (1973). The effect of vicarious reinforcement on attentive behavior in the classroom. Journal of Applied Behavior Analysis, 6(1), 71–78. https://doi.org/10.1901/jaba.1973.6-71
Kifer, R. E., Lewis, M. A., Green, D. R., & Phillips, E. L. (1974). Training Predelinquent Youths and Their Parents to Negotiate Conflict Situations. Journal of Applied Behavior Analysis, 7(3), 357–364. https://doi.org/10.1901/jaba.1974.7-357
LaFrance, D. L., & Tarbox, J. (2019). The importance of multiple exemplar instruction in the establishment of novel verbal behavior. Journal of Applied Behavior Analysis, 53(1), 10–24. https://doi.org/10.1002/jaba.611
Meichenbaum, D. H., Bowers, K. S., & Ross, R. R. (1968). Modification of Classroom Behavior of Institutionalized Female Adolescent Offenders. Behaviour Research and Therapy, 6(3), 343–353. https://doi.org/10.1016/0005-7967(68)90067-3
Nevin, J. A. (2012). Resistance to extinction and behavioral momentum. Behavioural Processes, 90(1), 89–97. https://doi.org/10.1016/j.beproc.2012.02.006
Redd, W. H., & Birnbrauer, J. S. (1969). Adults as Discriminative Stimuli for Different Reinforcement Contingencies with Retarded Children. Journal of Experimental Child Psychology, 7(3), 440–447. https://doi.org/10.1016/0022-0965(69)90006-x
Rosch, E. H. (1973). In Cognitive Development and the Acquisition of Language (pp. 111–144). essay, Academic Press.
Schnell, L. K., Vladescu, J. C., Kodak, T., & Nottingham, C. L. (2018). Comparing procedures on the acquisition and generalization of tacts for children with autism spectrum disorder. Journal of Applied Behavior Analysis, 51(4), 769–783. https://doi.org/10.1002/jaba.480
Schroeder, G. L., & Baer, D. M. (1972). Effects of Concurrent and Serial Training on Generalized Vocal Imitation in Retarded Children. Developmental Psychology, 6(2), 293–301. https://doi.org/10.1037/h0032121
Schwarz, M. L., & Hawkins, R. P. (1970). Application of delayed reinforcement procedures to the behavior of an elementary school child. Journal of Applied Behavior Analysis, 3(2), 85–96. https://doi.org/10.1901/jaba.1970.3-85
Skinner, B. F. (1953). Science and Human Behavior. Free Press; Collier-Macmillan.
Song, C. J., Vladescu, J. C., Reeve, K. F., Miguel, C. F., & Breeman, S. L. (2021). The influence of correlations between noncritical features and reinforcement on stimulus generalization. Journal of Applied Behavior Analysis, 54(1), 346-366. https://doi.org/10.1002/jaba.760
Sprague, J. R., & Horner, R. H. (1984). The effects of single instance, multiple instance, and general case training on generalized vending machine use by moderately and severely handicapped students. Journal of Applied Behavior Analysis, 17(2), 273–278. https://doi.org/10.1901/jaba.1984.17-273
Stokes, T. F., & Baer, D. M. (1977). An Implicit Technology of Generalization. Journal of Applied Behavior Analysis, 10(2), 349–367. https://doi.org/10.1901/jaba.1977.10-349
Stokes, T. F., & Osnes, P. G. (1989). An Operant Pursuit of Generalization. Behavior Therapy, 20(3), 337–355. https://doi.org/10.1016/s0005-7894(89)80054-1
Swan, A. J., Carper, M. M., & Kendall, P. C. (2016). In Pursuit of Generalization: An Updated Review. Behavior Therapy, 47(5), 733–746.
Walker, H. M., & Buckley, N. K. (1968). The use of positive reinforcement in conditioning attending behavior. Journal of Applied Behavior Analysis, 1(3), 245–250. https://doi.org/10.1901/jaba.1968.1-245
Williams, C. L., St. Peter, C. C., Perone, M., Aguilar, M., Cederberg, B. A., Gregersen, D. J., & Richardson, E. J. (2025). Using must‐have and can‐have features to improve conceptual learning. Journal of the Experimental Analysis of Behavior, 124(1). https://doi.org/10.1002/jeab.70037
Wunderlich, K. L., Vollmer, T. R., Donaldson, J. M., & Phillips, C. L. (2014). Effects of serial and concurrent training on acquisition and generalization. Journal of Applied Behavior Analysis, 47(4), 723–737. https://doi.org/10.1002/jaba.154