Why Maintenance and Generalization Matter
A behavior change that exists only during treatment sessions in one setting with one therapist is of limited practical value. The ultimate goal of applied behavior analysis is to produce behavior change that maintains over time (maintenance) and transfers across relevant settings, people, and stimuli (generalization). Baer, Wolf, and Risley (1968) identified generality as one of the seven dimensions of applied behavior analysis, recognizing that interventions must demonstrate effects beyond the immediate training context to be considered truly effective.
Despite the importance of maintenance and generalization, research consistently shows that behavior analysts often fail to program for them systematically. Instead, they adopt a "train and hope" approach, expecting that behavior change will automatically maintain and generalize without explicit programming. This approach frequently fails, resulting in lost treatment gains and the need for re-intervention. Understanding and applying evidence-based strategies for maintenance and generalization is essential for effective BCBA practice.
Defining Maintenance and Generalization
Maintenance
Maintenance refers to the persistence of a behavior change over time after the intervention has been reduced or withdrawn. A behavior is said to be maintained when it continues at acceptable levels without the original intervention procedures. Maintenance is concerned with the temporal dimension: does the behavior change last?
Generalization
Generalization refers to the occurrence of a behavior change in conditions different from those in which training occurred. There are several types of generalization:
- Stimulus generalization: The behavior occurs in the presence of stimuli or settings that were not part of training. A child who learns to greet people during therapy sessions also greets people at school and at home.
- Response generalization: Changes in the trained response produce changes in similar but untrained responses. Teaching a child to say "hello" may also result in the child using other greeting forms like waving or saying "hi."
- Setting/situation generalization: The behavior occurs in settings different from the training setting.
- Person generalization: The behavior occurs with people other than the trainer.
Strategies for Programming Generalization
Stokes and Baer (1977) identified several strategies for promoting generalization, and subsequent research has expanded and refined these approaches:
Train Sufficient Exemplars
Teaching a behavior using multiple examples increases the likelihood that the behavior will generalize to novel examples. The principle is that training across diverse instances helps the learner abstract the relevant features of the stimulus class, so responding comes under the control of the relevant dimension rather than irrelevant features of a single training example. For instance, teaching a child to identify "dogs" using only golden retrievers may produce limited generalization. Teaching with golden retrievers, poodles, bulldogs, and dachshunds promotes generalization to novel dog breeds.
Train Loosely
Training loosely means introducing natural variation into training conditions rather than maintaining strict consistency. By varying the instructions, materials, settings, and other aspects of training, the behavior becomes more resistant to disruption when conditions change. This is the opposite of strict stimulus control training, where consistency maximizes discrimination. When the goal is generalization, some "messiness" in training is beneficial because it prevents the behavior from becoming overly dependent on specific training conditions.
Use Indiscriminable Contingencies
When the learner cannot reliably predict when reinforcement will or will not be available, behavior is more likely to maintain across conditions. Intermittent reinforcement schedules make the availability of reinforcement indiscriminable, producing behavior that is more resistant to extinction. Similarly, delayed reinforcement and variable-ratio schedules prevent the learner from discriminating between training and non-training conditions based on the reinforcement schedule.
Program Common Stimuli
Include stimuli from the generalization setting in the training setting. If a child needs to use their communication skills at school, include school materials, peer models, and classroom-like activities in the training environment. When training and generalization settings share common stimuli, the discriminative stimuli that control the behavior in training also function as discriminative stimuli in the generalization setting.
Teach to the Natural Contingencies
Select target behaviors that will contact natural reinforcement in the client's environment. A social greeting response that consistently produces friendly responses from others is more likely to maintain than a response that is only reinforced artificially. When designing interventions, ask: "Will this behavior produce reinforcement in the client's natural environment?" If the answer is no, either modify the behavior to better fit the natural contingencies or modify the environment to provide natural reinforcement.
Mediate Generalization
Teach the learner a mediating response that can be used across settings. Self-management strategies, self-instructions, and portable visual supports function as mediators that the learner carries from one setting to another. When the learner uses the same self-monitoring checklist at school, at home, and in the community, the checklist serves as a common stimulus that promotes consistent behavior across settings.
Strategies for Programming Maintenance
Systematic Schedule Thinning
Gradually thin the reinforcement schedule from dense, continuous reinforcement to leaner, intermittent schedules that approximate the natural reinforcement environment. Use specific criteria for thinning (stability at each level) and avoid abrupt changes that could produce resurgence of problem behavior or loss of the trained response.
Transfer to Natural Reinforcers
Shift from contrived reinforcers (tokens, edibles, stickers) to natural reinforcers (social praise, activity access, natural consequences) that will continue to be available after the formal intervention ends. Pair natural reinforcers with contrived reinforcers during training, then gradually fade the contrived reinforcers while maintaining the natural ones.
Build Fluency
Behaviors that are performed fluently (accurately and at an appropriate rate) are more resistant to distraction, more likely to be maintained over time, and more likely to be applied in novel situations. Use precision teaching and fluency-building procedures to bring target behaviors to fluent levels before beginning maintenance programming.
Involve Natural Change Agents
Train parents, teachers, siblings, peers, and other people in the client's natural environment to reinforce and maintain the target behavior. When natural change agents are skilled implementers, the behavior is reinforced consistently even after professional services end.
Conduct Booster Sessions
Plan periodic follow-up sessions to assess whether treatment gains are maintaining and to provide brief retraining if needed. Booster sessions can prevent small declines from becoming complete losses and can address new challenges that arise as the client's life circumstances change.
Common Pitfalls
- Train and hope: Assuming generalization and maintenance will occur without explicit programming.
- Over-standardized training: Training in only one setting, with one instructor, using one set of materials, which produces behavior that is too narrowly controlled.
- Abrupt termination: Ending services without a systematic transition plan that includes schedule thinning, natural reinforcer transfer, and caregiver training.
- Failing to assess: Not measuring behavior in generalization settings or during follow-up periods, so problems are not detected until they become severe.
BCBA Exam Focus
Maintenance and generalization are heavily tested on the BCBA exam. Know the definitions and types of generalization, the Stokes and Baer (1977) strategies, the difference between programming for generalization (proactive) and assessing for generalization (reactive), and how to design interventions that build in maintenance and generalization from the start. Exam scenarios may describe a behavior that fails to generalize or maintain and ask you to identify the missing programming element or recommend a strategy to address the problem.