Why Generalization Matters
Teaching a skill in a clinic is only the first step. If a child can label colors with a therapist at a table but not with parents at home, the program has not achieved its goal. Baer, Wolf, and Risley listed generality as one of the seven dimensions of ABA: behavior change is general if it lasts over time, appears in a variety of environments, or spreads to a variety of related behaviors. These three outcomes correspond to maintenance, stimulus generalization, and response generalization.
Stimulus Generalization
Stimulus generalization occurs when a behavior that has been reinforced in the presence of one stimulus (or stimulus class) occurs in the presence of other, similar stimuli. The more similar the new stimulus is to the training stimulus, the more likely the behavior will occur.
Examples
- A child who learned to say "dog" to a golden retriever says "dog" to a poodle and a beagle.
- A learner who was taught to wash hands at the clinic sink washes hands at home.
- A student who raises a hand in math class raises a hand in science class.
- A client who learned to greet a therapist greets other adults.
Stimulus generalization is often described as setting or situation generalization when it refers to behavior occurring in new settings, with new people, or with new materials. The terms are used interchangeably in many contexts, but exam questions typically refer to stimulus generalization for the general phenomenon.
Generalization gradients
A generalization gradient is a graph that shows how responding changes as stimuli become less similar to the training stimulus. A flat gradient means responding occurs at a similar rate across many stimuli (broad generalization). A steep gradient means responding drops rapidly as stimuli differ from the training stimulus (narrow generalization, strong discrimination).
Overgeneralization and discrimination
Generalization is not always desirable. A child who calls a cat "dog" has overgeneralized. Discrimination training, which reinforces the response in the presence of one stimulus and not others, narrows generalization to the appropriate stimuli.
Response Generalization
Response generalization is the extent to which a learner emits untrained responses that are functionally equivalent to the trained response. When reinforcement of one response form produces an increase in other forms that serve the same function, response generalization has occurred.
Examples
- A learner is taught to request a snack by saying "Snack, please," and later says "I want a snack" and "Can I have a snack?" without training.
- A child taught to complete addition problems with a number line solves them using fingers or mental math.
- A student taught to ask for a break by raising a card begins to use a gesture or a spoken request.
- An employee trained to greet customers with "Hello" begins to use "Good morning" and "Welcome in."
Response generalization can also include response induction, where a change in one behavior produces changes in another related behavior, which can be desirable or undesirable. If a client learns a new sentence structure, response generalization means related structures also show up.
Maintenance
Maintenance (response maintenance) is the extent to which a learner continues to perform the target behavior after some or all of the intervention has been terminated. A learner who still uses a mand six months after therapy ends shows maintenance. This is different from generalization across settings or responses, but all three are part of generality.
Comparing the Types
- Stimulus generalization: The same response occurs in new stimulus conditions.
- Response generalization: New response forms occur under the same stimulus conditions and serve the same function.
- Maintenance: The response continues over time after intervention is removed.
A quick test: ask what changed. If the stimulus changed and the response stayed the same, it is stimulus generalization. If the response changed and the function stayed the same, it is response generalization. If time passed, it is maintenance.
Programming for Generalization: Stokes and Baer
In 1977, Stokes and Baer argued that generalization should be programmed, not left to chance (the "train and hope" approach). They described a set of strategies, which Stokes and Osnes later refined into a framework.
- Exploit current functional contingencies: Teach behaviors that will be reinforced naturally in the environment, and recruit natural reinforcers.
- Train sufficient exemplars: Teach with enough examples of stimuli (different dogs, different teachers) and responses (different ways to ask) so that the learner generalizes to new examples.
- Train loosely: Vary instructions, materials, and settings, and avoid rigid, repetitive teaching that makes the behavior depend on one specific condition.
- Use indiscriminable contingencies: Deliver reinforcement intermittently or with delays so the learner cannot tell when it will be delivered, which promotes persistence.
- Program common stimuli: Incorporate stimuli from the natural setting into training, such as having a parent present or using home materials in therapy.
- Mediate generalization: Teach a self-management strategy, a rule, or a verbal mediator the learner can carry across settings.
- Train to generalize: Reinforce generalization itself, such as reinforcing novel responses or performance in new contexts.
Measuring Generalization
Generalization should be measured, not assumed. Use probes: brief tests of the behavior in untrained conditions without reinforcement or with minimal reinforcement.
- Probe across settings (clinic, home, school, community).
- Probe across people (therapist, parents, peers).
- Probe across materials (different items, pictures, and real objects).
- Probe across instructions (different wordings of the same demand).
- Probe for response variations (different forms of the same function).
- Probe over time for maintenance.
If generalization is poor, add training in the untrained conditions, or choose a different generalization strategy and probe again.
Clinical Example
A therapist teaches a child to say "hi" to a peer. Training occurs in the clinic with multiple staff members and different peers across rooms, with greeting responses varied among "hi," "hello," and "hey." The parent joins sessions and practices at home. The therapist probes in the school lunchroom with an unfamiliar peer. The child greets the new peer with "hey" (an untrained form). The new setting and peer indicate stimulus generalization, and the untrained word indicates response generalization. Three months later, the child still greets peers, which is maintenance.
Exam Strategies
- Do not confuse stimulus and response generalization. Identify which aspect changed.
- Remember that discrimination and generalization are opposites in many ways, but both are needed.
- Know Stokes and Baer's strategies, especially sufficient exemplars, common stimuli, and natural contingencies.
- When a question asks what to do to promote generalization, choose answers that program for it from the beginning and measure with probes.
Key Takeaways
Stimulus generalization is the same response in new stimulus conditions. Response generalization is new, functionally equivalent responses under similar conditions. Maintenance is persistence over time. Plan for all three using strategies such as sufficient exemplars, loose training, common stimuli, and natural reinforcers, and verify them with probes.