Generalization and Maintenance: Making Skills Travel and Last

A skill that appears only at one table, with one person and one set of materials, is not very useful yet. Once a learner starts succeeding, two questions matter:

• Does the skill travel? This is generalization: the skill shows up with people, in settings, or with materials that differ from where it was taught.
• Does the skill last? This is maintenance: the skill still shows up later, after active teaching has moved on.

Some sources, including IRIS and AFIRM's reinforcement glossary, use "generalization" as a broad label that includes lasting over time. The RBT exam outline lists generalization procedures and maintenance procedures as separate tasks, so this lesson treats them separately.

As a technician, you carry out the supervisor's plan for both. You also record what actually happened.

By the end, you should be able to:

• Tell acquisition, maintenance and generalization apart in a short scenario.
• Explain how varied teaching supports transfer, and why transfer still has to be observed.
• Run a check the way the plan specifies, and record any help you gave.
• Record observations rather than conclusions, and avoid treating missing or differently measured data as equal.
• Choose a next step when performance changes, without changing the plan yourself.

Terms you need:

• A prompt is help that makes a correct response more likely, such as pointing or saying the first sound.
• Reinforcement is a consequence after a behavior that makes that behavior more likely in the future.
• A probe is a planned check of performance. The plan says how it runs, including whether any help is allowed.

Acquisition and maintenance ask different questions

Acquisition procedures teach a skill the learner cannot yet do reliably. In plain terms, we are still building it. Expect frequent practice, prompts that are faded, and reinforcement delivered as the plan specifies.

Maintenance procedures check and practice a skill the learner already performs to the supervisor's criterion. In plain terms, we are keeping it. Practice is less frequent, and planned reinforcement is often reduced.

Here is an example. Rio is learning to say "help" when a container won't open. During acquisition, the technician sets up several stuck containers each session and prompts as the plan directs. Rio meets the mastery criterion written in the plan. The plan now says to probe once a week.

The procedure changed because the question changed. It was "Can Rio learn this?" Now it is "Does Rio still do it?"

There is no universal mastery number, such as "80% across three sessions." The supervisor sets the criterion for each target, and you follow that criterion.

Now change one fact. On a weekly probe, Rio pulls at the jar and never asks. The procedure is still maintenance, but the data suggest the skill may not have lasted. You do not move Rio back to daily teaching on your own. You record what happened and report it. The supervisor decides whether to return to acquisition.

Checkpoint 1: A learner has met criterion for washing hands, and the plan says to check it every Monday. On Monday, the technician also starts teaching a new target: drying hands with a paper towel. Which part is acquisition, and which is maintenance?

Answer: Checkpoint 1

The handwashing check is maintenance, because an established skill is being checked. Drying hands is acquisition, because it is new. One session can contain both. Label each target by the question being asked about it, not by the day or the room.

Generalization across people, settings and materials

Rio learned "help" with one technician, at a clinic table, using a snack jar. Generalization would be Rio asking a parent for help with a lunchbox zipper at home. That one example differs on all three dimensions:

• Person: a parent instead of the technician.
• Setting: home instead of the clinic.
• Material: a zipper instead of a jar.

A plan supports transfer by varying these features during teaching. Several people take turns. The learner practices with jars, bags and bottles. Some practice happens inside real routines, such as snack time.

Varying the features that don't matter helps the learner respond to the one that does: something won't open. Otherwise the learner may respond only to one particular jar. AFIRM's naturalistic intervention guide also describes building teaching into everyday routines and using reinforcers that occur naturally in them.

The tempting mistake is to assume transfer. Varied teaching makes generalization more likely, but it does not prove it. Suppose Rio was never observed asking a grandparent at a park. Then you cannot record that Rio does. You can record only what was observed.

Natural consequences still count

IRIS describes maintenance as a behavior continuing over time "even when reinforcement is not available." The expert interviews on the same page explain what this means in practice. The aim is behavior that continues without anyone intentionally adding reinforcement. The experts recommend:

• teaching behaviors that pay off in real life, and
• fading extra supports gradually, but not so fast that the student stops using the skill.

So maintenance often means the extra, planned reinforcement can be reduced. A natural consequence is a payoff that comes from the behavior itself: Rio asks, the jar opens, and Rio gets the snack. Maintenance does not, by itself, require removing that payoff, and when it is available it often remains.

Real life will not pay off every time, though. The IRIS interviews describe planned exposure, after mastery, to times when the behavior goes unreinforced, along with teaching what to do then. The supervisor's plan sets how and when reinforcement is reduced and what happens when it is unavailable.

Checkpoint 2: A technician says, "Rio's 'help' is in maintenance now, so I'll make sure asking never gets the jar opened." What is wrong?

Answer: Checkpoint 2

Maintenance does not, by itself, call for blocking the natural consequence that makes asking useful. If the plan says so, maintenance may reduce added reinforcement, such as praise or tokens. If the plan includes times when asking does not pay off, the supervisor plans that; the technician does not decide it alone. If the plan is unclear, ask the supervisor.

Run probes as planned, and record any help

An independent probe asks one question: "What can the learner do without help, under these conditions?" Not every probe is run this way. The supervisor's plan says how each probe runs and is scored, including whether prompts or teaching are allowed.

Whatever the design, help changes what a response shows:

• If you prompt, the response shows prompted performance, not independent performance.
• If you reteach partway through, the trials that follow show performance after teaching.

So follow the plan, and record all the help you gave.

AFIRM's prompting guide records unprompted correct responses separately from prompted correct responses for this reason: they show different things. Write "correct with a pointing prompt," not just "correct."

Follow the plan's rules for errors during a probe. If the plan says to end the probe and teach, record the probe result first. Then record the teaching separately.

Record what you saw, and compare like with like

Three data mistakes can make generalization and maintenance look better or worse than they really are.

1. Recording a conclusion as an observation. "Rio has generalized" is a conclusion. Compare it with this: "At home, Rio said 'help' and handed the lunchbox to a parent within 5 seconds, with no prompt." That is an observation. Write the observation, and let the supervisor draw the conclusion.

2. Recording missing data as zero. If the home probe was canceled, the result is "not probed," not 0%. Zero means Rio had chances and did not respond correctly. A blank means we do not know.

3. Treating different measures as the same.

• At the clinic, Rio asked on 4 of 5 opportunities: 4 ÷ 5 = 0.8, which is 80%.
• At home, Rio asked on 1 of 1 opportunity: 1 ÷ 1 = 1, which is 100%.

The home percentage is correct arithmetic, but it is not "better." It rests on a single chance, so report it as 1 of 1. Similarly, "3 requests in a 30-minute session" counts per stretch of time, while "3 of 4 opportunities" counts per chance. Dates matter too. A clinic score from March and a home score from June differ in both setting and time.

Before you compare two numbers, check three things: what was counted, out of what, and when.

Checkpoint 3: During a probe at school, Rio stalls. The technician points at the jar, and Rio says "help." The note reads: "Generalized to school: 100%." Find the problems.

Answer: Checkpoint 3

First, pointing was a prompt, so the response should be recorded as correct with a pointing prompt, not as independent. Second, "generalized" is a conclusion. The note should record the observed response, the setting, the person and the prompt. Third, the note hides the denominator. One of one is 100%, but report it as 1 of 1, because one opportunity is weak evidence of reliable performance.

Checkpoint 4: Sam learned to say "my turn" during games. The plan had three staff members take turns and used board games, card games and a ball game. A teammate writes three notes:

• "Sam practiced with lots of people and games, so Sam has generalized to recess."
• "Thursday recess probe: 0%." (The probe was canceled because of rain.)
• "Home is worse than clinic." (Clinic: 8 of 10 opportunities, in May. Home: 4 times in one hour, in July.)

What is wrong with each note, and what should be recorded instead?

Answer: Checkpoint 4

• Recess: Varied people and games make transfer more likely, because Sam learns to respond to a chance to take a turn rather than to one game or one person. But no one observed Sam at recess. Record nothing about recess until it is observed; the supervisor can decide whether to schedule a probe.
• Thursday: The probe did not happen, so record "not probed." Zero would mean Sam had chances at recess and did not respond correctly.
• Home versus clinic: 8 of 10 is 80% of opportunities. "4 times in one hour" counts per stretch of time, not per chance, and it comes from a different month and setting. Neither number is higher or lower than the other in any meaningful way. Report both with what was counted, out of what, and when.

Put it together

For each case, name what is being tested and say what the technician should do next.

1. Six weeks after mastering greetings with clinic staff, Mia stops greeting during a scheduled monthly probe. The probe uses the same staff in the same room as when Mia was taught.
2. Theo learned to sort socks by color at the clinic. A caregiver reports that Theo sorted towels by color at home without help. The technician did not see it.
3. Ana has just started learning to wait for a turn on a swing, and the plan uses prompts.

Answer: Put it together

1. Maintenance. The staff and room match the teaching conditions; the new factor is time. Record the result accurately and report the change to the supervisor promptly. Do not restart teaching on your own.
2. Possible generalization across setting (home) and materials (towels). The report does not say who was present, so it does not show a change in person. It is also not your observation. Document it as a caregiver report and pass it to the supervisor, who may schedule a probe.
3. Acquisition. The skill is new and taught with prompts. Prompted responses are expected; record them as prompted.

Remember

• Acquisition builds a new skill. Maintenance checks and practices a skill that has met the supervisor's criterion.
• Generalization means the skill shows up with new people, settings or materials. Maintenance means it lasts over time.
• Varied teaching helps transfer, but only observation shows that transfer happened.
• The plan says how a probe runs. Help given during any probe changes what the data show, so record it.
• Maintenance often reduces added reinforcement, as the plan specifies. It does not by itself require removing natural payoffs; follow the plan when reinforcement is reduced or unavailable.
• Record observations, mark missing data as missing, and compare only like measures. Keep the numerator and denominator.
• Report changes in performance. The supervisor changes the plan.

Sources

• Behavior Analyst Certification Board. (2023, updated 09/2024). RBT Test Content Outline (3rd ed.). https://www.bacb.com/wp-content/rbt-outline-3rdEd/
• Behavior Analyst Certification Board. (2021). RBT Ethics Code (2.0). Technicians follow supervisor direction and document accurately. https://www.bacb.com/wp-content/rbt-ethics-code
• IRIS Center, Vanderbilt University. Behavior module page on generalization, including the Staubitz and Allen interview transcripts. https://iris.peabody.vanderbilt.edu/module/bp/cresource/q1/p06/
• Amsbary, J., & AFIRM Team. (2025). Naturalistic Intervention. UNC Frank Porter Graham Child Development Institute. https://afirm.fpg.unc.edu/wp-content/uploads/Naturalistic-Intervention-Brief-Packet-Amsbary-AFIRM-Team-Updated-2025.pdf
• Sam, A., & AFIRM Team. (2024). Prompting. UNC Frank Porter Graham Child Development Institute. https://afirm.fpg.unc.edu/wp-content/uploads/Prompting-Brief-Packet-Sam-AFIRM-Team-Updated-2024.pdf
• Sam, A., & AFIRM Team. (2024). Reinforcement. UNC Frank Porter Graham Child Development Institute. https://afirm.fpg.unc.edu/wp-content/uploads/Reinforcement-Brief-Packet-Sam-AFIRM-Team-Updated-2024.pdf

Independent study resource

This study resource is not affiliated with, sponsored by, or endorsed by the Behavior Analyst Certification Board (BACB). BACB, Registered Behavior Technician and RBT are registered trademarks of the Behavior Analyst Certification Board. Source citations do not imply endorsement by their authors or publishers.