Shape Severe Behavior Without an Extinction Burst

When a Board Certified Behavior Analyst (BCBA) takes a case where self-injurious behavior (SIB) means the kid goes to the hospital if a session goes sideways, an extinction burst is not a clinical event, it is an ambulance ride. The shaping process still has to move forward. The percentile schedule and a very high W value give you a way to dial the density of reinforcement up so high that the client never contacts extinction during the early cab steps in Skill-Based Treatment (SBT), and noncontingent reinforcement (NCR) keeps the room safe while you do it. This page sits where percentile shaping, NCR, and SBT cab steps actually meet, which is where the safety call gets made.

Why an extinction burst is sometimes a hospital trip#

For most learners, an extinction burst is a bad afternoon. The graph spikes, the team holds the line, the spike comes down, and the data file gets a note. For a kid with severe SIB, head banging, or eye gouging, that same spike is a trip to the emergency room. The clinical math has to change before the program starts. You do not get to "ride it out." The team cannot accept any path where the client briefly contacts extinction and tests the schedule with a louder version of the dangerous behavior.

That is the safety call sitting at the top of every cab step in SBT for these cases. Cab one is the easiest version of the contingency, where the client gets reinforcement for almost anything in the response class. Cab two adds a tiny piece of work. If the early cab steps run on a thin schedule, the client will contact extinction within seconds and the behavior the team was hired to reduce will get bigger before it gets smaller. That is the burst you cannot afford.

The fix is not a different intervention. It is a different dial setting on the same intervention. You keep shaping. You keep using the percentile schedule. You just start at a W value so high that contacting extinction is mathematically very unlikely.

How the percentile schedule lets you dial in safety#

The percentile schedule is the equation that tells you which responses get reinforcement based on the client's recent history. W is the slider. A high W means a very high probability that the client's next response gets reinforced. A low W means the bar moves up faster and the client contacts extinction more often.

A high W means a very high likelihood that reinforcement is acquired, think errorless learning at 90 to 100 percent. From the talk — Matt Harrington

A low W does the opposite, it makes the shaping bar move quickly and the client contacts extinction much more often. For a severe-behavior case, you start the early cab steps with W set near 0.9 or 1.0. That is errorless learning territory. The client basically cannot fail. The shaping process still moves, but it moves on a glide path, not a staircase. The team is buying time without giving up on progress.

This is the same equation the cluster covers on the percentile-schedule mechanics page. The difference here is the use case. You are not picking W to make learning efficient. You are picking W to keep a kid safe.

Setting W high for early CAB steps in SBT#

In the first two cab steps of SBT for a client with severe SIB, the team is teaching the kid that the room is now a place where good things happen and the dangerous behavior is no longer the path to those good things. That message only lands if the kid actually gets the good things, often.

If you have severe behavior that cannot handle an extinction burst, on cab one and cab two, you need a high density of reinforcement so it is slow and steady. From the talk — Matt Harrington

A practical starting point looks like this. On cab one, the client gets full access to the established reinforcer for any in-seat, calm response in the response class. W is at or near 1.0. On cab two, the team adds the smallest possible piece of work, a one-second tolerance window, a single touch on the worksheet, a single step toward the table. W stays high, around 0.9. The team is not trying to make the kid earn it. The team is teaching the kid that the contingency is rich, predictable, and worth staying in.

Two notes on running this in the room. First, the reinforcer has to be the actual functional one, not a candy bar the team grabbed because it was easy. Second, the team has to be ready to deliver it fast. A two-second delay at W of 0.9 is still a thin schedule from the client's point of view.

When to drop W and let the client move faster#

You do not leave W at 0.9 forever. That is not a shaping program, that is a sedation program. Once the data shows the client is calm, in the chair, and reaching for the reinforcer without any of the severe behavior, you start to drop W.

As we get up to cab five and cab six, we drop density of reinforcement because the client can handle it. From the talk — Matt Harrington

The decision rule is simple. You drop W when the data over the last several sessions shows two things at once. The dangerous behavior is at or near zero. And the client is contacting the reinforcer faster than the team can almost deliver it. That is the signal that the client has surplus capacity. They can handle the bar moving up.

Move W in small steps. Go from 0.9 to 0.8, then to 0.7, then to 0.6. Watch the data after each drop. If the dangerous behavior comes back, the team did not fail. The percentile schedule just told you that the client is not ready for that bar yet. Move W back up, run a few more sessions, and try again.

This is where the equation earns its keep. The team is not guessing. The team is reading a number and adjusting a dial.

Pairing the percentile schedule with NCR for self-injury cases#

For the most dangerous cases, the percentile schedule by itself is not enough. The room has to be saturated with reinforcement even when the client is not responding. That is NCR.

NCR is great as an intervention tool when the environment cannot sustain an extinction burst, like head banging. From the talk — Matt Harrington

NCR runs in the background. The functional reinforcer is delivered on a time-based schedule, not a response-based schedule. The kid gets it whether they earned it or not. While that is running, the percentile schedule with a high W runs on top of it for the actual shaping work. The client gets the rich, time-based reinforcement no matter what, and gets an extra hit of the same reinforcer for every response that meets the cab step.

That stacking is what makes severe SIB cases survivable in the room. The kid is never in a state where the room feels lean. The team has a constant supply of the functional reinforcer flowing. The shaping program adds the contingency on top without ever pulling the floor out.

The cluster covers the NCR-versus-extinction fidelity question on its own page. The reason it lives here is that the safety call is the same call. When the environment cannot sustain a burst, you do not run extinction. You run NCR plus a high-W percentile schedule.

Talking the tech through the density change#

The BCBA is rarely the person in the chair. The Registered Behavior Technician (RBT) or the trained tech is the person delivering reinforcement second by second. The density change is meaningless if the tech is not on board.

The conversation has three parts. First, explain why the density is high right now. Use plain words. Tell them the kid cannot afford to fail this week, so the program is set so the kid basically cannot fail. Second, tell them what the signal will be that the density is about to drop. Tell them what to look for in the data, calm behavior at zero, fast reaches for the reinforcer, no precursors. Third, tell them that if the dangerous behavior comes back, that is information, not a failure on their part.

A tech who knows the dial is being turned on purpose, and knows what data triggers each turn, will run the program clean. A tech who thinks the density is high because nobody trusts them yet will get tired of delivering reinforcement constantly and the schedule will drift.

What the data should look like before you change the dial#

You change W when the graph tells you to, not when the calendar tells you to. There are three things to look for before you turn the dial down.

The dangerous behavior is at zero or near zero across the last three to five sessions. The client is reaching for the reinforcer fast, which means the contingency is established. And the team can deliver reinforcement at the current density without delay. If any of those three is shaky, hold W where it is.

When you do drop W, change one variable at a time. Do not move W and add a new cab step in the same session. The whole point of the percentile schedule is that you know which change drove the result. Stacking changes throws that away.

The same logic runs in reverse. If the dangerous behavior comes back after a W drop, the program does not get scrapped. W goes back up. The shaping process keeps moving. The ethical guardrails of the talk apply here, cause no further harm, keep assent, build resistant repertoires. The dial is how you honor those guardrails on a kid who cannot afford a single bad afternoon.

Frequently asked questions#

What is an extinction burst and when is it dangerous?

An extinction burst is a temporary increase in the frequency, duration, or intensity of a behavior when the reinforcement that used to follow it stops. For most learners, it is a clinical event you plan around. For a client with severe SIB, aggression toward staff, or any behavior that can cause serious injury in seconds, the burst is a safety event. You do not run programs that produce a burst in those cases. You run programs that prevent the client from ever contacting extinction during the shaping process.

How do I pick a starting W value for severe behavior?

Start at or near 1.0 for cab one and around 0.9 for cab two. That is errorless learning territory. The client almost cannot fail, which is the point. Pick the W value based on what the environment can sustain, not on what the long-term program looks like. You can always drop W later. You cannot un-send a kid to the hospital.

Can I use the percentile schedule with SBT and CAB steps?

Yes, and for severe cases that is the cleanest way to run it. SBT gives you the cab progression, which is the shaping ladder. The percentile schedule gives you the rule for which responses on that ladder get reinforced. Set W high for the early cab steps, drop W as the client shows surplus capacity, and never change W and the cab step in the same session.

When should I lower the reinforcement density?

When the data shows three things at once. The dangerous behavior is at or near zero across several sessions. The client is contacting the reinforcer fast. And the team can deliver reinforcement at the current density without missing reps. If any of those is shaky, hold W where it is.

How is this different from errorless learning?

A high-W percentile schedule looks like errorless learning at the start, and that is on purpose. The difference is that the percentile schedule is built to move. Errorless learning often stays errorless. The percentile schedule has a dial. You start in errorless territory for safety, then turn the dial as the client earns the room to fail without hurting themselves.

Watch the full talk#

The talk walks through contingency space analysis, the matching law, and the percentile schedule as three foundational equations behind ethical behavior reduction. The shaping-without-a-burst use case is one slice of the percentile section. If you carry severe-behavior cases, this is the talk to put on before your next supervision call.