Progressive Prompt Delay for Mand Training: Step-by-Step

Progressive prompt delay is a teaching procedure where you start by prompting a mand at zero seconds (so the kid has no chance to fail), then slowly add wait time between the moment the kid wants something and the moment you give a model. The classic example, taught in Bowen and colleagues' 2012 study on functional communication, is a kid walking to a fridge and pointing at the juice. The therapist signs "juice" the second the kid points. The kid copies the sign. The kid drinks juice. That single trial is a zero-second progressive prompt delay, and it is the seed of every mand a kid will ever produce with this method.

This page walks the whole procedure: the first trial, the criterion to fade the prompt, when to step the delay back down because the kid is missing, the whiteboard tally trick that keeps your RBT (registered behavior technician, the person actually running the session) honest, and when to even start counting a trial in the first place.

Why progressive prompt delay beats generic least-to-most#

Least-to-most prompting (start with the smallest hint, add bigger ones if the kid does not respond) is the default in a lot of EI (early intervention) clinics. It sounds humble. The problem is that it lets the kid sit in silence, then guess, then get a partial prompt, then maybe guess wrong again. That is a lot of chances to practice the wrong response.

Progressive prompt delay flips the order. The first trial is errorless, meaning the prompt comes so fast the kid cannot get the answer wrong. The teaching only fades after the kid has stacked a few clean wins. You are not testing whether the kid knows the mand. You are building a strong, fast, accurate response and then slowly removing your help.

That matters most with manding. A mand is a request driven by what the kid actually wants right now, which behavior analysts call the MO (motivating operation, the thing that makes a reinforcer powerful in that moment). Juice when thirsty. Bubbles when bored. If the kid practices a wrong response while the MO is hot, you have just paired the wrong word with the strongest reinforcer in the room. Errorless teaching protects against that.

Step 1: the zero-second prompt (model and reinforce)#

The first trial sets the whole program. You wait for the MO to show up on its own. The kid walks toward the thing, looks at the thing, reaches. Then you prompt instantly, before the kid has time to do anything else.

the kid walks over, he maybe points to the juice and the therapist immediately says, immediately signs the sign for juice... And then the kid signs juice and they get their juice. That's a zero-second progressive prompt delay. From the talk — Matt Harrington

Three things have to be true on that first trial. The kid has to actually want the item (do not run trials with cold goldfish at the end of snack). The prompt has to be the full prompt, the one you are most confident the kid can copy (a motor imitation of the sign, a full vocal model, a button press, whatever modality you picked). And the reinforcer has to land within a second or two of the response. Juice means juice, not "good job, here is your juice in a minute."

Run this trial multiple times in the same session. You are not waiting for mastery yet. You are building a reliable two-step pattern: kid shows MO, then immediately produces the mand with help.

Step 2: when to move to a two-second delay#

The criterion for fading is small and concrete. After three correct independent or near-independent responses in a row, you add a two-second wait between the MO and your model prompt.

it would go from zero seconds, that immediate motor imitation prompt, the model prompt, if you will, and then three, I think it was three correct, and then you go to two seconds, and then three correct, and then you go to six seconds. From the talk — Matt Harrington

Two seconds feels short on paper. It is not. To a kid who has only ever heard the sign as a model, two seconds of silence after they reach for juice is a real opportunity to produce it on their own. This is called a transfer trial: the response transfers from prompted (you sign it) to independent (the kid signs it without you). When they produce it, you still reinforce immediately. Juice still means juice.

If the kid produces the mand independently inside the two-second window, count it as correct. If the two seconds run out and you have to model, count it as a prompted response. Three independent in a row is the gate.

Step 3: stepping up to five or six seconds#

After three independent responses at two seconds, you step up again. The Bowen procedure jumps to six seconds. Six is a long wait for an early learner. Most kids who can hold the response at two will surprise you and hold it at six too, but not always.

Six seconds gives the kid enough time to scan the environment, register the MO, retrieve the mand, and produce it. That is most of what we care about. A kid who can mand for juice with a six-second delay between wanting it and asking is functionally communicating. You can start to think about generalization, which is different therapists, different settings, different times of day, once you have stable independent responses at this final delay.

When to step the delay back down#

The Bowen procedure is a one-way staircase. Matt's clinical version is not. If a kid is at six seconds and starts missing (silence, wrong sign, reaching without manding), you do not keep banging on six. You drop the delay back down.

for every three independent or, you know, correct imitation, then I want you to progress the delay two more seconds... having a flexible criteria that both goes up and goes down and having a tech who really understands and is able to keep track of that arrangement. From the talk — Matt Harrington

The rule we use: three independents in a row, move up two seconds. Three misses in a row, drop back two seconds. That cuts the trip back to errorless teaching short and keeps the kid succeeding.

Some kids will yo-yo for a week before they hold six seconds. That is fine. The dance up and down is the program working. What you are trying to avoid is a long run of misses at the top of the staircase, because every miss is a chance to pair the MO with the wrong response.

Training your RBT with a whiteboard tally#

The procedure dies if the tech running the session cannot remember whether the last three trials were independent or prompted. You need a visible scoreboard.

I've used whiteboards before, that's been pretty helpful to just let the text quickly tally one dot, two dot, three dot, and then, all right, fade up, fade down, et cetera. From the talk — Matt Harrington

A small whiteboard in the corner of the therapy room works. One dot per independent trial. Wipe it clean every time you fade up or down. Some teams use clicker counters on a lanyard. Some use a sticky note on the back of a clipboard. The tool is less important than the rule: the criterion lives outside the tech's head.

When you train a new RBT on this program, do not start with the kid. Run the procedure on a stuffed animal with another tech. Let them get the rhythm of "kid wants thing, count to two, model if needed, mark the board." Then let them try it on a real session with you watching.

When to start the trial (and when not to)#

A common mistake is treating every kid behavior as a manding trial. Kid wanders toward the toy shelf, that's a trial. Kid looks at the window, that's a trial. Kid sneezes, somehow that's a trial. This wrecks your data because the denominator (number of opportunities) explodes and your percent-independent number swings around for no reason.

You need a clear cue for the tech that says "this counts." It could be the kid walking up and clearly orienting toward the item. It could be the tech holding the preferred item visibly. It could be a specific routine (snack time only).

The point is that not every glance is a trial. If your data sheet is full of "prompted" trials because the kid was barely interested, you are teaching to a weak MO and getting weak data.

Once the trial is started, the tech runs the delay, scores the response, marks the whiteboard, and moves on. No trial without a clear cue. No cue without a real MO.

Frequently asked questions#

What if my client never prompts independently even at zero seconds?

Check the modality. If you are running a sign and the kid does not have the fine motor skills to produce the sign, no amount of prompt delay will help. Same with vocal models for a kid who is not yet echoic, or a PECS-style picture exchange for a kid who does not yet scan an array. Drop back to prerequisites (imitation, echoics, matching) and pick a modality the kid can actually produce. The prompt delay procedure assumes the response is already in the kid's repertoire as an imitated or modeled behavior.

How many sessions before I should move from two seconds to six?

There is no fixed session count. The criterion is the criterion: three independent responses in a row at two seconds. Some kids hit that in one session. Some take ten. Use the data, not the calendar. If a kid is stuck at two seconds for more than a week, the problem is usually one of three things. Weak MO during the trials. The wrong modality. Or the tech accidentally prompting (eye gaze, head tilt) before the two seconds are up.

Should the prompt delay reset at the start of every session?

Not all the way to zero. Pick up where the kid left off the day before, but be ready to drop down two seconds if they miss the first couple of trials. Mornings are weird, the MO is different, the room smells different. Treat the first three trials of any session as recalibration data, not as the start of a fresh staircase.

Run the procedure with the rest of the program#

Progressive prompt delay is one of five days of content in "5 Days of Manding Mastery." The other days cover modality selection, mand-tact programming order, manding for information, and FCT (functional communication training, where you replace problem behavior with a mand). Watch the full course to put this procedure in context, then come back to this page when you are setting up the next program.