Chaining: Forward, Backward, and Total-Task Methods for Building Behavior Sequences

How a behavior chain works

Skinner's analysis of the rat in the operant chamber already treated the "simple" lever press as a chain: the rat faces the lever, approaches, raises a paw, presses, hears the food magazine operate, lowers its head to the tray, seizes the pellet, and eats. Each response changes what the rat sees, feels, or hears, and each change sets the occasion for the next.^[1] Keller and Schoenfeld drew the sequence out link by link in 1950.^[2]

The stimulus at each junction plays two roles. The sound of the magazine is the discriminative stimulus for going to the tray: in its presence, tray-approach is reinforced; in its absence, it is not. The same sound is a conditioned reinforcer for pressing: a stimulus that reliably precedes reinforcement acquires the power to reinforce the response that produces it, and Skinner showed in 1938 that the magazine sound alone, once paired with food, could condition a new response.^[1] Nothing but the last link touches food, yet every link is reinforced, because every link produces the stimulus the next link needs.

The laboratory version is the chained schedule: a pigeon completes several schedule components in sequence, each signaled by its own key color, with food only after the last. Kelleher and Gollub's 1962 review drew a lesson from these schedules that every trainer eventually rediscovers: responding is strongest in the link closest to food and weakest in the links farthest from it, and in long chains the early links can fail outright.^[4]

The idea is older than the vocabulary. In 1890 William James wrote that in habitual action "what instigates each new muscular contraction to take place in its appointed order is not a thought or a perception, but the sensation occasioned by the muscular contraction just finished."^[5] That is a chain described from the inside (the chapter is in the library); what Skinner added is that the sensation does not merely trigger the next movement, it pays for the last one.

Task analysis: the first step

A chain cannot be taught until it has been written down. A task analysis breaks a skill into its component responses in the order they occur, each ending in a detectable stimulus change. Cooper, Heron, and Heward describe three ways to build one: watch a competent person perform the task, ask people who do it well, or perform it yourself and note each thing that changes.^[3] The third catches the steps experts no longer notice.

The right grain depends on the learner: seven steps suit a preschooler, while for a learner with a severe disability "rub hands together" may need to become four steps. Cuvo, Leaf, and Borakove's 1978 restroom-cleaning study broke the job into subtasks — mirror, sink, toilet, floor — and each subtask into component responses with a definable start and finish.^[6] The test of a task analysis is whether a trainer who has never met the learner could score each step as done or not done.

A task analysis also makes a baseline probe possible. In a single-opportunity probe the trial stops at the first error and everything after it is scored as not done; in a multiple-opportunity probe the trainer quietly completes any failed step so the learner can attempt the next, which shows exactly which steps are already in the repertoire.^[3] The probe decides where teaching starts and is the baseline against which progress — usually the percentage of steps done independently — is measured.

Forward chaining, backward chaining, and total-task presentation

Once the steps are written there are three ways to link them; the difference is where the learner starts and where the reinforcer falls. In forward chaining the learner masters the first step, then the first two, and so on, with the trainer completing the rest. In backward chaining the trainer performs every step except the last, the learner performs the last step and contacts the natural reinforcer, and earlier steps are added one at a time, so every trial ends with the finished task. In total-task presentation the learner attempts every step on every trial and the trainer prompts the steps that fail. A variant, backward chaining with leaps ahead, skips steps the probe showed the learner can already do.^[3] Reviews of the applied literature have found all three effective and none consistently superior.^[7]

Backward chaining's advantage is structural: the new step always leads directly into steps already mastered and then into the reinforcer. Forward chaining's is that the learner initiates and practices the chain in the order it will be used; in the classroom, where most academic procedures are chains whose order carries meaning, it is the default.^[3]

What the research shows

Spooner and Spooner reviewed the chaining studies available in 1984, most with learners with severe disabilities, and found no method that consistently beat the others; the studies differed in learners, tasks, prompting, and measures.^[7] Slocum and Tiger taught four children arbitrary sequences of simple actions, some by forward chaining and some by backward chaining, then assessed which procedure each child preferred. Both methods worked, neither was reliably faster across the four children, and the children had preferences that differed from child to child.^[8] Ash and Holding, with adults learning a sequence of key presses, found that the group trained by backward chaining outperformed the group trained by forward chaining.^[9] The honest summary: all three methods teach chains, backward chaining has an edge in some tasks, and the learner's preference is a legitimate tie-breaker.

The applied literature is where chaining earns its keep. Cuvo and colleagues taught adolescents with intellectual disabilities to clean a restroom; the skills generalized to an untrained restroom and were maintained at follow-up.^[6] Bellamy, Horner, and Inman built a vocational program for adults with severe intellectual disabilities around task-analyzed assembly work presented as a total task.^[12] Wilson, Reid, Phillips, and Burgio taught residents of an institution family-style dining — passing serving dishes, serving themselves — and reported, as the title says, both effects and noneffects: the skills were acquired; some broader changes the authors looked for did not follow.^[13] Test, Spooner, Keul, and Grossi task-analyzed the public pay-phone call and taught it to adolescents with severe disabilities.^[14] Jerome, Frantino, and Sturmey used backward chaining with errorless prompting to teach adults with developmental disabilities to reach a chosen website.^[10]

The successful studies did not stop at acquisition: they probed generalization, checked maintenance, and trained where the chain would be used.^[6][13] More on applied behavior analysis ›

Chaining in dog and animal training

Animal trainers work with chains constantly, and they almost always build them backward. A formal retrieve is a chain: sit and wait, run out on cue, pick up the object, return, hold, release to the hand. Trainers teach the hold and release first, then the return, then the pick-up, then the run-out, so that every new link leads into links the dog already performs fluently.^[11] Karen Pryor recommends the same for people: learn a piece of music or a speech from the end, so that practice always moves from the unfamiliar into the familiar.^[11]

Trainers also see what happens when an unwanted response gets into a chain: if a dog barks and then sits, and the sit is reinforced, the bark is now the first link and will be strengthened with the sit. The cure is differential reinforcement of the clean sequence: the sit that follows no bark is reinforced, the sit that follows a bark is not. A marker signal — a click or a word — is useful in chains for the same reason: it is a conditioned reinforcer that can be delivered at the exact link that needs strengthening without stopping the chain to deliver food.^[11] Reinforcement-based dog training ›

How to build a behavior chain

1. Write the task analysis. List the steps in order, each ending in a stimulus change you can see. Check it by doing the task yourself and watching someone competent do it. If a step contains two stimulus changes, split it.^[3]
2. Run a baseline probe. A multiple-opportunity probe tells you which steps the learner already has; a single-opportunity probe tells you how far they get unaided. Record the percentage of steps performed independently; that is your progress measure.^[3]
3. Choose a method. Most steps already present: total-task presentation. A long chain, a learner who tires easily, or a reinforcer that is the natural end of the task: backward chaining. Early steps present and an order that carries meaning: forward chaining. If two seem equal, let the learner's preference decide.^[8]
4. Prompt and fade. Decide the prompt hierarchy before the first trial — most-to-least for a learner who makes many errors, least-to-most or time delay for one who can sometimes beat the prompt — and fade every prompt until the product of the previous step, not your voice or hand, is the cue.^[3]
5. Reinforce the link being taught, then let the learner finish. During acquisition the new step is too far from the end of the chain for the terminal reinforcer to reach it. Mark it immediately with praise, a token, or a click — positive reinforcement delivered where the link happens — then let the learner run the rest of the chain to the natural reinforcer, and drop the added reinforcers one at a time as the link stimuli take over.^[4]
6. Thin, vary, and check back. Once every step is independent, thin any added reinforcement, practice with different materials and settings so the cue is "wet hands" and not "this sink," and probe again after a few weeks.^[6]

Shaping vs. chaining

The shaping page gives the one-line contrast: shaping builds a new form of a single response, chaining links existing responses into a sequence. The deeper difference is what the trainer manipulates. In shaping, the criterion for reinforcement moves along a dimension of one response — louder, longer, closer. In chaining, the criterion is fixed by the task analysis; what moves is the number of steps the learner performs without help, and the tools are prompting, fading, and conditioned reinforcement.

The two meet constantly. A learner working through a chain will hit a step they cannot perform in any form — the child who can do every part of shoe-tying except pulling the loop through — and that step is shaped or physically prompted on its own, then put back into the chain. Ask, of any skill, whether the problem is form or order. Form problems are shaping problems; order problems are chaining problems; and attaching a new habit to the end of an old one — coffee, then vitamins — is chaining under another name, with the end of the old routine as the cue for the new one. Building habits ›

Common chaining mistakes

• Steps that are too big. A step that hides two or three stimulus changes stalls the learner in the middle, where nothing cues the next move. If a learner keeps failing the same step, it is usually two steps.
• Prompts that are never faded. The learner performs the whole chain — as long as the trainer says "now rinse." The trainer's voice, not the lather, has become the discriminative stimulus, and the chain will not run when the trainer leaves.^[3]
• Reinforcing only the end. A single reinforcer at the end of a long chain reaches the last few links well and the first few barely, as chained schedules show in the laboratory.^[4] Reinforce the link being taught where it happens; thin later.
• Reinforcing an unclean chain. Whatever happened just before the reinforcer is in the chain, wanted or not — the bark before the sit, the flourish before the signature. Reinforce only the clean sequence.
• Skipping the probe. Teaching steps the learner already has wastes sessions; assuming steps they do not have puts them on extinction at that step every trial.
• Changing the cues without noticing. A new soap dispenser, a different sink, a phone with the buttons elsewhere: each removes a stimulus the chain depended on. If a chain breaks at one step in a new setting, look for the stimulus that changed.^[6]
• Stopping at acquisition. A chain that runs on trainer-delivered praise and tokens is not finished until it runs on its natural end, where it is needed, weeks later.

What the evidence does not show

It does not show that one method is best. The comparative studies are few, small, and mixed: Spooner and Spooner found no consistent winner, Slocum and Tiger found the answer varied by child, and Ash and Holding's backward-chaining advantage was for one keyboard task in adults.^[7][8][9] Nor does the evidence say how to predict which method a given learner will do best with; pick one, measure, and switch if the data say so.

It does not show that the dual-function account is the whole story. That each link stimulus is both discriminative stimulus and conditioned reinforcer is inferred from performance, and Kelleher and Gollub's review made clear that conditioned reinforcing strength falls off with distance from the terminal reinforcer, so "every link reinforces the one before" is true in kind but not in degree.^[4]

It does not show that chains generalize or persist on their own. The studies that found generalization and maintenance had built them in, and Wilson and colleagues' "noneffects" are a reminder that teaching a dining chain teaches a dining chain; anything else hoped for has to be measured.^[6][13] And nearly all of the applied evidence comes from learners with developmental disabilities. Chaining is used every day in classrooms, sports coaching, and workplace training, but the controlled evidence for those uses is thin.

Key takeaways

• A behavior chain is a sequence in which each response produces the discriminative stimulus for the next and a conditioned reinforcer for the one before. Only the last link contacts the terminal reinforcer.
• Chaining begins with a task analysis — steps in order, each ending in a detectable stimulus change — and a baseline probe of which steps the learner already has.
• Forward chaining teaches from the first step, backward chaining from the last, and total-task presentation runs the whole chain with prompts on failed steps. Studies find all three effective and none consistently best.
• Fade prompts until the product of each step, not the trainer, is the cue, and reinforce the new link where it happens; a reinforcer at the end of a long chain barely reaches its beginning.
• Shaping changes the form of one response; chaining changes the order of many. Ask whether the problem is form or order, and combine the two when a chain contains a step the learner cannot yet perform.
• Generalization and maintenance are programmed, not automatic: vary materials and settings, thin added reinforcers, and probe again weeks later.

Check yourself

A child performs every step of hand-washing correctly, but only after a parent names each step aloud. Has the chain been learned?

Not yet. The parent's instructions, not the products of each step, are the discriminative stimuli, so the chain will not run when the parent is absent. The prompts have to be faded until running water cues wetting the hands and lather cues rinsing. Score a step as mastered only when it follows the previous step's result rather than a prompt.

A trainer teaching a dog a ten-obstacle agility sequence reinforces only at the finish. The last three obstacles are fast and clean; the first three are slow and often skipped. What is going on?

A reinforcer at the end of a long chain reaches the links closest to it and barely reaches the earliest ones, the same pattern seen in laboratory chained schedules, where responding is weakest in the links farthest from food. During training the early links need their own conditioned reinforcement, such as a marker, and back-chaining the sequence would let each newly added obstacle lead directly into obstacles that already pay off.

A teenager can lift the receiver, insert coins, and speak on a pay phone but cannot press the buttons firmly enough to register. Shaping or chaining?

Both. The order problem is a chaining problem, but the button press does not yet exist in a usable form, which is a form problem. That step is shaped, or physically prompted and the prompt faded, on its own, and then put back into the chain.

Want more? The 20-question quiz covers every page on the site with instant explanations.