Behavioural Science & Sociocultural Psychiatry · Paper A
Shaping (operant conditioning)
Shaping is training by 'getting warmer': you reward whatever comes closest to the goal today, then quietly move the goalposts tomorrow. Only the exact target behaviour eventually earns the reward. Nothing about it requires the learner to already know the behaviour, to have watched someone else do it, or to have the individual steps already linked together -- only a reinforcement criterion that keeps tightening.
You have almost certainly played the childhood game where a friend hides an object and calls out 'warmer... warmer... colder... HOT!' as you search the room. Nobody tells you exactly where to look. Instead, every step you take gets a little bit of feedback, and you gradually zero in on the exact spot. That children's game is, structurally, exactly what a psychologist means by shaping -- and once you can see the resemblance, half of applied behaviour analysis stops looking mysterious.
Shaping is a way of teaching a completely new behaviour -- one the learner has never done before, not even by accident -- by rewarding whatever comes CLOSEST to it right now, and then gradually raising the bar. At first, almost anything in the right direction earns a reward. As the learner keeps succeeding, the reward is only given for something a bit more accurate than before, and the cruder attempt that used to work no longer does. Repeat this enough times and the learner ends up reliably producing the exact behaviour you originally wanted, even though at the very start they could not do it at all.
Here is the trap almost every exam candidate falls into: assuming that ANY training programme broken into small steps counts as shaping. It does not. If the small steps are things the learner can ALREADY do, and training simply links them together in order (brushing teeth = wetting the brush, then applying paste, then brushing, then rinsing), that is chaining, a completely different procedure. Shaping is specifically about a behaviour that does not exist yet at all, built by rewarding closer and closer approximations of it. If nothing new is being created -- only existing pieces being sequenced, cued, or demonstrated -- you are looking at chaining, prompting, fading, or modeling instead, and mixing these up is the single most tested confusion in this topic.
B. F. Skinner's whole operant-conditioning framework rests on one idea: behaviour that is followed by a reward happens more often in future (this is reinforcement, in the plainest possible words -- a consequence that makes a behaviour more likely to happen again). The obvious problem is that reinforcement can only strengthen a behaviour that has actually occurred at least once. If a pigeon has never once pecked a particular spot on a wall, you cannot wait around for that exact peck to happen by chance and reward it -- it might never happen.
Skinner's solution was to reward something EASIER first: any movement toward the spot at all. Once that easier behaviour is happening reliably, you simply stop rewarding it and start rewarding something slightly closer to your real goal instead. The pigeon does not know the rules changed; it only knows that what used to work no longer earns the reward, so it tries something a little different -- and if that something happens to be closer to the target, it gets rewarded again. Repeat that pattern enough times, in small enough steps, and you arrive at the exact target behaviour, built entirely out of behaviours that were never the target at all.
Imagine a behaviour analyst teaching a child who has never made eye contact on request to do so reliably. Session 1: any glance toward the analyst's face lasting even half a second earns immediate praise and a small reward -- almost every attempt qualifies, so the child is succeeding constantly and staying engaged. Session 4: only glances lasting one full second are rewarded; the half-second glance, which used to work, no longer does. Session 7: the bar rises to two seconds.
Session 10: only eye contact of three seconds or longer is rewarded, and because the response is now well established, the schedule has also thinned -- reward is no longer given for every single success, but for roughly every third one, which makes the behaviour more durable over the long run than constant reward would. Notice what stayed constant throughout: nothing was demonstrated to the child (that would be modeling), nothing was physically guided (that would be prompting), and no separate steps were linked together (that would be chaining) -- only the criterion for what counted as 'good enough' kept moving.
| Looks similar | How it actually differs |
|---|---|
| Chaining | Links behaviours the learner can ALREADY perform into a sequence; builds nothing new |
| Prompting | Adds an extra cue to trigger an already-possible response; does not tighten a reinforcement criterion |
| Fading | Withdraws that added cue gradually; also does not touch the reinforcement criterion |
| Modeling | New behaviour acquired purely by observing a demonstrator; no reinforcement of the learner's own attempts needed |
| Classical conditioning | An involuntary reflex elicited by a paired stimulus, not a voluntary behaviour reinforced |
There is no test or checklist here, because shaping is a teaching procedure, not a clinical condition -- but a researcher or behaviour analyst still needs to show that shaping, specifically, produced a given change. The evidence is a data record showing the reinforcement CRITERION being deliberately and progressively raised over time, alongside the behaviour's accuracy tracking that rising criterion step for step. If the criterion never changed, or if the behaviour improved only after a demonstration was introduced or a physical prompt was added, the change is better explained by something other than shaping -- modeling, prompting, or simple practice.
A well-run shaping programme, as described in Cooper, Heron and Heward's standard applied-behaviour-analysis text, starts by clearly defining the terminal target behaviour and breaking the path to it into a series of realistic approximation steps (a task analysis). A genuinely motivating reinforcer is identified for that specific learner. Each step is held in place only until performance is comfortably reliable -- not just lucky once -- before the criterion advances, because advancing too soon on a fluke success sets the learner up to fail and can produce an extinction burst (a temporary flare-up of frustrated, off-target behaviour) instead of progress.
Reinforcement typically starts continuous, rewarding every qualifying attempt, and is deliberately thinned to an intermittent schedule once a step stabilises, because a behaviour maintained on occasional reward tends to be more durable than one that only ever ran on constant reward. Many real programmes combine shaping with prompting early on to jump-start the first approximation, then fade the prompt out as shaping's own criterion-tightening takes over — and every competent programme plans explicitly for generalisation, checking that the new behaviour transfers beyond the training room to real settings and real people, since a behaviour shaped only in one room with one trainer is not yet a finished skill.
Whenever a vignette describes a completely NEW behaviour being built with a reward criterion that gets stricter over time, the decisive question is never 'was reinforcement used' -- almost every operant procedure uses reinforcement. The decisive question is whether the target behaviour existed at all beforehand, and whether the CRITERION for reward is the thing changing. Before you move on: could you explain to a friend why teaching a child to say a whole word by rewarding closer and closer sounds is shaping, while teaching the same child to brush their teeth by linking five already-known actions together is not?
Shaping builds a genuinely new behaviour by rewarding whatever is closest to the goal right now and quietly raising the bar, never by linking, cuing, demonstrating, or reflexively pairing anything that already exists.