Reinforcement: Types, Schedules, Punishment and Behaviour Modification

Types and the schedules of reinforcement; escape, avoidance and punishment.

Official Syllabus: Paper I · Unit 6 Learning

“Of several responses made to the same situation, those which are accompanied or closely followed by satisfaction to the animal will… be more firmly connected with the situation.”

— Thorndike (1911), the Law of Effect

Classical conditioning explains how an existing reflex transfers to a new signal. It cannot explain how an organism acquires behaviour it never had. Operant conditioning answers that: behaviour is shaped by its consequences, and the technology built on this principle — reinforcement schedules, shaping, token economies, behaviour modification — is the most directly applicable body of knowledge in the subject.

Thorndike, Skinner and the operant principle

• Edward L. Thorndike (1898) placed hungry cats in puzzle boxes requiring a latch, loop or pedal to escape to food. Escape times fell gradually across trials as ineffective actions dropped out — trial-and-error learning, not insight.
• From this he stated the Law of Effect (Thorndike, 1911): responses followed by satisfaction become more firmly connected to the situation, those followed by discomfort less so. The connection is stimulus-response, strengthened by its outcome.
• B. F. Skinner (1938) rejected the mentalistic language of “satisfaction” and rebuilt the account on observables, introducing the operant chamber and the cumulative recorder to measure rate of responding continuously.
• His central distinction: respondent behaviour is elicited by a stimulus; operant behaviour is emitted and then selected by its consequences. Operant conditioning is therefore sometimes called Type R conditioning, for its emphasis on the response.
• A reinforcer is defined functionally and circularly on purpose — anything that increases the probability of the response it follows. Nothing is a reinforcer in itself; it is one only if the behaviour increases. This is why “rewards” chosen by adults often fail to reinforce children.
• Skinner (1948) showed how far consequence-learning goes with “superstition” in pigeons: birds reinforced on a purely time-based schedule acquired idiosyncratic rituals, having associated whatever they happened to be doing with the food. Adventitious reinforcement explains a great deal of ordinary human ritual.

Reinforcement and punishment: the four contingencies

The whole topic rests on one convention, and almost every error in answering it comes from missing the convention:

• Positive and negative describe whether a stimulus is added or removed. They do not mean pleasant and unpleasant, or good and bad.
• Reinforcement and punishment describe whether the behaviour subsequently increases or decreases.

• Negative reinforcement is not punishment. Negative reinforcement increases behaviour by removing something aversive; punishment decreases behaviour. They produce opposite effects.
  • The confusion arises because both involve an aversive stimulus, so the aversiveness is what gets noticed while the direction of effect does not.
  • A diagnostic that settles any example: ask what happened to the behaviour. If it went up, it was reinforcement, whatever was added or removed.
• Escape and avoidance are the two forms negative reinforcement takes — terminating an aversive stimulus already present, or preventing one signalled as coming. Avoidance is the more persistent, because the absence of the aversive event confirms the behaviour every time.

Types of reinforcers

• Primary reinforcers satisfy biological requirements and need no learning — food, water, warmth, relief from pain.
• Secondary or conditioned reinforcers acquire their power by pairing with a primary reinforcer. A light that precedes food in the operant chamber becomes reinforcing itself; so do money, grades, praise and tokens. This is where the account reaches human behaviour, since almost nothing that motivates adults is primary.
• Generalised reinforcers are secondary reinforcers effective across a wide range of situations and deprivation states. Money and social approval are the standard cases.
• The Premack principle (Premack, 1959) replaced the search for a list of reinforcers with a relation: a higher-probability activity will reinforce a lower-probability one.
  • Its force comes from the reversal. Premack (1962) showed the relation flips with deprivation — for a thirsty rat, drinking reinforces wheel-running; for a rat free to drink but confined, running reinforces drinking. Neither activity is inherently reinforcing.
  • The applied form is familiar: homework before television works, and the reverse does not.
  • Response deprivation accounts later refined this, showing restriction below baseline matters more than relative probability as such.
• Immediate reinforcement is markedly more effective than delayed, because a delay allows intervening behaviour to be reinforced instead. This is why bridging stimuli — a clicker, a token, a “well done” — are used: they are conditioned reinforcers that close the gap until the primary reinforcer arrives, and it is why long-delayed consequences govern human behaviour so poorly.

Types and the schedules of reinforcement

• Continuous reinforcement reinforces every response. It produces the fastest acquisition and the fastest extinction.
• Partial or intermittent reinforcement reinforces some responses only, and is defined by two crossed dimensions — whether the requirement is a number of responses (ratio) or a period of time (interval), and whether that requirement is fixed or variable. Ferster and Skinner (1957) mapped the resulting response patterns systematically.

Ratio schedules drive rate; variable schedules remove the predictability that permits pausing.

• The partial reinforcement extinction effect is the finding that matters for “strength of learning”: behaviour maintained on an intermittent schedule persists far longer without reinforcement than behaviour maintained continuously. The organism has learnt that unreinforced responses are normal, so their absence is not informative.
• Variable ratio is the strongest for persistence, which is the mechanism behind gambling and behind compulsive phone-checking — both deliver reinforcement on an unpredictable ratio.
• Schedule thinning is the practical consequence: begin with continuous reinforcement to establish a behaviour, then move to intermittent and increasingly variable schedules to make it durable. Any behaviour-modification plan that stays on continuous reinforcement will fail when reinforcement stops.
• The matching law (Herrnstein, 1961) describes choice between concurrently available schedules: the proportion of responses allocated to an alternative matches the proportion of reinforcement obtained from it.
  • It converts reinforcement from a strengthener of single responses into a principle of allocation among competing options, which is why an unwanted behaviour persists whenever it yields a better reinforcement rate than the alternative on offer.
  • The applied reading is direct: to reduce a behaviour, enrich the schedule for its alternative rather than attack the behaviour itself.

Differential reinforcement

Reinforcement can be arranged to reduce behaviour without any aversive component, which is why these procedures are preferred in applied work:

• DRA — differential reinforcement of alternative behaviour. Reinforce a specified acceptable behaviour that achieves the same outcome, while withholding reinforcement for the unwanted one.
• DRI — differential reinforcement of incompatible behaviour. Reinforce a behaviour that cannot physically co-occur with the unwanted one, such as sitting against running, or cooperative play against hitting.
• DRO — differential reinforcement of other behaviour. Reinforce the absence of the target behaviour across an interval, which is effective but teaches no replacement.
• DRL — differential reinforcement of low rates. Reinforce the behaviour only when it occurs below a set rate, used where the behaviour is acceptable in moderation.
• All four rest on the same logic the matching law makes explicit: behaviour follows the better available reinforcement rate, so changing the rate for the alternative changes the choice.

Shaping, chaining and new behaviour

• Shaping or successive approximation solves the problem of a behaviour that is never spontaneously emitted and so can never be reinforced. Reinforce whatever crudely resembles the target, then progressively restrict reinforcement to closer approximations.
• Chaining builds a sequence by linking separate responses, each one’s completion acting as the conditioned reinforcer for the previous step and the cue for the next. Complex routines — dressing, toilet training, laboratory skills — are taught this way, often in backward order so the final step is learnt first and always leads to reinforcement.
• Operant extinction withholds the reinforcer. Expect an extinction burst — a temporary increase in rate and intensity — before the decline, which is precisely when inconsistent adults relent and inadvertently reinforce the escalated form.

Stimulus control: discrimination and generalisation

A reinforced response does not occur at random. It comes under the control of the circumstances in which reinforcement has been available, and the two processes that establish this control are opposites that depend on each other.

• The three-term contingency. A discriminative stimulus (S^D, or S+) is one in whose presence responding is reinforced; an S-delta (S−) is one in whose presence it is not. Antecedent, behaviour and consequence together form the unit on which all applied work is built, and the antecedent is the term a learner must get right for a skill to be usable.
• Discrimination is learned by differential reinforcement: the response is reinforced in the presence of S+ and not in the presence of S−, until responding occurs reliably to one and not the other. A pigeon reinforced for pecking a green key and not a red one soon pecks only at green. The behaviour has not changed; when it occurs has.
• Generalisation is the spread of the response to stimuli resembling S+ without further training. Plotting response rate against stimulus similarity gives the generalisation gradient — a peak at the trained value, falling away as similarity decreases. A steep gradient means tight discrimination; a flat one means broad transfer.
• They are complementary, not opposed, and a skill requires both. Generalisation alone produces a response emitted in the wrong settings; discrimination alone produces a response usable only in the exact conditions of training, which is failure of transfer. A learner driver must generalise braking across vehicles, roads and weather, and must discriminate the situations that call for it from those that do not. Competence is the correct breadth of the gradient, and training works by widening it through varied practice while sharpening its edges through differential reinforcement.
• Over-generalisation and over-discrimination are the two failure modes, and each has a standard remedy: vary the training conditions to widen a gradient that is too steep, and introduce explicit S− trials to sharpen one that is too flat.

Does punishment work, and when does reward fail?

• Punishment does suppress behaviour, so a flat denial is wrong. Azrin and Holz (1966), whose experimental analysis remains the most thorough, defined it functionally as a reduction in the future probability of a response produced by a stimulus contingent on it, and identified the conditions for effectiveness:
  • delivered immediately, intense from the outset rather than escalated gradually, applied consistently on every occurrence, with escape from the punishment unavailable, and crucially with a reinforced alternative behaviour available.
  • Everyday punishment — delayed, mild, inconsistent, escapable, with no alternative taught — violates nearly all of these, which is why it suppresses weakly and temporarily.
• The costs are independent of the efficacy question:
  • It specifies what not to do without teaching what to do.
  • Suppression tends to be specific to the punishing situation and to the punisher’s presence, producing concealment rather than change.
  • It conditions fear and avoidance of the punisher, damaging the relationship that any later influence depends on.
  • It models aggression as a legitimate means of control.
  • Learned helplessness follows where aversive events are uncontrollable, a point developed in the escape-and-avoidance material.
• Reward fails too, and the reasons are examinable:
  • Overjustification. Lepper, Greene and Nisbett (1973) found that children rewarded for an activity they already enjoyed subsequently chose it less often. Deci (1971) showed the same undermining of intrinsic motivation. Reinforcing already-intrinsically-motivated behaviour can reduce it.
  • Instinctive drift. Breland and Breland (1961) found trained animals reverting to species-typical food-related behaviours that displaced the reinforced response, showing operant principles are bounded by biology.
  • Reinforcer relativity. What reinforces depends on current probabilities and deprivation (Premack, 1962), so a fixed reward list fails.
  • Delay and schedule errors: a reinforcer too late, or a schedule never thinned, produces behaviour that does not last.
  • Satiation erodes a reinforcer’s power with repeated delivery.

Applications

• Token economy (Ayllon and Azrin, 1968) — tokens as generalised conditioned reinforcers exchangeable for goods or privileges, developed with psychiatric inpatients and since standard in classrooms, de-addiction settings and institutional care.
• Behaviour modification in parenting. For a child’s aggression, the plan is differential reinforcement of incompatible behaviour — reinforcing cooperative play — combined with negative punishment as time-out and consistent non-reinforcement of the aggressive act, since adult attention is usually the reinforcer maintaining it.
• Workplace design. For habitual latecoming: define the target operationally, baseline it, then reinforce punctuality on a schedule thinned over time rather than punishing lateness, because punishment there produces concealment and resentment while leaving punctuality untaught.
• Road safety. The seat-belt case maps the full two-by-two: the alarm that ceases on buckling is negative reinforcement, a fine is positive punishment, insurance benefit for compliance is positive reinforcement, and licence points withdrawn is negative punishment.
• Toilet training is the standard worked case, and it uses almost every principle at once.
  • Establish readiness first: the physiological control the behaviour requires is a matter of maturation, and no contingency produces it early.
  • Task-analyse the chain — signalling the need, reaching the toilet, undressing, elimination, flushing, washing — then teach the links by chaining, with the completion of each acting as the conditioned reinforcer for the one before.
  • Shape each link by successive approximation, reinforcing approximations at first and then restricting reinforcement to closer ones.
  • Reinforce immediately with praise or a token, since a delay lets intervening behaviour be reinforced instead; use scheduled sitting so the opportunity to reinforce recurs.
  • Make the bathroom and the bodily cue discriminative stimuli for the chain, so control transfers to the natural antecedent.
  • Never punish accidents: punishment here conditions fear and avoidance of the toilet itself, suppressing the very behaviour being taught, and it teaches nothing about what to do.
  • Thin the schedule to intermittent once the chain is reliable, so it survives the withdrawal of adult attention.
• Programmed instruction is the educational application of the same principles — small steps, active responding, immediate reinforcement — and is treated as its own syllabus clause.
• Biofeedback extends operant conditioning to autonomic responses by making an internal state visible and reinforcing changes in it.

Criticisms and limitations

• Biological constraints. Breland and Breland (1961) showed that species-typical behaviour overrides reinforcement contingencies, so the behaviourist assumption that any response can be conditioned to any consequence is false.
• The circularity objection. Defining a reinforcer by its effect makes the Law of Effect unfalsifiable as a general claim. The defence is that it is a functional definition yielding testable predictions once a reinforcer is identified for an individual — but the objection has force against the law as stated.
• Cognitive mediation. Humans respond to rules and instructions about contingencies, often without contact with the contingency itself, which an account built on direct consequence-learning cannot explain. Expectancy and self-efficacy accounts arose to fill this.
• Observational learning is unaccounted for: behaviour is acquired by watching, with no reinforcement of the observer, which is the basis of the social-learning critique.
• Reductionism and ethics. The technology of control raises questions about autonomy that Skinner addressed directly and that remain contested, particularly where aversive procedures are applied to people who cannot consent.
• Intrinsic motivation. The overjustification findings show the account is incomplete about why people do things they are not reinforced for.

Indian context

• Corporal punishment is prohibited in Indian schools. Section 17(1) of the Right of Children to Free and Compulsory Education Act, 2009 (Ministry of Education) bars physical punishment and mental harassment, and 17(2) makes contravention punishable by disciplinary action. NCPCR guidelines classify it as physical punishment, mental harassment and discrimination, and the Commission monitors compliance under Section 31.
• The practice persists despite the prohibition. The Ministry of Women and Child Development’s Child Abuse in India study (2007) reported that two of every three school children had faced corporal punishment — the standing Indian figure for this topic, and the reason the gap is one of implementation rather than law.
• The evidence base. Gershoff (2002) and Gershoff and Grogan-Kaylor (2016), meta-analysing 88 and 75 studies respectively, found physical punishment associated with more aggression, more behaviour problems, poorer mental health, lower cognitive performance and worse parent-child relationships, with no reliable benefit even for immediate compliance once baseline differences were controlled. The psychological case against it is therefore not a value preference but a finding.
• The constructive implication for Indian classrooms is what a Psychology answer should supply where a Polity answer would stop at the statute: the alternative is differential reinforcement of incompatible behaviour, token systems, and operationally defined targets with baselines — techniques that teach the wanted behaviour rather than suppressing the unwanted one, and that teacher-training programmes under the Act are meant to provide.