29.04.02 · psychology / learning-memory

Classical and operant conditioning: schedules of reinforcement, extinction, latent learning

stub3 tiersLean: nonepending prereqs

Anchor (Master): Skinner, B. F. — Science and Human Behavior (1953)

Intuition Beginner

Learning is a lasting change in behavior that comes from experience. The clearest laboratory case is classical conditioning, which Ivan Pavlov stumbled on while studying digestion in dogs. A bell rung before feeding, repeated often enough, comes to trigger salivation on its own. The bell starts neutral and ends as a signal: a conditioned stimulus that predicts food and calls forth a conditioned response.

Operant conditioning, developed by B. F. Skinner, asks a different question. What happens after a voluntary act shapes whether it repeats. A rat that presses a lever and gets food presses again; one that gets a shock does not. Rewards strengthen behavior, punishment weakens it. Slot machines run on this logic: unpredictable, variable-ratio payouts produce play that is extraordinarily hard to extinguish.

Not all learning needs a reward at all. Edward Tolman let rats explore a maze unrewarded for days. The moment food appeared at the goal, those rats ran it almost perfectly. They had been building a cognitive map all along. This is latent learning: knowledge acquired silently and drawn on only when it pays.

Visual Beginner

Figure: The conditioning landscape in one view. Stimulus pairing (Pavlov), consequence-driven behavior (Skinner), the four canonical reinforcement schedules, and Tolman's latent-learning crossover together frame why the same reward, delivered differently, produces behavior with very different persistence.

Worked example Beginner

Acquisition and extinction of a conditioned response

A lab conditions a dog. The conditioned stimulus is a 1000 Hz tone; the unconditioned stimulus is food powder. Salivation is measured in drops on tone-alone probe trials.

Trial	Procedure	CR (drops)
1	tone + food	0
2	tone + food	2
3	tone + food	6
5	tone + food	14
8	tone + food	22
10	tone + food	24
12	tone + food	25

By trial 12 the response has plateaued near 25 drops. Acquisition is fast at first, then levels off — a negatively accelerated curve.

Now extinction begins. The tone sounds, but no food follows.

Extinction trial	CR (drops)
1	21
3	12
6	4
9	1
12	0

The response fades. But this is not erasure. After a 30-minute rest, a single tone probe elicits about 8 drops — spontaneous recovery — weaker than before, then fading again with repeated tones.

A variable-ratio slot machine

A machine pays out on average once every 25 plays (a VR-25 schedule), with an average jackpot of $20. E a c h pl a y cos t s$ 1. Over 100 plays the gambler expects about 4 payouts, roughly $80 ba c k o n$ 100 wagered — a $20 loss. Yet response rate stays high, because the next play could be the winner. The uncertainty, not the average value, is what locks behavior in.

What this tells us. Conditioning is not about reward strength alone but about schedule and expectation. The same reward, delivered on a different schedule, produces behavior that resists extinction in completely different degrees.

Check your understanding Beginner

Formal definition Intermediate

Conditioning names two families of associative learning that differ in what gets associated. The vocabulary below is standardized across the anchor texts ^{[source pending]}.

Classical (Pavlovian) conditioning. A learning procedure in which a biologically potent unconditioned stimulus (US) that reflexively elicits an unconditioned response (UR) is repeatedly paired with a previously neutral conditioned stimulus (CS). After sufficient pairings the CS alone elicits a conditioned response (CR) resembling the UR. The principal phases and phenomena:

Acquisition — the period of CS–US pairing during which CR strength rises, typically along a negatively accelerated curve.
Extinction — repeated CS-alone presentations; the CR weakens progressively.
Spontaneous recovery — reappearance of a weakened CR after a rest, smaller than at peak.
Stimulus generalization — CR spreads to stimuli similar to the CS; the steeper the generalization gradient, the sharper the discrimination.
Stimulus discrimination — differential responding when one CS is reinforced and a similar CS is not.
Higher-order conditioning — pairing an established CS with a new neutral stimulus, so the new stimulus also elicits a (weaker) CR, without further US presentation.
Counterconditioning — replacing an unwanted CR by pairing the CS with a stimulus evoking an incompatible response.

Operant (instrumental) conditioning. A learning procedure in which the consequence of a voluntary operant response changes the future probability of that response. Thorndike's law of effect is the ancestor: responses followed by satisfying effects are strengthened. The four-cell taxonomy of consequences:

	Stimulus added	Stimulus removed
Increases behavior	Positive reinforcement	Negative reinforcement
Decreases behavior	Positive punishment	Negative punishment

Primary reinforcers satisfy biological needs (food, warmth); secondary (conditioned) reinforcers gain power through association with primary reinforcers (money, praise, tokens). Shaping reinforces successive approximations toward a target response; chaining links shaped behaviors into a sequence, typically taught backward. A discriminative stimulus ( $S^{D}$ ) signals that reinforcement is available for a response; its absence ( $S^{Δ}$ ) signals that it is not.

Schedules of reinforcement govern how reinforcers are delivered. Ratio schedules depend on the number of responses; interval schedules depend on the passage of time; each may be fixed or variable.

Schedule	Rule	Response pattern	Extinction
FR (fixed ratio)	every $n$ th response	high rate, post-reinforcement pause	moderate
VR (variable ratio)	on average every $n$ responses	very high, steady, no pause	very slow
FI (fixed interval)	first response after $t$ seconds	scalloped — pause then accelerating	moderate
VI (variable interval)	first response after avg $t$ seconds	moderate, steady	slow

The partial reinforcement extinction effect: behavior maintained on intermittent (especially VR) schedules persists far longer without reinforcement than behavior maintained on continuous reinforcement, because non-reinforcement is hard to distinguish from a normal thin schedule.

Biological constraints. The generality of conditioning is bounded by evolution. Garcia and Koelling (1966) showed that rats readily associate taste with later illness but not with immediate shock, and associate audiovisual cues with shock but not with illness — prepared learning shaped by what mattered to the species' ancestors. Breland and Breland's "misbehavior of organisms" documented instincts (e.g., raccoons "washing" token coins instead of depositing them) intruding on operant contingencies. Constraints of this kind limit a pure stimulus–response account.

Latent learning. Tolman and Honzik (1930) showed that rats allowed unrewarded maze exploration nonetheless acquired a cognitive map, revealed the moment a food reward was introduced at the goal. Learning (acquisition of knowledge) was dissociated from performance (its expression), undermining the claim that reinforcement is necessary for learning rather than for its display.

Key model Intermediate

The Rescorla–Wagner model (1972) is the most influential formal account of classical conditioning, and the bridge from behavioral data to the prediction-error framework that now dominates learning theory and computational neuroscience ^{[source pending]}.

Core claim. Learning on a trial proceeds only when the US is surprising — when it exceeds what the CSs already predict. A reinforcer fully predicted by present cues supports no further learning. Formally, for a single CS paired with a US,

Δ V_{C S} = α_{C S} β_{U S} (λ_{U S} - V),

where $V$ is the current associative strength of the CS, $α \in (0, 1]$ measures CS salience, $β \in (0, 1]$ measures US drivability, and $λ$ is the asymptotic strength the US can support. The quantity $(λ - V)$ is the prediction error: the gap between what was expected and what occurred. When several cues $C S_{1}, \dots, C S_{n}$ are present together, each cue's error is computed against the aggregate prediction $V = \sum_{i} V_{C S_{i}}$ , so cues compete for the predictive space.

Iterating the delta rule drives $V$ toward $λ$ along a negatively accelerated trajectory, matching the empirically observed acquisition curve. Setting $λ = 0$ recovers extinction: with the US gone, each trial subtracts a fraction of $V$ until it approaches zero.

Blocking. Kamin's phenomenon is the model's signature success. Train a CS $_{A}$ → US to asymptote, then compound it with a new CS $_{B}$ and continue pairing the compound with the US. The new CS $_{B}$ acquires virtually no strength. The model explains why: at the start of compound training $V_{A} \approx λ$ , so the aggregate error $(λ - V_{A} - V_{B}) \approx 0$ , and there is nothing for $V_{B}$ to learn from. Learning is driven by surprise, not by contiguity, and a redundant cue that adds no predictive value is ignored.

What the model captures, and what it does not. Rescorla–Wagner accounts for blocking, overshadowing, overexpectation, and the dependence of conditioning on CS–US contingency rather than mere pairing. It fails on phenomena that depend on temporal structure or on within-compound associations: latent inhibition (pre-exposure to the CS retards later conditioning), second-order conditioning, the effect of CS–US temporal gap (trace versus delay), and recovery from extinction (spontaneous recovery, renewal, reinstatement). The last is decisive: because the model treats extinction as the simple erosion of $V$ , it cannot explain why an extinguished CR returns under context change or rest. This limitation motivates the inhibitory-learning and comparator accounts taken up at Master tier.

Exercises Intermediate

Exercise 3 (medium, application).

A teacher reinforces every correct homework submission for two weeks (continuous reinforcement), then switches to rewarding only after every fifth correct submission (FR-5), and finally moves to rewarding on a variable-ratio schedule centered near 5. Order the three phases by resistance to extinction (most to least) and justify the ordering using the partial reinforcement extinction effect.

Hint

Resistance to extinction rises as the schedule becomes harder to distinguish from non-reinforcement.

Answer

Most resistant to least: VR-5 > FR-5 > continuous. Continuous reinforcement extinguishes fastest because the learner detects the absence of reward on the very first unrewarded response. FR-5 is more resistant because several unrewarded responses are normal between rewards. VR-5 is most resistant of all because no number of unrewarded responses is ever diagnostic — the learner cannot tell whether the schedule has ended or merely thinned. This is the partial reinforcement extinction effect.

Exercise 5 (medium, application).

A Garcia-type experiment gives rats one of two pairings: (a) sweet taste followed hours later by nausea, or (b) bright light plus tone followed hours later by nausea. Predict which group will show a strong learned aversion, and explain why this is a problem for a pure contiguity-based account of conditioning.

Summary

Consider which cue the rat's biology is "prepared" to link to internal illness, and how long the delay is.

Answer

Group (a), the taste–nausea pairing, will show a strong aversion; group (b), the audiovisual–nausea pairing, will not. Conditioning across a delay of hours is normally impossible under contiguity-based theory, which requires the CS to closely precede the US. The result shows prepared learning: evolution has wired rats to link gustatory cues to later gastrointestinal consequences, because that pairing was fitness-relevant for an omnivore. A pure contiguity account cannot explain the selective, delay-tolerant association.

Exercise 6 (hard, critical evaluation).

A clinician claims that because a patient's phobia was eliminated by exposure therapy, the original fear memory has been erased. Drawing on what is known about extinction mechanisms (spontaneous recovery, renewal, reinstatement), evaluate this claim in 4–6 sentences.

Hint

If extinction were erasure, could the fear ever return? What does the evidence say?

Answer

The claim is not supported. The dominant evidence indicates that extinction is new, context-dependent inhibitory learning that masks the original excitatory association rather than deleting it. Three findings converge: the extinguished CR reappears after a rest (spontaneous recovery), returns when the subject is tested outside the extinction context (renewal), and re-emerges after unsignaled US presentations (reinstatement). If the original memory had been erased, none of these recoveries would occur. Exposure therapy is therefore best understood as strengthening a competing safety association, which is why treatment gains are context-specific and prone to relapse unless generalization is deliberately trained.

Advanced results Master

Rescorla–Wagner in full, and where it breaks

The Master-tier treatment treats $V$ as a vector of cue-specific associative strengths competing for a shared US. With cues $C S_{1}, \dots, C S_{n}$ co-present,

V_{total} = i = 1 \sum n V_{i}, Δ V_{i} = α_{i} β (λ - V_{total}),

so every cue present on a trial receives the same prediction error scaled by its salience. This competition formalizes overshadowing (a salient cue depresses learning about a simultaneous partner), blocking (a previously-trained cue drives the error to zero for a redundant partner), and overexpectation (two cues each trained to asymptote, compounded, produce a combined prediction exceeding $λ$ , so both must decrement on the next trial). Overexpectation is the most counter-intuitive prediction — compounding two strong predictors should, on a contiguity view, maintain both; Rescorla–Wagner insists both must weaken, and the data confirm it.

The model fails where conditioning depends on what is absent, on temporal microstructure, or on learned attention. Latent inhibition (CS pre-exposure retards conditioning) arises because the CS has learned to be irrelevant, a process the model cannot represent without an attentional gate ( $α$ that itself learns). Conditioned inhibition (a cue that predicts US absence) requires a negative $V$ , which the model admits, but recovery effects show that inhibition, like excitation, is context-bound. Second-order conditioning — a neutral cue paired with an established CS acquires strength without contacting the US — needs an associative chain the linear model can only approximate by relabeling.

Extinction: inhibitory learning, not erasure

The empirical signature that defeated the erasure interpretation is the recoverability of the extinguished response ^{[source pending]}. Three dissociable phenomena:

Spontaneous recovery. After a rest following extinction, the CR returns, weaker than at peak but reliably present.
Renewal. Extinction performed in context B does not transfer to context A (where acquisition occurred). Changing context after extinction reinstates the CR — most strongly in the ABA design, but present in ABC and AAB designs as well.
Reinstatement. Unsignaled US presentations after extinction partially restore the CR, especially in the context where the US was re-presented.

Bouton's account treats extinction as the acquisition of a context-specific inhibitory memory that competes with — but does not overwrite — the original excitatory trace. Retrieval of either depends on context, which is why return to the acquisition context (or any sufficiently distant context) favors the older, more generalizable excitation. This is the theoretical backbone of exposure therapy for anxiety and OCD: the goal is not erasure but the construction of a robust, broadly-generalized safety memory, which explains the clinical emphasis on varied contexts, expanded extinction stimuli, and occasional retrieval cues.

The matching law and choice

Herrnstein (1961) found that pigeons choosing between two keys distribute their responses in proportion to the reinforcement obtained on each:

\frac{B _{i}}{B _{1} + B _{2}} = \frac{R _{i}}{R _{1} + R _{2}},

the matching law. Matching is one of the most reliably reproduced quantitative regularities in operant behavior. It has been extended to concurrent VI–VI, concurrent ratio, and concurrent-chain schedules, and reformulated as the generalized matching law with bias and sensitivity parameters:

lo g \frac{B _{1}}{B _{2}} = s lo g \frac{R _{1}}{R _{2}} + lo g b .

Deviations ( $s \neq = 1$ , $b \neq = 1$ ) index sensitivity to reinforcement and inherent bias, and generalize to foraging, drug self-administration, and consumer choice. Melioration (Herrnstein, Vaughan) offers a process account: organisms shift toward the locally better-paying alternative rather than the globally optimal one, and the resulting steady state is matching — sometimes at the cost of overall reinforcement rate. Delay discounting — the hyperbolic devaluation of delayed reward, $V = A / (1 + k D)$ — connects the same framework to impulsivity, addiction, and intertemporal choice.

Dopamine as reward prediction error

Schultz, Dayan, and Montague (1997) recorded midbrain dopamine neurons in monkeys during conditioning and found that dopamine transients did not encode reward per se but reward prediction error: neurons fired to an unexpected reward, ceased firing when a fully predicted reward was omitted, and shifted their response to the predictive CS as learning proceeded. The Rescorla–Wagner delta is thus realized neurally as a phasic dopamine signal ^{[source pending]}.

This identification enabled the temporal-difference (TD) account of dopamine. TD learning (Sutton and Barto) propagates prediction error backward through time to the earliest predictive state, giving a single algorithm that learns both when and how much to expect, and whose error signal mirrors dopamine anatomy and dynamics. This convergence of behavioral theory (Rescorla–Wagner), electrophysiology (dopamine RPE), and machine learning (TD) is the founding result of computational reinforcement learning as a model of the brain.

Incentive sensitization and addiction

Robinson and Berridge (1993) dissociated "liking" (the hedonic impact of a reward) from "wanting" (incentive salience — the motivational pull of cues paired with reward). Repeated exposure to drugs of abuse sensitizes the wanting system (mesolimbic dopamine) while tolerance often reduces liking. Cues previously paired with drug use acquire exaggerated incentive salience, drawing attention and approach even in the absence of pleasure. This accounts for the clinical puzzle of relapse long after withdrawal: the patient does not necessarily enjoy the drug but is pulled toward it by sensitized cue-triggered wanting. The framework reframes conditioned place preference (an animal's preference for the environment in which it received a drug) as a measure of incentive salience rather than passive hedonic recall.

Social learning and the boundary of conditioning

Bandura's Bobo doll studies (1961, 1963) showed that children acquire aggressive repertoires by observation alone, without direct reinforcement of their own behavior, and that vicarious reinforcement (seeing the model rewarded) raises the probability of performance. Bandura's four-process model — attention, retention, reproduction, motivation — separates acquisition from performance, exactly as Tolman had separated learning from its display. Observational learning is not a counterexample to conditioning so much as a demonstration that reinforcement is one input among several to a representational system that can encode, store, and later deploy what it has witnessed.

Limits the framework imposes

Conditioning research is one of the most quantitatively developed bodies in psychology, and its formal models (Rescorla–Wagner, TD, matching) set a standard of precision that much of the field does not meet. The price of that precision is a narrow scope: the framework handles associative and procedural learning well, and says little that is operational about declarative memory, language, insight (Köhler's chimpanzees), or cultural transmission. The nontrivial empirical claim is not that conditioning explains everything but that its predictive-error, schedule-driven logic is the substrate on which higher-order learning is built — and that the substrate has turned out, under dopamine-RPE and TD, to be surprisingly computational.

Connections Master

Learning and memory 29.04.01 is the immediate prerequisite. This unit deepens the conditioning strand of that overview into formal models and mechanism; the memory-systems strand (encoding, storage, retrieval, working memory) continues in 29.04.03 pending (pending), to which this unit is the behavioral foundation.
Neuroscience [29.02.NN] (pending) supplies the substrate. Dopamine neurons of the ventral tegmental area and substantia nigra realize the reward prediction error (Schultz 1997); the amygdala mediates fear conditioning; the basal ganglia and cerebellum underlie operant and procedural learning respectively; the hippocampus encodes the cognitive maps Tolman inferred behaviorally. Long-term potentiation is the leading candidate for the synaptic plasticity on which all of this depends.
Psychological disorders [29.09.NN] (pending) inherits conditioning directly: anxiety disorders as persistent conditioned fear, PTSD as traumatic memory that resists extinction, substance use disorders as incentive-sensitized cue-driven relapse, and the rationale for exposure-based and contingency-management treatments.
Therapy and treatment [29.10.NN] (pending) applies the extinction machinery. Exposure therapy for phobias and OCD, token economies, contingency management for addiction, and behavioral activation for depression are all conditioning technologies, and their effectiveness (and their limits — renewal, relapse) is predicted by the inhibitory-learning account of extinction.
Motivation and emotion [29.11.NN] (pending) connects through incentive salience, delay discounting, and the dopamine account of reward. The matching law and melioration link operant schedules to the economics of choice and self-control.
Statistics and learning theory [26.NN, 25.NN] (pending) connect at the formal level: Rescorla–Wagner is a gradient step on a squared prediction-error loss, TD learning is approximate dynamic programming, and the Bayesian generalizations of conditioning (Kalman filters, hierarchical prediction) draw on the same machinery as state-space estimation and reinforcement learning in computer science.
Philosophy of mind 20.06.01 connects through the behaviorism–cognitivism debate and the free-will question Skinner pressed in Beyond Freedom and Dignity. Whether prediction-error models of conditioning illuminate or merely restate the mind–body problem remains contested, but the empirical work sets hard constraints on any theory of agency.

Historical & philosophical context Master

Pavlov was a physiologist, not a psychologist, and he came to conditioning from the study of digestion — for which he received the 1904 Nobel Prize. The realization that psychic secretions (salivation to the sight or sound of the feeder) obeyed lawful, measurable rules turned his laboratory toward associative learning for the remaining three decades of his career. Conditioned Reflexes (1927) presented the framework that would define the field's vocabulary ^{[source pending]}.

In America, Thorndike's puzzle-box experiments (1898) and his law of effect supplied the operant intuition before Skinner systematized it. Watson's 1913 behaviorist manifesto declared that psychology, to be a science, must study only observable behavior — a methodological austerity that dominated American psychology for forty years. Watson's Little Albert study with Rayner (1920) attempted to show that human emotion is itself conditionable, and became simultaneously a foundational result and a permanent ethical indictment.

Skinner rejected Watson's passivity for an operant analysis: behavior is emitted, not elicited, and is selected by its consequences in a process analogous to natural selection. The operant chamber, the cumulative recorder, and the schedules of reinforcement were the instruments by which Skinner made learning into a quantitative science. Science and Human Behavior (1953) and The Behavior of Organisms (1938) are the canonical statements; Verbal Behavior (1957), with Chomsky's (1959) hostile review, marks the front at which behaviorism began to give ground to cognitivism ^{[source pending]}.

The cognitive counter-revolution had two conditioning-internal sources. Tolman's latent-learning and cognitive-map experiments (1930, 1948) showed that animals learn more than their behavior reveals and that they represent space and means–end relations, forcing the postulation of intervening variables. Rescorla's (1968) contingency experiments showed that what matters in conditioning is not CS–US contiguity but the informational relationship between CS and US — a CS uncorrelated with the US fails to condition even when the two co-occur frequently. Both lines pushed the field from association toward representation and computation.

Rescorla and Wagner's (1972) model was the formal synthesis: a simple error-driven learning rule that captured a surprising range of phenomena, predicted overexpectation, and supplied the prediction-error concept that neuroscientists would later locate in dopamine. The subsequent absorption of conditioning into reinforcement learning (Sutton and Barto) and the discovery of the dopamine RPE (Schultz 1997) completed a line from salivating dogs to a computational theory of the reward system — a convergence none of the original participants anticipated.

The ethical record is part of the history and not a footnote to it. Little Albert (Douglas Merritte) was conditioned without consent and never deconditioned. Decades of animal research raised their own welfare questions. The same principles that produced effective therapies also produced, in some hands, coercive applications in institutions, education, and advertising. The science is real and the cost is real; neither cancels the other.

Bibliography Master

Pavlov, I. P. — Conditioned Reflexes: An Investigation of the Physiological Activity of the Cerebral Cortex (Oxford University Press, 1927; trans. Anrep).
Thorndike, E. L. — Animal Intelligence: An Experimental Study of the Associative Processes in Animals, Psychological Review Monograph Supplements 2, 1–109 (1898).
Watson, J. B. & Rayner, R. — "Conditioned emotional reactions", Journal of Experimental Psychology 3, 1–14 (1920).
Skinner, B. F. — The Behavior of Organisms: An Experimental Analysis (Appleton-Century, 1938).
Skinner, B. F. — Science and Human Behavior (Macmillan, 1953).
Tolman, E. C. & Honzik, C. H. — "Introduction and removal of reward, and maze performance in rats", University of California Publications in Psychology 4, 257–275 (1930).
Tolman, E. C. — "Cognitive maps in rats and men", Psychological Review 55, 189–208 (1948).
Garcia, J. & Koelling, R. A. — "Relation of cue to consequence in avoidance learning", Psychonomic Science 4, 123–124 (1966).
Breland, K. & Breland, M. — "The misbehavior of organisms", American Psychologist 16, 681–684 (1961).
Rescorla, R. A. — "Probability of shock in the presence and absence of CS in fear conditioning", Journal of Comparative and Physiological Psychology 66, 1–5 (1968).
Rescorla, R. A. & Wagner, A. R. — "A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and nonreinforcement", in Black & Prokasy (eds.), Classical Conditioning II: Current Research and Theory (Appleton-Century-Crofts, 1972), pp. 64–99.
Kamin, L. J. — "Attention-like processes in classical conditioning", in Jones (ed.), Miami Symposium on the Prediction of Behavior: Aversive Stimulation (University of Miami Press, 1968).
Herrnstein, R. J. — "Relative and absolute strength of response as a function of frequency of reinforcement", Journal of the Experimental Analysis of Behavior 4, 267–272 (1961).
Bandura, A., Ross, D. & Ross, S. A. — "Transmission of aggression through imitation of aggressive models", Journal of Abnormal and Social Psychology 63, 575–582 (1961).
Schultz, W., Dayan, P. & Montague, P. R. — "A neural substrate of prediction and reward", Science 275, 1593–1599 (1997).
Sutton, R. S. & Barto, A. G. — Reinforcement Learning: An Introduction (MIT Press, 2nd ed., 2018).
Bouton, M. E. — "Context and behavioral processes in extinction", Learning & Memory 11, 485–494 (2004).
Robinson, T. E. & Berridge, K. C. — "The neural basis of drug craving: an incentive-sensitization theory of addiction", Brain Research Reviews 18, 247–291 (1993).
Myers, D. G. & DeWall, C. N. — Psychology, 13th ed. (Worth, 2021), Ch. 7.
Gleitman, H., Gross, J. & Reisberg, D. — Psychology, 8th ed. (Norton, 2011), Ch. 3.