Higher Order Conditioning

Higher-order conditioning is a form of classical conditioning, first demonstrated by Pavlov. It happens when an already-conditioned stimulus is used to condition a brand-new stimulus. That new stimulus then triggers the same response, one step further along the chain.

In classical conditioning, higher-order conditioning, otherwise known as second-order conditioning, is a procedure in which the conditioned stimulus of one experiment acts as the unconditioned stimulus of another.

First-order conditioning happens when the conditioned stimulus (CS1) is paired with the unconditioned stimulus until CS1 alone elicits a conditioned response.

Second-order conditioning goes one step further. A new, previously neutral stimulus (CS2) is paired with CS1 itself, rather than with the unconditioned stimulus, until CS2 also elicits the response. The new stimulus never met the unconditioned stimulus directly.

A bell paired with food until it alone elicits salivation. This is a case of first-order conditioning. Pairing a light with that same bell, rather than with food, builds a second-order association instead.

If the light alone comes to elicit salivation, higher-order conditioning has occurred.

Pavlov (1927) provided the earliest and most influential demonstration, training dogs to salivate to a buzzer that had itself been conditioned to predict food.

Causes and Implications for Behavior

There are several theoretical models that attempt to describe how higher-order conditioning works.

These have a basis in associative learning theories. Among the most famous of these are (Honey & Dweyer, 2022):

• The Rescorla-Wagner Model which stipulates that higher-order conditioning is due to the conditioned stimulus (conditioned stimulus) and unconditioned stimulus (UCS) being associated with each other through a shared response.

  In other words, when the conditioned stimulus and unconditioned stimulus are both present, they share a common response — in the above example, salivation. This shared response makes it more likely that the conditioned stimulus will elicit the same response as the unconditioned stimulus, even when they are not presented together.

• The Model of Temporal Contiguity posits that higher-order conditioning occurs because the conditioned stimulus and unconditioned stimulus are experienced close together in time.

  This proximity strengthens the association between the two stimuli, making it more likely that the conditioned stimulus will elicit the same response as the Unconditioned Stimulus.

• The Model of Inhibitory Processes suggests that higher-order conditioning is the result of the conditioned stimulus (conditioned stimulus) inhibiting or suppressing the response to the unconditioned stimulus (unconditioned stimulus).

  In other words, when the conditioned stimulus is present, it prevents the unconditioned stimulus from eliciting its usual response.

  This suppression makes it more likely that the conditioned stimulus will elicit the same response as the unconditioned stimulus, even when they are not presented together.

• The Model of Stimulus Substitution posits that higher-order conditioning occurs because the conditioned stimulus (conditioned stimulus) comes to substitute for the unconditioned stimulus (unconditioned stimulus).

  In other words, the conditioned stimulus stands in for the unconditioned stimulus and elicits the same response.

The Role of Dopamine

Higher-order conditioning has also been studied at the level of brain mechanism, especially the role of midbrain dopamine signalling.

Seitz, Blaisdell and Sharpe (2021) reviewed evidence that dopamine is needed for both second-order conditioning and a related process called sensory preconditioning. In sensory preconditioning, two neutral stimuli are linked before either is paired with a reward.

These two processes rely on different brain circuits. Second-order conditioning uses the amygdala to carry emotional value forward, while sensory preconditioning depends on the hippocampus and orbitofrontal cortex.

Because dopamine contributes to both despite this split, Seitz and colleagues argue it supports a general capacity for chaining learned relationships together, not just signalling reward on its own.

Evaluative vs. Higher-Order Conditioning

Higher-order conditioning is often confused with evaluative conditioning, but the two work differently:

• Definition: evaluative conditioning is a change in liking or disliking that happens because an object has been paired with an already-positive or already-negative stimulus (Martin & Levey, 1978).
• Steps Involved: higher-order conditioning runs in two linked stages (CS1 to UCS, then CS2 to CS1); evaluative conditioning happens in a single step.
• What Transfers: a new stimulus inherits an existing CS’s response in higher-order conditioning, versus a neutral stimulus inheriting an already-meaningful stimulus’s evaluative tone directly in evaluative conditioning.
• Resistance to Extinction: evaluative conditioning is often more resistant to extinction than ordinary Pavlovian conditioning, partly because attitude-based pairings can happen outside conscious awareness (Martin & Levey, 1978).

Researchers measure evaluative conditioning through self-report, cognitive tasks such as lexical-decision tests, and physiological indices. It happens outside conscious awareness for both positive and negative stimuli, which is part of why it underlies so much advertising (see In Marketing below).

Examples

Critical Evaluation of Higher-Order Conditioning

Before looking at each point in detail, here is how the evidence for higher-order conditioning holds up:

1. Explanatory Range: it accounts for how stimuli that never met a primary reward, like money or brand logos, still shape behaviour and emotion.
2. Cross-Species Support: the effect appears from honeybees to humans, though that does not guarantee the findings transfer directly to complex human behaviour.
3. Fragility: higher-order associations are weaker than first-order ones and fade faster without reinforcement.
4. Theory Lagged Evidence: for most of the twentieth century, no single account of the mechanism was tested rigorously against its rivals.
5. Associative vs Cognitive Debate: the phenomenon fits both a simple associative account and a more cognitive, expectancy-based one, so it does not settle the debate.

Explains Stimuli With No Direct Reward

Money has no biological value on its own, yet people work hard to get it. Brand logos trigger real emotional reactions despite never delivering a reward directly. Higher-order conditioning explains both, because a conditioned stimulus can pass its learned power on to something that never met the original reward.

The pattern repeats everywhere. The applications section above shows this playing out in cue-reactivity, where drug-linked locations and objects trigger cravings despite never containing the drug itself. The Contemporary Research below extends the same logic to how children’s fears spread to animals they were never directly frightened by.

Neither example needed a direct reward.

Without higher-order conditioning, none of this would make sense. Direct conditioning alone cannot explain why cues several steps removed from any reward or punishment still move people to act.

Evidence Across Species, With a Caveat

Hussaini and colleagues (2007) showed second-order conditioning in honeybees, and the same pattern appears in rats, dogs, and humans. That spread across such different nervous systems suggests the underlying mechanism is a basic, ancient feature of how animals learn, not a quirk of mammalian brains.

This does not mean the findings transfer directly to people. Showing that bees and rats can chain associations together proves the mechanism is general. It does not prove it works identically, or carries the same clinical weight, in complex human behaviour such as attitude formation or anxiety.

Bees are not people, though.

Psychologists call this the animal-extrapolation problem, and it applies here as much as anywhere else in comparative psychology. Cross-species evidence supports the mechanism’s generality, but human-specific evidence is still needed before applying it directly.

Weaker and More Easily Lost

Pavlov found that the conditioned response grew weaker at each extra step in the chain, and in dogs it rarely survived past the third or fourth order.

The effect is genuinely fragile.

Because the higher-order stimulus is never directly reinforced, it is also extinguished more easily. This limits how much explanatory weight the theory can carry. A mechanism that fades within a few links cannot, by itself, explain conditioned responses to something many steps removed from any reward, without extra help.

That extra help might come from real-world repetition, direct experience, or verbal instruction, not from the original chain alone.

Not always, though. Williams (2002) notes some higher-order associations can still resist extinction under certain conditions, but this remains the exception rather than the rule.

The Missing Theory, Until Recently

For most of the twentieth century, psychologists offered several rival explanations for how higher-order conditioning works: shared response, temporal contiguity, inhibitory processes, and stimulus substitution.

No one could test them properly. Shared response, temporal contiguity, and inhibitory-process accounts all fit the same handful of findings equally well, so there was no obvious way to rule any of them out.

These accounts were argued informally, in words, rather than tested against each other using a precise, falsifiable model. That left the field in an odd position: everyone agreed the phenomenon existed, but no one could say with confidence which account of the mechanism was correct.

That changed recently.

The gap was unusual, given how central the phenomenon is to Pavlovian theory. The Contemporary Research below explains how Honey and Dwyer (2022) closed it.

Both Sides of the Learning Debate Claim It

Pavlov himself thought higher-order conditioning showed that conditioning is more than a simple reflex. Cognitively oriented researchers have used the same evidence to argue that even ‘simple’ classical conditioning involves learning about relationships between events, not just a mechanical stamping-in of stimulus-response links.

A purely associative account, such as the shared-response version of the Rescorla-Wagner model, and a more cognitive, expectancy-based account can both explain higher-order conditioning’s basic findings equally well. The phenomenon is used as evidence by both camps.

This makes higher-order conditioning a useful test case for the wider debate over whether learning is a mechanical process or an informational one.

Neither side has won this argument outright. The debate matters beyond this one phenomenon, since it shapes how psychologists interpret conditioning experiments more generally.

Contemporary Research

Recent work asks not just whether higher-order conditioning happens, but how it happens mechanistically, and how far it reaches into socially and clinically relevant human learning.

A New Computational Model

Aim: Honey and Dwyer (2022) set out to review the higher-order-conditioning literature and build the first fully specified computational model of how it works.

Method: they synthesised decades of findings across procedures and species, then extended their existing ‘HeiDI’ model of Pavlovian conditioning into a new model built specifically to simulate higher-order effects.

Results: no existing informal account explained the full pattern of results, including why some higher-order pairings excite a response and others inhibit it. Their model explained it all.

Conclusion: Honey and Dwyer (2022) argued that higher-order conditioning helps explain how everyday neutral cues gain emotional weight.

Realising that relevance needs a precise, testable theory like theirs, not just informal accounts.

Second-Order Fear in Children

Aim: Reynolds, Field and Askew (2015) tested whether fear that children pick up just by watching someone else react fearfully can chain onward to a second, unrelated animal.

Method: children aged 5 to 11 saw images of two novel animals, shown either with a fearful adult face or a neutral one. A further animal image was paired only with the first, fear-linked animal, never with a face at all.

The design was simple.

Results: fear beliefs rose for the animals paired directly with a fearful face, as expected. They also rose for the second animal, which the children had never seen alongside a fearful face at all.

The pattern held.

Conclusion: second-order fear conditioning is not limited to direct pairings with something frightening. Children’s fears can spread to objects they were never actually frightened by themselves (Reynolds, Field, & Askew, 2015).

Together, these two studies point the same way. Honey and Dwyer supply, for the first time, a formal account of why an established stimulus can pass a response onward.

Reynolds and colleagues show the same chaining at work in exactly the kind of real-world fear learning the theory is meant to explain.

Further Information

• Behaviorism
• Little Albert Experiment
• Pavlov’s Dogs
• Systematic Desensitization as a Counter-conditioning Process
• Aversion Therapy
• Little Peter (Jones, 1924)
• Some Practical Applications of Classical Conditioning