Canine Logic: Do Dogs Understand Cause and Effect?

1. Introduction

1.1 The Core Distinction

Associative learning is powerful and ubiquitous, and most of what a dog "knows" is built from it – a cue, a behavior, a consequence, strengthened by repetition and precise timing (how reward and prediction drive canine learning). Causal understanding is a stronger claim: it says the animal represents why one event brings about another, and can therefore act correctly in a novel arrangement it has never been reinforced on. The scientific task is to tell these apart, because a dog that succeeds at a task may be reasoning about causes – or may simply be running a well-worn association. The two look identical from the outside and require careful design to separate.

1.2 How to Read the Evidence

Two symmetrical cautions run through the whole field. Success does not prove causal understanding: a dog can pass a task through a simple perceptual rule that mimics reasoning. Failure does not prove its absence: a dog can fail a task it "understands" because of memory load, competing cues, or a confusing setup. Good comparative work tries to close both gaps, and the honest reading of most canine studies lands between them – dogs show real learning about regularities without clear evidence of abstract causal insight. Throughout, the leaner explanation is preferred until the richer one is earned.

1.3 Why This Question Keeps Coming Back

Few questions in canine cognition generate as much disagreement, and the reason is not that the data are unusually poor. It is that the two competing accounts predict the same behavior almost everywhere, so each new study can be read as supporting either.

The disagreement is therefore less about dogs than about what would count as settling it. Naming that at the outset makes the rest of the article easier to read, because the studies described are best understood as successive attempts to build a situation where the two accounts finally diverge.

1.4 What Is Not in Dispute

Some things are established well enough that no serious account denies them. Dogs learn contingencies quickly and hold them for a long time. They attend closely to human behavior and use it as information. They solve novel problems in ways that improve with experience.

The dispute concerns the machinery underneath those abilities, not the abilities themselves — a distinction that popular coverage collapses in both directions, either crediting dogs with reasoning they have not been shown to have or dismissing competence that is well documented (since the observable and the inferred are routinely conflated).

2. What Causal Understanding Means

In cognitive science, causal understanding is the ability to recognize relationships between events and use them flexibly to predict outcomes – not merely learning that B follows A, but detecting the structure that connects them. In dogs it is probed through three domains.

2.1 Means–End Understanding

Recognizing that an intermediate object – the "means" – can be used to obtain a goal, such as pulling a string to draw in food that is out of reach.

2.2 Inferential Reasoning

Drawing a conclusion from indirect evidence. If only one of two shaken cups makes a noise, an animal reasoning causally should infer that the noisy cup contains food.

2.3 Physical Causality

Sensitivity to physical principles such as contact, support, and connectivity – the intuitive "folk physics" that lets an agent predict how objects act on one another.

The central question across all three: do dogs go beyond associative contingencies to reason about causal structure?

2.4 Why the Distinction Is Hard to Test

Associative learning and causal understanding predict the same behavior in most situations, which is why they are difficult to separate experimentally. A dog that has learned that pulling a string produces food behaves identically whether it grasps the connection or has simply been reinforced.

The tasks that distinguish them are therefore artificial by necessity: they arrange configurations where the association points one way and the causal structure another (as the reinforcement literature sets out).

2.5 Neither Is a Deficit

The framing of this question often carries an implicit ranking, with causal understanding as the achievement and association as the consolation prize. That ranking is not warranted.

Associative learning is a powerful, flexible and fast system that solves most problems an animal faces. A species that tracks contingencies extremely well is not a species that failed at something else (with individual variation doing more work than group means suggest).

3. Inference: The Cups Task

3.1 The Paradigm

The cups task is the standard test of inferential reasoning. Two opaque cups are shaken; only one contains food and therefore rattles. An animal using inference should choose the cup that made the sound – reasoning "noise means contents."

3.2 What Dogs (and Wolves) Actually Do

A recent, careful study revisited this paradigm with both wolves and dogs and found that neither species reliably solved it by inference; choices were instead shaped by perceptual salience and the order of presentation, i.e. by surface cues rather than deduction (Rivas-Blanco et al., 2025). This fits a much-cited earlier result in which dogs chose a container that made a noise whether the sound was caused by the food shaking inside or by an arbitrary, non-causal source – they used the noise as a cue without grasping the causal link (Bräuer et al., 2006). That study gave the field its enduring shorthand, "social dog, causal ape": dogs excel at reading human communicative cues but lag apes on physical inference.

The phrase compresses more than it states. Apes are not uniformly strong on physical causality either – performance varies by species, by task and by rearing – and the contrast rests on a particular set of paradigms rather than on a general ranking. It is a useful label for a real pattern and a poor summary of primate cognition.

3.3 The Domestication Angle

Comparing dogs with wolves sharpens the picture. When dogs and equally raised wolves were tested on communicative, behavioral, and causal cues, the wolves outperformed the dogs at following causal cues, while developmental history (pack-raised versus pet) made surprisingly little difference (Lampe et al., 2017). The implication is that domestication did not raise general intelligence; it reshaped a specific profile – tuning dogs toward the social while, if anything, leaving physical-causal skills behind. Where dogs win is the human channel (how dogs read human gestures), not the physical one.

3.4 Why Wolves Are the Comparison

Wolves are used as the comparison group because they share the ancestry without the domestication, which makes them the natural test of whether a canine ability is a product of living with humans. The logic is sound and the practice is harder than it sounds.

Wolves in research settings are hand-raised, intensively socialized and few in number, while pet dogs are numerous and variably reared. Any species difference is therefore confounded with rearing, sample size and testing experience, and the studies that have tried to separate these do not fully agree (with individual variation doing more work than group means suggest).

A second confound is rarely mentioned. Research wolves live in packs at a small number of institutions and are tested repeatedly across many studies over years, which makes them unusually experienced subjects. Pet dogs recruited from the public are typically naive. Whichever way a comparison comes out, testing experience differs systematically between the groups.

3.5 What Above-Chance Performance Means

A group performing above chance does not mean the individuals in it solved the task. It is compatible with most animals choosing at random and a minority choosing correctly, and with a modest bias across all of them.

Reporting the spread alongside the mean would settle this, and it is done less often than it should be.

4. Means–End: String-Pulling

4.1 The Proximity Error

String-pulling tests whether an animal understands connectivity: only the string actually attached to the reward will work. In a foundational study, dogs succeeded when the connected string ran straight toward the reward but failed when it ran at an angle, committing a "proximity error" – pawing or mouthing at the string end nearest the food regardless of whether it was connected (Osthaus et al., 2005). This points to a spatial heuristic ("go for what's closest to the goal") rather than spontaneous means–end reasoning.

4.2 Learning Connectivity with Experience

The picture is not entirely negative. Testing 34 Border collies on string tasks that varied how close the reward sat to the correct string's end, researchers found dogs initially defaulted to proximity but that some individuals learned, with experience, to attend to connectivity instead – overcoming the proximity bias once the misleading cues were reduced (Riemer et al., 2014). And in "support" versions of the problem, where proximity cues are inherently less misleading than in classic string tasks, dogs perform noticeably better. The lesson is twofold: dogs rarely show spontaneous causal insight, but they can extract causal regularities through experience (a capacity tied to their broader behavioral flexibility), and task design heavily shapes what they appear to "understand."

4.3 What the Learning Result Adds

That dogs can learn to attend to connectivity is a different claim from spontaneous understanding, and it is the more interesting of the two. It establishes that the information is available to the animal and that the default strategy is simply cheaper.

A dog reaching for the nearest string is not failing to perceive the connection; it is using a rule that works nearly always for far less effort. Under that reading the proximity error is efficient rather than stupid (as the reinforcement literature sets out).

5. Overimitation: When Dogs Copy the Useless

5.1 Copying Causally Irrelevant Actions

A striking line of work turns the question around. When a human caregiver demonstrated an action that included a causally irrelevant step – a manipulation unnecessary for getting the reward – a reasonable number of dogs copied the useless step anyway (Huber et al., 2020). This "overimitation" is prevalent in human children and essentially absent in great apes, which makes its appearance in dogs notable.

5.2 Social Motivation over Causal Efficiency

Crucially, this is not read as a failure of causal reasoning but as evidence of its opposite priority: dogs are sensitive to which actions matter, yet in affiliative contexts they will subordinate causal efficiency to social alignment – doing what their person did because their person did it (a form of social learning). It is the same theme from another angle: the dog's mind is tuned to the social contingency, sometimes at the expense of the physical one.

5.3 Why Overimitation Is a Causal Result at All

Copying an action that visibly does nothing is informative precisely because it runs against causal efficiency. An animal reasoning about mechanism should drop the useless step; an animal treating the demonstration as a social instruction should keep it.

Dogs frequently keep it, which is a finding about what they weight rather than about what they can perceive (as the judgement-bias paradigm handles the same inference problem).

5.4 What It Means for a Handler

If demonstrations are read socially rather than mechanically, then everything a handler does during a demonstration is potentially part of the lesson, including the parts that were incidental.

That is worth knowing before concluding that a dog has learned something odd. It usually learned exactly what was shown (because transfer across contexts is not automatic).

6. Implicit Expectations: The Eye-Tracking Evidence

6.1 A Different Kind of Question

The tasks described so far ask what a dog will do. A second family of methods asks what a dog appears to expect, by presenting an event that either follows a physical regularity or violates it and measuring the reaction.

This matters because acting and expecting can come apart. An animal may fail a manual task for reasons of motor control, motivation or attention while still showing a surprise response when the physics goes wrong (which is what operationalizing a behavioral claim is for).

6.2 The Launching Event

Dogs were shown realistic three-dimensional animations of launching events — one object moving toward another, which then moves off — either with contact between the objects or without it. In both conditions the objects moved with the same timing and kinematic properties, so the only difference was whether they touched (Völter & Huber, 2021).

From an early age, humans perceive spatiotemporally contiguous launching events as causal. Whether anything comparable happens outside the primate order was largely unknown.

6.3 What Was Measured

The study combined eye tracking with pupillometry, a violation-of-expectation approach adapted from work with human infants. Pupil dilation is used alongside looking time because it responds quickly and is less vulnerable to test-order effects (Völter & Huber, 2021).

6.4 The Result

The dogs tracked the object movements closely throughout, but their pupils were larger in the no-contact condition, and they looked longer at the object that initiated the launch after the no-contact event than after the contact event (Völter & Huber, 2021).

The authors conclude that dogs have implicit expectations about contact causality. An object that starts moving with nothing touching it registers as something.

The direction of the looking result is worth noting, because it is not the obvious one. After the no-contact event the dogs looked longer at the object that started the movement rather than at the one that moved without being touched, which is the reverse of what a simple novelty account would predict and part of why the authors read the pattern as an expectation about the causal relationship rather than about the visual display.

6.5 What "Implicit" Is Doing in That Sentence

The word is carrying weight and the authors chose it deliberately. An implicit expectation is a regularity the perceptual system is tuned to, not a belief the animal could act on or report.

It is the difference between being surprised that a ball moved by itself and understanding why balls normally do not. The first is what was measured; the second is what a reader is tempted to infer (since the observable and the inferred are routinely conflated).

6.6 The Size of the Study

Fourteen dogs took part. That is normal for eye-tracking work, which is labour-intensive and requires animals that will sit still in front of a screen, and it is a small number on which to rest a claim about the species.

It is also a screen-based paradigm. Whether expectations demonstrated with animations extend to physical objects a dog could approach is a separate question that this design does not address (because transfer across contexts is not automatic). Dogs are known to treat screens differently from real objects in other contexts, which makes the question a real one rather than a formality.

7. When a Finding Does Not Replicate

7.1 The Original Claim

Solidity — the principle that one solid object cannot pass through another — is among the most basic physical expectations, and an earlier study had reported that dogs use it spontaneously. That result circulated widely, as basic-competence findings tend to.

7.2 The Follow-Up

A subsequent investigation presented 112 dogs with seven different versions of the task. Its conclusion stood in stark contrast to the earlier claim that dogs spontaneously show an understanding of the solidity principle (Müller et al., 2014).

One hundred and twelve dogs across seven task variants is a serious attempt at the question, and it is unusual in this literature for its size.

7.3 Why This Belongs in the Article

The pattern it illustrates recurs throughout canine cognition: a striking positive finding from a small sample, widely repeated, followed by a larger and better-controlled study that does not reproduce it.

The follow-up rarely travels as far as the original. Popular summaries of dog cognition were written from the first result and have not been updated (a problem early testing runs into as well).

There is no bad faith in this. A positive finding about a familiar species is interesting to write about, a null result is not, and the incentives that govern which papers get covered are the same ones that govern which get published. The consequence is a public picture of canine cognition systematically tilted toward the more flattering answer, assembled from studies that were each reported accurately at the time.

7.4 What It Does Not Show

A failure to replicate is not proof of absence. It establishes that the earlier evidence does not support the claim, which is a weaker and more useful statement than saying dogs lack the competence.

The honest position after such a result is that the question is open, not that it has been settled in the negative. That distinction is routinely lost in both directions.

7.5 Reading the Two Findings Together

Set beside the eye-tracking work above, a picture emerges that is more interesting than either result alone. Dogs may show implicit perceptual expectations about physical regularities while failing tasks that require acting on those regularities.

That dissociation between what an animal appears to expect and what it can use is well documented in comparative cognition, and it is the most likely shape of the answer here (with individual variation doing more work than group means suggest).

8. What These Paradigms Can and Cannot Establish

8.1 The Object Choice Task

Two containers, a hidden reward, a cue, a binary choice. The design is simple, quick and comparable across species, and it compresses a rich question into right or wrong.

Every conclusion drawn from it travels through that compression, and above-chance performance is compatible with several explanations that the task cannot separate (which is what operationalizing a behavioral claim is for).

8.2 String-Pulling

String-pulling asks whether an animal grasps that a physical connection is what makes pulling work. It is confounded by proximity: dogs frequently pull whichever string is nearest the reward, which is a spatial rule that succeeds in most natural situations without any understanding of connectivity.

Distinguishing the two requires configurations where proximity and connection point to different answers, which is exactly what the later studies in this area were designed to do.

8.3 Violation of Expectation

Looking time and pupil dilation measure a reaction to something unexpected. They do not identify what the animal expected, only that something registered as different.

A longer look at a no-contact launching event is consistent with an expectation about causality and also with an expectation about the visual pattern the dog has seen before. Ruling out the second requires further conditions, and the strength of a violation-of-expectation study lies in how many alternative readings its controls exclude.

8.4 The Perception-Action Gap

Perception-based methods tend to make animals look more competent than action-based ones, in dogs as in primates. That is not a reason to prefer one family of results; it is a reason to report which family a claim comes from.

Most confident statements about what dogs understand rest on one or the other without saying which (since the observable and the inferred are routinely conflated).

8.5 Sample Sizes and Where They Sit

The eye-tracking work here involved 14 dogs, the solidity follow-up 112. Both numbers are appropriate to their method, and they cannot carry the same weight.

A result from 14 animals establishes that an effect can be detected under those conditions. A result from 112 across seven variants establishes something about how robust an effect is. Treating the two as equivalent evidence is a common error in summaries (as the judgement-bias paradigm handles the same inference problem).

8.6 What Would Settle More of This

The design this field most needs is not another paradigm but the same dogs tested across several. Perception-based and action-based methods disagree systematically, and nobody knows whether that reflects two capacities within an individual or two populations of dogs behaving differently.

Testing the same animals on a violation-of-expectation task and a manual task addressing the same principle would answer that, and it is well within reach of an existing laboratory.

8.7 Why Nobody Has Done It

The obstacle is not conceptual. Eye-tracking work and manual task work tend to be run by different people with different equipment, and a study designed to satisfy both traditions is harder to publish than one aimed at either.

This is a familiar pattern in comparative cognition and it produces exactly the situation described above: two literatures about the same species, each internally coherent, disagreeing without ever meeting.

9. Neural Considerations

9.1 What the Neural Picture Suggests

Direct neural evidence for causal reasoning in dogs is essentially absent – no imaging study has isolated a "causal inference" network in the dog brain. What can be said is general: the regions implicated across mammals in flexible decision-making and procedural learning – the prefrontal cortex, the basal ganglia, and the cerebellum – are the plausible substrates for the experience-driven behavioral adjustments seen in these tasks. The prefrontal contribution to inhibiting a prepotent response (such as the pull-toward-proximity reflex) is especially relevant (the prefrontal cortex and canine self-control), and how these systems support learning is exactly the kind of question future canine neuroimaging may address (how the dog brain underpins behavior). At present, conclusions about canine causal reasoning rest primarily on behavioral evidence rather than direct measurements of neural activity

9.2 What Canine Neuroscience Can and Cannot Contribute Here

Awake canine imaging exists and has produced real findings, mostly about reward, social stimuli and odour. It has not been used to address causal reasoning, and the neural material in this chapter is accordingly drawn from other species.

Prefrontal contributions to inference are described from human and primate work. No canine study has related activity in any region to performance on a causal task, which means the neural section explains a plausible mechanism rather than reporting a measured one (which is what operationalizing a behavioral claim is for).

10. Research Gaps and Methodological Challenges

The interpretive difficulties here are not a footnote; they are the heart of the field.

Success versus understanding. Passing a task can reflect abstract causal representation or a simple rule that happens to work. Distinguishing the two is the central methodological problem, and it is rarely fully solved.

Failure versus absence. A dog that fails may lack causal understanding – or may be defeated by memory demands, competing perceptual cues, or an unfamiliar apparatus. Null results are genuinely ambiguous.

Fragile generalization. Dogs often stumble when a task is only slightly altered, which is the opposite of what robust, abstract causal knowledge would predict – and a reason for caution about rich interpretations.

Over-read behaviors. Even the appealing observation that dogs "look back" at humans when stuck has been challenged as over-interpreted, once motivation and task framing are controlled. Convenient narratives outrun the data easily here (the difficulty of measuring behavior objectively), and whether dogs monitor their own knowledge at all is itself unsettled (metacognition in dogs).

Small, narrow samples. Much of this work uses modest samples and an overrepresentation of a few breeds (Border collies especially), complicating generalization to dogs at large.

Perception-based and action-based results disagree systematically. Eye-tracking work suggests implicit physical expectations (Völter & Huber, 2021) while manual tasks frequently show failure. No study has tested the same dogs on both to establish whether the dissociation holds within individuals.

Sample sizes differ by an order of magnitude between methods. Fourteen dogs in the eye-tracking study against 112 in the solidity follow-up (Müller et al., 2014). Comparing conclusions across that gap requires care that summaries rarely take.

Replication is uneven. At least one widely repeated claim about basic physical understanding did not survive a larger, better-controlled test, and the corrected picture has not displaced the original in popular coverage.

The neural account is borrowed. No canine imaging study has related brain activity to performance on a causal reasoning task, so the mechanism described here is extrapolated from other species.

11. Practical Implications

11.1 Training as Structured Exploration

Because dogs can extract causal regularities from consistent experience, training can go beyond rote reinforcement. Letting a dog interact with a mechanism and observe reliable, repeated contingencies – rather than only drilling a fixed response – can build more flexible, better-generalized learning (which rests on the same reward machinery). The dog may not grasp the physics, but it can learn the regularity.

11.2 Predictability and Emotional Stability

Causal clarity creates predictability, and predictability is calming. When outcomes reliably follow specific behaviors, a dog experiences a sense of control over its environment, and reliable contingencies reduce chronic stress and support emotional regulation (why chronic unpredictability taxes the brain). Much of what makes training feel "fair" to a dog is simply that its world becomes legible.

11.3 The Punishment Problem

The associative-versus-causal distinction exposes a specific hazard of punishment. Because dogs often bind an outcome to whatever is perceptually salient rather than to their own behavior, an aversive consequence can attach to the wrong thing entirely – the owner's presence, a location, a nearby dog – instead of the action it was meant to address (the neurological cost of aversive methods). Clear, consistent, well-timed reinforcement produces far more accurate contingency learning, whereas ambiguous punishment often teaches a lesson no one intended (and suppressed behavior tends to return anyway).

11.4 Working With the Social Mind

The deepest practical point follows from "social dog, causal ape." A dog faced with a hard problem often looks to its human rather than to the mechanism – so the most effective training works with that social orientation rather than demanding a physical reasoning the dog does not have. This is the same profile that shows up across canine cognition generally (what dogs are and aren't good at thinking about).

11.5 Why This Matters for Troubleshooting

If a dog's competence rests largely on tracking contingencies rather than grasping mechanisms, then a behavior that fails in a new setting has usually not been forgotten. The contingency the dog learned included features of the original setting that nobody intended to teach (because transfer across contexts is not automatic).

That reframes most training failures as information about what was actually learned rather than as disobedience or as evidence that the dog cannot generalize.

11.6 What Not to Expect

Two expectations are worth setting aside. A dog is unlikely to work out why an arrangement produces an outcome from watching it once, and it is unlikely to infer the correct action in a novel physical problem by reasoning about mechanism.

What it will do reliably is notice what predicts what, including things the handler is not aware of doing. Designing for that is more productive than training against it.

11.7 Build the Contingency You Want

The practical consequence of a contingency-tracking animal is that the arrangement teaches more than the intention does. A behavior that is reinforced sometimes by the handler and sometimes by the environment is being shaped by both, and the environment does not take breaks.

Deciding in advance what should predict what — and then arranging it — is more effective than correcting the associations that form when nobody decided.

11.8 Where Causal Language Misleads Owners

Phrases such as "he knows what he did" and "he understands why" assume a kind of reasoning the evidence does not support and that shapes how households respond. A dog that returns to a torn cushion with a lowered posture is reading the handler, not confessing.

Replacing the causal story with a contingency one changes what an owner does next, which is the whole practical point of this article (since the observable and the inferred are routinely conflated). It also lowers the temperature of the conversation, because a dog that did not understand cannot have been defying anyone.

12. Summary at a Glance

Dogs show implicit expectations about contact causality — Pupils were larger and looking times longer after launching events without contact than with it, in animations matched for timing and kinematics (Völter & Huber, 2021).

That study involved 14 dogs and a screen — Normal for eye tracking, and a narrow base for a claim about the species, with no test of whether the expectation extends to physical objects.

An earlier solidity finding did not survive scrutiny — A study of 112 dogs across seven task versions reached a conclusion in stark contrast to the earlier claim that dogs spontaneously understand the solidity principle (Müller et al., 2014).

Failure to replicate leaves the question open — It establishes that the earlier evidence does not support the claim, not that the competence is absent.

Perception and action come apart — Dogs may show expectations they cannot act on, which is documented across comparative cognition and is the likely shape of the answer here.

Social cues outcompete physical ones — Where a human demonstration conflicts with physical efficiency, dogs frequently follow the demonstration.

The paradigms constrain the conclusions — A binary choice task, a string-pulling apparatus and a looking-time measure each answer a different question, and confident summaries rarely say which was used.

13. Conclusion

The balanced position is clear and worth stating plainly. Dogs do not demonstrate robust, spontaneous, flexible human-like causal reasoning of the kind seen in humans or great apes, though experience can teach them specific causal regularities; across many experimental conditions they rely on perceptual salience and proximity-based shortcuts rather than logical inference (Bräuer et al., 2006; Osthaus et al., 2005; Rivas-Blanco et al., 2025). But this is not an absence of causal sensitivity. 

Dogs can adjust behavior to experienced regularities, particularly once misleading cues are stripped away (Riemer et al., 2014), and they are exquisitely attuned to social contingencies – sometimes to the point of copying a person's useless actions (Huber et al., 2020). Their causal understanding is best described as limited, experience-dependent, and domain-specific, shaped by an evolutionary history that specialized them for life with humans (Lampe et al., 2017).

Dogs, in short, are not intuitive physicists. They do not build abstract theories of how the physical world works. What they are is superb observers of contingencies – above all those involving us. Recognizing that shifts the whole frame: instead of expecting human-like reasoning and being disappointed, we can work with the considerable contingency-tracking abilities the dog actually has.

Key Insights (Takeaways)

• The key distinction is between associative learning (B follows A, learned by reinforcement) and causal understanding (grasping the structure that makes A cause B, so the knowledge transfers to new situations). Most of what dogs do is the former, and telling the two apart is the field's central and hardest problem.

• On inference tasks, neither dogs nor wolves reliably reason their way to the answer; they follow perceptual salience and order (Rivas-Blanco et al., 2025), and dogs treat a noise as a cue without grasping its causal source (Bräuer et al., 2006) – the origin of the "social dog, causal ape" summary. Tellingly, wolves outperform dogs on causal cues (Lampe et al., 2017), so domestication reshaped the profile rather than raising general intelligence.

• On means–end string-pulling, dogs default to a "proximity error," going for the string nearest the reward rather than the connected one (Osthaus et al., 2005) – but some individuals learn to attend to connectivity with experience (Riemer et al., 2014). Spontaneous causal insight is rare; learning about causal regularities is real.

• Dogs will "overimitate," copying causally irrelevant actions from a caregiver (Huber et al., 2020) – not a reasoning failure but a sign that social alignment can outrank causal efficiency. Direct neural evidence for canine causal reasoning is essentially absent; the story is inferred from mammalian decision-making and procedural-learning systems.

• Practically: build predictable, consistent contingencies (they calm as well as teach); prefer clear reinforcement to punishment (dogs readily misattribute aversive outcomes to salient bystanders rather than to their own behavior); and train with the dog's social orientation rather than demanding physical reasoning it does not possess.

References

Bräuer, J., Kaminski, J., Riedel, J., Call, J., & Tomasello, M. (2006). Making inferences about the location of hidden food: Social dog, causal ape. Journal of Comparative Psychology, 120(1), 38–47. https://doi.org/10.1037/0735-7036.120.1.38

Huber, L., Salobir, K., Mundry, R., & Cimarelli, G. (2020). Selective overimitation in dogs. Learning & Behavior, 48(1), 113–123. https://doi.org/10.3758/s13420-019-00400-w

Lampe, M., Bräuer, J., Kaminski, J., & Virányi, Z. (2017). The effects of domestication and ontogeny on cognition in dogs and wolves. Scientific Reports, 7, 11690. https://doi.org/10.1038/s41598-017-12055-6

Müller, C. A., Riemer, S., Range, F., & Huber, L. (2014). Dogs' use of the solidity principle: Revisited. Animal Cognition, 17(3), 821–825. https://doi.org/10.1007/s10071-013-0709-9

Osthaus, B., Lea, S. E. G., & Slater, A. M. (2005). Dogs (Canis lupus familiaris) fail to show understanding of means–end connections in a string-pulling task. Animal Cognition, 8(1), 37–47. https://doi.org/10.1007/s10071-004-0230-2

Riemer, S., Müller, C., Range, F., & Huber, L. (2014). Dogs (Canis familiaris) can learn to attend to connectivity in string-pulling tasks. Journal of Comparative Psychology, 128(1), 31–39. https://doi.org/10.1037/a0033202

Rivas-Blanco, D., Krause, S. D., Marshall-Pescini, S., & Range, F. (2025). Inference in wolves and dogs: The "cups task," revisited. Animal Behaviour, 227, 123268. https://doi.org/10.1016/j.anbehav.2025.123268

Völter, C. J., & Huber, L. (2021). Dogs' looking times and pupil dilation response reveal expectations about contact causality. Biology Letters, 17(12), 20210465. https://doi.org/10.1098/rsbl.2021.0465

Explore more research articles on dog behavior and training in our Research Library.