Behavior Does Not Equal Emotion: Limits of Inferring Internal States from Observable Actions

1. Why Behavior Underdetermines Emotion

1.1 The Core Problem

The same observable behavior can arise from different internal states, and the same internal state can produce different behaviors. This is underdetermination, and it is not a technicality — it is the reason a single cue can never settle the question.

At least four factors sit between emotion and its behavioral expression. Learning history: a dog punished for growling has learned to suppress the signal, not to stop feeling what produced it. Context: lying down means one thing on a sofa at home and another in a veterinary waiting room. Physiological state: pain, illness, fatigue, and hunger alter behavior independently of emotion (and pain is the most frequently missed of these). Conflicting motivations: a dog can want to approach and want to withdraw at the same time, producing behavior that belongs cleanly to neither (approach-avoidance conflict in detail).

1.2 What Follows

Inferring emotion from behavior requires information the behavior itself does not contain. That information comes from context, from knowledge of the individual dog's history, from physiological measures where available, and from what the dog does when given a genuine choice.

This is not a counsel of despair. Behavior remains the primary route into canine emotional states. It is a counsel of method: single cues are hypotheses, not conclusions (which is precisely the case for operationalizing behavioral measurement).

2. Four Behaviors That Do Not Mean What They Seem

2.1 Tail Wagging

The most consistently over-read behavior in dogs. A loose, sweeping, whole-body wag frequently accompanies positive arousal — but wagging also occurs in uncertainty, conflict, and immediately before aggression. High, stiff, rapid wagging indicates high arousal without specifying its valence.

There is also a lateralization finding: Siniscalchi and colleagues (2013) reported that dogs respond differently to left- versus right-biased tail wagging in conspecifics, consistent with hemispheric asymmetry in emotional processing. It is a real effect and it is not usable in real time by an owner watching a dog across a park.

The practical statement is narrow and worth stating precisely: wagging indicates arousal, not valence. "But he was wagging his tail" is among the most common sentences after a bite.

2.2 Stillness

A still dog may be relaxed, may be freezing, may be executing a trained down-stay, or may have stopped offering behavior altogether. These are four different internal situations producing one observable output.

The differentiating cues are subtle but real: muscle tone (loose versus tense), breathing (slow and regular versus shallow, rapid, or held), eyes (soft with normal blinking versus wide, dilated, reduced blinking, sclera visible), and responsiveness (a relaxed dog can be reached by its name or a piece of food; a frozen or shut-down dog often cannot).

Food refusal in a dog that normally takes food is the single most useful marker in this cluster (as it also is in reactive episodes).

2.3 Approach

Approach is routinely read as sociability. It can equally be produced by conflict — a dog moving toward a person with a low tail, flattened ears, and a tense body is approaching, but the motivation may be appeasement or social pressure rather than interest.

Parr-Cortes et al. (2024) provide a pointed illustration. Dogs exposed to the odour of an unfamiliar stressed person were slower to approach one of three ambiguous locations in a cognitive bias test — but among dogs that did approach, the behavior itself looked the same. The internal state differed; the visible action did not.

2.4 Vocalization

Growling, whining, and barking are commonly mapped one-to-one onto aggression, anxiety, and excitement. The mapping is coarser than that: growls occur in play, in resource guarding, in fear, and in pain; whining accompanies anxiety, frustration, anticipation, excitement, and pain alike.

The research picture here is more encouraging than the training literature usually suggests, and it needs stating accurately. Pongrácz, Molnár and Miklósi (2006) found that human listeners categorized dog barks by recording situation well above chance, and that emotionality ratings correlated with measurable acoustic parameters — peak frequency, fundamental frequency, and inter-bark intervals — with almost no effect of dog ownership or breed familiarity.

The correct reading is therefore not that barks are uninterpretable. It is that humans extract coarse information from barks via general mammalian acoustic rules — low and harsh tends toward threat, high and tonal toward fear or affiliation — and that this coarse read supports arousal and rough valence, not fine-grained emotional labeling. Related work found that some categories, such as play barks, are considerably harder to identify than others.

3. Masking, Regulation, and Shutdown

3.1 Three States That Look Alike

Masking is learned inhibition of the outward expression while the internal state persists or worsens. Emotional regulation is genuine change in the internal state, as produced by counterconditioning. Shutdown, usually discussed under learned helplessness, is the passive state that follows repeated uncontrollable aversive experience.

All three can produce a quiet, compliant, easy-to-handle dog. Only one of them is a good outcome.

3.2 How Masking Develops

The mechanism is straightforward operant learning applied to signals rather than actions. A dog growls when frightened and is corrected. Growling declines. The fear does not — and may increase, because the correction confirmed that the situation was dangerous.

The canine evidence for this general process is strong. Schilder and van der Borg (2004) found that guard dogs trained with electric shock continued to show stress-related behavior in sessions where no shock was delivered, having learned that their handler's presence and commands predicted it. The behavior was under control; the emotional learning had gone somewhere else entirely (the wider fallout of aversive methods).

Masking itself is a clinical model rather than a directly tested canine construct — but it is consistent with that data and with the well-documented consequences of chronic unresolved stress (the physiology of which is covered separately).

3.3 Shutdown Is Not Calm

The modern account of learned helplessness differs from the classic one: passivity is the default mammalian response to prolonged uncontrollable adversity, and what is actually learned — via frontal circuitry detecting control — is the presence of control, which inhibits that default (Maier & Seligman, 2016).

The clinical consequence is specific. A shut-down dog does not need less stimulation; it needs restored controllability (the full revision of that account).

3.4 Why This Matters Practically

A dog that has been taught not to warn removes the signals that make a bite predictable. Suppression also fails under load: behavior held down rather than changed tends to return when the suppressing conditions weaken (the same competitive-memory dynamic that governs extinction).

4. Arousal and Valence as Separate Dimensions

4.1 The Two-Dimensional Model

Contemporary affective science treats emotional states along two largely independent axes: arousal, the intensity of physiological activation, and valence, the pleasant-unpleasant quality (Mendl, Burman & Paul, 2010). Four combinations follow, and each has a familiar canine version — low arousal with positive valence (settled, content), low arousal with negative valence (withdrawn, shut down), high arousal with positive valence (excited, playful), high arousal with negative valence (fearful, frustrated, defensive) (where it tips into clinical anxiety).

Behavioral observation alone frequently fails to separate high-arousal states of opposite valence. A dog throwing itself at the door may be delighted or terrified (which is why arousal regulation is its own subject).

4.2 What the Measurement Evidence Shows

Flint et al. (2024) tested this systematically in 60 dogs across six scenarios designed to vary in valence and arousal. Cortisol, ACTH, heart rate variability, panting, whining, and body shake all differentiated arousal levels — but only within negative-valence scenarios. Heart rate tracked arousal similarly across both. Qualitative behavioral assessment was the best available indicator of positive valence.

Two conclusions follow, and both are uncomfortable for confident behavioral reading. Heart rate is the more valence-independent arousal measure. And the highly aroused, apparently happy dog is precisely the case where the standard indicator set performs worst.

4.3 The State of the Field

Csoltova and Mehinagic (2020) reviewed the assessment of positive emotion in dogs and reached a conclusion worth quoting in substance rather than in words: no single indicator of positive emotional state in dogs has been validated. The field has invested far more in detecting negative states than positive ones, and the absence of negative signs is not evidence of a positive state.

5. Four Common Misreadings

5.1 The Guilty Look

Horowitz (2009) tested this directly. The posture owners read as guilt — lowered head, averted gaze, tucked tail, flattened ears — appeared in response to the owner's scolding, not to the dog's actual prior behavior. Dogs that had not misbehaved showed it when owners believed they had. It is appeasement directed at a present human, not an admission about the past.

5.2 "He's Calm Now"

A dog that stops pulling after a leash correction, or stops barking after a reprimand, is frequently described as having calmed down. The alternative reading — suppression under an unchanged or elevated internal state — produces identical behavior and requires separate evidence to rule out.

The default assumption should run the other way: if a behavior stops abruptly following something aversive, treat suppression as the working hypothesis until something else indicates otherwise.

5.3 "He Loves Hugs"

This one deserves an honest note about the evidence, because it is widely repeated with a confidence the underlying material does not support. The frequently cited claim that around 80% of dogs show stress signals while being hugged originates from an informal analysis of 250 internet photographs published as a magazine column, with a single rater, no control condition, no statistical test, and no peer review. It was substantially criticized on those grounds.

The underlying principle is nonetheless sound and does not depend on that analysis: restraint is not intrinsically rewarding, tolerance is not preference, and a dog that does not leave has not thereby consented. The right response is not to cite a percentage but to test the individual dog — see §6.2.

5.4 "He's Friendly, He's Wagging"

Covered in §2.1, and worth repeating because of its consequences. Wagging reports arousal. Valence has to come from elsewhere: body outline, facial tension, ear carriage, weight distribution, and what the dog does when given room to leave.

6. Assessing Emotional State More Accurately

6.1 Multi-Channel Assessment

No single channel is sufficient. Four should be combined.

Context and history — what immediately preceded this, and what has happened to this dog in comparable situations before. Whole-body reading rather than single cues — ear position, eye shape, mouth tension, tail position and movement quality, posture, respiratory rate, weight shift. Physiological indicators where available — heart rate is the most practical and, per Flint et al. (2024), the most valence-robust. Choice behavior — what the dog does when approach and withdrawal are both genuinely available.

A caution belongs here too: Bremhorst and colleagues have examined whether canine facial expressions function as accurate emotion indicators across contexts, and the answer is more qualified than the popularity of body-language charts implies. Concordance across multiple channels matters more than confidence in any one.

6.2 The Consent Test

The most useful practical tool, and the one that converts observation into evidence. Stop the interaction — petting, handling, play — and wait. If the dog re-engages, moves closer, nudges, or resumes contact, the interaction was likely positively valenced. If it moves away, turns its head, disengages, or freezes, it was not.

This works because it measures the dog's preference directly rather than inferring it from tolerance. It applies to grooming, to veterinary handling, to greetings from visitors, and to hugs.

6.3 Indicators That Lean Positive

Soft, rounded body outline; loose, whole-body movement; play bows; relaxed open mouth; normal blinking and soft eye shape; voluntarily seeking proximity; a wide, low, sweeping tail arc. These lean positive in combination, and none of them is diagnostic alone.

6.4 Indicators That Lean Negative or High-Arousal

Tense, stiff outline; rapid shallow breathing; dilated pupils; lip licking and yawning outside eating and fatigue contexts; sclera visible; ears back or flattened; tail tucked or held high and rigid; freezing or slowed movement; displacement behaviors such as scratching, sudden sniffing, or shaking off while dry.

6.5 Rule Out the Body First

Before attributing behavior to emotion, rule out pain, illness, sensory decline, and neurological change. The association between pain and problem behavior is well documented (Mills et al., 2020), and an irritable dog that avoids handling is a medical hypothesis before it is a behavioral one (including the less visible forms of pain).

7. Why Getting It Wrong Matters

Misreading is not an academic failure; each error has a predictable consequence.

Reading a frozen dog as calm leads to continued exposure to a trigger the dog cannot cope with, which worsens the underlying fear and can produce defensive aggression. Punishing a growl removes the warning while leaving the fear, raising bite risk. Reading pain as stubbornness or dominance leaves a treatable medical problem untreated (and dominance framing is unsupported in any case). Believing a dog is choosing to misbehave makes punishment feel justified, which for a fearful dog compounds the original problem. And a masked or shut-down dog often receives no help at all, because from the outside nothing appears wrong.

Handler state belongs in this list. Dogs are affected by human stress signals (Parr-Cortes et al., 2024), and secure attachment to the handler is associated with better cortisol regulation during stressful separations (Schöberl et al., 2016) (attachment styles in dogs in detail) (how emotional states transfer between species).

8. Summary: Behavior and Emotion at a Glance

Behavior — What it is: observable action. What it reports: the joint output of emotion, learning history, context, and physiology. What it cannot do: identify an internal state on its own.

Arousal — What it is: intensity of physiological activation. Read from: heart rate, respiration, pupil size, muscle tension, activity level. Limit: says nothing about whether the state is pleasant or unpleasant.

Valence — What it is: the pleasant-unpleasant quality. Read from: approach versus withdrawal given real choice, re-engagement after a pause, cognitive bias in research settings, concordant whole-body cues. Limit: no single validated indicator of positive valence exists in dogs.

Masking — What it is: learned inhibition of expression with the internal state intact. Read from: absence of expected warning signals, stillness combined with muscle tension, abrupt behavioral change following something aversive. Risk: false reassurance, elevated bite risk, chronic stress.

Shutdown — What it is: passivity following uncontrollable adversity. Read from: reduced offering of behavior, low responsiveness, apparent compliance without engagement. Response: restore controllability rather than reduce stimulation further.

9. Research Gaps and Critical Appraisal

Positive emotion in dogs has no validated indicator. This is the central limitation of the whole field (Csoltova & Mehinagic, 2020), and it means confident claims that a dog is "happy" are weaker than confident claims that a dog is stressed.

The standard indicator set is validated mainly for negative valence. Cortisol, ACTH, HRV, panting, whining, and body shake differentiated arousal only within negative-valence scenarios; heart rate performed across both (Flint et al., 2024).

Physiological measures are peripheral. Salivary cortisol and heart rate correlate with central processes rather than measuring them, and cortisol in particular responds to arousal generally, not to distress specifically.

Masking is a clinical model. It is coherent, consistent with the shock-collar data, and lacks a dedicated canine experimental literature.

Bark research is often cited backwards. Pongrácz et al. (2006) found humans categorized barks above chance with little effect of experience — the honest limitation is the coarseness of that read, not its failure.

Facial expression accuracy is contested. Whether canine facial expressions reliably indicate emotion across contexts is an open question rather than a settled foundation for body-language charts.

Individual variation is large. Breed, temperament, age, and history all shape how a given state gets expressed (as the temperament and coping-style literature describes), and early experience shapes the baseline (particularly during the sensitive period).

10. Conclusion

Behavior is the visible portion of something larger, and what makes it visible is not only emotion but learning history, context, physiology, and competing motivations pulling in different directions. The same tail wag can accompany delight or the moments before a bite; the same stillness can mean contentment, fear, obedience, or collapse. This does not make behavior useless — it is the primary route into a dog's internal state and will remain so. It makes single-cue reading indefensible. The workable alternative is unglamorous: read the whole body, know the history, weigh the context, measure arousal separately from valence, rule out pain before attributing anything to attitude, and where possible let the dog answer directly by stopping and seeing whether it comes back. Above all, treat a sudden improvement following something unpleasant as suppression until shown otherwise. A dog that has gone quiet has told you less than it seems to have, and a dog that has stopped growling may be closer to biting than it has ever been.

Key Insights (Takeaways)

• The same behavior can arise from opposite internal states. Learning history, context, physiology, and conflicting motivations all sit between emotion and its expression, so a single cue is a hypothesis rather than a conclusion.

• Wagging reports arousal, not valence. It occurs in uncertainty, conflict, and immediately before aggression as readily as in pleasure — which is why "but he was wagging his tail" is such a common sentence after a bite.

• No validated indicator of positive emotion exists in dogs. Cortisol, HRV, panting, whining, and body shake differentiated arousal only under negative valence, while heart rate held across both (Flint et al., 2024; Csoltova & Mehinagic, 2020). The absence of stress signals is not evidence of a good state.

• Suppression looks like success. A behavior that stops abruptly after something aversive should be treated as masked until shown otherwise; shock-trained dogs carried measurable stress into sessions where nothing happened to them (Schilder & van der Borg, 2004), and a dog taught not to warn has lost the signal that made a bite predictable.

• The guilty look is appeasement, not guilt. It appears in response to owner scolding regardless of whether the dog did anything (Horowitz, 2009) — and the practical alternative to guessing is the consent test: stop, wait, and see whether the dog chooses to continue.

References

Beerda, B., Schilder, M. B. H., van Hooff, J. A. R. A. M., & de Vries, H. W. (1998). Behavioural, saliva cortisol and heart rate responses to different types of stimuli in dogs. Applied Animal Behaviour Science, 58(3–4), 365–381.

Bunford, N., Csibra, B., & Gácsi, M. (2019). Individual differences in response to ambiguous stimuli in a modified Go/No-Go paradigm are associated with personality in family dogs. Scientific Reports, 9, 11067. https://doi.org/10.1038/s41598-019-47510-z

Csoltova, E., & Mehinagic, E. (2020). Where do we stand in the domestic dog (Canis familiaris) positive-emotion assessment: A state-of-the-art review and future directions. Frontiers in Psychology, 11, 2131. https://doi.org/10.3389/fpsyg.2020.02131

Flint, H. E., Weller, J. E., Parry-Howells, N., Ellerby, Z. W., McKay, S. L., & King, T. (2024). Evaluation of indicators of acute emotional states in dogs. Scientific Reports, 14(1), 6406. https://doi.org/10.1038/s41598-024-56859-9

Horowitz, A. (2009). Disambiguating the "guilty look": Salient prompts to a familiar dog behaviour. Behavioural Processes, 81(3), 447–452.

Maier, S. F., & Seligman, M. E. P. (2016). Learned helplessness at fifty: Insights from neuroscience. Psychological Review, 123(4), 349–367. https://doi.org/10.1037/rev0000033

Mendl, M., Burman, O. H. P., & Paul, E. S. (2010). An integrative and functional framework for the study of animal emotion and mood. Proceedings of the Royal Society B, 277(1696), 2895–2904.

Mills, D. S., Demontigny-Bédard, I., Gruen, M., Klinck, M. P., McPeake, K. J., Barcelos, A. M., Hewison, L., Van Haevermaet, H., Denenberg, S., Hauser, H., Koch, C., Ballantyne, K., Wilson, C., Mathkari, C. V., Pounder, J., Garcia, E., Darder, P., Fatjó, J., & Levine, E. (2020). Pain and problem behavior in cats and dogs. Animals, 10(2), 318. https://doi.org/10.3390/ani10020318

Parr-Cortes, Z., Müller, C. T., Talas, L., Mendl, M., Guest, C., & Rooney, N. J. (2024). The odour of an unfamiliar stressed or relaxed person affects dogs' responses to a cognitive bias test. Scientific Reports, 14(1), 15843. https://doi.org/10.1038/s41598-024-66147-1

Pongrácz, P., Molnár, C., & Miklósi, Á. (2006). Acoustic parameters of dog barks carry emotional information for humans. Applied Animal Behaviour Science, 100(3–4), 228–240.

Schilder, M. B. H., & van der Borg, J. A. M. (2004). Training dogs with help of the shock collar: Short and long term behavioural effects. Applied Animal Behaviour Science, 85(3–4), 319–334. https://doi.org/10.1016/j.applanim.2003.10.004

Schöberl, I., Beetz, A., Solomon, J., Wedl, M., Gee, N., & Kotrschal, K. (2016). Social factors influencing cortisol modulation in dogs during a Strange Situation Procedure. Journal of Veterinary Behavior, 11, 77–85. https://doi.org/10.1016/j.jveb.2015.12.003

Siniscalchi, M., Lusito, R., Vallortigara, G., & Quaranta, A. (2013). Seeing left- or right-asymmetric tail wagging produces different emotional responses in dogs. Current Biology, 23(22), 2279–2282.

Wan, M., Bolger, N., & Champagne, F. A. (2012). Human perception of fear in dogs varies according to experience with dogs. PLoS ONE, 7(12), e51775. https://doi.org/10.1371/journal.pone.0051775