Measuring Stress in Dogs: Cortisol, HRV, Behavior and Their Limits

1. What Is Being Measured?

1.1 Stress Is Not a Single Number

"The dog was stressed" can refer to at least three different claims: a physiological system was activated, the dog was in a negatively valenced state, or the dog's welfare was impaired. Those are related questions, not interchangeable ones.

Cortisol can tell us about activity of the hypothalamic-pituitary-adrenal (HPA) axis. Heart rate and HRV tell us about cardiovascular and autonomic regulation. Behavior tells us what the dog did in a context. Choice and judgment-bias paradigms are designed to get closer to how the animal evaluates that context. No single one of these measures supplies the whole answer.

1.2 Arousal and Valence

A useful affective framework separates arousal - the degree of activation - from valence - whether an experience is positive or negative (Mendl, Burman & Paul, 2010). A dog greeting a familiar person and a dog confronted by a threat may both show substantial physiological activation while differing profoundly in valence.

For that reason, physiological measurements should be described first in the system they actually assess: HPA-axis activity, heart rate, autonomic modulation, temperature or immune activity. Calling all of them "arousal measures" is convenient but can hide important biological differences (why behavior does not identify an emotion by itself).

1.3 Construct Validity

The central methodological question is construct validity: does the measure actually represent the construct that the authors say it represents? Cobb, Jiménez and Dreschel (2025) identify recurring problems in canine welfare physiology, including heavy reliance on single biomarkers, incomplete characterization of individual dogs, inconsistent sampling procedures and insufficient attention to individual variation.

A technically precise cortisol assay can therefore produce a biologically real number and still fail to answer a welfare question.

1.4 Acute and Chronic Questions Are Different

A transient response to a sudden event and the physiology of an animal living under long-term challenge are not the same research question. Acute measures are sensitive to timing. Longer-term measures integrate over broader periods but bring additional confounds. Low or unchanged values are not automatically signs of wellbeing, and high values are not automatically signs of harm.

1.4a Why the Word Is Used So Loosely

Stress is a convenient word because it sounds precise and carries a moral charge. A training method that "causes stress" sounds harmful; one that "reduces stress" sounds kind. Both claims often rest heavily on activation measures, and both then overreach in the same way (where the term carries the most weight).

That is not an argument for avoiding the word. It is an argument for asking, each time it appears, which meaning is intended and what evidence stands behind it.

1.5 Four Evidence Domains

A careful welfare assessment can draw on four domains:

Physiology: HPA-axis activity, cardiovascular and autonomic measures, temperature, immune or metabolic indicators.

Behavior: posture, movement, vocalization, interaction, avoidance, approach and behavior over time.

Choice and preference: what the dog chooses when meaningful alternatives are available.

Valence-sensitive tasks: for example judgment-bias paradigms, which infer affective state from responses to ambiguity.

The strongest welfare conclusions usually come from agreement across more than one domain rather than from a single "stress" value.

2. Cortisol

2.1 What Cortisol Measures

Cortisol is an end product of HPA-axis activation. It participates in energy mobilization, immune regulation and adaptation to challenge. Threat can increase cortisol, but so can exercise, novelty, social stimulation and other activating events. A cortisol increase is therefore evidence of HPA-axis activity, not a direct measure of distress or welfare (how the canine stress axis works).

2.2 Saliva: Useful, but Not Automatically Interchangeable With Blood

Saliva became attractive because blood collection can itself alter the response under study. Early work in laboratory dogs found useful correspondence between plasma and salivary cortisol responses (Beerda, Schilder, Janssen & Mol, 1996), and salivary cortisol subsequently became one of the most common non-invasive measures in canine research.

That validation does not generalize automatically to every dog population. Ferrans et al. (2025) compared paired serum and saliva samples in 34 assistance-dog puppies and 38 adult pet dogs. Serum and salivary cortisol were not significantly correlated in either group. In the puppy sample, an arousing manipulation increased serum cortisol but did not produce a corresponding salivary increase.

The appropriate conclusion is not that salivary cortisol is useless. It is that salivary cortisol should not be assumed to be an interchangeable proxy for circulating cortisol in heterogeneous pet dogs or puppies without population- and protocol-specific validation.

2.3 There Is No Universal 20-30 Minute Correction Rule

Salivary responses are often sampled after an event because endocrine responses take time to develop, but canine data do not justify a fixed rule that saliva simply follows blood by twenty or thirty minutes. Beerda et al. (1996) did not find a simple universal lag in repeated paired sampling, and Ferrans et al. (2025) identify response timing as one of the unresolved explanations for inconsistent serum-saliva relationships.

A good study therefore justifies its sampling window for the specific protocol rather than applying a generic offset.

2.4 The Large Salivary-Cortisol Meta-Analysis

Cobb et al. (2016) pooled 31 raw data sets containing 5,153 salivary samples from 1,205 dogs. Concentrations varied widely, and several individual and procedural variables were associated with the measured values, including age, sex and neuter status, living and testing environment, owner presence and collection medium.

This variability is not noise to be ignored. It is part of the biology and part of the measurement problem.

2.4a Owner Presence as a Variable

Owner presence during testing was among the variables with significant effects in the pooled salivary data (Cobb et al., 2016). In many applied settings the owner is part of the situation being studied, and a design that includes the owner in one condition and not in another has changed two things at once (how the dog–owner bond is studied).

Whether a familiar person buffers the dog's response is an interesting question in its own right, and a confound whenever it is not the question.

2.5 Time of Day: Standardize It, but Do Not Assume One Universal Rhythm

The pooled data in Cobb et al. (2016) showed lower early-morning values than values obtained later in the day. Other canine studies have not always reproduced a clear, stable diurnal pattern. The practical lesson is therefore methodological: sample comparable conditions at comparable times and report the timing, rather than treating one canine circadian curve as universal.

2.6 Collection Method Can Change the Result

Dreschel and Granger (2009) showed that collection materials and salivary stimulants can influence cortisol measurement. This matters because an apparently minor procedural choice - what the dog chews, whether saliva is stimulated and how the sample is handled - can alter the number before any biological interpretation begins.

2.7 The Dog-Park Study: Activation Without a Valence Answer

Ottenheimer Carrier et al. (2013) found higher salivary cortisol after twenty minutes in an off-leash dog park than after an on-leash walk in the same companion dogs. Most coded behaviors did not track cortisol closely; hunched posture was one exception associated with higher values.

The defensible conclusion is that the dog-park condition was associated with greater HPA-axis activation. The study does not establish that the dogs were enjoying themselves, and it does not establish that they were distressed. Valence remained unresolved.

2.8 Hair Cortisol: Longer-Term Integration, Not a Chronic-Stress Meter

Hair cortisol concentration (HCC) is used as an integrative measure of longer-term cortisol exposure. Bennett and Hayssen (2010) found a relationship between hair and salivary cortisol and also found a coat-pigment effect: black dogs had lower hair cortisol values than non-black dogs while the same difference was not seen in saliva.

Later work shows why HCC needs caution. Hair growth rate, sampling site, pigmentation, season, age, disease, medication and other biological factors can influence interpretation. Mariti et al. (2025), for example, found lower HCC in dogs with chronic gastroenteric disease than in healthy dogs, and associations with age-related variables were not simple. Hair cortisol is therefore useful as an integrative HPA measure, but its exact time window and welfare meaning cannot be read directly from a single value.

2.9 High Is Not Automatically Bad; Low Is Not Automatically Good

Two mirror-image errors are common. A rise is labeled harmful because cortisol is called a stress hormone. Or an unchanged or low value is labeled reassuring. Both interpretations are too simple. Chronic conditions can alter reactivity and regulation, and a single value cannot diagnose an "exhausted" stress axis.

Use terms such as altered, blunted or dysregulated HPA activity only when the design supports them; avoid the vague idea that the axis has simply become "exhausted".

2.10 Single Samples and Group Means

A single sample from an individual dog has limited interpretability. Repeated within-dog measurements, crossover designs and well-matched baselines can improve precision. That does not mean that every valid study must have an individual baseline: randomized between-group designs can also be informative when groups, timing and procedures are well controlled.

The important point is to match the strength of the conclusion to the strength of the design.

2.10a Neuter Status, Age and Habituation to the Setting

Sex and neuter status, age and time spent in the environment before testing all had significant effects in the pooled salivary data (Cobb et al., 2016). A comparison between groups that differ in any of these — young intact males against older neutered females, or dogs sampled on arrival against dogs sampled after settling — can produce a cortisol difference that says nothing about the intervention being studied.

Well-designed studies build in acclimatization and balance their groups. When a study does neither, the result is still data, but it is data about several things at once.

2.11 What Cortisol Is Good For

Cortisol is useful when a study asks a specific HPA-axis question, uses standardized sampling, accounts for relevant covariates and interprets the result together with behavior or other welfare measures. It is weak when a single post-event sample is translated directly into "the dog felt worse."

3. Heart Rate and Heart Rate Variability

3.1 Heart Rate Is Sensitive and Nonspecific

Heart rate changes rapidly with movement, posture, respiration, temperature, anticipation, physical effort and emotion. Beerda et al. (1998) showed prompt cardiovascular responses to acute challenges, illustrating why heart rate is useful for time-resolved physiology and poor as a stand-alone emotional label.

3.2 What HRV Adds

Heart-rate variability (HRV) describes variation in the intervals between successive heart beats. It provides information about autonomic modulation of cardiac activity and can reveal patterns that mean heart rate alone does not (von Borell et al., 2007).

It should not, however, be reduced to a single "sympathetic versus parasympathetic balance" number. The idea that common HRV ratios directly quantify sympathovagal balance is physiologically oversimplified, and interpretation depends on the parameter, recording conditions and underlying heart rate.

3.3 Behavior and Breathing Are Major HRV Confounds

Berg et al. (2026) directly examined short-term canine HR and HRV across behavioral states. HR and HRV varied strongly with what the dogs were doing, and high-intensity breathing was associated with higher HR and lower HRV. RMSSD and high-frequency power showed some of the more robust differences between behavioral states.

This is critical for welfare studies. A dog that is standing, moving or panting can produce a different HRV profile before any inference about emotion is made.

3.3a Why the Posture Confound Reaches Applied Studies

Where the behavioral outcome is how much a dog lies down, and the physiological outcome is HRV, the two measures are not independent: lying, activity level and breathing are associated with different HRV profiles (Berg et al., 2026). Several enrichment studies report exactly this pairing and treat the agreement as convergent evidence (an example from the auditory enrichment literature).

Agreement between two measures that share a mechanical cause is weaker than it looks. It is worth checking whether a study adjusted for activity before reading HRV as an independent confirmation.

3.4 Measurement Quality Matters

Useful HRV data require reliable beat-to-beat intervals, artefact detection, adequate recording length and documentation of posture, activity and respiration. Short noisy segments and undocumented movement can make mathematically precise output biologically ambiguous.

3.5 Wearables: Raw Measures and Proprietary Scores Are Different Questions

It is too broad to say that no canine wearable has been independently validated. Activity monitors and some collar-based pulse, respiration and HRV outputs have been studied independently, including PetPace and ActiGraph systems (Ortmeyer, Robey & McDonald, 2018).

But validation of a sensor for movement or pulse does not validate a proprietary "stress," "calm" or "wellbeing" score derived from those data. The algorithmic interpretation is a separate construct that requires separate validation.

3.6 Continuous Curves Are Often More Informative Than One Peak

For short events, the trajectory can be more useful than one absolute value: how quickly heart rate rises, what the dog is doing at the time, and how it changes afterwards. Even then, recovery should be treated as a descriptive outcome, not as a validated direct measure of valence.

4. Behavior

4.1 Behavior Can Discriminate Contexts That Physiology Does Not

Beerda et al. (1998) recorded multiple behavioral and physiological responses to acute challenges. Different stimulus types produced different combinations of behavior, heart rate and salivary cortisol. Importantly, the stimuli also differed in more than predictability, so the study cannot isolate predictability as the cause of those differences.

For a cleaner canine test of predictability and controllability, Dess et al. (1983) independently manipulated both during electric shock exposure. Uncontrollable shock produced a larger immediate cortisol rise; predictability did not significantly alter the immediate cortisol response. Prior unpredictability did, however, increase the cortisol response to later novel shocks. That study is a better reminder that predictability and controllability can affect different parts of the stress response (control and predictability in dogs).

4.2 Behavior and Physiology Often Disagree

Behavioral and physiological measures frequently show weak or inconsistent relationships. Beerda et al. (1998) found no simple correlation across their measured behavioral and physiological stress parameters, and Ottenheimer Carrier et al. (2013) found few behavioral correlates of cortisol in the dog-park study.

The correct lesson is not that disagreement is universal. It is that mismatches are common enough that neither domain should be used as a substitute for the other.

4.3 There Is No Universal "Stress Signal"

Yawning, lip licking, shaking, lowered posture, vocalizing, freezing, pacing and avoidance can all be relevant observations. None is sufficiently specific to identify one internal state without context. The same topography can occur after waking, during anticipation, after eating, during social interaction or under threat (how behavior should be operationalized).

4.4 Posture Is Informative, Not a Universal Best Marker

Very low or hunched posture was associated with physiological measures in some influential studies (Beerda et al., 1998; Ottenheimer Carrier et al., 2013). That makes posture worth recording carefully. It does not make lowered posture a validated universal "strongest stress cue."

A systematic review by De Winkel et al. (2024) examined 39 studies of observable canine welfare or emotional indicators. Only five reported some form of validity assessment, while inter-rater reliability was much more common. The field therefore has many candidate behaviors and relatively few fully validated observational instruments.

4.5 Context and Sequence Matter

A behavior becomes more informative when we know what preceded it, what the dog could do instead, what happened afterwards and whether the pattern changed over repeated exposures. Video often preserves that sequence better than a tally of individual "stress signals."

4.5a Video Over Tallies

A lip lick after eating and a lip lick while a stranger approaches are the same movement and different events. A recording allows the context to be reconstructed, the sequence to be checked and a second observer to disagree; a tally of yawns allows none of that (how reactive episodes are read).

4.6 Observer Expectancy

Behavioral scoring can be influenced by what observers expect to see. Blinding and explicit operational definitions reduce that risk in research. In practice, a useful correction is to write "the dog lowered its body, turned its head away and moved behind the handler" before writing "the dog was stressed."

4.7 Stillness Is Ambiguous in Both Directions

A still dog may be resting, freezing, behaviorally inhibited, tired, trained to remain still or showing another state entirely. Low activity does not prove calmness, and it does not prove learned helplessness. The internal state remains an inference unless additional evidence narrows the possibilities (why passivity must be interpreted cautiously).

4.8 When Stillness Is Not Calm

Studies of dogs left alone illustrate the ambiguity well. A dog that vocalizes and paces is plausibly distressed; a dog that lies still may be relaxed or may be inhibited, and neither cortisol nor a behavior count alone settles which (how separation-related behavior is assessed).

Video over the whole absence, the dog's behavior at the owner's return and its behavior in the days around the recording help narrow the interpretation considerably, although a dog lying still can remain ambiguous even on video. They are more work than a swab and considerably more informative.

5. Measures That Are More Sensitive to Valence

5.1 Judgment Bias

Judgment-bias tasks examine how an animal responds to ambiguous cues after learning positive and negative reference cues. More optimistic or pessimistic response patterns are treated as valence-sensitive evidence. The method was designed around affective evaluation, which makes it conceptually closer to the welfare question than cortisol.

It is still inferential. Performance depends on training, motivation, learning history and task design. "Valence-sensitive" is more accurate than saying that judgment bias measures valence directly (how canine judgment bias is tested).

5.2 Choice and Preference

Choice can provide valuable welfare information when the alternatives are meaningful and the animal can actually access them. A preference tells us what the dog selects under those conditions. It does not automatically prove long-term benefit, nor is every choice independent of learning history, effort or prior reinforcement.

Choice is therefore strong evidence about preference, not a direct readout of subjective wellbeing (how choice can be incorporated into handling).

5.3 Recovery Is Useful, but Not a Valence Test

Time to return to baseline behavior or physiology after a challenge is often informative. It can be tracked repeatedly in the same dog and can reveal whether responses are becoming larger or smaller over time. But recovery depends on challenge intensity, fitness, prior exposure, context and individual response style.

Use recovery as a longitudinal outcome, not as a validated shortcut to positive or negative valence.

5.4 Positive Welfare Remains Harder to Measure

Canine welfare science has more tools for detecting challenge and negative states than for demonstrating positive experience. Play, affiliative interaction, voluntary engagement, exploration, relaxed postures and positive cognitive-bias outcomes are all useful candidates, but no single one functions as a universal positive-welfare marker.

6. Other Physiological Indicators

6.1 Oxytocin

Oxytocin is relevant to social physiology, but peripheral measurement does not map simply onto central release or emotional meaning. Assays, sampling matrices and social context matter. It is a research variable, not a canine "happiness hormone" (social physiology in dogs).

6.2 Immune, Metabolic and Oxidative Measures

Immune and metabolic indicators may provide information about cumulative physiological load, disease and adaptation. Their weakness as simple welfare measures is the same as their strength biologically: they respond to many processes besides psychological stress.

6.3 Thermal Imaging

Infrared thermography can detect changes in surface temperature associated with autonomic and vascular responses. Interpretation depends on body region, ambient temperature, distance, angle, coat and other technical factors. It is non-contact and potentially useful, but it does not turn activation into valence.

6.3a Longer-Term Biological Signatures

Some research links chronic stress to cellular aging markers. That is a legitimate research direction, it integrates over far longer periods than any hormone sample, and it is a long way from anything that could inform a decision about an individual dog (what is known about stress and cellular aging).

6.3b Why Novel Indicators Are Not an Escape

Each new marker is introduced with the hope that it will measure welfare directly, and each so far has shared the basic limitation of the older ones: sensitivity to activation and to individual characteristics. That is not a reason for pessimism; it is a reason to treat each new indicator as a hypothesis until it has been tested against measures designed for valence (what physiology can and cannot show about behavior).

6.3c Why Valence Is Hard to Measure

Valence is a property of experience, and experience is not directly observable in any animal. Every valence measure is therefore an inference from something else: a choice, a pattern of responses to ambiguity, approach or avoidance. That is not a weakness peculiar to canine science; it is the condition of studying affect in any species that cannot report it.

What distinguishes good valence measures is that they are designed around the question rather than borrowed from another purpose. Cortisol was not developed to measure how an animal evaluates its situation; judgment bias tasks were.

6.4 Convergence Is Better Than Counting Biomarkers

Several measures pointing in the same direction can strengthen an interpretation, but only if they contribute partly independent information. Movement can simultaneously alter behavior, heart rate, HRV and surface temperature. Four affected outputs are not automatically four independent confirmations.

The goal is converging evidence from different domains, not the largest possible number of sensors.

A recent narrative review of 69 studies on free-roaming and shelter-housed dogs reaches the same conclusion from a different population: it examines cortisol measured in blood, saliva, feces and hair for its applicability in real-world settings and argues for multidimensional assessment that links physiological and behavioral indicators rather than relying on any single measure (Rotaru et al., 2026). As a narrative rather than systematic review, it adds breadth rather than a new mechanism, and its population differs from companion dogs in private homes.

7. How to Read a Canine Stress Study

7.1 Ask What the Measure Actually Measures

Does the study measure HPA activity, cardiovascular regulation, behavior, preference or a valence-sensitive cognitive outcome? Translate "stress" back into the actual variable before reading the conclusion.

7.2 Ask Whether the Population and Protocol Were Validated

Ferrans et al. (2025) is a useful warning: a saliva-blood relationship established in laboratory dogs did not replicate in assistance-dog puppies or heterogeneous adult pet dogs. Validation is not automatically portable across age groups, populations and sampling protocols.

7.3 Ask How Timing Was Chosen

Hormonal, cardiac and behavioral responses operate on different time scales. A useful study explains why samples were taken when they were, rather than relying on a generic twenty-minute rule.

7.4 Ask What Was Controlled

For cortisol: time, sampling medium, food, acclimatization, environment and relevant dog characteristics. For HRV: posture, movement, respiration, artefacts and recording length. For behavior: operational definitions, observer reliability and preferably blinding.

7.5 Ask Whether Valence Was Actually Tested

"Cortisol increased" is not equivalent to "welfare decreased." If the conclusion concerns how the dog experienced the condition, look for choice, avoidance, judgment bias, sustained behavioral patterns or other evidence that bears directly on evaluation.

7.5a A Worked Example

Suppose a study reports that dogs trained with method A had higher salivary cortisol after sessions than dogs trained with method B. Before concluding that method A was more aversive, a careful reader asks: were samples taken at comparable times, with the same material, after the same acclimatization? Were the groups comparable in age, sex and neuter status? Did each dog contribute a baseline? Was behavior recorded, and did it point in the same direction? Was anything measured that bears on valence?

If most answers are yes, the finding is informative about the measured HPA-axis response under method A and a reasonable prompt for a closer look. If most are no, it says little about either method. In neither case does cortisol alone establish that the dogs were worse off.

7.6 When Measures Agree

If a dog avoids the context, shows a sustained cluster of defensive or inhibited behavior, shows prolonged recovery and also displays physiological activation under a controlled design, concern is more strongly supported than by any single measure alone.

7.7 When Measures Disagree

Do not force them into one story. Rising cortisol with behavior that appears relaxed leaves valence unresolved. A lowered, inhibited posture with unremarkable cortisol likewise deserves attention without assuming that behavior is "more correct" than physiology. The systems may differ in timing, sensitivity or what they represent.

8. What This Means in Practice

8.1 Trainers Do Not Need a Hormone Assay

The practical value of this literature is methodological. Describe what the dog does, define the context, follow the sequence over time and avoid turning one observation into a diagnosis. Laboratory biomarkers are valuable for research but rarely improve an ordinary training decision by themselves.

8.2 Track Patterns Within the Same Dog

For applied work, repeated comparable observations are often more useful than comparing a dog with an imagined universal normal. Does the dog approach sooner over sessions? Recover faster? Spend more time exploring? Need less distance? Show fewer interruptions? Patterns are easier to defend than labels.

8.2a Recovery in Everyday Practice

How quickly a dog returns to its ordinary behavior after a challenge is observable without equipment, and a dog that recovers more slowly than usual is telling you something worth following up. Recovery is not a valence test, but a change in recovery within the same dog is one of the more useful practical signals available (how arousal is regulated).

8.3 Combine Rather Than Rank

Do not build a hierarchy in which posture always outranks cortisol or HRV always outranks heart rate. Use each measure for the question it can answer. In practice, context, behavior over time, recovery and meaningful choice can complement one another without pretending that one is the definitive stress signal.

8.3a Be Careful With Claims

"Our method reduces stress" is a claim almost nothing in the applied literature can support, because lower activation is not the same as better welfare. "Dogs in our program settle faster, recover sooner and choose to stay" is a claim that can be observed, recorded and defended.

The difference matters for trainers who describe their work publicly. The second kind of statement is both more honest and more persuasive to anyone who knows the literature.

8.3b Reports and Case Notes

In written case notes, the same discipline applies. Recording what the dog did, in which situation, and how long it took to return to baseline produces a document another professional can use. Recording that the dog "was stressed" produces a conclusion nobody can check.

8.4 Handling and Veterinary Settings

Predictability can matter, but Beerda et al. (1998) should not be used as a clean demonstration that announcing an event lowers cortisol because stimulus type and predictability were confounded. Dess et al. (1983) isolated predictability and controllability more carefully and showed that they affected different aspects of cortisol responding.

For practice, explaining or signaling what will happen may still be useful as part of structured handling, but the claim should be behavioral and procedural unless the specific physiological effect has actually been tested (the evidence on predictability and control).

8.5 Wearables

Activity trackers and smart collars can be useful for longitudinal monitoring of movement, rest, pulse or other raw variables when those outputs have been validated. A proprietary "stress score" is a separate claim. Unless that score has been independently validated against an appropriate construct, treat it as an algorithmic estimate rather than a measured emotional state.

8.6 A Useful Owner Explanation

A dog's body can activate during both pleasant and unpleasant events. One number therefore cannot tell you how the dog felt. Look instead for the pattern: what the dog did, whether it could choose, how the response changed over time and whether several independent indicators tell the same story.

8.7 A Short Checklist for Trainers

Before calling a dog stressed in a report or conversation: describe the posture, describe the context, note what happened afterwards and how quickly the dog returned to its usual behavior, and note whether it had the option to leave and used it. That description is more accurate and more useful than the label, and it leaves room for the possibility that the dog was simply excited.

8.8 Where Measurement Is Justified

Physiological monitoring earns its place in research, in clinical research or veterinary cases where physiological monitoring is relevant to the clinical question, and in program evaluation where decisions affect many animals. In each, the same principles apply: within-dog baselines when feasible, or another design that controls for individual differences, such as randomized or crossover comparisons; consistent collection; more than one kind of indicator; and a clear statement of what the measure can and cannot show.

9. Summary at a Glance

Cortisol is an HPA-axis measure, not a welfare score. High and low values both require context.

Salivary cortisol is not automatically interchangeable with serum cortisol. Ferrans et al. (2025) found no significant serum-saliva correlation in 34 assistance-dog puppies or 38 adult pet dogs.

There is no universal 20-30 minute saliva rule. Sampling timing has to be justified for the specific protocol.

Hair cortisol integrates over a longer period but has many confounds. Pigment, disease, age-related variables, sampling site and hair biology can influence the result.

HRV depends strongly on what the dog is doing. Berg et al. (2026) showed major effects of behavior and intensive breathing on canine HR and HRV.

No single behavior identifies stress. A 2024 systematic review found many candidate behavioral measures but limited formal validity testing.

Choice and judgment bias are valence-sensitive, not direct mind readers. They answer different questions from physiological activation.

Convergence matters. The strongest welfare interpretation combines partly independent evidence rather than repeating the same confound through several sensors.

10. Research Gaps and Critical Appraisal

Salivary cortisol needs renewed validation. The 2025 serum-saliva findings raise a direct replication and generalizability problem for one of the field's most common biomarkers.

Canine HRV standards remain incomplete. Behavior, respiration, posture, recording duration and artefact processing need more consistent reporting.

Behavioral welfare instruments need validation, not just reliability. De Winkel et al. (2024) found formal validity assessment in only a small minority of included studies.

Positive-welfare indicators are underdeveloped. The field remains better at identifying challenge than at demonstrating positive affect.

Wearable stress algorithms need construct validation. Validation of activity or pulse measurement does not validate an inferred emotional score.

Hair-cortisol interpretation remains biologically complex. Better information on growth, body site, pigment, disease, medication and time window is needed before HCC can support strong longitudinal welfare claims.

10.1 What Would Move the Field

Useful advances would include preregistered multisystem studies, standardized sampling and HRV protocols, repeated within-dog designs where appropriate, transparent reporting of dog characteristics, stronger validation of behavioral instruments, and direct comparison of physiological markers with preference or valence-sensitive paradigms.

The decisive question is not whether a marker changes. It is whether the change has been validated against the welfare construct assigned to it.

11. Conclusion

Canine stress measurement has become more sophisticated without producing a single "stress meter." Cortisol measures HPA-axis activity; HR and HRV reflect cardiovascular and autonomic regulation; behavior describes what the dog does; choice and cognitive tasks provide additional information about evaluation. Each is useful when interpreted within its construct and vulnerable when translated too quickly into welfare.

The 2025 finding that salivary cortisol did not reliably track serum cortisol in assistance-dog puppies or adult pet dogs is an especially important reminder that even familiar biomarkers require population-specific validation (Ferrans et al., 2025). The 2026 HRV data provide the same lesson from another system: breathing and behavior can shift the measure substantially before emotional interpretation begins (Berg et al., 2026). Behavioral science has its own limitation - many proposed canine welfare indicators are used more often than they are formally validated (De Winkel et al., 2024).

The practical conclusion is therefore not "ignore physiology." It is the opposite: use physiology precisely. Describe HPA activity as HPA activity, HRV as HRV, behavior as behavior, and welfare only when the design supports a welfare inference. The best evidence is convergent, longitudinal where possible, and explicit about what remains unknown.

Key Insights (Takeaways)

• No single cortisol, HRV, behavior or wearable value is a validated canine welfare score.

• Salivary cortisol should not automatically be treated as an interchangeable proxy for serum cortisol in pet dogs or puppies (Ferrans et al., 2025).

• Cortisol is best described as a marker of HPA-axis activity; it does not by itself identify positive or negative valence.

• HRV is strongly affected by behavior, breathing, posture and measurement quality (Berg et al., 2026).

• No single "stress behavior" is sufficiently specific; posture can be informative but must be read in context (De Winkel et al., 2024).

• Stillness proves neither calmness nor helplessness.

• Choice, preference and judgment bias are useful valence-sensitive tools, but they remain inferential.

• Validated raw wearable measurements do not automatically validate proprietary "stress scores."

• Welfare conclusions are strongest when partly independent physiological, behavioral and choice-based evidence converges.

References

Beerda, B., Schilder, M. B. H., Janssen, N. S. C. R. M., & Mol, J. A. (1996). The use of saliva cortisol, urinary cortisol, and catecholamine measurements for a noninvasive assessment of stress responses in dogs. Hormones and Behavior, 30(3), 272-279. https://doi.org/10.1006/hbeh.1996.0033

Beerda, B., Schilder, M. B. H., van Hooff, J. A. R. A. M., de Vries, H. W., & Mol, J. A. (1998). Behavioural, saliva cortisol and heart rate responses to different types of stimuli in dogs. Applied Animal Behaviour Science, 58(3-4), 365-381. https://doi.org/10.1016/S0168-1591(97)00145-7

Bennett, A., & Hayssen, V. (2010). Measuring cortisol in hair and saliva from dogs: Coat color and pigment differences. Domestic Animal Endocrinology, 39(3), 171-180. https://doi.org/10.1016/j.domaniend.2010.04.003

Berg, P., Koskela, A., Lipponen, J. A., Tornqvist, H., Kujala, J., Tarvainen, M. P., & Kujala, M. (2026). Behavior-related changes in canine heart rate and heart rate variability during short-term measurement. Applied Animal Behaviour Science, 296, 106899. https://doi.org/10.1016/j.applanim.2025.106899

von Borell, E., Langbein, J., Despres, G., Hansen, S., Leterrier, C., Marchant-Forde, J., Marchant-Forde, R., Minero, M., Mohr, E., Prunier, A., Valance, D., & Veissier, I. (2007). Heart rate variability as a measure of autonomic regulation of cardiac activity for assessing stress and welfare in farm animals - A review. Physiology & Behavior, 92(3), 293-316. https://doi.org/10.1016/j.physbeh.2007.01.007

Cobb, M. L., Iskandarani, K., Chinchilli, V. M., & Dreschel, N. A. (2016). A systematic review and meta-analysis of salivary cortisol measurement in domestic canines. Domestic Animal Endocrinology, 57, 31-42. https://doi.org/10.1016/j.domaniend.2016.04.003

Cobb, M. L., Jiménez, A. G., & Dreschel, N. A. (2025). Beyond cortisol! Physiological indicators of welfare for dogs: Deficits, misunderstandings and opportunities. Journal of Applied Animal Welfare Science. Advance online publication. https://doi.org/10.1080/10888705.2025.2572616

De Winkel, T., van der Steen, S., Enders-Slegers, M.-J., Griffioen, R., Haverbeke, A., Groenewoud, D., & Hediger, K. (2024). Observational behaviors and emotions to assess welfare of dogs: A systematic review. Journal of Veterinary Behavior, 72, 1-17. https://doi.org/10.1016/j.jveb.2023.12.007

Dess, N. K., Linwick, D., Patterson, J., Overmier, J. B., & Levine, S. (1983). Immediate and proactive effects of controllability and predictability on plasma cortisol responses to shocks in dogs. Behavioral Neuroscience, 97(6), 1005-1016. https://doi.org/10.1037/0735-7044.97.6.1005

Dreschel, N. A., & Granger, D. A. (2009). Methods of collection for salivary cortisol measurement in dogs. Hormones and Behavior, 55(1), 163-168. https://doi.org/10.1016/j.yhbeh.2008.09.010

Ferrans, M., Salomons, H., Moore, K., White, P., Hare, B., & Gruen, M. E. (2025). Salivary cortisol is an unreliable correlate of serum cortisol in adult pet dogs and assistance dog puppies. Scientific Reports, 15, 15986. https://doi.org/10.1038/s41598-025-00425-4

Mariti, C., Russo, G., Mazzoni, C., Borrelli, C., Gori, E., Habermaass, V., & Marchetti, V. (2025). Factors affecting hair cortisol concentration in domestic dogs: A focus on factors related to dogs and their guardians. Animals, 15(13), 1901. https://doi.org/10.3390/ani15131901

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: Biological Sciences, 277(1696), 2895-2904. https://doi.org/10.1098/rspb.2010.0303

Ortmeyer, H. K., Robey, L., & McDonald, T. (2018). Combining ActiGraph Link and PetPace collar data to measure activity, proximity, and physiological responses in freely moving dogs in a natural environment. Animals, 8(12), 230. https://doi.org/10.3390/ani8120230

Ottenheimer Carrier, L., Cyr, A., Anderson, R. E., & Walsh, C. J. (2013). Exploring the dog park: Relationships between social behaviours, personality and cortisol in companion dogs. Applied Animal Behaviour Science, 146(1-4), 96-106. https://doi.org/10.1016/j.applanim.2013.04.002

Rotaru, G., Hrițcu, T. D., Mălăncuș, R. N., Hrițcu, L. D., Soponaru, C., Nechifor, F., Boghian, V., Cherșunaru, A. A., Munteanu, A., & Spataru, M. C. (2026). Stress in free-roaming and shelter-housed dogs: A review of neurobiological, physiological and behavioral mechanisms relevant to welfare assessment. Animals, 16(14), 2183. https://doi.org/10.3390/ani16142183