Owner-Reported Data in Dog Research: How Reliable Are Owner Surveys?

1. Why Dog Research Relies on Owners

1.1 What Owners See That a Laboratory Cannot

Most of a dog's life happens where no researcher is present: at home, in the yard, on the same three walking routes, at night, while the household is out. That is also where most of the behavior that matters to owners and clinicians takes place — barking at the door, distress when left alone, reactions to visitors, to thunderstorms, to the vacuum cleaner. Very little of this can be reliably elicited during an hour in a test room, and some of it would be ethically questionable to provoke deliberately.

The person who lives with the dog has seen these situations hundreds of times. For behavior that is rare, context-bound, or tied to the home environment, the owner is frequently the only available source. A standardized behavior test shows how a dog responds on one day, in an unfamiliar place, to a stimulus chosen by the experimenter. An owner report describes how the dog has tended to behave over months in its own surroundings. Both are useful, and they do not measure the same thing (behavioral assessment in dogs).

1.2 Scale

The second reason is sample size. An observational study of thirty dogs is labor-intensive. An online questionnaire can reach tens of thousands. For some questions that difference is decisive, particularly in genetics, where effects of individual loci are small and only large samples can detect them.

A widely cited study of breed and behavior combined owner surveys on 18,385 dogs with genotyping of 2,155 of them. Most behavioral traits were heritable, with heritability above 25%, yet breed explained only about 9% of behavioral variation between individual dogs (Morrill et al., 2022). A finding of that kind is not attainable with laboratory observation alone. It is worth noticing what the behavioral half of that dataset is, though: owner answers to questions about their own dogs (behavioral genetics in dogs, breed versus behavior).

1.3 An Instrument Is Not the Same as a Survey

"Owner questionnaire" covers two very different things. At one end are psychometric instruments: item sets written with care, tested on large samples, examined for internal structure and reliability, and compared against other measures. At the other end are study-specific surveys whose questions were assembled by a research team for a single project and never examined in that way.

Both are called questionnaires, and both produce numbers. The difference between them resembles the difference between a calibrated scale and an estimate by eye. For anyone evaluating a study, the first question is therefore often the most informative one: was the outcome measured with a validated instrument, or with a purpose-built survey whose measurement properties are unknown?

Even validated instruments are not uniform in this respect. A systematic review of 38 studies reporting 37 canine behavioral questionnaires coded each item by what it actually asks about: temperament, emotion, literal behavior, or other. Depending on the group of items, between 2.3% and 40.5% used emotion-based descriptors, and between 26.3% and 64.4% were mixed or could not be assigned to a single category. The authors conclude that questionnaires frequently conflate observable behavior with inferred traits or emotional states (Johnson et al., 2026).

This distinction is a measurement issue before it is a statistical one. A study can apply impeccable statistics to a variable that does not capture what its name suggests (how behavior gets operationalized).

2. What Validation Establishes, and What It Does Not

2.1 The C-BARQ

The best-known validated owner instrument in dog research is the Canine Behavioral Assessment and Research Questionnaire. In its development study, owners of 1,851 dogs, recruited through a veterinary teaching hospital and national breed clubs, completed an initial 152-item questionnaire, along with owners of 203 dogs that had been seen by behavior practitioners for behavior problems. Factor analysis produced 11 factors from 68 of the original items. For 7 of those factors it was possible to check whether dogs with a corresponding clinical diagnosis scored higher than dogs with unrelated diagnoses, and they did. For the remaining 4 factors, validity could not be examined because there were not enough dogs with matching problems (Hsu & Serpell, 2003).

The structure of that validation is worth spelling out. It showed that owner reports on the C-BARQ line up with the judgments of behavior professionals for several important problem categories. It did not show that owner reports match direct behavioral measurement, and it left a third of the factors without criterion evidence in the original paper.

2.2 The Short Form

A 42-item short form was later compared against the full instrument in 435 owners from the Dog Aging Project who completed both within 60 days. Correlations between the long and short versions exceeded 0.60 for 12 of the 14 behavior and temperament domains. Agreement on individual items was only fair to moderate, with kappa values between 0.23 and 0.58 (Wilkins et al., 2024).

That is a reasonable result for its purpose, which is to show that a shorter instrument tracks the longer one closely enough for population research. It is a comparison of a questionnaire with a questionnaire. Neither version was compared with the dog's observed behavior in that study, and the item-level agreement is a reminder that domain scores are more robust than any single question.

2.3 Behavior After a Change of Home

One of the more demanding tests of owner reports comes from the shelter setting. Relinquishing owners completed a short C-BARQ at intake, and those reports were compared with evaluations by shelter staff and with behavior reported after adoption. Owners were randomly told either that their answers might be used by shelter staff, that they would remain confidential, or nothing about their use, and the study found no consistent evidence that this changed what owners reported. That is a narrower finding than general accuracy: it shows that the expected use of the answers did not detectably bias them. Owner ratings of aggression and fear correlated with staff assessments, but in a follow-up with 53 adopters only three behaviors — aggression toward unfamiliar people, chewing inappropriate objects, and urinating when left alone — correlated significantly between the pre- and post-adoption reports (Duffy, Kruger & Serpell, 2014).

Two readings are possible, and both are probably partly true. Owners who give up a dog are not necessarily inaccurate. But much behavior is specific to the environment in which it was learned, so a description of a dog in one household is an imperfect predictor of the same dog in another (temperament and coping styles).

2.4 How Stable Is What Gets Measured?

Reliability also depends on how stable the underlying trait is. A meta-analysis of 31 studies with 822 consistency estimates found an overall consistency of personality measures in dogs of r = 0.43, which the authors described as substantial. Consistency was higher in adult dogs than in puppies (r = 0.51 versus 0.30), higher over short intervals than long ones, and higher when the identical test was repeated than when different tests were used (Fratkin et al., 2013).

For survey research this has a practical consequence. A single owner report is a snapshot of something that changes, especially in young dogs. Correlations between that snapshot and later outcomes have a ceiling set partly by the instability of the trait itself, not only by the quality of the instrument.

2.5 Newer Instruments

Validated owner instruments now cover a growing range of constructs. The Canine Frustration Questionnaire was developed as a psychometric tool for frustration tendencies (McPeake et al., 2019), and the Lincoln Canine Anxiety Scale was developed and psychometrically validated for anxiety (Mills et al., 2020). A German-language impulsivity scale has been translated and validated in a large sample of dog–owner pairs (Kero, Dolezal & Affenzeller, 2026).

Each of these makes owner reports on a specific construct more comparable across studies. None of them converts an owner rating into a functional diagnosis for an individual dog. A high frustration score describes how an owner rates the dog's tendencies in general; it does not establish why the dog reacted in a particular encounter (the neurobiology of frustration).

2.6 A Screening Tool Done Well

Some owner instruments perform remarkably well. The Canine Cognitive Dysfunction Rating Scale consists of 13 items about everyday changes such as disorientation, altered sleep, and failure to recognize familiar people. On an independent sample it reached an overall diagnostic accuracy of 98.9%, with a negative predictive value of 99.3%. Test–retest reliability over two months was r = 0.73 (Salvin et al., 2011).

The less quoted figure is the positive predictive value: 77.8%. Roughly one in five dogs flagged by the scale did not have the condition. That is acceptable, even good, for a screening instrument. It is also why a positive result is a reason for veterinary workup rather than a diagnosis, particularly since many medical conditions produce similar behavioral changes in older dogs (canine cognitive dysfunction).

2.7 Rating Scales Are Ordinal

Most owner instruments ask for responses on a fixed scale, often from 0 to 4. Such responses are ordered categories, not interval measurements: the step from "never" to "seldom" is not necessarily the same size as the step from "often" to "always". This is not a defect, but it affects how results should be analyzed and reported, and it is one reason to read a mean such as "2.4 out of 4" with caution (Sullivan & Artino, 2013).

3. Observation Versus Interpretation

3.1 What Is Asked Matters More Than Who Answers

Owner reports are most dependable when the question concerns behavior that is conspicuous, frequent, and does not require inference. "Does your dog bark when the doorbell rings?" asks for an observation. "How stressed is your dog when the doorbell rings?" asks for an inference about an internal state. The first can be answered from memory of events; the second depends on whether the respondent recognizes stress and on what they believe stress looks like.

Well-constructed instruments are written to keep questions on the observable side of that line, asking about specific behavior in specific situations. Study-specific surveys often are not, and questions about emotion, motivation, or intention are where owner reports become least reliable (why behavior is not the same as emotion).

3.2 What Owners Name as Stress

A questionnaire can only capture what the respondent perceives. In a survey of 1,190 dog owners, the signs most often named as indicating stress were trembling and whining, followed by aggression, excessive barking, and panting. Subtler behaviors — looking away, turning the head, yawning, and nose licking — were named considerably less often (Mariti et al., 2012).

That study measured what owners say they recognize, not what they actually detect when watching a dog in a real situation. The pattern is still informative. Owners frequently identified conspicuous signs of stress, while subtler, context-dependent behaviors were mentioned much less often. The study did not examine in which order such signals appear, so it does not support a fixed sequence of "early" and "late" signs. It does suggest that surveys asking owners about stress may miss states that are expressed only in subtle behavior. How large that gap is will depend on the sample and the wording.

The opposite caution applies as well. Yawning, nose licking, and head turning are not in themselves proof of stress. Dogs yawn when tired and lick their noses after eating. These behaviors become informative only in combination with the situation and the rest of the dog's behavior (canine body language).

3.3 Underestimation in the Home

A study of household noises took the question further. Owners of 386 dogs were surveyed about their dogs' responses to common household sounds, and 62 videos of dogs reacting to such sounds, taken from a public video-sharing platform, were coded for behavior and for the reactions of the people in them. Responses were stronger to high-frequency intermittent sounds than to low-frequency continuous ones. Respondents appeared to underestimate their dogs' fearfulness, and most of the people in the videos responded to visible fear with amusement (Grigg et al., 2021).

The videos cannot be used to estimate prevalence, since posted videos are a selected sample. They do show the distance that can lie between a dog's behavior and the way the person next to it reads that behavior (noise sensitivity in dogs).

3.4 Familiarity Is Not Expertise

It is natural to assume that owners, having more experience with dogs, read them better than non-owners. One study tested that assumption directly. An online sample of 71 adults viewed three videos of small children interacting with dogs that an expert panel had classified as fearful or anxious and lacking confidence. Most respondents described the dogs as relaxed (68.4%) and confident (65.1%), and most commonly described their behavior as play or friendly. Participants who did not own dogs were more successful than dog owners at classifying the dogs' emotional states (Demirbas et al., 2016).

The sample was small and the videos few, so the result should not be generalized carelessly. It fits a broader point, however: experience with dogs without systematic knowledge of their behavior can produce confident interpretation without corresponding accuracy.

Interpretation also varies with characteristics of the observer that have nothing to do with the dog. Forty-seven veterinary students rated five videos of dogs on 19 adjectives. Among students who had never been responsible for a dog, those with lower animal empathy scores rated the dogs as more aggressive in four of the five videos. Self-reported skills with dogs had no effect on interpretation (Meyer, Forkman & Paul, 2014). The same dog, in other words, was described differently depending on who was watching.

3.5 Inference Presented as Observation

The most familiar example of an owner inference is the "guilty look". Many owners report that their dog knows when it has done something wrong. In an experiment with 14 dogs, owners were asked to forbid a treat and then leave the room. Whether the dog had actually eaten the treat made no difference to the frequency of guilty-look behaviors. What did make a difference was whether the owner scolded the dog on returning, and the effect was stronger in dogs that had obeyed (Horowitz, 2009).

The study shows what elicits the behavior, not what the dog experiences, and it does not show that dogs feel nothing. It does show that the interpretation "he knows what he did" is not derived from the behavior itself but added to it. For questionnaire research the lesson generalizes: any item that requires an interpretation measures, to an unknown extent, the respondent's beliefs rather than the dog (the guilty look in dogs).

4. Expectation and the Caregiver Placebo Effect

4.1 Two Different Placebo Effects

Placebo effects are well documented in human medicine. In animals, two distinct phenomena need to be kept apart. The first is a real change in the animal under an inert treatment, for example because participation in a study brings more attention, handling, or routine. The second is a change in the assessment without a change in the animal: the person completing the outcome measure expects improvement and reports it. The second is usually called the caregiver placebo effect, and it is the one that matters most for owner-reported outcomes.

4.2 Osteoarthritis: Owners and Veterinarians Alike

The clearest demonstration comes from the placebo arm of a prospective, randomized, double-blinded, placebo-controlled multicenter trial in 58 dogs with lameness from osteoarthritis. Owners reported improvement in their dog's lameness in 39.7% of assessments. Veterinarians examining the dogs reported improvement in 44.8% of assessments at a walk and at a trot, and in 43.1% when evaluating pain on joint palpation. The effect increased over time. Force-platform gait analysis showed no change in mean ground reaction forces under placebo; of the 58 dogs, 46 were unchanged, 7 improved by at least 5%, and 5 worsened by at least 5% (Conzemius & Evans, 2012).

Two points stand out. The effect was not limited to lay owners; trained veterinarians showed it at a similar rate. And it occurred in a double-blinded trial. Blinding prevents assessors from knowing which treatment a dog received, but it does not prevent them from expecting that the treatment, whatever it is, will help. Blinding is necessary; it is not sufficient. What protected the conclusions in that trial was the combination of a placebo control and an objective outcome measured on the dog.

4.3 Cognitive Decline: Owner Ratings Changed Without Corresponding Changes in Test Performance

A study of aging dogs with cognitive decline separated the placebo effect from the effect of trial participation. It compared 21 dogs from the placebo group of a randomized controlled trial with 17 dogs from an observational longitudinal study, using two remotely completed owner questionnaires and three in-house cognitive tests. After six months, one owner questionnaire showed a strong and significant improvement in the placebo cohort (Hedges' g = 0.76), while the other questionnaire showed no change. The observational cohort showed a small deterioration on the second questionnaire. None of the in-house cognitive tests showed a significant change in either group (Simon et al., 2025).

The groups were small, and the authors' own conclusion was that one instrument was more robust against caregiver placebo than the other — which is itself a finding about owner instruments. The broader point is the same as in the osteoarthritis trial: when the people who expect a benefit are also the people reporting the outcome, owner ratings can change without a corresponding change in measured performance. That does not prove that nothing changed in the dogs; it shows that the owner rating alone cannot establish it.

4.4 Beyond Medication

Nothing about this mechanism is specific to drugs. It applies to any study in which owners know whether their dog received a training program, a diet, a supplement, a piece of equipment, or a behavioral intervention, and then report on the result. In behavioral medicine, where many outcomes are necessarily owner-reported, this is a structural vulnerability rather than an occasional flaw (psychopharmacology in dogs).

When reading such studies, three questions help: Was there a control group that received something that looked like an intervention? Were outcome assessors blinded? And was there at least one outcome not dependent on the expectations of the people involved?

4.5 What Owners Believe Helps

Owner beliefs also shape what gets reported as effective. In a study of 48 dog–owner pairs, dogs were exposed to an unfamiliar object, a remote-controlled car, while owners provided one of several kinds of support: verbal reassurance, food, approaching the object together with the dog, or remaining passive. None of the strategies reliably reduced stress-related behavior compared with passive owners. Approaching together, which owners themselves rated as the most useful strategy, brought dogs closer to the object but showed no advantage in stress-related behavior (Miller et al., 2026).

The stimulus was mild and exposure brief, so this cannot be transferred to genuinely fearful dogs in real situations. It is a useful reminder that "owners consider X helpful" and "X helps" are separate empirical claims, and that surveys typically measure the first (social referencing and social buffering).

5. Agreement Between Instruments Is Not Accuracy

5.1 Osteoarthritis Instruments and the Force Plate

Pain research offers a useful case. Several owner-completed clinical metrology instruments exist for dogs with osteoarthritis. In a prospective evaluation of 222 dogs, three of these instruments correlated moderately with each other, but only weakly with force-platform measurement of limb loading (Walton et al., 2013).

When several questionnaires agree, that shows that owners answer consistently across instruments. It does not show that the instruments capture what an independent measure captures. Convergence among methods that share the same source of information is weaker evidence than convergence across sources (chronic pain and aggression).

5.2 Why the Instruments Are Still Used

None of this makes owner pain instruments useless. They capture everyday function — reluctance on stairs, stiffness after rest, changes in play — that a force plate does not see, and clinical guidelines treat owner evaluation as an important component of chronic pain assessment, alongside veterinary examination (Gruen et al., 2022). The point is about what kind of evidence they provide. They tell us how owners perceive change in their dog's function. That is clinically relevant, but it is a different quantity from measured limb loading, and the two should not be used interchangeably (why pain is missed in dogs).

6. Systematic Sources of Bias

6.1 Who Responds

Most owner surveys rely on volunteers recruited through social media, online forums, breed clubs, veterinary clinics, or approaches on walking routes. People who choose to complete a long questionnaire about dog behavior are likely to be more interested in dog behavior than average, more likely to have a dog with a problem, or more engaged with their dog in general. Such a sample does not represent the population of dog owners.

The consequences differ by question. Associations within a self-selected sample can still be informative, since the comparison is internal. Prevalence estimates are much more vulnerable: a statement that a given percentage of all dogs have a particular problem cannot be derived from a sample in which people with that problem were more likely to take part. The fireworks literature illustrates both sides. A New Zealand survey returned 1,007 valid questionnaires out of 8,966 distributed, covering 3,527 animals, and owners recognized fear of fireworks in 46% of them; the authors noted that respondents were not randomly selected (Dale et al., 2010). A later survey covering 1,225 dogs found fear of fireworks in about half of the sample (Riemer, 2019). Such figures describe the respondents. Whether they describe dogs in general is a separate question that the design cannot answer.

6.2 What People Prefer Not to Report

When people are asked about their own behavior, their answers are shaped by how they would like to be seen as well as by what they do. In training research this applies most obviously to punishment. A respondent asked whether they shout at their dog, jerk the leash, or use physical correction knows which answers are currently regarded as better.

None of the studies cited here measured the size of this effect in dog owners, and it should not be quantified by assumption. What can be said is that many owners do report aversive methods. In a British survey, 72% of owners reported using some form of positive punishment, and only 16% reported using positive reinforcement exclusively (Blackwell et al., 2008). Whether and to what extent social desirability affected those figures was not examined, and the figures come from a small convenience sample that cannot be generalized to dog owners today.

6.3 One Respondent, Both Variables

A problem rarely mentioned in media coverage is that in many studies the same person reports both the presumed cause and the presumed effect. The owner describes how the dog is trained and describes how the dog behaves. Both answers pass through the same perceptions, attitudes, and memory.

An owner who regards their dog critically may report more problem behavior and also use more punishment, without either causing the other. An owner well versed in learning theory may describe their training more favorably and their dog more charitably. Patterns of this kind, sometimes called common-method or shared-rater bias, can create or inflate associations that do not exist in the dog's behavior. The remedy is independent measurement: assessors who do not know how a dog was trained, or physiological and behavioral data collected by researchers rather than reported by owners.

6.4 Memory Is Not a Record

Many questionnaires ask about the past weeks or months; some ask about the dog's entire history. Recall is selective. Dramatic episodes are remembered better than routine ones, recent events better than earlier ones, and the current state colors memory of the past. An owner who has just had a difficult walk rates their dog differently from one who has had a good week. Questions about puppyhood answered by owners of adult dogs are especially vulnerable, because memory of the puppy is filtered through knowledge of the adult it became.

6.5 How the Questions Are Worded

Before any analysis, the wording of a question shapes the answer. Three problems recur in study-specific surveys. Double-barreled items ask about two things at once — "Does your dog bark or lunge at other dogs?" — so that a yes cannot be assigned to either behavior. Vague frequency terms such as "sometimes" or "often" mean different things to different respondents, which is why validated instruments anchor their scale points with descriptions. And questions without a defined reference period leave each respondent to choose their own, so that one owner answers for the past week and another for the dog's whole life.

Leading wording is a fourth problem, and the subtlest. A question that asks how much a dog "suffers" when left alone presupposes suffering. A question that asks whether an owner "still" uses a particular method implies that they should not. None of these issues requires bad intent, and all of them are largely avoidable with pre-testing. Whether a survey was pre-tested is rarely reported, which makes the wording of the items, where the paper provides them, worth reading in full.

6.6 How the Answers Are Analyzed

Analytic decisions can matter as much as the data. A recent example is unusually clear. An analysis of owner data on 43,517 dogs from the Dog Aging Project concluded that 99.12% of dogs had at least one behavior problem, with category prevalences of about 55.6% for aggression, 49.9% for fear and anxiety, and 85.9% for separation and attachment behaviors combined (Beaver, 2024). The figure rested on a composite criterion: a dog counted as affected in a category if it had either one rating indicating marked problems or two ratings of at least moderate level, with individual items treated as independent.

A re-analysis of the same data by James A. Serpell, one of the developers of the C-BARQ, and Lauren R. Powell arrived at substantially lower estimates for the individual categories — roughly 1% to 30% for different forms of aggression, 14% to 28% for fear, and about 9% for separation-related behavior. The category definitions themselves differed: the re-analysis, for example, treated separation-related behavior separately from attachment and attention-seeking. Their criticism was that the original analysis ignored the factor structure of the instrument, pooled items from different subscales, combined weakly related items into categories, and used a threshold low enough that a dog growling at a delivery worker would count as showing moderate aggression (Serpell & Powell, 2026).

The respondents and the answers were identical. The same dataset produced substantially different prevalence estimates depending on how behavioral categories and severity thresholds were defined. Any prevalence figure drawn from owner data should be read together with the definition of a case used to produce it.

7. Association Is Not Causation

7.1 What a Cross-Sectional Survey Can Show

Most owner surveys are cross-sectional: they record, at one point in time, how a dog is trained and how it behaves, and test whether the two are related. In the British survey mentioned above, the training methods owners reported were associated with the total number of potentially undesirable behaviors reported, as well as with attention-seeking, fear, and aggression scores (Blackwell et al., 2008). The sample was 192 returned questionnaires from owners approached in person, among other places while walking their dogs.

What such a study cannot establish is direction. Punishment may promote problem behavior. Owners of dogs with problem behavior may also turn to punishment when other approaches have failed. And both may depend on a third factor, such as the owner's experience. The design shows co-occurrence, nothing more.

7.2 Reverse Causation in Practice

The fireworks literature provides a clean example of how this plays out. In the New Zealand survey, comforting a frightened animal was associated with greater severity and duration of fear over time. The authors described this as an association that may indicate a negative effect on longer-term welfare; they did not claim that comforting caused the worsening (Dale et al., 2010).

Caution is warranted. It is at least as plausible that owners comfort more intensely the dogs that are more frightened, so that severity drives comforting rather than the reverse. A cross-sectional association between comforting and fear is consistent with both explanations and cannot decide between them. Claims that comforting a fearful dog reinforces fear, often presented as established, rest on this kind of correlational evidence.

7.3 Longitudinal Designs Help, Within Limits

Following the same dogs over time strengthens inference. In the Generation Pup longitudinal study in the United Kingdom, with around 1,130 dogs in the main model, a change to a more aversive category of training method between 9 and 12 months of age was associated with more owner-reported problem behavior (Owczarczak-Garstecka et al., 2025). Because the change was observed over time, this is more informative than a single snapshot. The authors nevertheless state that the direction of causation remains open, since emerging problem behavior could itself prompt a change in method. And both method and behavior were still reported by the owners.

7.4 Design Matters More Than Sample Size

Neutering research shows how much design can change a conclusion. An older telephone survey of owners of 57 castrated male dogs reported substantial reductions in mounting, urine marking, and roaming, without a comparison group (Neilson, Eckstein & Hart, 1997). A later prospective study surveyed owners of 48 gonadectomized and 48 intact dogs before surgery and nine months afterward. Mounting decreased in the castrated males, but marking did not change (Palestrini et al., 2021).

Both studies relied on owner reports, and the second was smaller in the treated group than one might expect a definitive study to be. It is still more informative, because it has a baseline and a control group. A modest owner survey with a sound design can outweigh a larger one without them (neutering and behavior in dogs).

7.5 When the Overall Evidence Carries Weight

None of this means that owner surveys cannot support conclusions. A single survey proves little. When many surveys with different samples, countries, and instruments point in the same direction, and when studies using other methods agree, the picture becomes more robust.

The evidence on aversive training methods is a case in point. A two-part systematic review of human-side risk factors for human-directed aggression by dogs, using narrative synthesis because meta-analysis was not feasible, found that among the factors examined only the association with physical punishment and confrontational methods could be rated as robust (Howell et al., 2025). That is still an association, not proof of causal direction. Studies that do not rely on owner reports add to it: in 92 dogs from training schools, dogs trained with aversive methods showed more stress-related behavior during training and larger increases in salivary cortisol, measured by researchers rather than reported by owners, although dogs were not randomly assigned to methods (Vieira de Castro et al., 2020).

The conclusion drawn here goes deliberately beyond what any single one of these studies measured. Punishment can stop behavior in the short term. On our reading of the combined evidence, the risks and side effects of aversive methods outweigh what they achieve, and reward-based methods allow learning without the specific risks of aversive stimuli (what aversive methods do neurologically, punishment fallout).

8. Reading an Owner-Survey Study: A Checklist

Faced with a headline of the form "Study of thousands of dogs shows …", six questions go a long way.

Who was asked? A random or population-based sample, clients of a single clinic, or volunteers responding to an online call? The more self-selected the sample, the more cautious one should be about prevalence figures.

With what? A validated instrument such as the C-BARQ, with its structure respected in the analysis, or a purpose-built survey? If the paper does not say, that is itself informative.

Observation or interpretation? "Barks at visitors" is an observation. "Is stressed", "is jealous", or "knows it did wrong" are inferences.

Did respondents know what was being tested? If owners knew whether their dog received an intervention and also reported the outcome, expectation effects are likely, blinded or not.

Is there an independent measure? Physiological data, coded video, assessments by blinded observers, or objective instruments such as force plates provide a check that owner reports cannot provide on their own.

Is an association presented as a cause? "Dogs whose owners do X show more Y" is a different claim from "X causes Y". A cross-sectional design cannot establish direction.

None of these questions makes a study worthless. They indicate how much weight a result can bear (measuring stress in dogs).

9. Research Gaps

Direct comparisons between owner reports and observed behavior are scarce. Most validation compares owner instruments with clinical diagnoses or with other questionnaires. Studies that compare owner ratings of specific behaviors with systematic observation of the same dogs in the same contexts are rare.

Social desirability has not been quantified in dog owners. It is a well-known mechanism in survey research generally, and it plausibly affects reports of training methods, but its size in this population is unknown.

Caregiver placebo effects are documented mainly for pain and cognition. Their size in behavioral interventions, where outcomes are almost always owner-reported, has hardly been studied.

Many questionnaire items mix observation and inference. A recent systematic review found that a quarter to nearly two thirds of items, depending on the group, were mixed or could not be categorized. Instruments built strictly around literal, observable behavior are still the exception.

Representative samples are almost absent. Nearly all large datasets are convenience samples. Population-based prevalence estimates for behavior problems in dogs remain largely unavailable.

Reporting standards are inconsistent. Many papers do not report response rates, recruitment channels, or the measurement properties of their survey items, which makes it difficult for readers to assess the risks described in this article.

10. Conclusion

Owner surveys are neither unreliable by nature nor a gold standard. They are a tool, and their reliability depends on the question. Where owners are asked about concrete, observable behavior using validated instruments, they provide data no other method can provide: behavior in the home, over long periods, in large numbers of dogs. Where they are asked for interpretations, about their own behavior, or about outcomes they expect, owner reports are vulnerable to systematic bias.

The strongest examples in this article make that concrete. Under placebo, owners and veterinarians reported improvement in osteoarthritic dogs whose gait did not change. Owners frequently named conspicuous signs of stress and seldom subtle ones. Dog owners were less accurate than non-owners at identifying fear in videos of dogs with children. And the same answers from the same 43,517 owners produced substantially different prevalence estimates, depending on how behavioral categories and severity thresholds were defined.

The practical implication is not to discard owner data but to read it for what it is: a record of what owners perceive and report, which is often closely related to what their dogs do and sometimes is not. Robust conclusions come from convergence — many surveys, different designs, and independent measures pointing the same way.

Key Insights

Owners see what laboratories cannot. For rare, home-based, and context-dependent behavior they are often the only source.

Validated instruments are not the same as surveys. The C-BARQ was checked against clinical diagnoses; most study-specific surveys were not checked at all.

Observations are more reliable than interpretations. "Barks at the door" can be reported well; "is stressed" much less so.

What owners do not recognize is missing from the data. Subtle stress signals are rarely named, and fearfulness tends to be underestimated.

Expectation changes reports, and blinding alone does not prevent it. In a double-blinded osteoarthritis trial, caregivers reported improvement under placebo while force-plate data did not change.

Agreement between questionnaires is not accuracy. Instruments that share a source can agree with each other and still diverge from independent measurement.

The same data can produce very different numbers. Prevalence estimates from one dataset changed substantially with the definition of categories and severity thresholds.

Cross-sectional surveys show association, not direction. Whether method shapes behavior or behavior shapes method remains open in a snapshot.

The overall pattern carries more weight than any single study. Conclusions become robust when many surveys and independent measures converge.

References

Beaver, B. V. (2024). The prevalence of behavior problems in dogs in the United States. Journal of Veterinary Behavior, 76, 34–39. https://doi.org/10.1016/j.jveb.2024.11.001

Blackwell, E. J., Twells, C., Seawright, A., & Casey, R. A. (2008). The relationship between training methods and the occurrence of behavior problems, as reported by owners, in a population of domestic dogs. Journal of Veterinary Behavior, 3(5), 207–217. https://doi.org/10.1016/j.jveb.2007.10.008

Conzemius, M. G., & Evans, R. B. (2012). Caregiver placebo effect for dogs with lameness from osteoarthritis. Journal of the American Veterinary Medical Association, 241(10), 1314–1319. https://doi.org/10.2460/javma.241.10.1314

Dale, A. R., Walker, J. K., Farnworth, M. J., Morrissey, S. V., & Waran, N. K. (2010). A survey of owners' perceptions of fear of fireworks in a sample of dogs and cats in New Zealand. New Zealand Veterinary Journal, 58(6), 286–291. https://doi.org/10.1080/00480169.2010.69403

Demirbas, Y. S., Ozturk, H., Emre, B., Kockaya, M., Ozvardar, T., & Scott, A. (2016). Adults' ability to interpret canine body language during a dog–child interaction. Anthrozoös, 29(4), 581–596. https://doi.org/10.1080/08927936.2016.1228750

Duffy, D. L., Kruger, K. A., & Serpell, J. A. (2014). Evaluation of a behavioral assessment tool for dogs relinquished to shelters. Preventive Veterinary Medicine, 117(3–4), 601–609. https://doi.org/10.1016/j.prevetmed.2014.10.003

Fratkin, J. L., Sinn, D. L., Patall, E. A., & Gosling, S. D. (2013). Personality consistency in dogs: A meta-analysis. PLOS ONE, 8(1), e54907. https://doi.org/10.1371/journal.pone.0054907

Frontiers Production Office. (2026). Correction: Measuring canine impulsivity: Translation and validation of a German dog impulsivity assessment scale. Frontiers in Veterinary Science, 13, 1981727. (Correction concerns the funding statement only.) https://doi.org/10.3389/fvets.2026.1981727

Grigg, E. K., Chou, J., Parker, E., Gatesy-Davis, A., Clarkson, S. T., & Hart, L. A. (2021). Stress-related behaviors in companion dogs exposed to common household noises, and owners' interpretations of their dogs' behaviors. Frontiers in Veterinary Science, 8, 760845. https://doi.org/10.3389/fvets.2021.760845

Gruen, M. E., Lascelles, B. D. X., Colleran, E., Gottlieb, A., Johnson, J., Lotsikas, P., Marcellin-Little, D., & Wright, B. (2022). 2022 AAHA pain management guidelines for dogs and cats. Journal of the American Animal Hospital Association, 58(2), 55–76. https://doi.org/10.5326/JAAHA-MS-7292

Horowitz, A. (2009). Disambiguating the “guilty look”: Salient prompts to a familiar dog behaviour. Behavioural Processes, 81(3), 447–452. https://doi.org/10.1016/j.beproc.2009.03.014

Howell, H., Baslington-Davies, A., Mills, D. S., & Hogue, T. E. (2025). Risk factors for human-directed aggression by dogs: The human side of the problem. A two-part systematic review and narrative synthesis. Applied Animal Behaviour Science, 284, 106552. https://doi.org/10.1016/j.applanim.2025.106552

Hsu, Y., & Serpell, J. A. (2003). Development and validation of a questionnaire for measuring behavior and temperament traits in pet dogs. Journal of the American Veterinary Medical Association, 223(9), 1293–1300. https://doi.org/10.2460/javma.2003.223.1293

Johnson, A., Schenke, K. C., Foster, J., & Stephens-Lewis, D. (2026). A systematic review of the characterization of behavior in canine behavioral questionnaires. Journal of Applied Animal Welfare Science, 1–47. https://doi.org/10.1080/10888705.2026.2693645

Kero, D., Dolezal, M., & Affenzeller, N. (2026). Measuring canine impulsivity: Translation and validation of a German dog impulsivity assessment scale. Frontiers in Veterinary Science, 13, 1869578. https://doi.org/10.3389/fvets.2026.1869578

Mariti, C., Gazzano, A., Moore, J. L., Baragli, P., Chelli, L., & Sighieri, C. (2012). Perception of dogs' stress by their owners. Journal of Veterinary Behavior, 7(4), 213–219. https://doi.org/10.1016/j.jveb.2011.09.004

McPeake, K. J., Collins, L. M., Zulch, H., & Mills, D. S. (2019). The Canine Frustration Questionnaire—Development of a new psychometric tool for measuring frustration in domestic dogs (Canis familiaris). Frontiers in Veterinary Science, 6, 152. https://doi.org/10.3389/fvets.2019.00152

Meyer, I., Forkman, B., & Paul, E. S. (2014). Factors affecting the human interpretation of dog behavior. Anthrozoös, 27(1), 127–140. https://doi.org/10.2752/175303714X13837396326576

Miller, J., Cavalli, C., Azadian, A., & Protopopova, A. (2026). Immediate behavioural outcomes of owner-delivered support during novel object exposure in dogs. BMC Veterinary Research. https://doi.org/10.1186/s12917-026-05837-6

Mills, D. S., Mueller, H. W., McPeake, K., & Engel, O. (2020). Development and psychometric validation of the Lincoln Canine Anxiety Scale. Frontiers in Veterinary Science, 7, 171. https://doi.org/10.3389/fvets.2020.00171

Morrill, K., Hekman, J., Li, X., McClure, J., Logan, B., Goodman, L., Gao, M., Dong, Y., Alonso, M., Carmichael, E., Snyder-Mackler, N., Alonso, J., Noh, H. J., Johnson, J., Koltookian, M., Lieu, C., Megquier, K., Swofford, R., Turner-Maier, J., White, M. E., Weng, Z., Colubri, A., Genereux, D. P., Lord, K. A., & Karlsson, E. K. (2022). Ancestry-inclusive dog genomics challenges popular breed stereotypes. Science, 376(6592), eabk0639. https://doi.org/10.1126/science.abk0639

Neilson, J. C., Eckstein, R. A., & Hart, B. L. (1997). Effects of castration on problem behaviors in male dogs with reference to age and duration of behavior. Journal of the American Veterinary Medical Association, 211(2), 180–182. https://doi.org/10.2460/javma.1997.211.02.180

Owczarczak-Garstecka, S. C., Casey, R. A., Cooper, B., Da Costa, R. E. P., Kinsman, R. H., Lord, M. S., Tasker, S., Williams, A., & Murray, J. K. (2025). Association between training methods and owner-reported problem behaviours in dogs enrolled in the ‘Generation Pup’ longitudinal study in the UK. Journal of Veterinary Behavior, 77, 52–60. https://doi.org/10.1016/j.jveb.2024.10.001

Palestrini, C., Mazzola, S. M., Caione, B., Groppetti, D., Pecile, A. M., Minero, M., & Cannas, S. (2021). Influence of gonadectomy on canine behavior. Animals, 11(2), 553. https://doi.org/10.3390/ani11020553

Riemer, S. (2019). Not a one-way road—Severity, progression and prevention of firework fears in dogs. PLOS ONE, 14(9), e0218150. https://doi.org/10.1371/journal.pone.0218150

Salvin, H. E., McGreevy, P. D., Sachdev, P. S., & Valenzuela, M. J. (2011). The canine cognitive dysfunction rating scale (CCDR): A data-driven and ecologically relevant assessment tool. The Veterinary Journal, 188(3), 331–336. https://doi.org/10.1016/j.tvjl.2010.05.014

Serpell, J. A., & Powell, L. R. (2026). Prevalence and severity of behavior problems in dogs in the United States: A re-assessment. Journal of Veterinary Behavior, 83, 8–13. https://doi.org/10.1016/j.jveb.2025.11.005

Simon, K. E., Russell, K., Mondino, A., Yang, C.-C., Case, B. C., Anderson, Z., Whitley, C., Griffith, E., Gruen, M. E., & Olby, N. J. (2025). Quantifying placebo and trial participation effects on cognitive outcome measures in aging dogs. GeroScience, 47(6), 6909–6926. https://doi.org/10.1007/s11357-025-01822-3

Sullivan, G. M., & Artino, A. R. (2013). Analyzing and interpreting data from Likert-type scales. Journal of Graduate Medical Education, 5(4), 541–542. https://doi.org/10.4300/JGME-5-4-18

Vieira de Castro, A. C., Fuchs, D., Morello, G. M., Pastur, S., de Sousa, L., & Olsson, I. A. S. (2020). Does training method matter? Evidence for the negative impact of aversive-based methods on companion dog welfare. PLOS ONE, 15(12), e0225023. https://doi.org/10.1371/journal.pone.0225023

Walton, M. B., Cowderoy, E., Lascelles, D., & Innes, J. F. (2013). Evaluation of construct and criterion validity for the ‘Liverpool Osteoarthritis in Dogs’ (LOAD) clinical metrology instrument and comparison to two other instruments. PLOS ONE, 8(3), e58125. https://doi.org/10.1371/journal.pone.0058125

Wilkins, V., Evans, J., Park, C., The Dog Aging Project Consortium, Fitzpatrick, A. L., Creevy, K. E., & Ruple, A. (2024). Validation of the shortened version of the Canine Behavioral Assessment and Research Questionnaire (C-BARQ) using participants from the Dog Aging Project. PLOS ONE, 19(4), e0299973. https://doi.org/10.1371/journal.pone.0299973