The 9 Percent: What Your Dog's Breed Actually Predicts

Picture the dog you know best — the one whose moods you can read, whose quirks have become a running joke, whose demands on your schedule are non-negotiable. Now ask yourself: how much of that dog was written in the breed label, and how much is just... him? That question, which has occupied dog behaviourists for decades, received its most rigorous answer in April 2022, when a team led by geneticist Elinor Karlsson at the Broad Institute published a study in the journal Science. The headline figure was a small number: nine percent.

Nine percent is the share of behavioural variation in individual dogs that can be explained by breed. Not zero — but not the majority. The study, which analysed genetic data alongside owner surveys for 18,385 dogs across 128 breeds, found that for most traits we associate with specific breeds — sociability, biddability, aggression, howling — there is enormous variation within breeds, and considerable overlap between them. A Border Collie is more likely to be biddable than a Beagle. But that is a population-level statement. It tells you almost nothing about the specific Border Collie you are meeting for the first time.

Why the label became a promise

Breeds are a recent invention. The modern concept of a pure-bred dog — a stable type with a written standard, a closed stud book, a predictable form — is mostly a Victorian project. The Kennel Club was founded in 1873; the American Kennel Club in 1884. Before that, dogs were broadly sorted by function (hound, terrier, gundog, drover's dog) and selected for working performance, not documented lineage. The result was enormous local variety and a great deal of genetic mixing. What the Victorian registries did was freeze certain functional and aesthetic types, close the books, and — over generations of selective breeding — amplify particular traits while reducing others.

What they did not do was produce behavioural uniformity within breeds. Even highly selected working lines — dogs bred for generations specifically for performance at a task — show wide variation. The Karlsson study found that howling, for instance, was more heritable and more breed-predictable than most traits; but sociability with strangers and engagement with toys showed much weaker breed signals. Physical traits (body shape, coat type, size) are far more strongly predicted by breed than any behavioural measure.

The C-BARQ evidence

The Canine Behavioral Assessment and Research Questionnaire — C-BARQ, developed by James Serpell at the University of Pennsylvania — is one of the most widely used standardised instruments for measuring dog behaviour. It asks owners to rate their dog across more than 100 behaviours on validated scales, producing scores on dimensions like stranger-directed aggression, owner-directed aggression, trainability, chasing, and separation anxiety. When C-BARQ data is sorted by breed, patterns do emerge: certain breeds cluster toward higher trainability scores; others toward higher chasing or fear scores. But the within-breed variance is consistently large — often larger than the between-breed variance.

The Karlsson team used C-BARQ-style survey data as one of their behavioural inputs. Their genomic analysis found that even traits with the strongest breed signals — biddability, for example — showed only moderate heritability. For the most part, individual behavioural outcomes appear to be shaped by a combination of many genetic variants (each contributing a tiny effect), developmental experience, training history, and environment. Breed heritage is one input among many, and not the dominant one.

The stakes: breed-specific legislation

The nine-percent finding has implications that reach beyond the individual owner trying to choose a family pet. Breed-specific legislation — laws that restrict or ban ownership of dogs based on appearance or breed — is premised on the assumption that certain breeds are inherently more dangerous. The United Kingdom has banned pit bull terrier types since 1991 under the Dangerous Dogs Act. Several Australian states restrict or ban certain breeds. Singapore restricts certain large breeds in HDB public housing.

The Karlsson team was direct: their data does not support using breed as a proxy for individual behaviour in policy decisions. The evidence base for breed-specific legislation has been contested by veterinary and animal behaviour organisations including the American Veterinary Medical Association, which reviewed the literature and concluded that breed is a poor predictor of individual dog behaviour and that bite risk is better understood through individual history, owner behaviour, and situational factors.

What breed does predict — and what it doesn't

So what, practically, can a prospective owner learn from a breed label? Quite a lot about body. Size, coat type, exercise requirements, and predispositions to certain heritable health conditions vary meaningfully by breed. A Border Terrier will reliably be small. A Siberian Husky will shed. A French Bulldog faces well-documented respiratory risks. Physical breed characteristics carry a far stronger genetic signal than behaviour.

Functional tendencies — the residue of selective breeding for a specific job — also have some predictive value, but with important caveats. A working-line Malinois from a herding background is more likely to show high arousal and intensity than a toy breed from generations of lap-dog selection. But this is a probabilistic tendency, not a guarantee. And it operates at the level of a working line or careful breeding programme, not the breed label alone — which may encompass both working and show lines with very different behavioural profiles.

Individual dogs, not archetypes

The most useful reframe the nine-percent finding offers is this: stop asking "what do Labrador Retrievers do?" and start asking "what does this Labrador Retriever do?" The difference matters enormously in practice. A shelter dog labelled a "pit bull mix" carries no predictive information about whether he will be gentle with children. A purebred Golden Retriever is not guaranteed to be the sociable, biddable dog the standard describes. An individual's history — early socialisation, training, the quality of attachment to their owner — shapes behaviour far more powerfully than any breed average.

This is not a counsel of randomness. It is a counsel of observation. Spend time with the actual dog. Meet its parents if possible. Learn its history. Talk to the trainer or foster carer who knows it best. Those sources of information will tell you more about the dog you are about to share your life with than any breed profile — however confidently written.

The road ahead

Genetics research on canine behaviour is accelerating. As more dogs are sequenced and more behavioural data collected — through instruments like C-BARQ, through platforms like Darwin's Dogs, and through emerging wearable and AI tools — the picture will sharpen. Researchers are already identifying specific genetic variants associated with particular behavioural tendencies: for example, variants near genes involved in neural development and social cognition. But even as the resolution improves, the pattern is unlikely to reverse. Dog behaviour is a deeply polygenic, developmentally plastic trait. The genome is a tendency, not a script.

Nine percent is a small number. It is also, in its way, a hopeful one. It means that the dog in front of you is not fully predetermined. It means that how you raise, train, and engage with a dog matters enormously. And it means that a shelter dog of uncertain background — dismissed because of a label, feared because of a silhouette — deserves the same chance to surprise you as any pedigree dog ever did.