How Much of a Dog’s Behavior Is Predicted by Its Breed?

Source: Andrea Stöckel/Public Domain Pictures

You may have seen some recent media reports with headlines like "Study Shows Dog Breed Explains Very Little about Dog Behavior and Personality." These articles are supposedly summarizing a new study that investigated the influence of genetics on dog behavior and physiology. The problem is that these reports are based upon a misinterpretation of the data.

A huge new data set

The lead author of the new study is Kathleen Morrill from Bioinformatics and Integrative Biology at the University of Massachusetts Chan Medical School. This is a massive investigation that contains a wealth of useful information. In fact, it is so large that it involved 24 authors and required a 112-page supplementary data package to go online upon publication. Data were collected from surveys of 18,385 dog owners and the sequencing of the DNA of 2155 dogs. About half of the dogs were not purebred but were mixed breed. This information was gathered using a website called Darwin's Ark. It is an example of "citizen science," and it has been accumulating surveys and genetic data on thousands of dogs across the United States since 2015.

Participating dog owners completed an inventory of 117 questions to specify their dog's demographics and physical characteristics. The inventory also included behavior questions ranging from how friendly the dogs are with strangers to whether the dogs typically circle before they poop. Following completion of the questionnaires, participants send in a cheek swab from the dog which can be used for DNA sequencing.

Physical traits are more heritable than behavioral traits.

Combining the genetic and survey data for 1,967 of these dogs, researchers found that the physical traits of canines are "exceptionally heritable," often above 85 percent heritability. This analysis was simplified by the fact that some physical characteristics, such as whether a dog has a prick ear or a floppy ear, are the result of variations of a single gene.

In some instances, there are also simple genetic predictors of particular behaviors. For example, whether or not a dog tends to howl was mapped to a specific area of the genome which in humans is involved in the development of speech and language.

However, it is much more usual to find that dog behavioral traits are "polygenic," meaning that several genes or genetic loci each contribute some small effect (along with the environment) to shape the final behavior. Thus, one would expect smaller inherited effects for behaviors. Nonetheless, the researchers report more than 25 percent heritability for some behavioral dimensions.

The problem

Modern dog breeds began to appear in the 1800s. Before that, humans had been intentionally breeding dogs to perform specific functions, such as hunting, guarding, and herding—paying no attention to what the dogs looked like. With the advent of dog breeds, people began to sort the dogs on the basis of appearance and certain notions of aesthetics. The result was that less emphasis was placed on behavioral characteristics.

The researchers in this new study suggest that as a result, breed identity became an unreliable predictor of canine personality and behavior. Thus, although Border Collies are characteristically described as clever and trainable, occasionally some people will find that their new Border Collie pup seems to have the learning ability of a river rock. The researchers suggest that instances like this mean that breed is not a reasonable predictor of behavior.

Understanding what went wrong

Unfortunately, their conclusion is based on a misunderstanding of basic psychological principles. It is possible to have an inherited behavioral tendency which is a good predictor when we look at groups of individuals; however, that does not necessarily indicate that every individual in that group will have the predicted characteristics. L

et me give you an example based on physical characteristics, which may be easier to understand. Consider the physical height of men and women. The average American man, 20 to 30 years of age, has a height of just over 5 feet 9 inches (69.1 inches or 175 cm). The average height for American women in their 20s is just under 5 feet 4 inches (63.7 inches or 162 cm). So obviously, on average, men are taller than women. Therefore, it would make sense if for some reason you wanted a taller individual for some task (e.g., playing basketball or picking tree fruit) that in the absence of any further information you should select a male. However, 6 percent of women are taller than the male average, and 4 percent of males are shorter than the average woman. So if you use the difference in the distribution of heights by sex as your basis of selection, and you find that you end up with a short male, that does not invalidate the usefulness of sex as a predictor of height.

Genetics Essential Reads

Genetics and Addiction: What We've Learned

Decoding Nature and Nurture: Insights From Twin Studies

Reasoning in the same way, if you find a not-very-trainable Border Collie, it does not invalidate the usefulness of breed identification in predicting dog behavior. The reality is that dog breed, and all of the genetic characteristics that describe it, are good predictors when you consider groups of individuals, but may not be accurate in predicting the performance of any one single individual.

How well does dog breed predict dog behavior?

These researchers developed a computational engine that allows us to calculate the effect of breed on behavior, based on their data. Given any measured characteristic, it computes how many dogs from a selected group of breeds are expected to fall into the top quartile (highest 25 percent) compared to all dogs in the study (which, conversely, fall into the lowest 25 percent). This allows us to see how reliable breed is for predicting behaviors.

Let's consider just two of the behavioral dimensions that they measured as examples. The first is "Human Sociability" which the researchers define as how comfortable a dog is around people, especially if the people are unfamiliar. Sorting for the highest degree of sociability, we find that 62 percent of golden retrievers will fall into the highest quartile. This doesn't mean that you will never find an unsociable golden retriever since 18 percent of them will fall into the lowest quartile. However, your odds of getting a highly sociable dog if you choose a Golden retriever are more than 3 to 1. Some other highly sociable dogs with a better than 50 percent chance of falling in the top group are the Siberian husky, pug, and Labrador retriever.

Another dimension with high breed predictability is what these scientists refer to as "Biddability," but which other researchers have referred to as "Working and Obedience Intelligence." It has to do with how easily a dog can be trained and how well it responds to human direction. The dog breed that scores highest on this is the Border Collie, with 72 percent of these dogs falling in the top quartile for all dogs measured. Again, the breed is not a perfect predictor, since 16 percent of this breed of dogs will be in the lowest quartile; however, your odds are better than 4 to 1 that you'll be getting an intelligent and trainable dog if you get a border collie. Other dog breeds with a better than 50 percent chance of falling in the top group for biddability are the golden retriever, German shepherd, Australian cattle dog, and Australian shepherd.

There are several behavioral dimensions in which the predictive ability of breed was found to be considerably weaker. One example is the "Agonistic Threshold," which these researchers define as "how easily the dog is provoked by frightening, uncomfortable or annoying events." They interpreted this poor predictability as demonstrating that environmental factors can also influence this behavior.

So, contrary to the headlines, using the computational engine provided by these genetic researchers, we find that they have actually demonstrated that breed is a pretty good indication of some behavioral differences between groups of dogs. However, it is certainly not a guarantee for the behavior of any one individual.

Copyright SC Psychological Enterprises Ltd. May not be reprinted or reposted without permission