ResearchBlogging.orgLately, a paper to be published in the June edition of the American Naturalist has been getting some attention. The findings that are getting reported out of this paper didn't make sense to me, but I wondered if this was an issue with accuracy in reporting. So I went and found the paper. Turns out that the reporting is accurate, its the actual findings from the paper that confuse me. I really wanted to make sense of this paper, so I've been waiting a while to blog about it. But I can't make sense of one key finding.

confused otter.jpg

Figure 1: An artist's rendition of me, being confused. If, you know, I were an otter.

But before we get to the confused, let's just review the paper, shall we?

The basic finding is that personality, energy expenditure, and longevity are correlated in domestic dogs. Dogs are a really good species to study the potential relationships between psychological attributes (like obedience), physical attributes (like body mass), physiological attributes (such as energy expenditure or metabolism), and average length of life.

Why are dogs ideal for addressing this question? Because there are over 400 different breeds which have been explicitly bred for various purposes. And It is unlikely that different breeders explicitly selected dogs on the basis of longevity or physiology; instead, most dogs were bred for specific jobs on the basis of morphology and trainability (i.e. personality). For example, Great Danes was bred as guard dogs, German shepherds as herders, and Pomeranians as companions. And different breeders in different locations bred different dogs for ostensibly similar purposes (tracking, herding, guarding, fighting, etc). And the different breeds are genetically distinct - there are more genetic differences between breeds than within breeds because breeders try really hard to keep their breeds genetically "pure" (i.e. purebred). (Keep in mind that even though the different breeds are relatively genetically distinct, all domesticated dog breeds still belong to the same species.)

The researchers picked out 56 breeds about which a lot is known about breed-specific personality, longevity, and energy expenditure. They labeled these traits "activity" which relates to general activity and excitability, "aggressiveness," relating to territorial defense and aggression towards other dogs, and "trainability," which refers to obedience, and they define as falling along a shy-bold dimension. They did a literature search and assigned scores for these three variables to as many of those 56 breeds as possible. They correlated these scores for each breed with adult body mass, mortality (risk of dying in the first 10 years of life), and "metabolizable energy intake," which means how much the dog eats.

Figure 2: Tassie, who is apparently "not particularly obedient or friendly" belongs to bonobo expert Vanessa Woods.

They found three statistically significant correlations (say it with me: "correlation does not equal causation"):

(1) Activity correlated negatively with body mass. This means that bigger dogs are less active and excitable.

(2) Trainability correlated negatively with mortality. This means the more obedient breeds tend to outlive their more disobedient cousins. This is the one that's been getting all the attention in mainstream media.

(3) Aggressiveness correlated positively with energy intake. Breeds that eat more are more aggressive.

Figure 3: Scatterplots. Pretty.

Longevity and metabolism are not explicitly selected for in breeding, nor are they included in breed standards. Additionally, different breeds were bred in different places by different groups of people. If you find a systematic relationship between two traits, one of which was explicitly selected for, like obedience, then it is reasonable to infer that the second trait varies as a correlated by-product of selection for the desired trait, instead of being simultaneously explicitly selected as well. So add longevity and metabolism to the list of things that are potentially correlated with personality, at least in dogs. This does, however, generate interesting hypotheses that are easily testable in humans, and that would be interesting.

Here's a big problem I have, and why this study confuses me. The authors say, in their abstract:

We found that obedient (or docile, shy) breeds live longer than disobedient (or bold) ones.

They say that obedience falls along a shy-bold dimension, with more obedient dogs being more shy. The experiments on the domestication of silver foxes in Russia showed that it was the more bold, less afraid foxes that were more willing to interact with humans. The more shy foxes were more afraid of humans and more aggressive. It doesn't follow that more shy breeds would be more trainable, because more trainable dogs need to be more willing to interact with humans. They need to prefer approach over avoidance. I expect that the more shy breeds would tend to be less obedient simply because they would prefer avoidance over approach. I'm not certain that these researchers are measuring what they intend to measure. So the finding that more obedient dogs live longer might suggest that more bold dogs - dogs who prefer approach over avoidance - are the ones that live longer. And if this is the case, there are some very interesting implications and potential physiological mechanisms that should be investigated, like the hypothalamic-pituitary-adrenal axis, which is involved in the stress response and approach/avoidance. To recap: it's not the finding (obedience related to longevity) that I have an issue with, its the measurement (shy dogs = more obedient) and the interpretation that arises from it.

I have one other big problem with this. Despite the fact that they conducted their literature search on 56 breeds, several of the correlations included far fewer breeds. Mortality rate data was only included from 19 breeds, and energy intake from only 9 breeds. So I'm less that impressed by these results. They did have body mass data for all 56 breeds. While the scatterplots look good and the regressions are highly significant, it only takes ONE datapoint to make or break significance in a correlation, even if you have a bajillion datapoints. If you DID have a bajillion datapoints, you could be reasonably certain that your correlation is legitimate, but with only 9 or 19 datapoints, it is uncertain.

So what are the take-home messages?

(1) These correlations are based on extremely small sample sizes, so interpret with caution.

(2) Domestication often has unintended consequences. As we continue to investigate the effects of domestication, we will likely find more things that are correlated by-products of selection for physical or psychological attributes.

(3) I'm confused. Or perhaps the paper authors are confused. I'm not sure. Does anyone have a different understanding than I do?

Careau, V., R