Shelter dogs, movement, and stress

The always-awesome group of researchers at the Center for Shelter Dogs (associated with the Animal Rescue League of Boston, MA) has just published a new paper. They took a bucketful of different kinds of stress measurements of dogs in their shelter and looked to see if there were correlations between the different kinds. They are working on the same problem that I tackled in my Master’s work: it is awfully hard to tell which dogs in a population are stressed; can we use some kind of easy marker (like observing behavior) to do it? They got similar results to mine: yes, it’s super hard! There is no silver bullet answer. But they provide some interesting insight into how to move forward in the quest to improve stress detection methods.

Jones S., Dowling-Guyer S., Patronek G.J., Marder A.R., Segurson D’Arpino S. & McCobb E. (2014). Use of Accelerometers to Measure Stress Levels in Shelter Dogs, Journal of Applied Animal Welfare Science, 17 (1) 18-28. DOI:

The biggest contribution this paper makes, I think, is the use of an accelerometer to test activity levels in shelter dogs. They attached this device to the collars of dogs when they first came in to the shelter, and got a report of how much the dogs moved around over the course of 24 hours. Because cortisol is such a difficult measurement of stress to interpret, it’s important to supplement cortisol measurements with other measurements, to sort of triangulate your answer. Movement in the kennel is one that I haven’t seen measured before in shelter dogs, and is something that might provide some interesting answers: do dogs move around a lot when they are stressed (pacing) or very little (depression)? I’m really happy to see this new measurement entering the shelter dog literature.

The researchers also attempted to provide a stress score of the dogs by watching them and subjectively assessing stress levels based on a minute’s worth of behavior. This is the holy grail of stress studies: can we look at a dog and tell how stressed it is? If so, we wouldn't need to do all this correlation with cortisol levels. I think we all want to say that someone who really understands dog body language can tell if a dog is stressed by looking at it. I know that I believe that I can estimate the average dog's stress level by looking at it. If you see a dog with low body posture, refusing to meet your eyes, maybe shaking, it’s stressed, right? How hard is that?

Well, if you try to compare your observations (particularly over a very short period of time — in my work, I had more luck with 20 minutes than with 2), the answer is, it’s extremely hard. In this study, they tried to assess an average level of stress in the dogs by taking multiple cortisol measurements, both salivary and urinary. Salivary cortisol changes so very fast that I wouldn’t expect it to correlate well to a day’s worth of exercise, or even to behavioral observations unless they were taken within just a few minutes of the saliva sample, but as this study involved multiple saliva samples over time, I was interested to see if a better average measure was obtained. The researchers also compared salivary cortisol samples with urinary cortisol samples, and since urine builds up in the bladder over time, I’d expect urinary samples to produce a better average as well.

So what did they find? To the question of movement (measured by the accelerometer) vs cortisol (salivary and urinary):
  • Maximum activity level correlated with salivary cortisol (p = 0.025)
  • Maximum activity level did not correlate with urinary cortisol
  • Mean (average) activity level correlated with mean urinary cortisol (p = 0.028)
  • Mean activity level did not correlate with salivary cortisol
To the question of the behavioral scoring: no correlation with cortisol of either kind.

So how we interpret all this?

First off, this study ended up enrolling only 13 dogs, taking quite a few measurements of each dog (I haven’t actually covered all of their findings here), and trying to draw conclusions. On the one hand, I want to emphasize that this is how stress studies in dogs are done. I did exactly the same thing in my stress study of hospitalized dogs. It is just extremely hard to enroll enough dogs. If you read the paper itself you get a feel for what the primary researcher went through, as she lists the reasons she had to exclude dogs from the study. It reminded me of my intense frustration during my Master’s work as I had to give up on dog after dog for a variety of reasons. This is par for the course in all these studies: it is almost impossible to get the time and funding to enroll enough subjects to have solid statistical findings. So I am not in any way criticizing these researchers, who did a great job introducing some interesting findings.

But on the other hand, it’s important to recognize that with so many questions and such a small sample set, it is almost impossible to trust the statistical significance of the results. If you do a study in which you ask 100 questions, and you set your p value at 0.05 (which is usual), then you are saying that you expect 5 of your answers to look significant even though they are not. That's what a p value is: setting the bar at which you accept a few false positives. One way around this is to have more subjects, which will lower your p values. Then you can say you'll only accept p < 0.01 or something even more stringent. This makes your findings a bit more trustworthy.

For the number of questions this study asked and the size of their sample set, I would take their findings with a grain of salt. Does activity level correlate with cortisol level? I think it's likely that it does. But I also think that what this study tells us is “this is an interesting area which is worth more study,” not “you should trust that these findings are absolutely true.”

Moreover, what are high cortisol levels telling us about dogs who move around a lot? Exercise itself can increase cortisol levels. This can be a good stressor. So are these dogs distressed or not? This is a problem which is going to be very hard to pick apart. I think a lot of the dogs in a shelter who pace incessantly are indeed very distressed, but some, as this paper points out, are coping with their distress by means of that exercise and are doing better than the dogs who don’t move. The paper also asks the question about how to interpret movement in small versus large dogs. Small dogs simply have relatively more room to move in a little shelter kennel. So how does that change the equation?

As for the lack of correlation between the behavioral stress score and cortisol levels: I feel pretty confident interpreting that one despite my comments about statistics above, because this is a question that has been asked before, and is always answered the same way. We can’t tell what a dog’s cortisol level will be by looking at its behavior. Why not? Is it because we don’t know what the dog's inner experience is (we don’t know if the dog is feeling stress)? Or is it the cortisol level that is lying to us, doing a terrible job of telling us about stress levels, and we’re interpreting the behavior just fine? We don't know, but we do keep trying to find out. Hopefully one day we will.

In the end, I enjoyed this paper. Kudos to the researchers for exploring stress levels in a variety of ways, instead of just one or two. This study was well designed, in my opinion; it just needed a lot more dogs. Hopefully this group or another one will be able to pull together a bigger study going forward.
Labels: , ,