{"id":7604,"date":"2024-09-19T07:04:25","date_gmt":"2024-09-19T05:04:25","guid":{"rendered":"https:\/\/illis.se\/en\/?p=7604"},"modified":"2024-12-30T21:54:43","modified_gmt":"2024-12-30T20:54:43","slug":"shock-collar-critique","status":"publish","type":"post","link":"https:\/\/illis.se\/en\/shock-collar-critique\/","title":{"rendered":"A critique of the “banana” shock collar study"},"content":{"rendered":"\n

I find the latest scientific contribution on the issue of shock collars to be frustrating \u2013 even outrageous.<\/p>\n\n\n\n

Lots of problematic articles are published every day, but this one potentially has far-reaching consequences, and so I feel a lengthy, occasionally nerdy and also somewhat rambling blog post is in order.<\/p>\n\n\n\n

I won\u2019t describe the paper<\/a> down to a T, but lift specific topics that I take issue with, so I encourage you to skim it before reading this post \u2013 or not\u2026 <\/p>\n\n\n\n

In short, the authors found that teaching dogs to refrain from chasing a fast-moving lure was effective when shock-collars were used, however attempting to train the behaviour in a slightly shorter time frame using \u201cfood rewards\u201d (50 minutes as opposed to 60 for the shocked dogs) was essentially useless. They also saw no signs of distress in the shocked dogs, except that all of them yelped in pain at some point.<\/p>\n\n\n\n

I\u2019m paraphrasing here, obviously.<\/p>\n\n\n\n

And I\u2019m using quotation marks to draw your attention to the fact that I\u2019m not quite sure that they actually ever used those \u201cfood rewards\u201d as reinforcement for correct behaviour in those treatment groups \u2013 the training setup was absolutely bananas, if you\u2019ll pardon the pun (that last bit will make sense in a moment).<\/p>\n\n\n\n

TL;DR? Here are the main points of this blog post:\n\n- The \u201cfood-reward\u201d training was a travesty with a multitude of problems (e.g. dubious conditioning, unwanted behaviour reinforced, no shaping, no assessment of engagement, no calibration of the reward value, adding distractions and distance way too soon, not using a marker, etc etc); no learning occurred\n\n- The food-reward dogs got less training time than the shocked dogs\n\n- The type of training needed to be successful in the tests was not in the protocol for the food-reward dogs\n\n- The \u201cwelfare measures\u201d were inadequate and possibly confounded\n\n- Generalized fear learning occurred in the shocked dogs, a potential concern for their long-term wellbeing\n\n- There was conflict of interest: shock-collar trainers training the \u201cfood-reward\u201d dogs\n\n- The authors conclude that shock collars may occasionally save lives without considering the risks of shock collars costing<\/em> lives, which I suspect is on a different order of magnitude\n\n- The problems are serious enough to warrant retraction of the paper<\/pre>\n\n\n\n
We\u2019ll get to my detailed objections in a minute, but first off: I think the chosen approach, comparing several treatment groups, where essentially the only difference in training set-up is whether dogs receive a shock for doing the wrong behaviour, or a treat for doing the correct behaviour, is flawed.<\/p>\n\n\n\n
It\u2019s based on the proud tradition from other scientific fields such as biomedicine, whereby you compare two different treatment options, subjecting one group to treatment A and the other to treatment B, keeping all other variables identical in the two groups. And so, the theory goes, you can be sure that any differences between the groups will be a result of different effects of the two treatments, A or B, and not some other random factor such as the weather.<\/p>\n\n\n\n
\"\"<\/a>
Typically, a biomedical study involves genetically identical rodents housed under standardized conditions, so that presumably, the only difference between the two study groups is which research treatment option (A or B) they\u2019ve been given.<\/em><\/figcaption><\/figure>\n\n\n\n\n\n\n\n
The problem is that training using aversives<\/em> versus reward-based training <\/em>(what I\u2019ll refer to as R+ training throughout this post) <\/em>live in completely different universes. The difference doesn\u2019t simply boil down to the moment the consequence is delivered, whether that\u2019s appetitive or aversive, but there\u2019s large differences in a multitude of other areas as well. These two vastly different approaches simply don\u2019t lend themselves to the standard recipe for comparing two treatment options, the keep-everything-identical-except-the-variable-under-study-approach, what\u2019s referred to as standardization<\/em>.<\/p>\n\n\n\n
In other words, this paper is trying to fit a square peg into a round hole.<\/p>\n\n\n\n
There were also many many many other problems with this paper. I\u2019ll touch on the more egregious ones below.<\/p>\n\n\n\n
Let\u2019s grab the bull by the horn and look at the \u201cfood reward\u201d treatments.<\/p>\n\n\n\n
The \u201cFood reward\u201d training.<\/h5>\n\n\n\n
As far as I can tell, after allowing the dogs to spend up to 40 minutes chasing a lure, on the following day:<\/p>\n\n\n\n