Predictability and Control in Horse Training

In this literature review, I wanted to discuss Bassett & Buchanan's 2007 article entitled Effects of predictability on the welfare of captive animals. The authors are primatologists but their findings have a lot of relevance for horse training. The contradictions exposed and discussed in their paper make it difficult to follow the wealth of studies they examined, but when it is all said and done, they summarize their findings in seven guidelines for implementing what is definitively known in respect to animal management and husbandry. I present their recommendations in the following block quotes with annotations about how I see them applying to horse training.

---

(1) There should be a thorough evaluation of both positive and negative husbandry and management routines in terms of temporal and (often unintentional) signalled predictability, and welfare should be assessed prior to positive and negative events.

This principle speaks to the issue of developing an awareness of what is happening between your animals and yourself. If you offer consistent habits and patterns, your animals will experience you as predictable. Do you always train at a certain time? Do you dress differently when it's time to train? Do you ask for responses in predictable ways? What do you do erratically? Just recognizing the difference is the important part. If your animals are showing anticipatory behaviors in unexpected times, you should thoroughly analyze the situation to see what is signaling them to initiate that behavior.

2) If possible, negative (aversive) events should be made temporally predictable, and any unreliable (often unintentional) signals should be removed where feasible.

If you are using negative reinforcement (also known as pressure/release in the Natural Horsemanship world), then you want to always signal that the pressure is going to start with a very light and non-aversive cue. This cue, besides giving the animal a chance to prevent the aversive stimulus, takes a lot of stress out of the pressure part. If you don't signal the beginning of an aversive application of pressure, the animal is less likely to learn the lesson because they don't learn well if they are under stress. It's subtly different to think of a cue as being a tiny request versus thinking of it as a signal that something aversive is about to happen. Cues do both things, don't they?
Now think about the second part of the recommendations. An unreliable cue/predictor would be the fail to predict what was about to happen. An unreliable cue might fail to always precede the aversive event or might be given when the animal can't stop the aversive event it predicts from happening. For example if rein contact signals a stronger tug on the reins unless the animal offers a stopping response, it becomes less predictable for the animal if we take up contact then boot the animal into a "go" response. The predictors get mixed up. Quick sequential cues can come along, but not until the animal learns the behaviors separately.

3) A unique and reliable signal to indicate the onset of an aversive event should also be introduced to decrease stress, but the beneficial effects should be scientifically demonstrated before widespread practice is adopted. Care must be taken to ensure the signal only refers to the individual for whom the event will take place, and not be a (unreliable) signal for the whole group.

If you are working with a group of animals, precede cues with their names. This goes whether or not they are predictive of positive or negative consequences. They will learn to ignore cues that are not directed at them, but if you don't make them unique to the animal, then they will learn to ignore the general cue.

4) Training animals can be a useful management tool to provide improved care of captive animals. Positive reinforcement training increases the control and predictability the animals have and should be used to provide welfare benefits.

Having some degree of control over their world is very reinforcing for animals. Using a straight flooding (sacking) technique to get them "used to" something is far less effective than giving them a cue to predict the onset of contact with the stimulus. and much much less effective than giving them a way to respond to the stimulus in such a way that they feel they have some control over it. When they feel like they have control, the stimulus becomes much less threatening. Turning something into a negative reinforcement scenario instead of direct sacking-out session is good standard operating procedure. With a mustang, the human focus and attention is an aversive event to begin with. You can just heap it on them and eventually they get used to it, but you can train smarter and remove focus and attention every time they perform a particular desired behavior (like face-up). Watch Bryan Neubert or Leslie Neuman to see masters of this practice.

5) Every effort should be made to avoid delays to events occurring on a (usually) reliable temporally predictable schedule, as delays (loss of temporal predictability) have a negative impact on welfare.

Here is the value of using the rhythm of a cue and applying it a set number of times before upping the ante to the aversive event. I like the tempo setting practice of saying "Please (cue), do it (low level aversive), NOW! (high level aversive).
In positive reinforcement training, the bridge can help keep the cue, the response, and reinforcement all connected.

6) FAA (some of which is undesirable) is primarily seen with animals which are food restricted (and fed on a predictable schedule), constituting an additional welfare concern to this practice.

FAA means Food Anticipatory Activities, which translates into donkeys braying for dinner, or pacing along the stall gate. A predictable schedule may increase FAA. I am not sure that this means for training beyond what is dicussed under the next item.

7) If possible, temporally unpredictable feeding should be introduced (although note that the benefits have only been demonstrated in a limited range of species). Unreliable signals relating to feeding should be eliminated if possible, and a unique reliable signal introduced.

I interpret this as evidence that within a positive reinforcement protocol a variable reinforcement schedule is better than a fixed schedule, which would make the food temporally unpredictable. Then, to make a unique reliable signal a bridge (sound marker for the correct response) should be reliable. The bridge should not be used when the animal is not to be rewarded but should only be used to predict the arrival of the primary reinforcement.. The use of an intermediate bridge should be used instead of taking away the predictive power of the bridge.
Principle number 7 is about feeding and so cannot be applied to negative reinforcement for training.

---

Control and predictability get intricately tied together in animal management and training. Control is so important to animals that to if they can control their world they will forgo their natural preferences to do so. Think about it then watch some good horse training. Hmmm.... you might not have noticed it before now, but it's there.

Reference: Bassett L. and H. Buchanan-Smith. 2007. Effects of predictability on the welfare of captive animals. Applied Animal Behavior Science 102: 223-245.

Thanks to Simon Mulholland for his suggestions to improve this review.