Behavioural modeling

There you are, looking for a job, and one of the requirements is "experience in behavioural modeling".  What does that actually mean? 

The first thing to determine is what are behaviours.  Behaviours, or behaviors for the Americans among us, are easily understood but harder to define. Purely based on the linguistic definition, Oxford Languages says they are "the way in which one acts or conducts oneself, especially towards others" but also "the way in which an animal or person behaves in response to a particular situation or stimulus." Lee Alan Dugatkin agrees with the latter, formulating that "behavior is the coordinated responses of whole living organisms to internal and/or external stimuli." A behavior happens where there is a need "If an animal has a need, its motivational state is affected so that behavioral and physiological responses that should result in remedying that need can be made." (D. M. Broom).

It seems to me that behaviors are coordinated movements and acts, directed to addressing a need. Some behaviors are learned (styles of playing, replies to human commands), and some are inherited (searching for a teat after birth). 

Now that we know about behaviors, let's see what are models? Modeling means to build models, abstract representations of complex truths. A model can be a simplification, and can fit to all instances of a repeating concept. A model can be used to compare representations across species or in time.  Schank, Joshi and May write that "Models can be physical, symbolic, mathematical, or computational, but they are always simpler than the animal systems they represent." Clearly then there are several methods of modelling, and the best fit can be chosen based on the goal of the research. If we intend to use behavioral data in a computer system, a mathematical or computational description would seem the most suitable. If we intend to demonstrate the behavior to others or even mimic it, physical modelling is the way to go.  

Schank et. al. also give models nine dimensions:

1. Realism
2. Detail
3. Generality
4. Match
5. Precision: how quantitatively precise a model is in it's predictions  
6. Tractability: how analyzable or manipulable a model is
7. Integration: Can it be used together with other models?
8. Level: Cellular, population, strain...
9. Medium

Let's take a look at some models to better understand how they really could look like.  The first example is a hidden Markov model (HMM) from the publication of Leos-Barajas, Gangloff, Adam et. al (2017).  This is a mathematical model. It looks complicated, and will require understanding about the mathematical notation and statistics to understand and to apply. 

Another beautiful example is from Ellen Evers from the University of Utrecht. In her presentation she explains the differences of Agent-based models (ABM) and Ordinary differential equations (ODE). 

An ABM looks at individuals: how each individual moves and acts during the behavior. It is less effective in modeling the group. An ODE looks at a group and determines the behavior of the group as a whole, losing sight of each individual. Her example uses the balance of wolves and sheep. Sheep are increased by births, and decreased by predation by wolves. Wolves are increased by predation, and decreased by natural death. 

In ODE terms the  models are

Sheep = + birth*Sheep – pred.*Sheep*Wolves  

Wolves = + pred.*Sheep*Wolves – death*Wolves

But for ABM the rules are for each individual:

SHEEP: If I meet wolve: die!

With chance = birthrate: Reproduce!

WOLVES: If I meet sheep: energy +1

If energy (from sheep) = 0: die!

With chance = birthrate: Reproduce

The big difference is that ODE model is deterministic. It can be used to predict situations when some variables are known. It could change what happens when there are 100 wolves and 50 sheep in comparison to a situation with 100 sheep and 50 wolves (in both cases, it's not looking good for the sheep.) ODE is not spatial (measuring things in terms of movements in space) and requires homogenity. ABM, on the other hand, just models the situation. It is applicable to heterogenous populations. 

I hope that this short introduction has given you some insight into modeling animal behavior. It is a complex field, and the lesson from Ellen Evers is heartily recommended as a good starting point for understand the differences of ABM and ODE.

More information

Lee Alan Dugatkin: What is "behavior", anyway?

D. M. Broom Animal welfare: concepts and measurement

Schank, Joshi, May et. al. Multi-modeling approach

Leos-Barajas, Gangloff, Adam et. al: Multi-scale modeling of animal movement and general behavior data using Hidden Markov models with hierarchical structures

Ellen Evers: ABM and ODE

Types of animal welfare studies

 When talking about animal experiments, it's easy to start thinking along the lines of mice, needles and electric shocks. But there are naturally a lot of other kinds of experiments as well. Consider a simple case where an animal is offered two kinds of food to see which one they choose first. That is also research on animals!

The study of animal welfare is a very complex field. Just to start with: what IS welfare and how to measure it when the animal itself cannot talk or describe it's emotional states? Obviously a complicated set of research methodologies is needed to achieve robust results. The measurements taken to achieve the results can be roughly divided into two: physiological and behavioral.

Physiological	Behavioral
Temperature Heart rate Blood glucose Fat percentage Weight Bacterial count	Time spent sleeping Frequency of a specific behavior Posture changes Social contacts Time taken to approach objects Distance traveled to forage

As you can see, welfare can actually be quite specific!

Next, let's have a look at some types of studies which are conducted to find out more about animal welfare. All of the measurements listed above are just data, but it's of course important to know how to gather the data to make sure - you guessed it - that the data is correct. So what kind of experiments are used?

Preference studies

Preference and motivation studies are both examples of empirical study, where the researcher gathers data on the target of their study. In preference studies the animals choose what they prefer and what factors influence the preference.

An example of preference study is to offer chickens feed with and without NSAIDs (painkillers). Chickens with healthy feet eat more of the feed without medication, while chickens with leg sores and ulcerations prefer the feed with the medication (self-medication). Factors affecting the preferences can be available time, diet, health and pregnancy.  

(c) Louse Buckley, UFAW

Motivation studies

Motivation study measures the motivation or willingness of an animal to reach a specific goal. The aim is to find out the value of things  and what impacts the motivation. For example, an animal might be very motivated to get a treat when alone, but less motivated when they're in a group and know they might need to share their treat. Measurements can be speed and distance traveled to the goal,  latency before the animal tries to reach the goal and energy spent to reach the goal.

In classic examples animal can be trained to press a lever to get a specific treat. Then the treat is disabled, and the amount of times the animal presses the lever trying to get the treat is the "price" they're willing to pay for it. Things like temperature, eating, health and group size can all affect the value of the reward. 

Metastudy

Metastudy is different from the other types, because this is the study of studies. In metastudies the researchers gather a vast amount of previous researches, carefully select the most relevant ones and summarize their findings. The process of selection and omission must naturally be well documented and argumented. The aim is to see wider trends and generalizations. Commonly seen phrases like "many studies prove" or "there is little scientific base on the claim that" arise from metastudies. Metastudy goes further than just being a literature review. It uses for example statistical methods to compare the results of the studies once comparativeness is established.   

More information:

Animal welfare science - Wikipedia

Consumer demand tests (animals) - Wikipedia