Predator-prey models to model users
Predator-prey models are helpful and are often used in environmental science because they allow researchers to both observe the dynamics of animal populations and make predictions as to how they will develop/ change over time.
I have been quiet as we have been unpacking an idea that with a specific data set, we can model user behaviour based on a dynamic competitive market. This Predator-prey method, when applied to understand why users are behaving in a certain way, opens up a lot of questions we don’t have answers to.
As a #CDO, we have to remain curious, and this is curious.
Using the example of the rabbit and the fox. We know that there is a lag between growth in a rabbit population and the increase in a fox population. The lag varies on each cycle, as does the peak and minimum of each animal. We know that there is a lag between minimal rabbits and minimal foxes, as foxes can find other food sources and rabbits die of other causes.
Some key observations.
The cycles, whilst they look similar, are very different because of externalities - and even over many time cycles where we end up with the same starting conditions, we get different outcomes. Starting at any point and using the data from a different cycle creates different results; it is not a perfect science even with the application, say Euler's method, or bayesian network models. Indeed we appear to have divergence and not convergence - between what we expect and what we see, even with the actual reality showing that over a long time, the numbers remain within certain boundaries.
Each case begins with a set of initial conditions at a certain point in the cycle that will produce different outcomes for the function of the population of rabbits and foxes over a long period (100 years) - or user behaviours.
This creates a problem, as the data and models look good in slide form, as we can fix one model into a box that makes everyone feel warm and fuzzy. With the same model and different starting parameters - the outcome does not marry the plan. Decision-making is not always easier with data!
As a CEO
How do you test that the model that is being presented flexes and provides sensitivity to changes and not wildly different outcomes? It is easy to frame a model to give the outcome we want.