Simulation as Learning
You may hear marketing scientists or data scientists talk about simulation.
Simulation, at its core, means the imitation of something real.
The purpose of a simulation is to understand a system or a model. A simulation enables the analyst to take something very complex, program it, and run it again and again hundreds, thousands, or even tens of billions of times.
A good simulation takes in many independent variables, and produces a single number that is meaningful to a human. (Strong recommendation to web analysts: resist the urge to produce multiple dependent variables.)
- Can be fed figures that are observed in nature.
- Can imitate a system.
- Can produce figures that can be compared to figures that are observed in nature.
If a simulation produces predictions that are generally in agreement with what is observed in nature, we can start to really pick apart the assumptions contained within the model. This process of aggressive inquiry and adjustment is learning.
“But a simulation isn’t real!”
You’re right! A simulation isn’t real.
But neither is a negative number. The fact that it isn’t real doesn’t mean it can’t be useful.
What’s the point?
This question represents a very major fault line.
You have to make up your own mind. Are simulations useful for a sub-set of problems that you’ll confront?
I find them useful in a handful of circumstances. It’s a hammer. And not every problem is a nail.
It’s my hope that we’ll have a rational discussion in analytics about the applicability of simulation.
I’m Christopher Berry.
I tweet about analytics @cjpberry
I write at christopherberry.ca