On modelling the world

When trying to understand the world, we apply models to it, in order to quantify what we see. Different models are used to look at different things. A crucial point is understanding what models to apply at what resolution. A model which is well suited to say something about the population of a country may break down when increasing the resolution to looking at a city, or it may still hold true, but break down when trying to apply it to residents in a street, the inhabitants of a building, or individual men and women.

One example of this is the much used, and much debated, body mass index – BMI for short. When using BMI, it is important that we understand both what it is – a tool for statistical modelling – and isn’t – a universally applicable indicator of personal health. While increasing BMI numbers on a population level is certainly cause for concern, when applying it to individuals, the model breaks down.

I know people who, according to BMI, are obese, but whose amount of body fat is so low that their GPs are alarmed, and would like to see the ratio of body fat increase. How did the model break down? It didn’t take into account individual differences, and did not compensate for muscle mass (keep in mind; muscles weigh considerably more than fat). Even so, BMI is used on a daily basis to bracket people. The simplest example is insurance companies, who charge people of high BMI more for insurance than they do people of normal BMI.

They would counter that they need to predict their risk somehow, and that BMI, while imperfect, gives them a good starting place. That starting out point is, however, utterly worthless if it’s also where the evaluation ends. You might find yourself wondering what my point is. It is simply this: when applying a model, you must justify the application of the model, and you must make this justification transparent. That way, the application of the model may be challenged, and better models applied, as appropriate.