During the past few months, we have heard a lot about “models” that predict what’s going to happen during the Covid-19 pandemic. We have all become familiar with phrases like “flattening the curve” and “will there be a second peak?” The models are designed to predict such things as the number of people that will be infected by the virus or the number of deaths from the virus. That is a somewhat gruesome business. But what are models and how are they made? Perhaps more importantly, how reliable are they?
For the last 20 years, I spent my professional life developing pharmacological models. These models help predict what happens to drugs in the body and how effective they are. Some of the models are useful, some are not. A brief look into the basics of modeling will help clarify this confusing field.
A model is typically a mathematical prediction that is based on information related to what you are trying to predict. Some models are very simple, some are very complex. They can be divided into two major types: “top down” and “bottom up” models.
Let us take a simple example to gain a better understanding of what is going on. By searching the internet, you can easily find a calculator that will predict how tall an infant will grow to be. How do they do that? One way is to collect the height of infants (say 2 years old) and check their height later (say when they are 18 years old). After gathering this clinical data, a mathematician then would create a relationship between a typical two-year-old’s height and a typical 18-year-old’s height. Wait a minute – what is a “typical” person? The more limited you make your model by defining what you mean by “typical” or breaking this down into different groups (for example, one model for males, another for females), the better the model. This is a top down model – using actual measured height data to generate a prediction tool.
On the other hand, another way you could predict how tall an infant will grow is to look at their genetic makeup, the environment they live in, what you expect their diet to be, and so forth. With the knowledge of which genes control height and how the environment and diet typically affect height enables one to make predictions of how tall someone will grow. The better you understand how this information affects height, the better the model. This is a bottom up model – using the factors that affect the phenomenon (in this case, height) to generate a prediction tool.
Anyone who has worked with models very long learns that the accuracy of models can vary greatly. It is a tricky business and creating good models typically takes a long time. Biological variability (we are not all the same!), unknown factors, and inaccurate data are just a few things that lead to poor models. In reality, models are an educated guess and might work even if they have nothing to do with the phenomenon (height in the examples above). This has led to the common aphorism that “all models are wrong, but some are useful” (first attributed to the statistician George Box).
Models can be useful, sometimes very useful, in making predictions. In our current situation, most of the models have been created quickly and changed frequently as more data is collected. There are many factors at play and honest experts in the field state the obvious – they are not sure what is going to happen. As much as possible, avoid stressing over models – but do not ignore advice from scientists and medical professionals! And never forget that the Lord is in control.