In this pandemic, a lot has been said and written about models – their usefulness, their limitations, their inaccuracy and more. If nothing else, we have become familiar with epidemiological models – at least, the ‘curve’!!

The most famous model being used today is the COVID-19 Projections developed at the Institute for Health Metrics and Modeling at the University of Washington. Public health experts have disputed its forecasts, even claiming it is purely statistical and has not epidemiological basis. Other models have been developed at the Northeastern University, Los Alamos National Lab, MIT, Imperial College London, Columbia University. These models use different techniques and assumptions, leading to differing projections about the trajectory of the pandemic. Some of them have been consolidated at the University of Massachusetts Amherst – one can clearly see the varied projections.

At the same time, these are being used by policymakers in making decisions and by others to criticize those very decisions because of the disparity in these models. For example, the New York State, these models predicted, would need much more ICU beds and ventilators than it has actually turned out to need. Does it mean that the models were wrong? Or has the social distancing policies guided by those models worked such that less of these medical equipment are needed?

We don’t know yet.

So, what is a model?

A model is, by design, a fragment of reality – developed to study specific features of it. It explains some specific aspects of the world, not the whole. It has assumptions built into it, that are reasonable in some contexts, not so in others.That is why we have many different models, even for the same narrow issue we try to understand. Features of the system that are important today maybe redundant tomorrow. What is crucial for some people may not be so for others.

Another reason we have so many models is there is a healthy disagreement among those who develop these models. So each model has its strengths and weaknesses.

The question then is, how should judge at a model?

An initial question to ask is: What is the model trying to explain? Looking for an answer to a question that is not the focus of the model defeats the purpose, doesn’t it?

A second step might be to see if the assumptions built into the model are reasonable for the question it is trying to answer. Does it explain the reasoning behind those assumptions? Are those assumptions intuitive? Do they align with what is already known about the world?

Next, we should look at the specification of the model itself. Given the question it is trying to address and the assumption it makes, are the elements of the model logically plausible? Does it explain the intuition behind each new step? Does the explanations it provides make sense intuitively to someone with some idea of the subject at hand?

Finally, does the model fit the data reasonably? To what extent? Does it address reasonably well, the limited question it was trying to address? Does it leave open possibilities to be tweaked to a different context? Can it be adapted as we get new data, wouldn’t that be fantastic?

So, what are we left with?

Hopefully, more understanding, both about the pandemic and the practice of modeling itself. We would be better off realizing that different models explain different things and hence, we should look at a multitude of them before we make up our mind.

Is uncertainty the only certainty? At least for now. In time, we will have a clearer picture. About this pandemic. Then, we will have a different uncertainty. Which is uncertain.