This past spring, dominated by the coronavirus, has been full of statistics: tests, cases, patients in hospitals and numbers of deceased. Suddenly the whole media stream is only statistics! A statistician, however, gets excited only after we start to clarify what is behind the innumerable numerical values. What is the basic reproductive rate or total mortality rate? How many infected persons do we have in total if we have found 6,380 cases in the tests? Why do antibody studies give many different percentages and what on earth is an “erroneous positive”?
From the perspective of a statistician, it has been especially interesting to see how uncertainty has been discussed in this spring. Many think it has been talked about too much, but I claim it has been discussed far too little.
When simplifying and crystallising a message, it is easiest to talk less about uncertainty. Or is it so that the message is a bit more convincing when you don’t stray into long-winded explanations about the accuracy of information?
In the long run, truthful and precise communication of uncertainty would be beneficial. Then public officials or experts would not lose credibility if and when earlier information must be modified in the light of new data. But can you even talk about uncertainty to an average citizen? It may be difficult at the beginning, but what if we just start talking, persistently and precisely. At least keeping quiet does not improve the ability to read statistics. And aren’t we already used to the idea of the “margin of error”?
For many it may have been news this spring that statistics even still exists as a field of science. Why haven’t we heard about the new methods of machine learning during the crisis? Couldn’t the much-discussed artificial intelligence find the best way to solve the crisis?
Statistics is a living discipline, but methods that have been tested in practice and ironed out by theory have their own charm in the middle of a crisis. Machine learning has a lot of potential but the goals and application targets are often different than in statistics.
Before this spring, my knowledge of epidemiology was hardly passable. My own special field deals with basic research in the calculation methods of Bayes statistics, especially in relation to modern time series analysis. Plainly said, it is working for some so that imaginary adapters in the future can get their calculations done, preferably quickly and reliably.
When the pandemic started, I familiarized myself with basic epidemiological concepts and models. It was surprising how close to my own work the virus came. The Bayes models have been applied right from the start of the epidemic. And it seems that the epidemiological questions about progress refer to the modelling of time-dynamic phenomena. Hey, wait a minute – I can estimate the fluctuating basic reproductive rate directly with the calculation software I wrote!
Nevertheless, I did not peddle my analysis directly to media; there are much more experienced statisticians who are specialised in epidemiology in Finland. Instead, I asked the modelling group of the Finnish Institute for Health and Welfare (THL) if we could assist them in coronavirus modelling. Now the cooperation has started. Maybe our basic research will play some role in beating this epidemic.
Department of Mathematics and Statistics
Subscribe to the JYUnity newsletter
Get latest articles from The University of Jyväskylä’s stakeholder magazine into your email. You can cancel your subscription at any time.