Discover more from Growing Meta
An unfinished story of a statistical debate
Frequentist vs. Bayesian paradigms and how they impact the world
I stopped writing this newsletter throughout March to experiment with my mental health. It appeared that the effects of stopping were negative, as writing Growing Meta gives me a bulk of synaptic activity that burns the calories I gain from an unhealthy information diet. I am back once again, reading research papers and giving you the nuggets.
Here is an important statistical/philosophic debate that shapes our world: How best should we model the world — with what we believe to be true, or with what we are sure of?
It may seem like a confusing question, but this is the gist of the Bayesian/frequentist debate. As you well know, modeling the world affects almost every aspect of our lives and our economies, from pharmaceutical companies designing the best clinical trials, to weather agencies trying their best to give the most accurate predictions, to municipalities planning how their cities should grow.
When we model, we build knowledge from knowledge. Let’s take the problem of estimating how many people will book a hospital appointment and not show up for it in April. Frequentists will use only what they are sure of, i.e. they might infer it from the number of people who booked an appointment and the historical percentages of people who didn’t show up in past Aprils. They assume there is only one best answer. Bayesians, on the other hand, would utilize a different form of knowledge — belief , on top of data, and they would output a distribution of numbers that gives a ballpark estimate of that number.
For a very, very long time, the two schools of thought fought over what is the best way to model things to get an answer. It almost became a cult, if we were to define a cult as a system of epistemic devotion. At a certain period you were either a Bayesian or a Frequentist, and there was no middle ground.
The debate slowly grew to be more rational and reasonable towards the 21st century. Today, it is still alive but less dogmatic. The literature is ripe with ways to combine both paradigms in solving problems, such as improving the way drugs are being tested.
I am not sure as to what extent did this debate shape our world, but I do believe that the speed in which we converge to epistemic agreement** is positively related to our ability to build very effective, explainable models.
** i.e. when all arguments are addressed both philosophically and empirically
Growing Meta is a weekly letter tackling real-world topics from meta angles, such as complexity, analysis or the future. Subscribe to get more epistemic stories.