StatsDetective 2 | Predicting the 2024 NBA playoffs

In this episode, I put my statistical skills to the test, and attempt to predict the outcomes of NBA games before they happen. While I was not successful (spoiler alert) it was a great way to prove some of the theoretical concepts of statistical prediction and future pitfalls.

Prediction is not a new idea for the methods I use. Both logistic and linear regression are well suited to the task and have some built-in mechanisms by which to test and improve on predictive models. There is, however, a big difference between prediction in the sports world and prediction in the world of academia.

In a research article, you may see a logistic regression model validated by reports of predictive power, accuracy, predictive values, and other metrics.This is typically in the context of known physiological values that predict some medical outcome. For instance, if you know someone's blood pressure and age, you can predict whether or not they will have a heart attack in the next five years. In sports, it is not so easy. Mostly because in the blood pressure example, we just simply do not have that information. It is more like knowing someone's age, and weight, and then predicting their blood pressure tomorrow, but not today if that sounds like a weird example, it's because it is. Sports put is in a very interesting spot because we know certain things like past performances, but not others, like if Jayson Tatum is feeling hot tonight or if Luka Doncic has a sore foot. The fact is, I can tell you a lot about the outcome of a game if I know how well the 3-point percentage goes, but that is impossible before the game starts.