Predictive models open up a bunch of possibilities when it comes to implementing statistics in a basketball setting. We can use linear regression to predict continuous variables, like how many points a team will score, and logistic regression to predict categorical variables, like whether a player makes a shot. So far, we’ve only looked at how a model performs with data we trained it on; what we really want to know is how well our models perform on unseen data. We can do this through various forms of model validation.

With linear regression under our belt, we have a method to predict a numeric dependent variable. But what if the response isn’t numeric? If this is the case, you need to start playing around with classification. Logistic regression is one of the more basic forms of classification. It is used for binary predictions (i.e. two class responses like yes/no), predicting the probability of an observation being a certain class. We’ll explore logistic regression today by trying to predict whether a team wins or loses a game based on different box score statistics.

In the last tutorial, we went over simple linear regression with a continuous (numeric) predictor. We can expand on this idea by going over regression with categorical variables as well as regression with more than one variable. We’re going to be expanding on the same dataset we used in the last tutorial, so go back to that one if you want to go over how to scrape that again.

In this post, we’ll go over the basics of simple linear regression. Linear regression will be our first foray into predictive modeling. The models we’ve looked at in the past (t-tests, ANOVA) have focused solely on inference; although this is useful, regression will be a building block for a lot more use cases.

In the last post, I went over using t-tests to compare means between two groups. When we’re trying to compare more than two groups however, we move on to an analyis of variance test, also known as ANOVA. ANOVA tests follow the same general procedures as t-tests, but allows for the user to test for differences in the means of more than two groups.