Methodological Review of Basketball Models – Part 1

by André Vizzoni

Introduction

The series beginning here will be my first foray into topics beyond my model for predicting sports results. Specifically, I plan to undertake a methodological review of a few well-known basketball models which have been influential in the analytics community. The review will examine the “how” and “why” of those models, and may level criticism at parts of them. As there are a number of subjects that must be broached prior to the review proper, this introductory post will be shorter than its successors.

Motivation

The motivation to do this post came from my experiences reading responses to this sort of work. Maybe Twitter (a place where the tone can be…adversarial, to put it mildly) has broken me, but the idea of reviewing anything leads immediately to concerns about people becoming angry and accusing me of being unfair (among other choice adjectives). I am not one to let any of that keep me from having and/or disclosing my opinions, even if they are to be controversial. Nevertheless, I do have to admit that I feel the need to make sure all my bases are covered.

That happens to tie in quite nicely with the goal of this series. In this series, I will don my “I am an impartial judge of methodological quality” hat. I will also be putting myself in a similar position to a “subject supposed to know”. For that reason, I should be as unbiased in my analyses as possible. To that end, it seems wise to admit to having multiple biases when it comes to basketball analysis. Such admissions will also serve the purpose of sating the “base-covering” desires generated by my self-preservation instinct.

Brazilian Biases

Before getting into any reviews, I would like to talk a little about what shaped my basketball opinions. First, the fact that I am shaped by the fact that I am from Brazil, a country that loves basketball and whose populace is deeply proud of the fact that we have won two world titles. Moreover, Brazilians talk about the sport a bit differently than in the U.S. The players I heard the most about when growing up were Oscar Schmidt, Magic Paula, and Hortência. The way people venerated them affected my views of the sport. More to the point, Paula’s passing and Oscar’s three-point shooting influenced me toward the opinion that passing and three-point shooting are the two most important skills in basketball. (it didn’t hurt that Oscar played for my childhood team, Flamengo)

I got older and more invested in the fortunes of Flamengo in the national and continental tournaments in which they competed. In so doing, I learned about different Brazilian players, as well as the facets of basketball that they represented. Marcelinho Machado‘s three point shooting (he might be the most beloved player in Flamengo history), Alex Garcia‘s rebounding (I hated him so much, since he played for Flamengo’s rival, Brasília), Helinho‘s passing, and Nezinho‘s defensive play were my archetypes.

American Discourse

Reading and listening to Americans discussing basketball was a culture shock. At first, I did not understand the love for dunking, or for finishing at the rim in general. To me, shooting from three seemed easier and worth more points. I did not care one lick about rebounding, nor did I understand what was so interesting about post play (both offensive and defensive), among other things. There are other parts of the general American basketball discourse that I still do not comprehend, like the appeal of post fadeaways or America’s love for midrange jumpers and isolation plays. I actually used to pass up open mid-rangers on NBA video games because I did not want Steve Nash (my favorite player) to take any jump shots from inside the three point line. I avoided isolating any player on a defender in favor of keeping the ball moving by pass at all times.

Personal History

My introduction to American basketball and statistical analysis of the sport also seems to have an important impact on how I will discuss the models to be reviewed. Thus, I feel like a brief recounting of my personal history in this spaces is necessary, as well. I only began following NBA basketball (and not just about the performance of Nash and the Suns) in 2011. In 2012, I was in my first (and only) year of law school. I quite disliked my boring classes, and was looking for things to do other than actually pay attention. Consequently, I was on the lookout for a blog where I could read about advanced basketball stats with the same depth I found at ProFootballFocus. A Google search took me to Wages of Wins, and I loved what I found there.

The work done by Dr. David Berri on that site made me want to be an economist, as did Steve Levitt’s work in Freakonomics. Here is where the second culture shock came, the one that made me want to write about biases in the first place: it seemed to me that many reviewers not only did not support Dr. Berri’s Wins Produced model, but also looked down on those who did. Not only that, but the opponents frequently used statistics that seemed inherently feeble to my mind. For three to four years, I did not really venture outside of this comfort zone in basketball analytics.

Change in Perspective

Then, Arturo Galetti came on Danny Leroux’s Real GM podcast. I was finally forced to look outside of my bubble, which led me to Nylon Calculus and to Layne Vashro. Expanding my reading list combined with what I had learned during my Bachelor’s in Statistics led to substantive development in my views on analytics in general, and my views on basketball in particular.

Conclusion

Having satisfied the demands of intellectual honesty, I will move on to the actual business of reviewing models. In the forthcoming posts, I will dedicate time for each model. In some cases, I will only review one model. In others, the analysis of different models will be sufficiently similar to warrant a single post.

In every post, I will address the model’s strengths, what I took from the modeler’s explanation of what they intended to do with said model, what may be missing in the process of building or applying the model, an evaluation of the overall quality of the model, and an acknowledgement of what biases (if any) seem to influence the model. The second post of this series will be about the model I with which I am most comfortable: the Wins Produced model. Check back soon!