This is my first analytics-related blog post, so thank you to whoever may be reading this! With the NBA regular season right around the corner — it actually started before this was posted — I decided to try my hand at predicting total season wins for each team.
First, I want to shout out a fellow University of Tennessee MSBA student — Louis Pipkin — who posted his own NBA win projections for this season about two weeks ago. I read his blog, and it is what inspired me to undertake this, so thank you Louis! If you want to read his projections — and you definitely should — you can find them here.
Alright, now to get into why you’re all here — NBA win predictions. To start, I’ll give a brief overview of what variables I’ve used in my model. These variables include:
- The number of wins, winning %, and SRS each of the past four seasons
- A “winning consistency” metric
- The number of lottery picks for each team in each of the past four drafts
- Total coach experience and tenure with current team for each of the past four seasons
- The number of players with 5+ Win Shares in each of the past four seasons
- A “roster quality consistency” metric
For those who may not know, to effectively make predictions, you must make sure that all the data used to make those predictions would be available prior to making the prediction. For example, to make predictions for the 2018–19 season, I couldn’t use any information that wasn’t available prior to the start of the season.
That — along with Louis paving the way — is why I chose to look at the past four seasons. Giving the model a good estimate at the consistency of a team’s results, the consistency of a team’s roster talent, how many top-tier rookies a team brings in each season, and how stable a team’s head coaching position is allows the model to get a decent glimpse into the state of affairs for a given team. My hypothesis was that if a team is very consistent in results, coaching, roster, and drafting, then we should expect to see similar results — whether that’s good or bad.
The final variable I added into the model is something I’m calling the LeBronFactor™. The trademark may make it seem overly complex and mysterious, but it’s quite literally just whether a team had LeBron James on its roster or not. My reasoning behind including this is that LeBron is capable of lifting a team to heights that it shouldn’t really be able to reach. So, I felt that it provided useful insight.
Well, here are my model’s predictions. I feel I should point out that I don’t expect these to be particularly great — this is my first attempt at making these predictions, after all. Hopefully, throughout this season and the ones to come, I can continue to improve my predictions.
In addition, here is an overall conference/playoff breakdown:
A few things I find most interesting about these predictions:
- My model doesn’t seem to like the Pacers much this season. After looking around at some other projections, my prediction is definitely on the bearish side. The additions of Doug McDermott and Tyreke Evans could absolutely propel Indiana into a top-half playoff spot — but, I can also envision them missing the postseason altogether.
- Most of my predictions go under the Bovada over/under point — 22 out of 30 to be exact. I think this is indicative of how tough the league as a whole will be this year, but only time will tell if my win projections are too conservative.
- The Pacers aren’t alone in the “Wow, Adam’s model really doesn’t like us that much” Club™. Please welcome the second member — the Los Angeles Lakers! It’s true that they are still projected to make the playoffs, but just barely. LeBron is a force to be reckoned with, but this Lakers squad has some serious retooling to do before they’re ready to contend.
Before I end this entirely-too-long explanation, I would like to provide another graphic that I find interesting:
This table provides a quick look at each team’s predicted wins, Bovada’s over/under for each team, along with what my projections suggest will happen, and the payout if my projections end up being correct. I ordered them based on a level of confidence in my prediction that was a blend of statistics and intuition.
Well, I hoped you enjoyed my blog thing here. Hopefully I’ll be able to continue making these on interesting topics!
Data used in this analysis are courtesy of Basketball-Reference.com
Modeling was completed in R