Highly Plausible NBA Win Projections 2015-16
For the third year in a row I am rolling out my pre-season win predictions for each team in the NBA. It’s a fun and useful exercise as it forces me to look in depth at each roster and answer burning questions like what exactly happened to Maurice Harkless.[1. Went to Portland it seems.] It also poses a number of analytic challenges such as projecting expected player ability, playing time, aging curves and fit issues, as I will discuss below.
Last year my predictions, using a blend of my Player Tracking Plus Minus (PT-PM) and a Regularized Adjusted Plus Minus (RAPM)[2. Via Jerry Engleman at APBR] player metrics, finished at near the top of the APBR metric predictions, whether by luck or by skill. So this year I mostly repeated the same process, relying of the same two metrics. One possible advantage this year is that I now have two years of the PT-PM metric to measure players’ ability, though my research indicated that the more recent year should be weighed more heavily. The raw scores were then adjusted based time played and the age of the player with the expected peak performance at age 27.
Rookies coming out of the NCAA game are rated based on Kevin Ferrigan’s RPM rookie model, while European rookies are rated based on their draft slot. Unknown veterans coming into the NBA are rated based on a general replacement level player rating of -2.
In terms of the actual win projections, if you’re a championship or bust fan, it looks like it’s Golden State’s league to lose, with the rest of the teams are just filling out the schedule on NBATV. In fact both of last year’s conference champs look like significant favorites in their divisions with only the San Antonio Spurs in a position to challenge them for the title.
The table below shows the wins projected for each team, last year’s record and the projected change from last season to this year.
As the far right column shows, beneath the three true contenders I am projecting a good deal of movement. The Timberwolves’ MVP this year just might be Reversion T. Mean[3. Killer crossover on that guy…] along with the effects of the natural aging curve, as they are projected to gain the most wins after their league worst record last year. After the Wolves, the Celtics and Hornets are projected with the biggest gains followed by the Knicks and a hopefully healthy Oklahoma City Thunder.
On the loss side, no surprise that Portland is projected to fall off the most by some distance, followed by the Nets and the Hawks, who will miss DeMarre Carol and are expected to suffer the downside of reversion towards the mean.
Playing time, fit, and To Do Lists
After estimating player value, the next major challenge is to distribute projected playing time. In the offseason I looked back at my projections fRon last year and found modest improvements in my predictions correcting for playing time. There are a mixture of models that provide estimates useful as starting guides, including one I developed for veterans that weighs prior playing time, age and player performance measured by box score statistics and another for rookies based on draft position.[3. The rookie formula adjusts based on the log of the slot drafted, 2440 + -1195 * log10(draft slot), so the number of minutes expected to play drops fairly quickly in the top spots then more slowly between later picks:
]
Player movement alone guarantees that none of the minutes played models are in and of themselves reliable enough to estimate playing time without adjustment. The Trailblazers, for example, lost four of their five starters from last year adding a host of new players that are expected to take on much bigger roles than their prior numbers would indicate, so some subjective adjustment is needed. Part of that adjustment involves positional balance; making sure each of the approximate positions add up to a fifth of the team’s playing time on the court. Balancing position time is an effort to approximate a minimum competency that most coaches would replicate.
The estimated team Plus/Minus is then run through the team schedule to estimate their Pythagorean odds of winning each game.
Other than balancing the cardinal positions, to date I have mostly punted on fit issues. There are a few generally accepted fit adjustments: spacing with three point shooters, rim protection, issues with low usage line ups and the diminishing returns on additional rebounders. But, as challenging as it is to predict individual playing time, projecting line ups is even more problematic. Last year, for example, only five line ups played over 500 minutes together over the entire season according to Basketball Reference. Undoubtedly, there is possible value there, but for now I am leaving is on my to do list for next year.