Ok, so Dave’s recent post on this Season’s Win/Loss prediction got me thinking. The consensus in the post’s comments is that the Blazers will win significantly more games this year than last year (40-50). This view mostly points to the bench being filled with productive veterans (plus having a real starting center). But is there a statistical method to explain this viewpoint, putting aside subjective opinion (for a sec)?
I’m sure there are very considered and complicated ways in the Sports statistics field to do this. However, I don’t attend the Sloan Conference, and I haven’t seen this on Bedge or the internet generally.
So I’ve applied what I call "PERM". This is simply PER x Minutes Played. While not perfect, PER seems to be a solid way of comparing players. Also by multiplying by the minutes played you diminish the value of 10 minute players with unusually high PER and elevate the real superstars playing 40 minutes a game.
For my team wins prediction, I take the Opening Night roster and sum each individual’s prior season PERM (whether that was on the Blazers team or not) to get a team total.
Here's the PERM data for this season's roster.
And here is the team summary data dating back to 1983-84.
The PER and Minutes played data are all sourced from Basketball-Reference.com. It was a little tricky finding starting day rosters but I think I’ve got it about right. Lockout-shortened seasons have been pro-rata adjusted.
The outliers that really stand out are the 2011-12 and 2004-05 seasons. In 2004-05, key players missed major minutes, Mo Cheeks was fired and then Kevin Pritchard played his kiddie corps. In 2011-12, we all know the team didn’t gel, then Wallace and Camby were traded and Nate McMillan was fired. The team then brought out the tanks. Obviously you can’t rule out that possibility this season (especially with the uncertainty surrounding LMA’s situation).
Using the Power function to determine a line of best fit, the R-squared of 0.38 (i.e. 38% of the variation in wins is explained by the team’s collective prior season PERM) is not bad but not very good. However, this is weighed down by the aforementioned two seasons that the Blazers were set up to do well but then entered tank mode mid-season. Excluding these seasons gives a much better R-squared of 0.56. This is shown on the scatter chart with current season wins on the Y-Axis and the prior seasons Team PERM on the X-Axis.
Using the regression formula predicts 50 wins including all seasons and 52 wins excluding the two outlier seasons. 52 wins feels high but I guess that’s why we should always take these things with a grain of salt.
What are the limitations?
First, the system penalizes rookies – they are allocated a prior season PERM of 0. So while the Total PERM of last season’s team was probably harsh, I prefer using historical performance rather than finding a way to project individual player performances. And really, no team is going to register lots of wins with 5 rookies on the roster. And that’s how 2012-13 turned out.
Second, the system also penalizes injured players. But I feel this is appropriate. Would you rather a player with PER of 20 but only played 40 games the season prior versus a player with a PER of 18 that played the whole season? And a prior season injury suggests a chance of more lost time this season. The one exception I’ve made is when a player misses the entire season prior, I have allocated the PERM from two seasons before. It would have been inappropriate to assign Arvydas Sabonis a PERM of 0 after he sat out the entire 2001-02 season and then returned for the 2002-03 season.
Third, the system only accounts for the starting season roster. This is the only way to predict the upcoming season since we have no way of knowing what trades will take place throughout the year.
Fourth, there is no way of factoring in team chemistry since this is really intangible. But it obviously does matter.
Who has the best Individual PERM?
As an aside, I thought it would be interesting to note the top 10 Individual PERM seasons by a Blazer in the last 30 seasons.
No surprise here that Clyde is the clear standout. I was mildly surprised to see Kenny Anderson make the top ten.
Anyway, thank you for reading. I’m sure this analysis can be improved with your input. PER is weighted more to Offense than Defense so I could incorporate a defensive measure into the analysis. Certainly the sample size should be increased but the whole process is very time consuming.