Instead of projecting the Celtics record, which I plan to do, I thought I would start by evaluating the Celtics roster overall. Projecting wins means projecting minutes, which is particularly tough on a team with turnover and a new coach.
The approach I took was to compare how the advanced metrics view the C's roster compared to a replacement player in the NBA. This gives a partial idea of how good the players look over the entire roster and gives a range of their expected value. Take a look at the brief explanation of the metrics used in the extended Grantland style footnote**, or go here for Evan Zamir's primer.
I used 'replacement player' rather than 'average player' because when I looked at the stats I found that most players have below average production. That's because the best players on most teams get more minutes, so the production numbers are skewed upward. Replacement player also gives one an idea of the value you could expect with a vet minimum type player. The level I chose is also at about the rating of players in the 25th percentile in the NBA.
All that said, any player with less than 4 wins over replacement has below average production.
So, below is each of the metrics ratings comparing playing that player playing starter level minutes, 36 minutes for every game, to a replacement level player win terms of added or subtracted wins and losses.
|Name||xRAPM Wins Compared to Replacement Player per 36||ASPM Wins Compared to Replacement Player per 36||Win Shares Compared to Replacement per 36||Wins Produced Compared to Replacement per 36||EZPM Wins Compared to Replacement player per 36||Average Wins Over Replmnt Player||Range|
A couple things come out here:
- Avery Bradley- The metrics don't like his game too much, in part because he didn't play very well last year on offense. xRAPM, which as a plus/minus stat, is probably the best at accounting for defense, likes him the best.
- Gerald Wallace- All the metrics like him. I think fans are under valuing him because he is over payed and his contract is too long. Over payed does not equal can't play. If he bounces back, he could be valuable to a contender at the deadline.
- Jared Sullinger- The metrics think he earned his starting minutes. He could be the most fun player to watch this year if the back is indeed healthy.
- Jeff Green doesn't do great or too bad. The metrics don't know which Green will show up any more than anyone else does.
- Crawford and Bogans- None of the metrics like them particularly, and they're rated below replacement level overall.
- Kris Humphries is the most controversial according to the metrics with a range of 10 wins. Win's Produced, which is a rebound heavy metric, likes Humphries a lot.
- Rajon Rondo- Not surprisingly, Rondo is the key to the season. Coach and locker room drama aside, Rondo's production is the key to being competitive for this team at all with a 10 win swing from Rondo to a replacement level point guard, which about what the team has as options.
- Faverani, Pressey, and Olynyk are based on a combo of the Wages of Wins projections based on their Euro or NCAA numbers and my selection of rookie 'comps' for other metrics. In other words, semi-educated guesses.
So, that's it there're talent on the team, but at this point one star and some potential.
ASPM- This a box score based stat developed by Daniel Meyers. I like this one, in part because it did the best in an independent study in terms of both consistency and predicting, though only narrowly. It's also pretty intuitive. I used two years of ASPM ratings weighted by minutes and giving more weight to last year than the prior year. I also added an aging factor.
Win Shares- Box score stat developed by Justin Kubatko based Dean Oliver's 'four factors'. I used two years of ratings and an aging factor.
Wins Produced- Box score stat developed by Dave Berri that loves rebounds. I used a projection by Arturo Galleti at Wages of Wins, using three years of ratings weighted by minutes and an aging factor.
EZPM- Evan Zamir's Play by Play data based stat. This one is sort of complicated, but it's the only one using play by play data. I only had one year of data and applied an aging adjustment.