Bucks Diary

Friday, September 26, 2008

How I came up with the Defensive Win Score metric


I keep using this "Defensive Win Score" statistic, I'm not sure if I ever laid out my justification for it. Nobody else uses it, so I guess the burden is on me to lay some foundation for it as a legitimate evaluation tool. I'll try to here. I will also highlight some of its weaknesses, and just let you take it or leave it after that. (I'm not, however, making a case for the Win Score or Wins Produced metric here. I think that's already been made. If you need it, turn to Professor Berri's website).

The Story in Slightly Rambling Form

Here's the story of how I stumbled on the concept of "Defensive Win Score". I owe it all to last season's Boston Celtics, and their unusual emphasis on defensive basketball.

For some reason, I was calculating Team Win Score averages (for the Bucks blog... I'm not a nerd!), in other words the cumulative Win Scores per 48 minutes for each NBA team, and something curious arose. I got to the Boston Celtics and punched up their average. I expected a mindblowing number. Didn't get it. Even though the Celtics had far and away the most wins in the NBA at the time, their team Win Score average was, comparatively, somewhere near the middle of the pack... very mediocre.

How could that be? That wasn't logical. Win Score above average correlated with wins, right? The "NBA Team Average" ought to simply be the sum total of the average Win Score at each of the five positions. And indeed it is. So why can't you determine the team's "Wins Produced" by taking the team's Win Score above average, just as you would a player? Its not logical.

If my brother's allowance is usually $5 a week, and he receives $10, and mine is usually $10 but I receive only $5, then my brother's account is five dollars over average, but our joint account is absolutely unchanged. The same should hold for a basketball team and its Win Score. You should be able to add up all of the player's results against average and that should tell you the number of wins the team produced. But it doesn't.

Here's an example. Take the aforementioned Celtics. Their collective Win Score average per 48 last season was 46.4. If you go to the "Season Summary" page on basketball-reference and calculate the NBA team Win Score average per 48 (using, obviously, the "league averages" line), you come up with the number 43.6. If you translate that into one collective player (dividing the difference by 5), then the average Celtic was only +0.5 Win Score points above average per 48 minutes. Pretty mediocre. In fact, that only translates into 46 wins. WTF!! The Celtics won 66 games.

Then I turned to the Phoenix Suns, the team with the top Win Score average (54.9). If you do the math, the Suns averaged +2.26 Win Score points per 48 per player, which translates into 69 wins produced by the team. Yet the Suns only won 55 games. So their total wasn't close to their Win Score production either. Something was amiss.

It got me thinking. Most of the teams whose wins produced missed badly were heavily weighted on one side of the court (the Celts) or the other (Phoenix's mediocre defense was bringing their offense down -- btw, the same phenomenon that caused the collapse of the George Karl Bucks earlier this decade). Somehow I got the idea: I needed to take into account Opponents Win Score. I did. And when I gave half credit to the team and half credit to its opposition, the formula works. I then made the rough conclusion that the former represented the team's "Offensive" efficiency and the latter represented the team's "Defensive" efficiency.


As you can see, if you give one half value to each team's Win Score above average per player, and one half value to -(Opponents Win Score average per player), and then plug that into Professor Berri's formula, you get a 92.6% accurate account of each team's Wins Produced (it would be even more accurate except for some reason the Heat were way, way off. According to Win Score, they should have won about 21 games... they won what 15? I suspect a tank job).

Thus, the Celtics sort of mediocre Win Score per player is augmented by the fact that their opponents, as a whole, averaged an incredible -2.76 Win Score below average per player, making the Celtics' "effective" Win Score per 48 per player average (-2.76 + 0.56/2) = +1.66. That's a big difference from +0.56.

+1.66 translates into 65.4 wins, much closer to their actual win total: 66. Defense mattered. It made it possible for the Celtics to win more games with a less efficient offense. As for Phoenix, if you factor in their oppositions Win Score, which is only slightly below average (41.0), and give that half weight, then the Suns win total comes 54, much closer to their actual total of 55 wins.

And on and on. Anyway, you can look at the chart. I have to wrap this discussion up, so let me cut to the chase.

Having established the legitimacy of defensive win score on a team level, as soon as I discovered 82games.com's "Opponent Counterpart" 48 minute production statistics, I used the same logic and method to apply defensive win score to individual players.

I think its a fairly legitimate measurement of defensive performance with several caveats:

1. I necessarily have to debit each player for things they may or may not have had control over (they don't guard the same player every second).

2. Certain statistics are double counted (a steal on offense is also a turnover created on defense).

3. Which brings up another weakness: the two categories are misnomers. What I call "Offensive Win Score" actually includes several traditional defensive statistics. But I'm keeping the misnomers because the preponderance of statistics on each side are accurately named.

4. Finally, defense in basketball is a really weird thing. Its value is not well understood, and thus it is not well compensated. Consequently, the motivation to play defense is a lot like the motivation instilled in a Marine -- you do it because you don't want to be the one to let your buddies down. But, that also means defense is dependent upon the faith. As soon as that breaks down, effort breaks down, and a disastrous domino effect takes place. What is the purpose of playing hard defense for a 31-51 team? What will you get in return for the effort? Thus, often times a player's defensive performance is situational and does not necessarily reflect his abilities.

Conclusion

With all that said, I am going to continue to track this statistic. Even with all of its weaknesses, I think it still captures the essence of defense: preventing your counterpart from playing efficient basketball. And the Boston Celtics proved last season that if enough players from the top of the roster to the bottom buy into the concept... look out!

2 Comments:

At September 28, 2008 at 6:16 PM, Blogger radar said...

brilliant blog, love this...the sports world should know about you!

 
At October 1, 2008 at 5:25 PM, Blogger TCW said...

Thank you Radar!

 

Post a Comment

Links to this post:

Create a Link

<< Home