Sunday, August 15, 2010

Testing the Four Factors

For a beginner's guide on the four factors discussed in this post read this.

The four factors certainly work effectively to determine a team's offensive/defensive ability. In order to take a closer look, I decided to "build" an expected offensive efficiency for each team in 2009-2010 based solely on the four factors. The major problem was determining how to weight each factor. For the NBA, Dean Oliver recommended 40% shooting, 25% turnovers, 20% rebounding, and 15% free throws. However, these numbers are somewhat arbitrary. When using those weights, the correlation coefficient between Dean Oliver's weighted four factors and actual offensive efficiency was .94. I then decided to play around with the weights. I was able to raise the correlation coefficient to just barely over .99 by using the following weights:
  1. Shooting: 58%
  2. Turnovers: 23%
  3. Rebounding: 16%
  4. Foul Shots: 3%

(NOTE: The spreadsheet used must have been created slightly before the end of the '09-'10 season. Stats for individual teams are slightly off, but the overall study remains useful.)

Why this weighting system correlated so well to offensive efficiency is a bit of a mystery. Getting to the line is barely accounted for, but this is what worked. The next step was to turn my new four factors rating into an expected offensive efficiency. The results were good. Every team's expected offensive efficiency was within 4.7 points per possession of their actual offensive efficiency. Prairie View A&M was the most inaccurate projection (91.8 expected OEff vs. 87.1 actual OEff).

The Expected OEfficiency tends to underestimate highly efficient teams and vice versa for inefficient teams. Here is a look at the teams most undervalued and overvalued:

Note: Final column may appear off by a tenth due to rounding.

  • From top to bottom these teams ranked 1, 5, 3, 13, and 4 in unadjusted OEff.
  • All five teams shot two pointers, three pointers, and free throws well above average with the exception of Duke's well-documented two pointers.
  • As expected, excellent ball control is a common theme among these teams.

  • From top to bottom these teams ranked 343, 342, 341, 345, and 336 in unadjusted OEff.
  • Each team was 330 or below in eFG%.

Clearly, the four factors work. The weights that seemed to work the best are the interesting part. It would definitely be interesting to see other work on this topic.

1 comment:

  1. Hoopvision,

    Did you ever expand these stats to include more seasons? I'd love to see your results if you did.