With the offensive rating/usage rate scatterplot phenomenon rapidly spreading through the Big Ten blogsophere (OK, one other guy decided to make one), I was inspired to produced another scatterplot of my own. Here it is:
The scatterplot includes the top two scorers on every Big Ten team (except for IU’s Devan Dumes, whose offensive rating of 89.0 doesn’t show up on the scale I’ve been using).
I won’t comment extensively, except to note that the graph confirms the reason I like this approach: it identifies the best offensive players in a way that’s intuitive. I think most observers would say Robbie Hummel, Talor Battle, Kalin Lucas, DeShaun Sims, and Manny Harris have been the top offensive performers in the league this year. And those are the five players nearest the upper, right-hand corner on the scatterplot.
I apologize for how cluttered the graph is. Raymar Morgan’s datapoint is the one to the right of his name.
[...] Can’t stop scatterplotting. SpartansWeblog scatterplots the whole conference… Best offensive players: Robbie Hummel, Talor Battle, Kalin Lucas, DeSha(w)n Sims, and Manny Harris. [...]
I visit your site regularly and really appreciate your analysis!
But I think you are misinterpreting the results of your own method. The best players aren’t those nearest the upper-right corner, but those farthest from the lower left. Print out the scatterplot, pull out your protractor, and draw an arc at a distance from the lower left that leaves five players to the upper right of the arc. You’ll find that the best are Gatens, Hummel, Battle, Harris, and, nearly tied for fifth, Meachem and Lucas.
Perhaps I’ve oversimplified, but I’d stick by my statement that those five guys have been the five best players.
Meachem and Gatens have been very good players–but in a role player capacity. They’re very efficient at what they do (namely, shoot the three), but they aren’t the guys who make their team’s offenses go (hence the sub-20% usage rates).
At the other end of the arc you suggest, you give Westbrook and Turner credit for bearing a lot of the burden of their team’s offenses, but neither has been extremely efficient at converting the possessions they use into points.
You could probably give each quadrant a label: efficient role players, efficient stars, inefficient role players, inefficient stars.
I think kj has it right: in fact, if my understanding of the offensive rating and usage rate stats is correct, the product of the two gives a measure of how many points an individual contributes per 100 team possessions while on the floor. That lends itself to arcs drawn in the other direction – specifically, hyperbolas centered at the origin with the focal points at 45 degrees. This puts a really high emphasis on usage, though – Morgan comes out ahead of Lucas in this rating, and Westbrook ahead of both. There might be a useful way to combine these into a single number that accurately captures offensive value, but I’m not sure what it is.
Dan’s explanation is better than mine. It’s not quite as simple as the team offensive/defensive efficiency scatterplot.
I’ll have to think some more about how to combine the two numbers into a single stat.
KJ-
In sabermetrics, there is a statistic called RC27, or runs created per 27 outs, which measures how many runs a team would score if they had 9 of the same player. Is that the kind of thing you are looking for here?
Sort of, but that’s really what Offensive Rating is. You have to work in Usage Rate, too. In baseball, everyone’s usage rate is basically the same, right?
In some sense, it is. But, once could argue that a player batting 1st or 2nd in the line-up will have more opportunities to created runs, than a player batting 8th or 9th. Even one extra at bat every 2 games would be 81 extra at bats over the course of the season.
I wonder if the best measure might be something like (ORtg – 90) * (%Poss), somewhat analogous to baseball’s VORP (value over replacement player). Not many players below 90 get significant playing time in the major conferences (in the Big Ten, there are eight with 30% or higher minutes and below-90 ORtg, but five play for Indiana), so 90 might be considered “replacement level”. This formula would give the increase in team points per 100 possessions relative to having a player with ORtg 90 taking all of your possessions. The lower you set the replacement level, the more emphasis usage gets. (If replacement level is 90, then players with ORtg/usage 110/28% and 130/14% are equal; bump it down to 70, and a player with 28% usage only needs an ORtg of 100 to match a 130/14% player.)
Using that formula, the players you list as being top would have the following rating:
Battle 8.38
Harris 7.48
Hummel 7.43
Lucas 7.20
Sims 6.40
For comparison, a few other players from the scatterplot:
Gatens 7.03
Meachem 6.25
Tucker 6.02
Morgan 5.81
Moore 5.03
Westbrook 4.80
Turner 4.50
Coble 4.41
And other MSU players:
Suton 5.85
Allen 4.54
Thornton 3.50
Green 3.13
Summers 2.92
Roe 1.86
Gray 1.77
Walton 1.24
Lucious 0.25
Ibok (-1.05)
I should clarify, the 5.03 is for NW’s Moore. Purdue’s Moore is at 1.50.
I think you’re on to something, Dan. I think you could take that number and convert it to an absolute value–which might have more meaning for POTY discussions–using % of minutes played and an average pace number. You’d have something like “Pace-Adjusted Points Over Replacement Level per 40 Minutes.” I’ll try to work on this.