Sunday, August 10, 2008
One of my recent posts used a convoluted metric to try to predict who would be the best senior prospects for the 2009 WNBA Draft. One of my commenters suggested that the metric be applied to the 2008 WNBA draft class. This way, we could determine if the metric could isolate who would be the most successful players out of that group of college seniors (and in the case of Candace Parker, juniors.)
I limited my analysis to the first three rounds of the draft. This involves 43 players (Sacramento was given a bonus third round pick - 14 x 3 + 1 = 43). Here is who the Senior Prospects Metric (SPM?) predicted for the draft.
At least we have the top four draft picks - Parker, Fowles, Wiggins and Hornbuckle - in the first round. The metric also brings a lot of third round draft picks into the first round, including two - Lauren Ervin of Arkansas ans Marscilla Packer of Ohio State - who ended up not playing a minute of WNBA ball. Ervin's season ended in December or January of 2008 with an ACL tear if I'm correct, which would explain why no one has heard of her - her position in the metric is based on the 16 or so games she played with the Razorbacks.
Let's look at how these player ended up in the actual draft. Furthermore, let's see how well each of these players have done in the 2009 season so far.
The last sentence where we "see how well" players have done is a loaded one. By what standard are you going to decide that players did "well" and players did not do "well"? There are several metrics you could use to evaluate players. I decided to use the same metric that SPM uses, the Wins Score metric. I like Wins Score because I think it evaluates shooting better than Efficiency.
Furthermore, using Wins Score solves the problem of players like Ervin who for whatever reason never made it into the league. Since Wins Score is an additive metric as opposed to some sort of rate, I can give those players who never made it into the WNBA( a Wins Score = zero. It works for me. These players have no positive stats, and no negative stats. It all balances out and furthermore strokes my ego by telling me that I'm a better player than Matee Ajavon).
The new picture:
It looks like Candace Parker and Nicky Anosike were the big wins of the draft. Both the SPM and the actual draft put them in the same locations.
Now, we need to know how much "correlation" the SPM and the Draft have to eventual Wins Score. Correlation implies a linear relationship between two lists of numbers.
A "zero" in correlation implies that there is no relationship.
A "one" in correlation implies that there is a strict relationship in the positive direction. As one list of numbers gets larger, the other list gets larger.
A "minus one" in correlation implies that there is a strict relationship in the negative direction. As one list of numbers gets larger, the other list gets smaller.
What we're hoping for is a negative relationship between draft position and overall performance. As position increases (#1, #2, #3....) the Wins Score associated with these numbers should start big, then start to drop.
Here are the correlation results:
These meanings of correlation depend on a lot. Let's look at the first one: the correlation between the SPM Predicted Order and the Actual Draft Order. 0.1220 is a very weak correlation, fairly close to random. The relationship between what the metric predicts and the actual draft pick order have little to do with each other.
Looking at the correction between the SPM Predicted Order and the Wins Scores of the Draftees, we get -0.2528. This is at the border of a small and medium correlation, more on the small side. At least the correlation is negative, which is what we want -- if it were positive, we would know that the SPM had no value.
Now, let's see how the actual GMs of the WNBA do against Wins Score. We get a correlation of -0.5925, a medium-large correlation on the large side of the border. The negative correlation is expected - as draft position goes up, performance should drop.
On the other hand, the -0.5925 number could be misleading. Remember that the Wins Score metric is linear additive - it gives you points whenever you do something well and takes points away whenever you do something poorly. It's not a "rate", but a sum. If a player is more likely to produce points than have points removed, then Wins Score and time played will have a strong positive correlation.
It is definitely likely that early draft picks will get more playing time than late draft picks - the draft pick has to be justified, or the player has to "get experience" and is more likely to be left in despite a low rate of Wins Score point production than a lesser pick. In short, early draft picks get better chances to build their Wins Score, so some of that good correlation might simply be a function of playing time.
(* * *)
Am I happy? so far I am. My next goal is to build in all of the height caveats. Hollinger stated that players of certain heights should be able to do certain things, and have points removed for failure. However, Hollinger is working with male height, and working in a different kind of game. This little side project of mine, it seems, is ever expanding - but at least, it's fun for a number cruncher!