NCAA Basketball Kansas Basketball Analysis: Grading Self's Playing Time Management
NCAA Basketball Kansas Basketball Analysis: Grading Self's Playing Time Management
| |
| Analysis: Grading Self's Playing Time Management |
|
|
| Dec 6, 2005 | |
|
There has been a lot of discussion about Self's substitution patterns and use of young players vs veterans. So, I thought I'd take a crack at using numbers to look at the situation. (surprised?) Many of you know that I use boxscore statistics to create a numerical rating (SANZ) for each player that represents how well they contribute to the team's stats. So, I decided to see how well those ratings correlate to the average minutes per game for each player. Here's what I did: - Used two comparisons: Composite SANZ & SANZ using only Per Minute. My composite SANZ ratings (the ones you see normally) are weighted to use a player's per game and per minute performance. But I also wanted to see what it would look if the only thing that mattered was per minute performance. - Although all players' stats were used to create SANZ ratings, only those who average more than 5 minutes per game were used for the correlation part of it. - Correlation is expressed as a number between -1 and +1. If two sets of numbers correlate at +1, it means that as one increases, the other also increases ... in perfect tandem. If they correlate at -1, it means that as one increases, the other decreases ... opposite in perfect tandem. A correlation of 0 would imply there is absolutely no pattern between one number and the other. Thus, in this model I am correlating each player's rating with his minutes played per game. In theory, the better a player is, the more he should be playing. So, a correlation of +1 would indicate that Self is really playing the best players the most (or getting the most out of those who are playing the most). This isn't perfect, folks. But it's an attempt to examine the situation with some objectivity. So, here are the results, cumulative after each game so far this season. Cumulative thru Game # - Opponent Composite *** Per Minute Thru Game 1 - Idaho St. 0.71 *** 0.62 Thru Game 2 - Arizona 0.65 *** 0.49 Thru Game 3 - Arkansas 0.54 *** 0.36 Thru Game 4 - Chaminade 0.39 *** 0.13 Thru Game 5 - Nevada 0.44 *** 0.21 Thru Game 6 - W Illinois 0.34 *** 0.09 Thru Game 7 - St. Joseph's 0.37 *** 0.03 I was disappointed when I saw these results. Seems like the correlation has gotten worse as the season has progressed, whereas I'd have guessed Self would start getting to know the players and playing the right ones. Now, I haven't done this analysis before, and I haven't seen it done anywhere else. So, I wanted to benchmark this against some of the top coaches in the country. (Took me a couple of hours to do this, but it was fun.) I did the same exact analysis for the other teams, but only used data through their last game (didn't break it out by game). This is through games of Tues Dec 6. I've put in Kansas for comparison and ranked them all by average of composite SANZ and per minute only (not really a meaningful number, but the only way I could think of to use both correlations): Team Comp *** Per Min *** Avg Uconn 0.91 *** 0.73 *** 0.82 Okla St. 0.8 *** 0.63 *** 0.715 Arizona 0.84 *** 0.57 *** 0.705 Illinois 0.77 *** 0.54 *** 0.655 Syracuse 0.79 *** 0.47 *** 0.63 Duke 0.59 *** 0.39 *** 0.49 UNC 0.49 *** 0.14 *** 0.315 Kansas 0.37 *** 0.03 *** 0.2 Yep, we're dead last no matter which correlation I use. Though UNC and Duke have seemingly low correlations, you have to remember that Duke has won most of its games handily, allowing it to play weaker players for more of the game. And UNC should be allowed some leeway because of its youth. But KU's number (esp when only per minute is used) is disturbingly low. Why? What's wrong? Two hypotheses: 1) My SANZ ratings system could be flawed. Obviously, a boxscore can't capture the whole value of a player. Lots of players stuff the stat sheet but don't show up when it's important. So many actions aren't even captured on paper (screens, defense, leadership, boxing out, etc). I know my player ratings system isn't perfect, but for most of the teams above, it seems to correlate pretty well (especially my composite SANZ rating). If it were a system flaw, then all the teams should have low correlations. So, I don't think this is the likely culprit. 2) Jeff Hawkins. Let me explain. When I look at the KU SANZ ratings and their min/gm, the player that sticks out most is Hawkins. He's dead last in the ratings but plays the 2nd most minutes on the team. I'm sure every coach has a couple of players he believes in and plays a bit more than their skills would merit, but according to my numbers above, they don't do it nearly as much as Self does. And I think Hawkins is his little treasure. When I remove JHawk's SANZ rating and mpg from the correlation analysis, KU's numbers jump considerably: Correlation between KU player SANZ ratings and minutes per game through SJU game: Using Composite Rating: With Hawkins (0.37) -- Without Hawkins (0.66) Using Per Min Only Rating: With Hawkins (0.03) -- Without Hawkins (0.32) So, without the unusual situation with Hawkins, KU's correlation numbers look similar to that of other top coaches, and would look good especially in light of the fact that it's a young team that Self needs to figure out. Bottom Line: I believe Self is doing a good job of playing-time management, with the strong exception of Jeff Hawkins. Based on boxscore stats and what I've seen personally on TV, I think Self should limit JHawk's playing time considerably. |
| |