There is a better way of representing it... it's called adjusted +/-. Again, if you take an often-flawed statistic and start dividing it and multiplying it by various numbers you still get an often-flawed statistic that happens to be a different number. The flaw in +/- is that it doesn't take into account who was playing at that time with the player in question. Through regression, that flaw is reduced a bit. You can't necesarily reduce this flaw easily using linear methods like multiplying by something or dividing by something. Linear manipulation just means you're still not fixing the problem just manipulating it.
I know this is completely random, but does anyone still have the "chicken with its head cut off" play (picture) that Aaron Brooks ran last year?
Honestly I don't see alot of value added trying to come up with a modified +/-. Now if I was trying to decide who to spend millions on...
Commodore, if you're bored, I just found this. You can read this about how the guy over at countthebasket.com approached adjusted +/- calculations. He gives a walkthrough of the steps he took. He used Excel for loading the data and R for the regression analysis : http://www.countthebasket.com/blog/2008/06/01/calculating-adjusted-plus-minus/ It may be beyond the scope of anything you want to do, but you, or others may find it interesting.
Cool. I think it's very cool to judge the condition of player by clicking the 'play' button. Obviously, Landry and Lowry are red hot right now. I wonder if Adelman using the same chart. This chart shows playing Landry and Lowry to close the game is a pretty samrt decision. Landry and Lowry rising up to the top that indicates how good was our bench in recent run. Surprisingly Arize and Chuck gave most + as a starter. Once again defense may not be shown in numbers but +/- would tell us the differents.
Well, yeah. If your aim is to find a stat that estimates the complete bottom line of how a player helps a team (and I'm not necessarily a fan of that) adjusted +/- trumps anything right now. But, it isn't as simple, and I didn't think Commodore wanted to get into that. The +/- is pretty meaningless as it is given in the boxscore but if it is given as a ratio it would mean something - not as grand as adjusted +/- but a decent boxscore type stat. Manipulating the +/- does not solve any of the flaws you describe but it does solve the problem of being inflated with the amount of points scored while a player is on the court. By the way, this is a nice thread, one people might want to check throughout the season - could it become a sticky thread?
2 points that I thought people might be interested in. Q: What is the correct way to calculate adjusted +/- from regular +/-? A: I agree with durvasa that using minutes played is a reasonable guess to player quality, which would allow one to get adjusted +/-. However, a better way to do this requires a self-consistency calculation (your adjusted +/- is itself the best measure of player quality). This can be thought of as a regression problem, but there may be multiple solutions of player quality that give a good fit to the data. That issue imo makes the problem too difficult to solve for every player in the league simultaneously. You could probably simplify by assuming that the Rockets opponents average out and then solving the regression for just the Rockets. So then you only have to solve the regression for the Rockets roster (15 players), which could be doable. The interesting thing would be to compare the regression solution to durvasa's suggested approximation. That comparison would provide a good way of figuring out who is undervalued. Presumably this is a useful approach for evaluating other team's players for trades as well, as it shows when other teams are misallocating their minutes. Q2: Is there a single +/- stat that correctly accounts for minutes played? A2: Yes. The issue is that one should normalize not only for the average value, but also the variance. There is more variance when you play little, which is why players like Brian Cook screw up the stats. One way to account for both effects is to use a z-score. This is an introductory-level stats concept. Morey I am sure is sophisticated enough to use this or other similar measures.
Updated to include game 24 along with T-Mac's stats. According to the chart, if McGrady plays the entire game we will win by 72 pts! :grin: :grin:
i already gave the counterexample of finley, i'm sure you could find dozens of similar examples. i just think looking at per-minute +/- is a futile exercise.
Stats for game 27 vs. OkC added. What do you guys thing of changing some of the stats to rolling averages over the player's last 7 games or so? For example, Gallup and Rasmussen have a daily presidential approval tracking poll, and they use a 3 day rolling average. As opposed to averaging it over the entire presidential term (which wouldn't tell you much about current trends). Seems like if you took the players average over their past 7 games you would still be able to eliminate the volitility and at the same time get a better idea of their current performance and how they are or aren't improving.
Yes, but then it gets cluttered with too many options. It would still move through time, just the average would no longer be over a season, just that game and the six before it.
