Discussed on Blog-a-Bull and Bulls Confidential There's a lot of weirdness in even adjusted plus minus ratings. I think it’s very hard to draw meaningful insights from the numbers without making a lot of assumptions about what’s behind them. For example, why does Kobe look worse than Lamar Odom and Pau Gasol? Or why does Kirk look so good? If you really think out the details of how APM adjusts, it’s intuitive that it’s not going to be able to adjust away a situation like that. Salmons, I’d argue, is one of the cases where you can see a clear difference in a player with simply raw plus/minus because there are large enough samples holding the rest of his teammates constant. If you look on 82 games, look at the Kings last year Udrih/Martin/Garcia/Thompson/Hawes played 126 minutes and was -8 Udrih/Martin/Salmons/Thompson/Hawes played 110 minutes and was -70 The year before Udrih/Martin/Artest/Moore/Miller played 376 minutes and was +10 Udrih/Salmons/Artest/Moore/Miller played 202 minutes and was -12 And in 06-07 Bibby/Martin/Artest/Thomas/Miller played 393 minutes and was +22 Bibby/Martin/Salmons/Thomas/Miller played 140 minutes and was -31 Bibble/Salmons/Artest/Thomas/SAR played played 46 minutes and was -5 Bibby/Martin/Artest/Thomas/SAR played 237 minutes and was -17 Bibby/Martin/Salmons/Thomas/SAR played 110 minutes and was -37 In all of these cases, the same lineup performed quite a bit better by making one change. Salmons out and Martin, Artest or Garcia in. Now obviously, I think Martin and Artest are better players than Salmons, but this isn’t really a defense of Salmons. I think Gordon is a better player than Salmons too, and this is further evidence that Salmons won’t be an adequate replacement. For projecting the Bulls: 1. I think Kirk’s excellent number probably reflects in part the fact he missed the time of the year when the rest of the team was struggling the most, and his return was largely coincident with the trade and Deng’s brief stretch of solid play. I can buy that Kirk has a positive APM, but especially his defensive APM really looks like an outlier and despite the fact that the A in APM tries to compensate for this sort of stuff, it ain’t perfect and there are obvious issues here we know APM doesn’t wholly compensate for. 2. Building off my stats from below, the only remaining Bulls that rated out as “above the average for their position” for guys playing big minutes were… Miller and Noah. This tells me that absent an absolutely quantum leap from Rose, we’re going to struggle quite a lot offensively. At the meaningful positions, we were below average across the board offensively aside from Gordon. Rose should be a positive this year, with improvement, but that'd still just get us back to one-dimensional and below average. I also like to look at things in terms of what players look like by position. For example, what does the average player’s APM look like, by position. So I took the spreadshoot this stuff was based on and analyzed it. Here is APM by position, for players with >1000 minutes and players with >2000 minutes > 1000 minutes Code: Pos Count Min O.APM D.APM APM G 70 2,016 0.99 (1.12) (0.13) W 55 1,831 0.43 (0.52) (0.09) F 74 1,888 (0.05) 0.77 0.72 C 50 1,739 (1.44) 1.48 0.04 >2000 minutes Code: Pos Count Min O.APM D.APM APM G 42 2,519 1.96 (0.82) 1.13 W 24 2,589 1.11 (0.61) 0.50 F 39 2,458 0.62 0.52 1.14 C 22 2,396 (0.35) 1.43 1.08 The positions – the way I think about them, are sort of a shorthand for defensive positions. I acknowledge up front it’s not a precise way of doing things, but no system really is. G- Guard. Most point guards and smaller shooting guards (Rose, Gordon, Hinrich, Wade) W- Swingman. A guy like Salmons or Iggy, who could be considered a big guard or a small forward. F- A bigger forward who might play either forward position or both (Deng, Jamison, Millsap), but who you probably don’t want matching up against either guards or centers on a regular basis. C- A guy who you feel can guard the average center (with this I’d include guys like Aldridge, who at 6’11 and 250lbs or whatever, is centerish, even if his position is usually PF). Anyway, the really clear trends everyone should take away from this are that, as players get “bigger” they contribute less offensively and more defensively. Even the better guards in the league “contribute” based on their offensive effects, not their defensive effects, which are, on average, negative. This more or less goes along with how I think about basketball (you build your defense from the inside out), but it’s interesting that the trends show up that way.
The key to the season is going to be Rose's defense and the defense of whoever is playing next to him. The offense isn't going to struggle as much as the team might struggle; there is plenty of offensive talent there. Rose is going to be better offensively, but if the backcourt is a sieve again, it won't matter. I really hope Salmons isn't that bad. He was hurt for the last dozen games or so, but those numbers are awful. And Kirk's numbers aren't misleading. He's a good defender who facilitates other player's offense. The team played better when he ran the offense. Just because he didn't do the scoring doesn't mean he didn't have a positive impact.
Code: Player O Minutes OAPM DAPM Kidd, Jason 2,814 3.31 3.35 Paul, Chris 2,888 9.36 3.35 Wade, Dwyane 3,048 10.66 2.96 West, Delonte 2,152 -0.39 2.92 Westbrk, Russ 2,668 1.03 2.41 Hinrich, Kirk B 1,338 1.12 2.38 Fernandez, Rudy B 1,957 3.14 2.06 Allen, Tony 788 -5.14 1.98 Gibson, Daniel 1,794 1.60 1.66 Fisher, Derek 2,320 -3.25 1.62 Murry, Ro (Flip) 1,964 1.32 1.43 Here's an ordered list of guards with the highest DAPM. Does that inspire confidence in the stat?
I really don't know what the hell that guy is doing....It's far from clear he knows either. This adds up to 50%. So maybe it's actually given double the weight. I don't think of this as a 6 year average by any streatch. Seems like last year got 16x the weight as 6 years ago.
Blogabull says "(70% weighted towards this past season and weighted greater for the most recent years of the remaining 30%)." That's dumb as hell. With all the noise in +/- and adj. +/- why bother to take years of data but completely discount everything but last year.
Um, yeah. Kidd is known for his defense, and he doesn't guard pgs anymore. West and Gibson were on a team that stressed defense. Fisher's job is to play defense and hit the occaisional shot, in that order. Fernandez and Westbrook were kind of surprising. But none of the rest really were. I don't see how you can both knock a stat and claim that it backs up your theory almost in the same breath.
Really, you could take the 5 guys after hinrich and no one in the world would say "Oh, i suspect those are the 2nd half of the top 10 PG defenders"
You don't understand how a fact can be meaningful in one context but not another? Like. Suppose you take a sample of 2000 people. Measure their heights. Does it tell you anything about height in general? Sure. Can it give you a meaningful estimate of the average height of a man and a woman. Sure. Can it tell you how tall a particular person is? Nope. Picture adjusted plus-minus as something like measuring peoples' height, except you can only measure people's height 5 people at a time. And you're measuring it at five random points during their life. And unlike height, you're measuring something that's variable based individual effort and the abilities of the other four guys you're thrown together with.
I don't think you really get the point of APM. APM was calculated to address all of those concerns. What you've described is an accurate criticism of regular P/M. And there's nothing random about the measurements. In fact they are weighted to reflect the most recent time period. Think of it as trying to figure out one guys height by comparing the averages of a collection of five guys with and without him over as large a sample of collections as you can get. The large sample size addresses the issues of whether or not a guy was giving maximum effort on a given night. It's obviously not a perfect way to do it, but it's a lot closer to the truth than you seem to understand.
I understand the point of APM perfectly well- I'm pretty well versed in doing regression analysis - but the fact that it addresses those concerns doesn't mean it perfectly addresses those concerns. Consider the idea of weighting to reflect the most recent time period. It's sensible enough, but there's very little research to indicate what the exact weighting should be. And changing it would make a significant different in the results. That is, to some extent these methods build very precise looking numbers and use lots of data, but on crucial and somewhat arbitrary assumptions. The sample sizes, in fact, aren't all that large when one looks at the high covariances one would expect in a team situation. In english, adjustments can only be made given enough relevant data. There's a problem the other way too. A person's height stays the same, but a player's productive ability doesn't just depend on his effort, but the efforts and abilities of his teammates. Which is why, for example, Kevin Durant looks like an absolutely dreadful player and while I don't think Salmons is all that great, he's not as bad as this makes him look. How "close to the truth" do you think it is? How "close to the truth" do I understand it to be? What sort of scale is involved? Miles? Inches? Percentages of an unknown unit? Does it only speak truth about Kirk Hinrich, but tell damned dirty lies about Kevin Durant and John Salmons? These are the things inquiring minds want to know.
I'm not sure what you're hung up on. The numbers pass the sniff test across the board, even for a guy like Durant. I like Ilardi's approach and the "crucial and arbitrary assumptions" are pretty sound.
The numbers are as close arbitrary than they are to "the truth". In the top 35, you have guys like: Young, Thaddeus Johnson, Amir Jamison, Antawn Cardinal, Brian Foster, Jeff Fernandez, Rudy Moon, Jamario And way down the list, you have guys like Deron Williams Dwight Howard Butler, Caron West, David Rondo, Rajon You may need to get your nose checked.
Guys in top 25% G Sessions, Gibson G Rudy. F, Foy F Young, Moon F Odom, Bonner C Foster, Przybilla VS Guys not in top 25% G Rose, Calderon G B. Gordon, Rip F Durant, T. Prince F Amare, Jefferson C D. Howard, Okafor
There is so much noise in +/-. There are different ways to derive a adj +/-. It looks like based on the raw numbers that Rashad Lewis was just slightly better with Howard last year and both were quite good. It sure looks like this guy's dervivative gives Lewis a big chunk of the credit and penalizes Howard. Per my earlier posts, the shortchanging of the prior years seems totally arbitrary. Back b/f he joined ESPN, when Hollinger seemed to add odd tweaks to his formula (and i can't think of any this random), he always showed what the order looked like prior to the tweak and then after the tweak. I'm really curious what the numbers would look like if you gave roughly equal weight to the last two years. Looking at the raw numbers, I'm thinking Howard moves up quite a bit in the order.
The "logic" for short-changing the prior years is to give an idea of last year's performance and use prior years as "noise reduction". Statistically that doesn't make a whole lot of sense. An equally weighted data set would give a more accurate result but you'd have to say it was APM for the entire period. By the logic guys use to do this, you could go back and create an "improved" 2007-2008 APM by including data from 2008-2009. Pretty crazy, huh?
