Here's what I did... I pulled in all big minute lineups (that they calculate out the complicated stats for) at BasketballValue.com. If you look through the numbers, you see interesting stuff, but the numbers come in small bites. Most of the lineups have only played together for a few minutes. So you've got small sample sizes that make drawing conclusions dangerous. For example, look at the Rose/Hinrich/Deng/Tyrus/Noah lineup. Looks pretty good, right? In 60 minutes, it outscored the opposition 127-107. Rose/Gordon/Deng/Tyrus/Noah would appear an inferior lineup for example. I copied the main lineups (all of them over 20 minutes- the cutoff, I guess, for the adjusted +/-) into Excel and started adding things up. They played almost 320 minutes, and had a net +/- of -7.2 per 48 minutes. By contrast, the Kirk lineup was 15.91 per 48 minutes. So a clear cut “win” for the Kirk lineup, right? Not so much to my mind. * First, the sample size is extremely small. * Second, we know that Deng/Tyrus/Noah at the beginning of the year, when Kirk was out, were quite a bit different than they were later in the year when Kirk returned to the court. I daresay Vinnie has even coached better. So how much of the difference is due to Kirk vs. Ben, and how much is due to the three other guys and the coach looking better? Probably quite a bit of the latter. So lets figure out how to get a more even comparison. Here's totals I added up for Rose/Hinrich, Rose/Gordon, and Hinrich/Gordon across all their frontcourt pairings: Code: Line Min Pts For Pts Ag Net/48 RG 1417.42 2953 3032 (2.68) RH 185.22 385 384 0.26 HG 271.52 631 543 15.56 The most obvious difference is RG has played a huge amount more together, and over a longer period than RH or HG. The overall numbers indicate to me that despite being a hell of a player, Rose's rookie struggles are not a myth. Now for the RG vs. RH comparison in more detail: The Rose Hinrich backcourt has about 185 minutes in the sample, with 60 coming in that lineup you mention. The others are pretty equally spread amongst: Code: Line Min PF PA Net/48 RHSTiM 40.73 82 83 (1.18) RHSTM 43.63 95 98 (3.30) RHSNM 40.53 81 96 (17.76) The same lineup, but with Kirk out and Gordon in show: Code: RGSTiM 10.23 24 14 46.92 RGSTM 56.15 107 123 (13.68) RGSNM 58.27 129 124 4.12 The frontcourts are Salmons-Tim Thomas-Miller (STiM) Salmons-Tyrus Thomas-Miller (STM) Salmons-Noah-Miller (SNM) The Rose-Hinrich lineups are all slightly negative, while the Rose-Gordon lineups tend to fair slightly better. Oddly the Hinrich-Tyrus-Miller combo is better than the Hinrich-Noah-Miller combo, while the Gordon-Tyrus-Miller combo is worse than the Gordon-Noah-Miller combo. So much for transitivity. What does this lead me to conclude? Well, nothing really… the sample sizes are all quite small. Around a single game’s worth of minutes in every case. That’s like trying to divine something about dice in only a couple of rolls. If we’re going to conclude anything at all, I hope it’s that the numbers are highly variant over a small sample. But since I’m going on about small samples, why not add the Hinrich-Gordon backcourt to the mix: Code: HGSTiM 44.67 124 97 29.01 HGSTM 45.27 95 99 (4.24) HGSNM 35.33 77 66 14.94 For the little its worth, Hinrich-Gordon generally tends to fair better in this sample than either RG or RH. Now, one way to improve things a bit is to sum up the backcourt pairings over these samples. If we look at recent (post trade) performance, our sample sizes get down into single-game territory, but the main thing that comes across is how far Rose has to go. To give a little bit more depth, I summed up the various backcourts when playing with those three frontcourt lineups I broke out. Interestingly, each of the three backcourts has played about the same minutes with those three frontcourt lineups, so if you add them up, you get about 124 minutes played for each of them. That’s about 2.6 full games. That is, it’s still a very small sample, but it’s over a very similar timeframe. Code: L G PF PA Net RH 2.6 99.2 106.5 (7.3) RG 2.6 100.1 100.5 (0.4) HG 2.6 113.4 100.4 13.0 So my basic conclusions 1. There’s some hope for the argument that Kirk can still be a starter on a good team. The team has functioned fairly well (over about 6 games worth of data) with the Kirk-Ben backcourt. Use these numbers when selling him to Minnesota and Portland! 2. If we discount for the early season suckitude likely biasing the Derrick-Ben backcourt, it’s seemes that that the RG (about 30 games of data) and RH (about 4 games of data) backcourts have performed relatively close. But obviously a 30 game stretch is a more solid basis of comparison than a 4 game stretch. 3. Our post-trade sample is smaller than I'd want to make a conclusion from, but it suggests to me that RG outperforms RH. HG outperforms both, but I'd expect that's due to experience and continuity issues that Rose can improve on. 4. Again, Rose's flaws show here, but I don't think that should detract from the sort of prospect he is (a great one). I'm comfortable living with the flaws in the short-run because I think he'll improve dramatically. 5. Yes, small sample sizes, but I don't see a lot to be encouraged about with Tyrus Thomas there.
You're missing one key front court combination, which is Salmons-Tyrus-Noah, that's started most of the last 15 games or so. My guess is that since they represent the predominant starting lineup with RG, it's going to be very negative and make the RG numbers much more negative.
Bulls are 11-10 since the trade, 11-9 in games Salmons has played. Considerably better than the overall record.
I looked it up a few games back and found they were 7-5 in the 12 games before the trade and 5-7 in the 12 after. Perhaps they were coming around in any case?
I will say one thing that makes advanced stats like this nearly worthless to me is the lack of year-in & year-out numbers. The data is there. If +/- is so f....ing improved by advanced mathmatics, why not show Tim Duncan's or Rasheed Wallace's or Deng's or any current player's career numbers year by year comparing +/- to super +/-. Edit: To basketball value's credit, they do have multiple years of data there for anyone to pull down and analyze. If I didn't have a day job, that would be great! Edit 2: Maybe the include it in the download files, but I don't see why they don't include the raw +/- in the same view as the adjusted +/-. Not cool that i need to go to 82games to see how the adjusted adjusts.
Standard Error. You can use this in two basic ways. 1. The +/- there is a mean (average) +/- for that lineup. Actually its adjusted +/- or APM. Mean divided by SE will give you a rough guess of whether the APM can be confidently said to be different than zero. EG, if your calculated APM is 48.44 and the SE is 28.19, then you get 2.3. The lower your SE the better. 2. Add and subtract 2*SE from your APM mean and you get a basic confidence interval between which you can expect most observations to fall. That is, if you roll out a lineup that has a 48.44 mean APM and a 28.19 SE on 100 different occasions, you can expect that on 95 of those occasions, their resulting APM will be between 104 and -8. The 48.44 mean APM with 28.19 SE is the best Bulls lineup posted on Basketball value. It's for Kirk/Ben/Deng/Noc/Gordon. The kicker, of course, is that lineup sucks, and only played about 29 minutes total. So basically, you can't tell much from it because the sample size is extremely small. If
