Charting Duke vs. Illinois

[Jumbo]

OK, this one was slightly more meaningful in terms of keeping track of +/-, since the main guys played more. Still, K substituted quite often, switching lineups 27 times. He used four lineups more than once: starters (four times); Paulus-Scheyer-Nelson-Singler-Thomas (five times); Paulus-Scheyer-Nelson-Singler-Zoubek (twice); Paulus-Scheyer-Nelson-Henderson-Zoubek (twice).

One interesting factoid before we hit the numbers. Duke scored 47 points when Henderson was in the game, and he scored 23 of them. I'm honestly not sure if that's a good thing or a bad thing. Anyway, on to it...

Individuals
Kyle Singler 53-38 (+15)
Jon Scheyer 63-50 (+13)
Brian Zoubek 41-31 (+10)
Lance Thomas 46-37 (+9)
Nolan Smith 23-15 (+8)
Greg Paulus 58-53 (+5)
DeMarcus Nelson 60-56 (+4)
Gerald Henderson 47-44 (+3)
Taylor King 4-6 (-2)

Lineups
Paulus-Scheyer-Nelson-Singler-Thomas 21-11 (+10)
Paulus-Scheyer-Henderson-Thomas-Zoubek 9-3 (+6)
Smith-Scheyer-Nelson-Henderson-Zoubek 7-4 (+3)
Smith-Scheyer-Nelson-Singler-Zoubek 5-2 (+3)
Smith-Scheyer-Henderson-Singler-Thomas 2-0 (+2)
Smith-Scheyer-Henderson-Singler-Zoubek 3-1 (+2)
Paulus-Scheyer-Nelson-Singler-Zoubek 6-4 (+2)
Paulus-Nelson-Henderson-Singler-Zoubek 5-3 (+2)
Paulus-Smith-Scheyer-Nelson-Henderson 1-0 (+1)
Smith-Scheyer-Henderson-King-Zoubek 4-4 (0)
Paulus-Scheyer-Nelson-Thomas-Zoubek 0-1 (-1)
Paulus-Smith-Scheyer-Henderson-Thomas 1-2 (-1)
Smith-Scheyer-Nelson-Singler-Thomas 0-2 (-2)
Paulus-Scheyer-Nelson-Henderson-Singler 0-2 (-2)
Paulus-Scheyer-Nelson-King-Zoubek 0-2 (-2)
Paulus-Nelson-Henderson-Singler-Thomas 11-13 (-2)
Paulus-Scheyer-Nelson-Henderson-Thomas 2-5 (-3)
Paulus-Scheyer-Nelson-Henderson-Zoubek 2-7 (-5)

[Sir Stealth]

These stats are really interesting. Right away it jumps out that Singler had the best +/- when on the surface he appeared to have a (relatively) quiet game. From watching the game I would have thought that Paulus-Scheyer-Nelson-Henderson-Zoubek would have been one of the more successful lineups, yet clearly that's not the case. More than meets the eye I guess. First reaction is that it makes me more encouraged about Singler more than anything else.

[mepanchin]

Singler had a good overall game. I have been charting defensive stats for players, and every game Singler is involved in a lot of players, forcing a lot of bad shots, etc. He had a few too many bad fouls tonight and didn't rebound as well as he had been, but otherwise had a good game.

[tux]

Thanks Jumbo for compiling these data. I find it difficult to draw any meaningful conclusions. Maybe it's the large number of line-ups. Maybe it's the fact that the data run counter to the impressions I formed watching the game. I.e., both Markie and Gerald had great games, especially on the offensive end. They were the only guys who could generate their own shots. Yet they seem to be close to a wash according to the +/-.

These data will probably become more useful mid-season, when we can compile them over many games (to state the obvious). Anyway, it's great to have this additional filter through which to post-process the game...

[CDu]

Thanks Jumbo for compiling these data. I find it difficult to draw any meaningful conclusions. Maybe it's the large number of line-ups. Maybe it's the fact that the data run counter to the impressions I formed watching the game. I.e., both Markie and Gerald had great games, especially on the offensive end. They were the only guys who could generate their own shots. Yet they seem to be close to a wash according to the +/-.

These data will probably become more useful mid-season, when we can compile them over many games (to state the obvious). Anyway, it's great to have this additional filter through which to post-process the game...

Agreed. There are just so many variables that occur over such a short period of time (especially with lots of substitions) in a single game that can skew these numbers. It's very possible that these numbers are meaningful from a single game, but it's not likely.

The numbers will get a lot more meaningful, though, as they are summed over a sample of 20+ games. That won't account for all the variables in play, but it will help a lot. I do like the +/- approach to analysis, but I think it's going to take a lot more minutes to glean anything worthwhile out of it.

That said, a single game's +/- CAN be useful for generating hypotheses to test. Jumbo has already done so with regard to Henderson. It will be interesting to see if a trend develops in which Henderson has seemingly good performances on both ends of the floor but has relatively low +/-. I suspect it won't hold (either he'll play worse, or his +/- will go up).

In other words, I don't believe he was the cause of his relatively low +/- last night. He could very well be the future cause of a mediocre/bad +/- (if he stops scoring as well as he did last night), but I think last night's numbers are indicative of some other factors that have not been controlled for.

[phaedrus]

Interesting that Smith was +8 and Paulus only +5.

[Troublemaker]

Agree with everyone that +/- is more useful over the long haul. For example, if a player is keeping a team afloat during a period in the game when the team is struggling and otherwise would be losing ground on the scoreboard, his +/- for that period of time might be close to zero but his effect on the game was great since without him, his team's +/- could be negative.

[Troublemaker]

That said, I LOVE looking at +/-.

[Saratoga2]

Interesting that Smith was +8 and Paulus only +5.

Smith didn't play for long and did seem lost with some of his passes. He also hit a three. Maybe his turnovers didn't result in points because the rest of the team picked up the defense. I just think there are so many variables that charting has limited value in describing what really happened on the floor.

[phaedrus]

Right. The reason I think it's interesting is that a) it didn't seem like Smith played for long, but the 23-15 represents a significant portion of the game (a quarter or so), and b) it didn't seem like he played that well, but we actually did pretty well when he was on the court.

Obviously the plus/minus is limited, but it still shows how the team did when a certain player was on the court.

[dcarp23]

Another "issue" with the limited sample size is that Paulus, Henderson and Nelson were all on the floor at the beginning of the game when Illinois jumped out to a six point ead and at the end of the game when they cut the lead from 19 to 13. Thus, those guys were dealt a -12 out of the gate. Take away those two periods of the games, and those guys are at the top of the leaderboard in +/-.

Obviously, you can't do that. But it does show how the effect of such a limited sample size.

[devildeac]

Agree that this another nice addition to our post-game analysis. Takes a lot of effort and time to keep up with all these numbers and then share with us. Don't know whether Jumbo mentioned previously, but this kind of format is utilized in the NHL and helps managers, coaches and the fans keep track of who does well/poorly both individually and as part of the team. I know when the Raleigh Hurricanes play(I refuse to call them the carolina Hurricanes as I refuse to refer to the NFL team in Charlotte as carolina Panthers), and the N&O posts their +/-, it helps me understand the game a bit better.

[CDu]

Another "issue" with the limited sample size is that Paulus, Henderson and Nelson were all on the floor at the beginning of the game when Illinois jumped out to a six point ead and at the end of the game when they cut the lead from 19 to 13. Thus, those guys were dealt a -12 out of the gate. Take away those two periods of the games, and those guys are at the top of the leaderboard in +/-.

Obviously, you can't do that. But it does show how the effect of such a limited sample size.

Correct. Similarly, the spurt of offense by Illinois near the end of the half can skew the data as well. Sometimes the opponent just catches a few breaks, and sometimes the opponent happens to hit a few shots that they weren't hitting at other times. Good defense doesn't always result in good outcomes, and bad defense doesn't always result in bad outcomes. Because of this, it's hard to tell whether the +/- indicate a causal effect of a player or a coincidental result when the sample size is just one game.

These issues do tend to wash out (somewhat) with a larger sample size, so I hope the +/- is summed over the course of the season. It will be very interesting to see those numbers. If used correctly, they can be a very valuable tool for evaluating an individual player or particular lineup with respect to their value to the team.

[Jeffrey]

Hi,

You guys seem to find more conclusions (even over longer periods) from this info. than I. Does this data cross-measure the opposing team's line-up? For example... if the opposition's best bigs are not in the game the majority of the time that Zoubek is going +10 in this measure, then what is the measure actually revealing? Wouldn't Weber have awakened to Gerald's great game and put his best defender on him? And wouldn't Weber have also rested his best defender when Gerald came out? IMO, that would completely skew this data to the point of little true value. No offense Jumbo, just my view.

Best regards,
Jeffrey

[CDu]

Hi,

You guys seem to find more conclusions (even over longer periods) from this info. than I. Does this data cross-measure the opposing team's line-up? For example... if the opposition's best bigs are not in the game the majority of the time that Zoubek is going +10 in this measure, then what is the measure actually revealing? Wouldn't Weber have awakened to Gerald's great game and put his best defender on him? And wouldn't Weber have also rested his best defender when Gerald came out? IMO, that would completely skew this data to the point of little true value. No offense Jumbo, just my view.

Best regards,
Jeffrey

These are very good questions, and all are questions that cannot be addressed with +/- from a single game. Because of this, analyzing these numbers is (or should be) a challenging science. Part of the reason +/- works in hockey is because there are a LARGE supply of games (81 over the course of a season) from which to sample data.

We ultimately can't control specifically for the "matchups" variable (especially not in a single game) with +/-. As such, you just hope that, over a large enough sample, the concern is mitigated by a diffusion of lineups.

[Kilby]

Even over the longhaul these statistics can be very misleading. If Henderson continues to play as well as he has, he is going to command double teams and defensive help whenever he has the ball. This would open it up and create opportunities for everybody else on the floor. Shooters would have open shots (and higher stats) that they would never have otherwise. I got the feeling, as I often have, that Henderson could take the ball and score, or at least have a high chance of scoring, anytime he had the ball. Invaluable for a team to have one player like this. If you have two you can go deep in the tourney.

[CDu]

Even over the longhaul these statistics can be very misleading. If Henderson continues to play as well as he has, he is going to command double teams and defensive help whenever he has the ball. This would open it up and create opportunities for everybody else on the floor. Shooters would have open shots (and higher stats) that they would never have otherwise. I got the feeling, as I often have, that Henderson could take the ball and score, or at least have a high chance of scoring, anytime he had the ball. Invaluable for a team to have one player like this. If you have two you can go deep in the tourney.

What you're describing absolutely WOULD be accounted for in the +/-. If what you say occurs, then other players should be getting easy shots (i.e., easier paths to score points) when Henderson is in there, which should make Henderson's +/- go up (along with everyone else's +/- who's on the floor). Conversely, when he leaves, those openings would go away, theoretically bringing everyone else's +/- down but not affecting Henderson's +/-.

What it can't account for (and would in fact invert the +/-) is if the opponent brings in a specific defense (or defensive player) to match up with a specific player. If that specific defender is guarding a specific player exclusively when he's in and that defender is not in the game when the player is not in, then there is a variable that is highly correlated with both the minutes played AND the expected +/-. So we'd have trouble discerning whether the player's +/- was due to the player's direct impact on the team or due to the influence of that specific matchup. Luckily, with a larger sample, this becomes more diluted and hopefully less of an issue.

[tux]

Agree with everyone that +/- is more useful over the long haul. For example, if a player is keeping a team afloat during a period in the game when the team is struggling and otherwise would be losing ground on the scoreboard, his +/- for that period of time might be close to zero but his effect on the game was great since without him, his team's +/- could be negative.

That's what I thought re: Gerald (and DeMarcus) when I saw the +/-; there was a stretch early where IU raced out to a lead. Both of those players were keeping us in the game and then helped push us over the top... So, +/- would be close to nil, but they were instrumental... I remember Scheyer coming in for Gerald right before the time we pushed out to a lead and the score was still close; he needed a breather, no doubt.

However, I have noticed looking at the box scores that G is taking quite a few shots relative to the other guys. I don't see that as a bad thing, as he is the best we have at creating his own shot and every great team needs a guy (or two) like that. He's definitely not as efficient (so far) as Singler and Scheyer. The team has a good mix of athleticism and headiness (if that's a word...) with Kyle having the best balance between the two, imo.

I thought K had an interesting comment about getting killed on the O-glass: Duke was switching quite a bit on D in order to disrupt the motion offense; as a result, Duke players were having a hard time finding their man to block out. K said essentially that "you can't stop everything" and they would do it the same way next time, given the outcome... I think that's something us fans have a hard time with. If the other team hits a couple of 3s, then it's: We're not playing good defense, etc. If Paulus has a bad TO: There's the old Paulus. But, you just can't stop everything out there; you pick your poison sometimes and hope positives outweigh the negatives.

[Jumbo]

Correct. Similarly, the spurt of offense by Illinois near the end of the half can skew the data as well. Sometimes the opponent just catches a few breaks, and sometimes the opponent happens to hit a few shots that they weren't hitting at other times. Good defense doesn't always result in good outcomes, and bad defense doesn't always result in bad outcomes. Because of this, it's hard to tell whether the +/- indicate a causal effect of a player or a coincidental result when the sample size is just one game.

These issues do tend to wash out (somewhat) with a larger sample size, so I hope the +/- is summed over the course of the season. It will be very interesting to see those numbers. If used correctly, they can be a very valuable tool for evaluating an individual player or particular lineup with respect to their value to the team.

Obviously, sample size is a key issue. I'm not saying otherwise. But the "spurt" argument is silly -- should Duke's players not get credit for a positive spurt?

[dw0827]

Hey Jumbo, does any of your data indicate how much we played zone last night? I watched the game . . . didn't tape it or chart it . . . so I only have impressions. And my impression is that, in the first half, we went zone for a few minutes (2 to 3). It was successful for 4 to 6 possessions, then they hit a 3, and we got out of it for the rest of the half. Is my impression correct? Any empirical data available?

Second half . . . used it sparingly to protect players and for a change of pace. Again, just an impression.

[CameronBlue]

A more useful measurement may be to chart differentials before and after a player enters the lineup to see what points of inflection show up in the data. Call it the player's "spark" potential.

[CDu]

Obviously, sample size is a key issue. I'm not saying otherwise. But the "spurt" argument is silly -- should Duke's players not get credit for a positive spurt?

You misinterpreted me. Obviously players should get credit for spurts (both Duke and opponent). My point is that spurts can be random events. Therefore, within a single game, +/- may reflect random spurts. These spurts are likely to trend toward the mean over time, which is why sample size matters.

My point was simply to say that making conclusions based on +/- from a particular game can be lead to some faulty conclusions, due to the fact that within game variability is so great. I wasn't trying to discredit the fact that the spurt happened - it happened. My point was that it may not have been influenced by anything other than random chance, whereas over time that random chance should be distributed more evenly across lineups.

Don't get me wrong: I'm a big proponent of +/-, provided it's used correctly. I just think it's much more useful over the course of a season than from a single game because in-game variability can skew results in a manner over which a player may have little/no control.

[Jumbo]

Hi,

You guys seem to find more conclusions (even over longer periods) from this info. than I. Does this data cross-measure the opposing team's line-up? For example... if the opposition's best bigs are not in the game the majority of the time that Zoubek is going +10 in this measure, then what is the measure actually revealing? Wouldn't Weber have awakened to Gerald's great game and put his best defender on him? And wouldn't Weber have also rested his best defender when Gerald came out? IMO, that would completely skew this data to the point of little true value. No offense Jumbo, just my view.

Best regards,
Jeffrey

Plus/minus and efficiency stats are far from an exact science. And cross-referencing is obviously a very good example of an area where they fall short. I don't know of any statisticians who have developed a matrix to measure who is on the court for the OTHER team, and how that affects things.

That said, something is better than nothing, and as the sample size increases, you can gain knowledge about lineup combos and player effectiveness through plus/minus.

[Jumbo]

Even over the longhaul these statistics can be very misleading. If Henderson continues to play as well as he has, he is going to command double teams and defensive help whenever he has the ball. This would open it up and create opportunities for everybody else on the floor. Shooters would have open shots (and higher stats) that they would never have otherwise. I got the feeling, as I often have, that Henderson could take the ball and score, or at least have a high chance of scoring, anytime he had the ball. Invaluable for a team to have one player like this. If you have two you can go deep in the tourney.

If that's the case, it would be reflected in his plus/minus, since the team would thrive while he's in the game. Problem is, as I pointed out in the post-game thread, G has a long way to go in developing his court vision to be a successful playmaker for others.

[Jumbo]

A more useful measurement may be to chart differentials before and after a player enters the lineup to see what points of inflection show up in the data. Call it the player's "spark" potential.


I have the full chart which shows the score every time a lineup change was made. Should I post the full game log?

[pfrduke]

I have the full chart which shows the score every time a lineup change was made. Should I post the full game log?

Alternatively, anyone interested in the full game log can just look at the box score (http://www.goduke.com/ViewArticle.dbml?SPSID=22726&SPID=1845&DB_OEM_ID=4200&ATCLID=1324874) on GoDuke. It provides the full play-by-play, including who subbed in for whom and when. It gives a box score for each half of play at the bottom.

[jipops]

Interesting that lineups giving away -3 and -5 were our smallest lineups last night. This may have to do with Illinois getting offensive putbacks, therefore the numbers could be indicative of the opponent we faced. Illinois was already a very strong offensive rebounding team. Will be interesting to see what the next sample give us.

In regards to the comment about Henderson being most of the points when he was in, I did get the feeling that every time Gerald got the ball it was going up. Good thing for us last night was that it also went in. But if this indicative of his play it could be a possible issue in the future because defenses will start keying on him and he is still not the strongest ball handler. He does need to be more comfortable with dishing out when his drive is cut off or if he doesn't have a decent look.

[throatybeard]

Obviously, sample size is a key issue. I'm not saying otherwise. But the "spurt" argument is silly -- should Duke's players not get credit for a positive spurt?

Now here we go--what we need is a sabremetric for quantifying spurtability.

Bobby Hurley had a hi SPRT, I know that.

[darthur]

I don't know of any statisticians who have developed a matrix to measure who is on the court for the OTHER team, and how that affects things.

As a math/CS guy, I think this would be a bad idea unless you have full +/- stats on the opposing team throughout an *entire season*. And of course you don't. If you try to adjust Duke's +/- based only on the opponent's +/- from the one game, you will basically be doubling the random noise, and with such a small sample size, that's a bad thing.

The two main problems I have with +/- are:
1. It doesn't account for time in the game. For example, Nolan Smith probably gained Duke more points per minute than Jon Scheyer did this game. But that isn't reflected in +/-.
2. It doesn't measure how good a player is in an absolute sense. It measures how they compare to the people they are sharing minutes with. If Zoubek were injured, for example, it would hurt everyone's +/- *except* Thomas's, so you'd see Thomas climb the +/- list through no action of his own.

It's still pretty interesting though.

[Jumbo]

1. It doesn't account for time in the game. For example, Nolan Smith probably gained Duke more points per minute than Jon Scheyer did this game. But that isn't reflected in +/-.
That's easily fixed with with a per-minute ratio. A lot of websites that track this stuf do that.


2. It doesn't measure how good a player is in an absolute sense. It measures how they compare to the people they are sharing minutes with. If Zoubek were injured, for example, it would hurt everyone's +/- *except* Thomas's, so you'd see Thomas climb the +/- list through no action of his own.

It's still pretty interesting though.

Of course it doesn't measure how good a player is, just how completion percentage doesn't measure how good a QB is or shooting percentage doesn't actually measure how good a shooter is, per se. It's just a bit more information in a complex puzzle. It's helpful, but far from definitive.