Posted: Tue Jan 22, 2008 3:22 pm Post subject: PBP Analysis: Offense by Starters in Game
When looking at the +/- differentials of low minutes players, it's hard to know how much weight to put on those numbers: are they measuring the effect of the player specifically? How much do they change because of the quality of competition? For example, we've discussed Renaldo Balkman, and how much of an impact he seems to have based on the 82games numbers. The question back then was: how much of his playing time is he playing in garbage time? And we got close to a resolution on that point. But I think the question we failed to answer was: how much does that matter?
When evaluating these numbers, it helps to have a baseline. What would we expect the numbers to look like for a typical deep bench player? If they spend a lot of time against non-starters, what impact should that have on their numbers?
I decided to investigate (spurring my earlier question) by going through the play by play and counting possessions and points produced from those possessions, based on the total number of starters in the game. If a possession had a sub in the middle (for example, someone is fouled on the floor, subs come in, and then the possession is continued) I used the lineup that finished the possession, under the assumption that that was the part that had an effect on the production of the possession. If subs were made in the middle of free throws, the subs were not counted as "in the game" until after the last free throw (since they weren't on the floor when the foul was committed).
As always, I'll make the disclaimer that my analysis may have some errors in it, and I'll try and fix it if it does, but for now I think it's error free. The data is through January 16th's games.
And a graphical representation (the gray line represents the league average offensive rating):
So you can see by the numbers and the graph, there's a steady increase where more starters in the game means better offensive production. This in itself is interesting, because while it seems intuitive, I don't think it's so obvious. You'd expect that lower quality players in the game would mean both a decrease in offensive and defensive talent, canceling each other out. Instead it seems to be mainly a decrease in offensive talent (or an increase in defensive talent, or both).
There are a number of arguments I could come up with as to why this could be so:
because coaches can evaluate offensive talent more easily than defensive talent - meaning that if a player is above average defensively but below average offensively, they're more likely to be pushed to the end of the bench than if they were above average offensively and below average defensively
there is less offensive depth than defensive depth - in other words, there are more good defensive players than offensive players, so when a team goes to its bench it's much more likely to drop off the level of offensive talent in the game than the level of defensive talent
defense is mainly effort based, whereas offense takes skill, knowledge of the offense and experience - players coming off the bench are giving it their all to try and earn playing time, and because defense is largely effort based, you don't see a huge drop off. But because offense can't be improved merely with hustle, you see a dip on offense but not on defense.
How many of these arguments are real? I think most likely all three play roles in the reason for the dropoff (although I think the last argument is dangerous reasoning, to argue that defense is mainly or only effort, as I've heard in some places).
In any case, there's also a second fascinating piece about this data: where the offensive dropoffs take place. Between 5-9 starters in the game gives you an offensive rating above the league average, in the 107-110 range (with an odd peak at 7 starters). But go down to 4 starters or (and here's the weirdest part) go up to 10, and you see about a 4 point/100 poss dropoff.
Why do the dropoffs occur at those two points? Why isn't it more gradual? 4 combined starters in the game would mean that both teams have hit their 8th man on the bench, and that I suppose means that most teams really only run 7 deep, at least offensively speaking. It could also mean that in relative garbage time, teams range from having 0-2 starters in the game, and the starters know the game is in hand and so give less of an effort. Still, the idea that it's a huge drop instead of something more gradual is odd.
And from the other side, why does offense decline with everyone's starters in the game, and why does it then jump back up when only 1 starter on either side comes out? This seems to make no sense, and I don't really have any kind of good explanation for it. I can come up with 2 guesses, though:
that "10 starters" only really occurs during the opening minutes of each half, and that it takes time for players to get (back) into a game rhythm. That time to get adjusted means they shoot slightly worse.
that the first player out of the game is often a big man who gets a couple of quick fouls. Starting centers are responsible for a huge share of the defense, and with them out of the game offense takes over more. (This feels like a bit of a stretch.)
I'd invite anyone else's attempted explanations of this data, presuming, as always, that it is truly error-free. I might follow up in a bit with a team-by-team breakdown, to see if that reveals anything interesting.
Edit: Yep, I found a small error. It was miscounting the data at the end of quarters. It didn't change the conclusions, though, changing the numbers for lineups with more than 0 starters by at most 0.2 points per 100 possessions. It changed the 0 starter lineups a lot, though, down from 95 to 84. All fixed now, however, I made sure of that.
Last edited by Ben F. on Wed Jan 23, 2008 12:19 am; edited 1 time in total
Posted: Tue Jan 22, 2008 4:00 pm Post subject: Re: PBP Analysis: Offense by Starters in Game
Ben F. wrote:
There are a number of arguments I could come up with as to why this could be so:
because coaches can evaluate offensive talent more easily than defensive talent - meaning that if a player is above average defensively but below average offensively, they're more likely to be pushed to the end of the bench than if they were above average offensively and below average defensively
there is less offensive depth than defensive depth - in other words, there are more good defensive players than offensive players, so when a team goes to its bench it's much more likely to drop off the level of offensive talent in the game than the level of defensive talent
defense is mainly effort based, whereas offense takes skill, knowledge of the offense and experience - players coming off the bench are giving it their all to try and earn playing time, and because defense is largely effort based, you don't see a huge drop off. But because offense can't be improved merely with hustle, you see a dip on offense but not on defense.
How many of these arguments are real? I think most likely all three play roles in the reason for the dropoff (although I think the last argument is dangerous reasoning, to argue that defense is mainly or only effort, as I've heard in some places).
Here's an additional reason: Defense is more important, so you don't get any minutes at all if you can't play defense. This means that the nba is selecting for players that can all play defense (which is another way of saying that offensive ability has a larger variation among players than defensive ability, but that this is selected for by the league).
So the difference between starters and bench players is that the starters are those players with both offensive and defensive ability, whereas the bench players are those with just defensive ability.
In additon to total number of starters on court it might add value to see the results of specific combinations of x starters vs y starters or under different levels of net starter advantage.
Going from 10 starter to 9 offensive rating jumps 4 pts. 10 was 5 on 5, 9 means 5-4 and someone had a starter advantage.
8 could be 5-3 or 4-4, 7 could be 5-2, 4-3. Depending on distributions the higher offensive rating of 7 over 8 could be coming from the greater average starter advantage. Whereas in 10 compared to 9, 9 always had an advantage and 10 never did , here 8 is sometimes balanced and sometimes shows advantage to one side. So the pts edge falling from 4 to 3 doesnt surprise me because the average advantage fell?
6 starters can be more combinations- from 5-1 to 3-3. Maybe there are less instances of advantage or the average size of advantage is lower? Just asking, can't tell in this form of roll-up. 5 starters isn't much different in possible combinations or perhaps distributions and therefore not much different in the results? Going from 6 to 5 the results aren't much different relative to the previous steps because the change in advantage was smaller?
4 and below are clearly bench dominated situation in some fashion with weaker offense than seen in heavier starter situations given the starter's presumed advantage on offensive skill you stated?
Maybe the results would be easier to read using one the alternative methods I suggested for grouping the data?
Last edited by Mountain on Tue Jan 22, 2008 5:21 pm; edited 2 times in total
Joined: 14 Jan 2005 Posts: 1712 Location: Delphi, Indiana
Posted: Tue Jan 22, 2008 5:15 pm Post subject: Re: PBP Analysis: Offense by Starters in Game
Ben F. wrote:
...the first player out of the game is often a big man who gets a couple of quick fouls. ...
Some teams start a designated fouler, ... I mean, tough inside defender who doesn't do much else. Sometimes he's a center.
Meanwhile, this is fascinating work. How's about you show us the breakdown of fouls, FTA, TO, etc, per possession, in each of the 10 groups? _________________ 40% of all statistics are wrong.
In addition to total number of starters on court it might add value to see the results of specific combinations of x starters vs y starters or under different levels of net starter advantage.
How's about you show us the breakdown of fouls, FTA, TO, etc, per possession, in each of the 10 groups?
I can do the "possession enders" - that is shots, FTs, TOs, but fouls become problematic because then you end up with the "split possession" problem I referenced in my first post: fouls often occur in one half of a possession, then subs, then a different outcome. So the grouping would have to be different. That's a problem, unless you only care about fouls that lead to FTs.
Last edited by Ben F. on Wed Jan 23, 2008 12:58 am; edited 1 time in total
Thanks for this breakout which leads to a new question: Is the offensive rating shown for 5 starters on 3 for example the sum of the data for the 5 on offense against 3 and the 3 on offense against the 5 combined? If so, can that be disaggregrated as well?
Posted: Tue Jan 22, 2008 9:03 pm Post subject: Re: PBP Analysis: Offense by Starters in Game
Ben F. wrote:
In any case, there's also a second fascinating piece about this data: where the offensive dropoffs take place. Between 5-9 starters in the game gives you an offensive rating above the league average, in the 107-110 range (with an odd peak at 7 starters). But go down to 4 starters or (and here's the weirdest part) go up to 10, and you see about a 4 point/100 poss dropoff.
An additional possible explanation: a lot of 6th/7th men off the bench are shooters and defensive liabilities.
With 8-9 starters on the floor: the offense might be better, AND, both marginally better offenses may be playing against poorer team defenses.
Kind of like a Steve Kerr in for Ron Harper effect. The shooting specialist isn't a guy you'd want out on the court without at least 3-4 starters, because it takes those 3-4 starters to mask some of the defensive deficiencies of the shooter (from a team defense standpoint).
Yep, I found a small error. It was miscounting the data at the end of quarters. It didn't change the conclusions, though, changing the numbers for lineups with more than 0 starters by at most 0.2 points per 100 possessions. It changed the 0 starter lineups a lot, though, down from 95 to 84. All fixed now, however, I made sure of that.
I edited the main post to reflect that. Again, it doesn't really change anything but I thought I should make it clear. I'm a lot more confident in the following data now, however, since it's double and triple checked.
Now, however, I've also added the data Mike G asked for. So we can analyze things a bit more in depth.
FG2M = made 2 point shots
FG2X = missed 2 point shots
FG3M = made 3 point shots
FG3X = missed 3 point shots
FTM = made free throws
FTX = missed free throws
So there go my theories about why "10 starters" don't play well on the offense end. In fact, not only does it not have to do with shooting at all, but my "foul" theory is completely wrong. It seems that the problem with having all starters in is that nobody gets to the line. The only real variability among the 5-10 starter lineups is in FTM/FGA, and you can see that FT% remains basically constant throughout those lineups. So for whatever reason, when the starting lineups are playing nobody gets to the line. Could this be because refs swallow their whistles in the early portion of the game? Or are players tentative taking it to the rim?
Here's the further breakdown by how many starters each team had:
Maybe you are going to get to it later but to followup on my previous post if 5 starters facing 5 have an offensive efficiency of a bit under 105 then I would assumed that 5 facing zero would be higher. The average shown is 96. Would that perhaps be 105+ for the 5 facing zero and say less than 87 or less for the zero starters facing 5 starters to produce an average of 96?
If I am misinterpreting something let me know.
I am not seeing a way to make good statements about offensive and defensive effects without seeing both sides of the court for each starter vs starter combination separately.
So for whatever reason, when the starting lineups are playing nobody gets to the line. Could this be because refs swallow their whistles in the early portion of the game? Or are players tentative taking it to the rim?
It could be any of these. Or it could be that 5x5 tends to happen at the beginning of each half more frequently. Fresh players = players that move their feet better, don't reach, and don't commit stupid fouls.
I'm with mountain on splitting out the data for each team. I think more could be learned by looking at 4s stats separately in a 4x1 and then looking at 1s stats rather than aggregating the data into a combined Ortg, eFG%, etc.
For one, it would tell us whether the subs tend to lack more offensively, defensively or evenly vs. starters.
So for whatever reason, when the starting lineups are playing nobody gets to the line. Could this be because refs swallow their whistles in the early portion of the game? Or are players tentative taking it to the rim?
It's simpler than that, isn't it? At the start of a quarter, teams aren't in the bonus. Reserves are in by the time they get in the bonus in the first and third quarters, and I'm willing to bet teams play their full starting lineup more often in these quarters than in the second and fourth quarters, when the situation is reversed.
Joined: 14 Jan 2005 Posts: 1712 Location: Delphi, Indiana
Posted: Wed Jan 23, 2008 6:01 am Post subject:
Kevin beat me to it. That's kinda why I asked about both PF and FTA .
So, the players in the game at the beginning of the possession are those responsible for fouls, and I'd also think they are the ones responsible for FTA. Yet your dilemma is that sometimes it's the possession-starters that determine a possession (foul+FT), sometimes the possession-enders (foul/no-FT). That certainly skewers things.
The turnover breakdown is quite weird:
Start - TO%
5 v 0 - 22.1%
5 v 1 - 20.5%
5 v 2 - 15.3%
5 v 3 - 16.2%
5 v 4 - 15.7%
5 v 5 - 15.5%
I would love to play with some others of these. Maybe if you repost, just the totals? I think I can copy/paste these things if the lines don't wrap. Placing a spreadsheet somewhere to download would also work. I'm also wondering about steals and blocks.
You could probably stick all the totals here; just drop Strt, Pts, ORTG, eFG%, TO%, FTM/FGA, and FT% (we can figure those). Add Stl, Blk, PF? And reduce the spaces between columns. That would be awesome. _________________ 40% of all statistics are wrong.
When the number of starters increases to an odd number, the off. increases more than when the number of starters increases to a par number, I think that has sense with the Mountain's one starter-matchup-advantage approach.
Joined: 05 Sep 2007 Posts: 18 Location: University of Miami-Florida
Posted: Wed Jan 23, 2008 10:24 am Post subject:
This is really interesting. Though, I'm not even sure this discussion is being based n a correct premise, that "offensive rating goes down as starters come out." Really, you don't see a change in offensive rating until you get below 5 total starters. That only accounts for about 20% of all game time, anyway. When there are 5 or more total starters, offensive rating varies up and down. I think analysis there should be "in what situations during a game would neither team have at least 3 starters on the court?" Perhaps a second level to this play by play analysis could be what the score was at these times. We could also use what period of the game this tended to be. I'd guess that those last minutes, with no starters for anyone, take place at the wrong end of a blowout where nobody in the building even cares what happens at that point. Also, I would guess that teams with worse rotations and generally worse players would be the types of teams to experiment with lineups that do not involve starters. So, this dip in production could be a reflection of that.
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
