Posted: Wed Aug 15, 2007 9:24 pm Post subject: Garbage Time
I was thinking that we've had a lot of discussions lately about garbage time. From the Boston Celtics discussion where it's important to see how many players log "impactful" minutes, to the Renaldo Balkman discussion where we were trying to determine how many minutes a player has to play such that we know he's not just a garbage player. In that thread cherokee and Mike G were able to derive a formula that's a good approximation of whether the player was playing against starters or bench players, but, while that's a very good measure, there's a difference between playing against Antonio Daniels off the bench and Donnell Taylor.
Anyway, I decided to do a pretty simple study to see how much garbage time was really being played, and hope it helps people in anything from more in-depth evaluations to back of the envelope calculations (something like: Balkman played over 1000 minutes, and the Knicks only faced about 400 minutes of garbage time, so he can't solely be a garbage player).
At first I struggled with coming up for the definitions of garbage time, but then I decided to take Ed's great within game win expectancy formula (helped along by the file that daviswylie posted) and apply it to this. I decided that if the win expectancy of either team was equal to or above 95%, it qualified as garbage time. (This was somewhat arbitrary, based on a couple observations, and what looked right, but it definitely could be argued with.) However, if at any point later in the game the win expectancy went back below 95%, that count would be wiped out. This gets rid of situations like 18 points leads in the 3rd quarter, that are then cut down to 10. If they stay at 10, or balloon back up, I will count the point that they become garbage time again as the start of the garbage time. But the point is that garbage time must be a period lasting to the end of the game.
Additionally, the win expectancy got a little funky towards the end of the game, since with 7 seconds left and a 2 point lead, the opposing team only has about a 4% chance of winning. As such I made sure garbage time had to start with > 30 seconds left in the game.
The results are as follows. SAS lead the league with 844 minutes and 11 seconds of garbage time. GARB% is my calculation of percentage of the team's minutes that were garbage minutes. I did not use the actual team figures for this, instead I simply divided the totals by 3936 ( 82 * 48 ). I did this because OT is not included in this study. While it could easily be argued that OT is never garbage time, I didn't even evaluate it as part of the study and as such ignored it in the percentage calculations.
Code:
Team MIN SEC GARB%
SAS 844 11 21.4%
DAL 794 45 20.2%
PHX 729 54 18.5%
GSW 709 08 18.0%
ATL 654 00 16.6%
MEM 641 00 16.3%
CHI 627 09 15.9%
HOU 593 51 15.1%
LAC 568 58 14.5%
POR 559 42 14.2%
IND 558 27 14.2%
MIA 548 08 13.9%
PHI 542 14 13.8%
MIL 539 57 13.7%
SEA 537 33 13.7%
CLE 527 59 13.4%
ORL 517 47 13.2%
TOR 508 37 12.9%
NJN 505 30 12.8%
CHA 505 11 12.8%
DET 474 35 12.1%
NOK 472 21 12.0%
SAC 463 52 11.8%
UTA 454 23 11.5%
DEN 449 45 11.4%
WAS 446 54 11.4%
MIN 427 33 10.9%
NYK 402 43 10.2%
BOS 383 24 9.70%
LAL 350 42 8.90%
Getting to garbage time would seem a team goal.
Do you have interest in also measuring coaching behavior in garbage time including but not necessarily limited to speed and extent of pulling starters? How do the Spurs compare on other teams on that? Is their deep rotation just getting more garbage time or they getting more regular time too? Does Phoenix ride starters heavier than average for strong teams in comparable garbage time situations? Is that foolish from standpoint of being fresh for playoffs? Do the starters love the opportunity to play in the lead, have fun, hear the crowd... and get the financial rewards of big stats? Same spin for coach and owner and surely it makes fans happy too. Is championship all or the whip topping?
Joined: 14 Jan 2005 Posts: 1506 Location: Delphi, Indiana
Posted: Thu Aug 16, 2007 5:27 am Post subject:
Wow, this looks pretty amazing. It sounds like you touched all the bases here. I expect to see the best and worst teams at the top, medium teams at the bottom. After the Big3, mediumite GS pops in. I guess they had so many guys miss games, they were sometimes an elite team (like the playoff version), sometimes a lousy team?
Seeing some of the worst teams at the bottom -- NY, Bos, Min -- seems to imply that these teams tended to be competitive, but they'd lose anyway?
Now I wonder if there is an approximating formula for these %ages. Something along the lines of y + x*abs(PtDiff). Looks like y might be around .09, x close to 1.3 .
This doesn't account for GSW's high garb%, but maybe one could factor in 'games missed by starters', or G*(PER-9), ala Justin's new stat. Such pocket approximations could estimate past seasons (for which there is no bbvalue data). _________________ 40% of all statistics are wrong.
Thanks for the kudos, guys. Any comments regarding picking 95% as the cutoff point?
Mike G wrote:
After the Big3, mediumite GS pops in. I guess they had so many guys miss games, they were sometimes an elite team (like the playoff version), sometimes a lousy team?
I think this is a good theory. Another theory is that their style is a highly volatile one - gambling on threes and steals - and as such could produce extreme results.
Mike G wrote:
Seeing some of the worst teams at the bottom -- NY, Bos, Min -- seems to imply that these teams tended to be competitive, but they'd lose anyway?
It seems that way. I can think of a couple explanations for this: I wonder if perhaps this is a measure of coaching ability - that is, Doc Rivers and Isiah Thomas probably should have won more games than they did given their situations. Or, it could be that the Knicks' and Celtics' garbage players were not nearly as bad as a typical garbage unit, thus keeping the score closer than expected or even making a rally or two to wipe out previous garbage time. This could also be seen, however, as a strike against these coaches since it would seem that players that were worthy to play when the game would be decided did not.
So I wonder if you could create some measure of coaching based on the expected garbage time (from pythagorean win%) versus the actual, and see how that ends up.
Mountain wrote:
Do you have interest in also measuring coaching behavior in garbage time including but not necessarily limited to speed and extent of pulling starters? How do the Spurs compare on other teams on that? Is their deep rotation just getting more garbage time or they getting more regular time too? Does Phoenix ride starters heavier than average for strong teams in comparable garbage time situations?
I don't trust the substitution data on the play by play enough to do this. There are often subs that are out of place in the PBP, and lots of other mistakes that can confound the results. I can try and see what I can pull out, but take the results only with a grain of salt.
Something else to consider - I factored in the relative strength of each team when calculating win expectancy, but I think this may have been a mistake. Looking back we know that Dallas, trailing by 6 with 2 minutes left has a better chance of winning than ATL does, but I'm not sure that should factor into the evaluation of garbage time at that moment. A coach of either team will most likely make the decision to pull players based on time and score, not based on the strength of his team.
There's that issue, and the issue that trailing by 6 with 2 minutes left isn't really garbage time. Even trailing by 8 with 2 minutes left isn't garbage time, really (although if you were down 18 before and cut it to 8, I'm not sure the coach would stick his starters back in). Yet down 8 with 2 minutes left shows a 2.6% chance of winning. I think the idea of garbage time is being out of contention with a fair amount of time left in the fourth quarter. Any ideas on what that time should be (4 minutes? 6 minutes?) or a way to perhaps find it by mining the PBP?
OK, last post for the day so that I give everyone else a chance to respond. Here are the new results, ignoring the teams' pythagorean winning% and with a new wrinkle: garbage time can't start until the start of the 4th quarter.
Code:
Team MIN SEC GARB%
GSW 465 11 11.82%
DAL 423 24 10.76%
SAS 415 09 10.55%
ATL 406 35 10.33%
MIA 401 36 10.20%
CHA 374 52 9.52%
PHX 372 05 9.45%
LAC 371 11 9.43%
MIL 365 20 9.28%
CHI 364 45 9.27%
ORL 362 09 9.20%
MEM 360 24 9.16%
IND 359 44 9.14%
PHI 354 35 9.01%
HOU 346 58 8.82%
TOR 345 47 8.79%
POR 342 48 8.71%
CLE 337 09 8.57%
DEN 336 17 8.54%
SAC 334 06 8.49%
SEA 319 17 8.11%
DET 315 59 8.03%
NJN 312 58 7.95%
WAS 312 09 7.93%
UTA 310 06 7.88%
NYK 305 17 7.76%
NOK 281 28 7.15%
MIN 281 04 7.14%
BOS 277 24 7.05%
LAL 273 13 6.94%
As you can see, the order doesn't change a whole lot. GSW vaults to the top of the list, but besides that the same general rankings stay the same (DAL, PHX, SAS, ATL, MEM near the top, BOS, NYK, LAL near the bottom). The totals went a lot farther down, which I think is more accurate. While it's interesting that SAS played over 1/5th of their minutes with the game pretty much decided, I'm not sure it's an accurate representation of true garbage time. And as I brought up in my previous post, I'm not sure this really is either, because you have situations where the game is still winnable involved as well.
Anyway, to help evaluate whether or not this is truly capturing garbage time, and to try and answer Mountain's question, I decided to find the garbage minutes totals for every player in the league.
Now, the disclaimer is the same as above. All of this should be taken with a grain of salt because there are plenty of PBP errors and there's always the chance that I could have had a programming error as well. But here are the top 25 garbage time players:
While I wasn't 100% sure of my definition of garbage time before this, the top 25 really strengthens my confidence in it. Maggette and Felton are the only real starters on this list. Barbosa leading the way doesn't surprise me too much because PHX played a lot of garbage time and doesn't like to go too deep into their bench. Barbosa plays most of their minutes with Nash off the floor, from observation.
I do, however, think it's still capturing a lot of time in "winnable" games, or at least times when the coach thinks the game is still winnable, situations like I described in my previous post. There are still stars father down on the list with large chunks of garbage minutes.
Joined: 03 Jan 2005 Posts: 664 Location: Washington, DC
Posted: Thu Aug 16, 2007 1:41 pm Post subject:
Not sure if you can program this, but you could approach the garbage time definition differently. Instead of arbitrarily choosing the 95% cutoff, you could look at the pbps and see when coaches actually removed their starters, their stars, or some substantial number of starters. Still kinda hazy, though.
Or, you could look at a sampling -- at what point do coaches think garbage time has started (based on sub patterns)? Then use that to refine the 95% cutoff point. _________________ My blog
Joined: 14 Jan 2005 Posts: 1506 Location: Delphi, Indiana
Posted: Thu Aug 16, 2007 1:47 pm Post subject:
Ben F. wrote:
.. Any comments regarding picking 95% as the cutoff point?
I was thinking about this while driving today. If I'm the coach, unless playoff seedings are not too relevant, I'm still strategizing at 5% (or 95%). Maybe 97% is garb, and 94-95% is scratch-n-sniff.
Late-season games might be considered differently. Whole games are garbage when a team is resting its starters. _________________ 40% of all statistics are wrong.
Not sure if you can program this, but you could approach the garbage time definition differently. Instead of arbitrarily choosing the 95% cutoff, you could look at the pbps and see when coaches actually removed their starters, their stars, or some substantial number of starters. Still kinda hazy, though.
Or, you could look at a sampling -- at what point do coaches think garbage time has started (based on sub patterns)? Then use that to refine the 95% cutoff point.
This is a good idea, and I've been thinking about how to approach it. It gets really thorny when you have games that are complete blowouts, and with a 30 point lead the coach doesn't want to play bottom of the bench players for 20 minutes so he still plays starters.
But I could see using the same definitions that we have now (95% and in the 4th quarter) as initial cutoffs and then seeing what percent of the time a lineup has 2 or less starters.
Mike G wrote:
If I'm the coach, unless playoff seedings are not too relevant, I'm still strategizing at 5% (or 95%). Maybe 97% is garb, and 94-95% is scratch-n-sniff.
Yeah, I suppose the question is how to come up with a cutoff that isn't just based on estimation but on some better rule. For example, something like what kjb suggested above, trying to determine at what point coaches put in lineups that have 2 or less starters. If we looked at the win expectancy every time that happened we might be able to get a better cutoff.
Mike G wrote:
Late-season games might be considered differently. Whole games are garbage when a team is resting its starters.
One thought I had early on when I saw BOS at the bottom of the list was I wonder if you can look at the difference between garbage time for the first 75% of the season (or some other cutoff point that marks when teams start to tank) and garbage time for the last 25%. Because my guess is that in those throw away games, win expectancy falls to one side early and then doesn't change.
At the same time, I've already started to limit my sampling to just the 4th quarter, because I don't think coaches consider any lead outside of the 4th quarter safe. For example, there was a WAS vs IND game early in the year where the Wizards lead by 30 with 22 minutes left in the game. But the starters played for the next 7 minutes, came out, and then at the start of the 4th quarter came in for another 2 minutes. Even though the game was out of reach for all 22 minutes, I don't think you can consider it all garbage time when starters are playing.
So I think that garbage time and tanking are two separate studies. That being said, you do raise a very good point that teams may play "garbage time" players at the end of the season in games they don't care about and in situations that aren't garbage time by any definition. I don't really know what to do about that.
Now, the disclaimer is the same as above. All of this should be taken with a grain of salt because there are plenty of PBP errors and there's always the chance that I could have had a programming error as well. But here are the top 25 garbage time players:
Code:
RNK Player MIN SEC
1 Barbosa,Leandro 279 09
2 Dunleavy,Mike 276 54
...
Nice work Ben. This is interesting. One question, can you also provide the list of top garbage minutes as a % of total minutes played? I'm guessing that while Barbosa apparently played a lot of garbage minutes, he also played enough non-garbage minutes that he'd rank low on that list. I don't think of him as the quintessential garbage time player, after all.
Also, I think I caught most of the pbp errors and corrected them, but a few slipped through. I think the accuracy for minutes played should be closer to 99% than 90%.....
You also got situations like those that Sam Mitchell used to advocate, i.e. in some games, he'd be so pissed at the general effort of his starters, he used to leave them in just to humiliate them. It happened many times, and I won't be surprised if some other coaches use this tactic. How many times have we wondered why so-and-so coach is still playing the starters?
My suggestion is to establish cut-offs, i.e. garbage time is when a team is down:
1. 30 pts down with 1 quarter to go
2. 25 pts with 10 minutes to go
3. etc.
Those are obviously arbitrary and should be refined, but they make sense in that even if the starters are still in the game, they probably know it's a useless task and basically just fold the tent.
To recapitulate, I don't think you can define garbage time by when the starters are taken out of the game.
Nice work Ben. This is interesting. One question, can you also provide the list of top garbage minutes as a % of total minutes played? I'm guessing that while Barbosa apparently played a lot of garbage minutes, he also played enough non-garbage minutes that he'd rank low on that list. I don't think of him as the quintessential garbage time player, after all.
Yes, this was the next step, but I wanted to come to a better definition of garbage time before I linked it up with actual NBA minutes.
basketballvalue wrote:
Also, I think I caught most of the pbp errors and corrected them, but a few slipped through. I think the accuracy for minutes played should be closer to 99% than 90%.....
My 90% comment was not directed completely at the BasketballValue data. While I have had to clean some stuff up that you didn't catch, the reason I'm not 100% confident in my lineup data is due to possible errors in my own programming. To find who's in the game my program goes through the PBP and finds people who have committed an action. But there are situations where a rebound at the end of a quarter will be mistakenly placed in the next quarter by the NBA PBP and as such my program thinks that player is in the game. There are also subs that have the wrong timestamp next to them, etc. Usually my program should catch this and fire an error so I know to take a look and fix it. But I think there are situations where it slips through undetected so the lineup data I have doesn't actually reflect what was on the floor at that time.
supersub15 wrote:
To recapitulate, I don't think you can define garbage time by when the starters are taken out of the game.
This is a fair argument but then what do you suggest is the best way to define garbage time? As I defined it above, it's based on the chance the team has to win the game, which is essentially what you were trying to do with the cutoffs you proposed. But choosing a percentage is arbitrary if not backed up by some kind of data. As such we proposed to look at starters to see if we could refine the percentage. I understand your argument, and I don't think it will give us the exact garbage time count, but it could be more accurate.
You also got situations like those that Sam Mitchell used to advocate, i.e. in some games, he'd be so pissed at the general effort of his starters, he used to leave them in just to humiliate them. It happened many times, and I won't be surprised if some other coaches use this tactic. How many times have we wondered why so-and-so coach is still playing the starters?
My suggestion is to establish cut-offs, i.e. garbage time is when a team is down:
1. 30 pts down with 1 quarter to go
2. 25 pts with 10 minutes to go
3. etc.
That's very far above 100% win expectancy, but I think 95% could be a little low. Maybe a combination of 99% or 100% with no more than 2 starters (4 between the two teams) on the floor (the last quarter of course, or maybe any quarter)
I also think a garbage% team list must be some correlated with the "absolute or net" points-margin team list, with exceptions to those teams that are both extreme winning and losing.
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