The effect of playing X consecutive/total minutes

[J.E.]

I took some time to research whether
- the amount of time a player has been on the court without being substituted out
- the amount of total minutes a player has been on the court
has any influence on player performance.

For the first scenario I reset 'time on the court' at halftime, but not between quarters or timeouts. There certainly are arguments for doing it differently, but for now, that's how it is

I did it by adding variables to my standard RAPM-framework that capture the number of players (currently on the court for home and away) that have played a certain amount of consecutive/total minutes.

As such, player skill is accounted for. 'Effect of being up X' is also accounted for

Here's the table/chart for table 'Effect of playing X consecutive minutes'

Code: Select all

╔═════════════════╦═════════╦═════════╗
║ Cons.Min played ║ Offense ║ Defense ║
╠═════════════════╬═════════╬═════════╣
║ “<2”            ║ -1.2    ║ -0.9    ║
║ “2-4”           ║ 0.2     ║ -0.2    ║
║ “4-6”           ║ 0.6     ║ -0.2    ║
║ “6-8”           ║ 0.8     ║ -0.1    ║
║ “8-10”          ║ 0.9     ║ -0.3    ║
║ “10+”           ║ 0.8     ║ -0.3    ║
╚═════════════════╩═════════╩═════════╝

It's interesting to see that players perform below average in the first two minutes of their stint. Since the effect on defense seems to be very similar to the effect on offense I can't blame the offensive performance in the "<2" minute window on the fact that there had to be at least one dead ball situation (to make the substitution). I'm guessing players have to 'get into the game', or something. Letting the players play 6-10 consecutive minutes appears to be optimal

and the table for 'Effect of X total minutes played'

Code: Select all

╔═══════════════════╦═════════╦═════════╗
║ Total Min. played ║ Offense ║ Defense ║
╠═══════════════════╬═════════╬═════════╣
║ “less than 40”    ║ -0.4    ║ 1.0     ║
║ “40-44”           ║ 1.0     ║ -1.2    ║
║ “44+”             ║ 1.5     ║ -1.8    ║
╚═══════════════════╩═════════╩═════════╝

One would have probably guessed that there's a negative effect on offense and defense when playing 40+ minutes, but it seems the negative effect seems to be for defense only. There's a strong inverse correlation for offense and defense here.
Obviously this analysis also suffers from survivor bias. My guess is that players only play 40+ minutes in close games, and probably moreso when their team is slightly behind. This might make them (incorrectly?) think they have to spend all/most their energy on offense and their defense subsequently suffers. Not sure though

Please note that I was trying to capture effect on performance only. Injuries are a whole nother story

[Mike G]

Are all players thrown into these bins?
Obviously most players do not play 40 minutes; most elite players do, at least sometimes. So, if your <40 sample is everyone but elite players in close games -- as would be expected, they're slightly below avg.

Likewise, there are many weak players who only go in for 5 minutes or less. Are they disproportionately represented in the 0-2 and 2-4 minute samples?
Interesting in any case.

I wonder how player performance changes with total minutes in 36 hours, i.e., in back-to-backs.

[J.E.]

Are all players thrown into these bins?
Obviously most players do not play 40 minutes; most elite players do, at least sometimes. So, if your <40 sample is everyone but elite players in close games -- as would be expected, they're slightly below avg.

Likewise, there are many weak players who only go in for 5 minutes or less. Are they disproportionately represented in the 0-2 and 2-4 minute samples?
Interesting in any case.

I wonder how player performance changes with <total minutes in 30 hours>.

Mike, player skill/strength is accounted for in all cases due to this being the standard framework that I use to compute RAPM. Weak players are certainly disproportionately represented in the 0-2 and 2-4 bins, but we "know" these players are weak - unless they play a very tiny amount of minutes in my 17-year sample and the mean regression RidgeRegression leads to them being overvalued. I don't think that's very likely, though

[SportsTrip]

Hi,

I don't know how complicated it would be, but couldn't you compare each player against himself and then take the mean and standard deviation of all players as the information? (e.g. does LeBron have a better rating if he gets subbed after 6-8 or after 10+ minutes. Or the same for a benchwarmer and low minutes. And then merge everything together)
Would probably at least be a good comparison.

One problem I see, is that players probably sometimes get more minutes because they have a good game. So in hindsight they get put into the 44+ minute corner because their first 40 minutes were good. Or that players get pulled after less than two minutes because they were terrible.
Not sure if it is possible, but you could as well take time bins for players that play 44+ minutes and then show in which of these bins they perform well. Maybe then you also see a dropoff for the last 4 minutes which would now be hidden behind stellar play beforehand.

It's definitely interesting and cool stuff, but I'm not sure if the correlation/causation part is very clear. I'm also jealous that I'm not capable of such things

Cheers,
Hannes

[J.E.]

Hi,

I don't know how complicated it would be, but couldn't you compare each player against himself and then take the mean and standard deviation of all players as the information? (e.g. does LeBron have a better rating if he gets subbed after 6-8 or after 10+ minutes. Or the same for a benchwarmer and low minutes. And then merge everything together)

That's so similar to what I'm doing already, I can't imagine that it would produce different results

One problem I see, is that players probably sometimes get more minutes because they have a good game.

I thought about that, too. But I haven't seen any research that suggests a player's past in-game performance has an influence on his future in-game performance. In fact, on a team basis, the opposite effect seems to exist. What seems likely (and doesn't contradict the effect when teams as a whole play well) is that the player with 40+ minutes had played well, and most others of his team had played badly. Then you're in the situation I described earlier: Close game, probably slightly behind - and the coach decides to go with the few players that had a good game so far
I probably don't care enough to dig that deep into it, though. I'll probably look at 'effect of foul trouble' and then call it a day

[v-zero]

One problem I see, is that players probably sometimes get more minutes because they have a good game.

I thought about that, too. But I haven't seen any research that suggests a player's past in-game performance has an influence on his future in-game performance.

If we assume then that a player who has had a good game up until time T will on average play to their mean from that point onwards, then their overall performance for that game will still be better than their average, so the fact that players who have good games tend to get more minutes in those games is likely to be a confounding issue.

[kggk]

Does it mean anything that the defense line is always below zero? The way I'm reading it now it seems like players are always below average on defense, which doesn't make any sense.

[J.E.]

Does it mean anything that the defense line is always below zero? The way I'm reading it now it seems like players are always below average on defense, which doesn't make any sense.

No it doesn't. Only the relative difference is important. Absolute coefficient values can be wacky sometimes with Ridge Regression, as they (sometimes) get shifted around in weird ways

[J.E.]

A quick word on the effect of foul trouble on player performance: I can't really find too much of an effect. Yes, having 5 fouls in the 4th appears to have a negative influence on defensive performance, but most other constellations (X fouls in quarter Y) are all over the place*, and none of them has a higher absolute value than 2 (p100p)

*by that I mean that the order often doesn't make sense, e.g. the coefficient for '3 fouls in Q2' is higher than for '2 fouls in Q2'

It's important to note that I did this analysis with all players. I have no doubt that "star" players (players that play lots of minutes, or have a high PPG) may be more concerned about their foul trouble and defend significantly worse than a 15MPG bench player who's in the same foul trouble

[Crow]

Bump. This topic came up once or several times recently but I don't think this thread was referenced. fyi, if interested in topic.

[Nate]

Is the data set big enough to control for in-game time? As the stints get longer, the interval that they might cover gets more restrictive, and I'd expect that the players who get the longer stints are mostly starters.

[jaynopp]

Good stuff, but I don't like the 'get into the game' explanation for the initial drop in performance and your quotation marks tell me you don't really buy it either. I think this could touch on an issue I've had with basketball for a while now, the inactivity of players on the bench. It's almost universal across sports that an athlete will engage in some physical activity before competing. Soccer players will warm up on the sidelines a few minutes before subbing in, it seems like NFL players are always riding the stationary bike, and Olympic swimmers who compete in multiple events will be swimming laps in between races. Yet the only physical activity basketball players do on the bench is getting a massage or the occasional cheer after a good play. Without getting too into the specifics of human biology, generally speaking it takes about 2-3 minutes of physical activity before the body is able to consume its maximum amount of oxygen, a time frame which would be consistent with the data you collected. It also takes 20-30 minutes of rest for oxygen consumption to return to resting levels, thus I would expect a more significant drop in the performance of players who have spent a long time on the bench and a minimal drop in players who take shorter rests. It would be tough to measure this precisely in terms of real time vs game time because of stoppages, but it might be something to look into.

[AcrossTheCourt]

This says some important things about low minute players, doesn't it? And this goes back to the Millsap effect and how people can underrate bench guys by dismissing them entirely. My counter was always that it's tough coming off the bench cold and performing in four minutes before being yanked again.

Some of this is mitigated by playing against a weaker lineup or in garbage time. But it's an interesting adjustment: a guy playing 500 minutes a year in spurts and fits moving up to 1800 minutes with well-defined stretches should produce more.

[NateTG]

... It would be tough to measure this precisely in terms of real time vs game time because of stoppages, but it might be something to look into.

There are other plausible explanations - players need to find their role on the court too.

Regardless, this seems like something that could be tested with biometric data and in-practice simulations.

[jaynopp]

Regardless, this seems like something that could be tested with biometric data and in-practice simulations.

With a little digging I came up with this study http://www.ncbi.nlm.nih.gov/pubmed/22262012 (abstract only) which shows some of the physiological effects of coming in cold off the bench. I wouldn't attribute the drop in initial performance exclusively to this, but compared to other possible factors (like getting into the flow of the game) warm-up time would be much easier to control.