APBRmetrics

I did something similar to this thread, where I posted "influence on teammates' PPS", but this time with turnovers.
This analysis gives us an estimate for a player's expected 'turnovers per 100 possessions, when playing with average teammates', and an estimate for 'influence on the chance that a teammate will turn the ball over'

The estimates for '12-'14 (weighed for recency) are here.

A small guide how to interpret these numbers: A 5 man unit turns it over ~15 times in 100 possessions. One player, thus, turns it over 15/5 = 3 times per 100 possessions on average. Westbrook 'leads' the league with 6.2 expected turnovers per 100 possessions. At the same time, though, his teammates, who together would be expected to turn it over 15-3 = 12 times with an average teammate, are only expected to turn it over 12 - 2.6 = 9.4 times with Westbrook on the court. So, Westbrook decreases each teammate's chance to turn it over by 2.6/4 = 0.65.
That's better than, say, DeMarcus Cousins, who turns it over 5.5 times, but only decreases each teammate's chance to turn it over by 0.2

Topping the list with good influence on teammate's turnovers are the high usage players - the other players simply aren't asked to handle the ball as much, decreasing their turnovers.
R^2 for 'individual turnovers' and 'influence on teammates TO' is 0.26
Some players manage to decrease their teammates' turnovers while at the same time not turning it over themselves at a high rate: Conley, Lawson, Lillard, J. Johnson. Others have a bad influence on teammates' turnovers and turn it over a lot themselves: Thabeet, C. Zeller, Gobert, Asik..

I wanted to this on the 'touches' level but since I, unfortunately, don't have SportVU data I had to do it on the possession level. With this framework, every possession gives us five rows in the X matrix (instead of just 1), blowing this matrix way up, to the point where I can't run all data together, not with 16G of RAM and using sparse matrices, so here is 'just' '08-'14

awesome work J.E. Im assuming you are working on some 4 factors RAPM which is why you have posted adjusted turnovers and adjusted shots. Are you working on an update of the adjusted rebounding study that you did a few years ago?

colts18 wrote:Are you working on an update of the adjusted rebounding study that you did a few years ago?

Not at this moment, but 'adjusted rebounding with influence on teammates' is on my to-do list.

I'm not going to do a FT/FGA RAPM though. I've included FTs in 'adj PPS'

Very very nice. Along with the effect on PPS work that you did, it's a natural logical application of RAPM techniques to look at players' influences in more detail.

Is rebounding next? That would not only give you coverage of the Four Factors (as you already noted, the FTs are subsumed within the PPS calculations), but also shed light on the diminishing returns question: are players helping their teams get more rebounds, or are they simply taking rebounds away from teammates?

mtamada wrote:Very very nice. Along with the effect on PPS work that you did, it's a natural logical application of RAPM techniques to look at players' influences in more detail.

Yeah I suppose we could now look at 'influence on teammate ..' for pretty much anything. Steals, dunks, blocks.. whatever. Not all of them make too much sense though (I'm guessing the 'influence on teammate' for some of these is 0). It makes sense to do for the 4(3) factors but beyond that... I don't know. Let me know if you think of something that could be interesting

Is rebounding next? That would not only give you coverage of the Four Factors (as you already noted, the FTs are subsumed within the PPS calculations), but also shed light on the diminishing returns question: are players helping their teams get more rebounds, or are they simply taking rebounds away from teammates?

I think we're definitely going to see that good rebounders reduce the number of rebounds their teammates get. That, I think, is not necessarily a bad thing though. Imagine a player that grabs 100% of the defensive rebounds - he's going to a have a negative influence on teammates' rebounding numbers fore sure, but no one would consider that to be a problem.
I think the problem is more with players that are like Asik with turnovers: Lots of individual turnovers and making others turn it over more - although for rebounds it would obviously be reversed: few individual rebounds but also reducing teammates' rebounds (i.e. this player rebounds the misses his team would have gotten anyway, but doesn't rebound the misses the opponent might have gotten)

In theory, will these numbers: http://stats-for-the-nba.appspot.com/ra ... _2014.html

sum to these numbers (when centered): http://stats-for-the-nba.appspot.com/ra ... dj_to.html ?

Seems like they ought to, but they don't seem to at the moment.

DSMok1 wrote:In theory, will these numbers: http://stats-for-the-nba.appspot.com/ra ... _2014.html

sum to these numbers (when centered): http://stats-for-the-nba.appspot.com/ra ... dj_to.html ?

Small differences can appear here and there because I made some changes to the penalty values. Before I had the same penalty value for offense and defense, now I have one penalty value for 'individual turnovers', 'influence on teammates' and 'forcing turnovers' each. The biggest change is probably in 'forcing turnovers' where the old penalty value was probably a little to low (not enough shrinkage).

That being said, I don't see crazy differences (>50) in rank. If I had to choose I'd probably trust the new version a little more

J.E. wrote:[I think we're definitely going to see that good rebounders reduce the number of rebounds their teammates get. That, I think, is not necessarily a bad thing though. Imagine a player that grabs 100% of the defensive rebounds - he's going to a have a negative influence on teammates' rebounding numbers fore sure, but no one would consider that to be a problem.

Right, for rebounding the interesting statistic would be "net rebounds gained". If the Trailblazers generally get 67% of the defensive rebounds, but when Bill Walton is on the floor they grab 100% (and he grabs all of them) then he has (to put it simplistically) a -67% impact on his teammates, but he's grabbing 100% himself, for a net +33% impact.

More realistically, conceivably two players could have identical defensive rebounding percentages, but one of them might be having more of a negative impact on his teammates' rebounding. Another extreme example: Dennis Rodman insisting on grabbing all missed opponent FTs himself -- his teammate lined up across the key would never get a def rebd off a missed FT, whereas most players would grab defensive rebounds aplenty if stationed next to the basket during FTs.

Looking at individual turnovers + influence on teammates on offense, 58% of players have a combined TO rate above league average. About 18% are more than 1 turnover per 100 possessions higher than average. About 5% are more than 1 turnover per 100 possessions lower than average.