SSAC Papers

[EvanZ]

http://www.sloansportsconference.com/?page_id=462

Some interesting stuff. Thought we should have a thread for discussion when people get around to reading the papers. I'm making this sticky at least through the conference.

[kjb]

Thanks for posting the links and starting this thread.

Just went through the Spatial Analytics paper. Thought it contained really interesting data and took an interesting approach. I thought it was a bit preachy about the need for spatial analytics -- especially since Goldsberry didn't acknowledge shot location charts that many NBA teams use as part of their scouting process. I also thought his summary of how court locations are talked about by "analysts" was strawman stuff.

Also, I know it's minor, but it bugged me to see the author refer to efg as "Enhanced Field Goal Percentage."

As for the actual research -- I thought it was good, albeit unsurprising. Guys who shot from a lot of locations were basically perimeter players who shot a lot period. I liked his Range measure, which looked at effectiveness from a variety of locations. I wonder, though, if dividing shot locations into 1x1 slices might be a tad small. Looking at the shot charts for the 4 best "Range" guys, there appear to be definite patterns by location.

Do Range scores change if you expand the "locations" to 2x2 segments? 4x4?

Finally, I'd love to see this kind of analysis applied to positioning of players off the ball, passing, screen-setting, defense, etc., etc., etc. This is interesting research, though.

[EvanZ]

That's the paper I noticed immediately. Haven't read it yet though. If I had those data I would calculate adjusted shot location.

I can do something similar, if I code inside shots as, say 3ft, mid-range shots as 15 ft (or wherever the centroid of those shots is), and 3-pt shots as 23 ft.

[mystic]

Well, I agree with kjb, the results are hardly surprising at all. But what I find a bit odd is the conclusions he is drawing from various things. How will he make a difference between a player who is able to get to his sweet-spot on the court more often and a player who just shots whenever he likes? Bryant takes shots all over the place, because he is often not able to get a better shot location. Bryant is inefficient in 688 segments in which he takes shots. Those charts aren't telling us a lot more about his shooting abilities than by just looking at his amount of midrange or longrange shots. It is also rather odd to pick >1 as determining factor, given the fact that this means 33% on 3pt shots are making a player "effective" and implying that someone who can hit 40% from 3pt range is as good from the same spots.

The efficiency part is not highlighted enough. And someone who takes a lot of shots from different location is better be a good "Range" player. How about the ratio Range/Spread. That might as well tell us something about the ability of a player to get to spots with a higher success rate rather than just taking a bad shot from everywhere on the court and actually implying it would be per se a good thing. Someone who is able to convert shooting opportunities at a higher rate might as well be a better shooter overall.

[kjb]

I think one of the issues with the way he's slicing the locations is the size of the box. It might be better to use something like this:

http://www.rawbw.com/~deano/images/scorloc.gif

Than 1x1 squares. Or to analyze shot location groupings. Even with Kobe, there appear to be shot location groupings when I look at the chart.

[kjb]

Just got through the Big 2's and Big 3's paper, which is looking primarily at fit between a team's best players. I think he has a good approach, but...I'd like to see a similar approach using per minute stats instead of per game. Also, he named the groups of players Cluster 1, Cluster 2, Cluster 3... and so on. Might have been better to come up with titles that characterize the type of player in each cluster. He gives definitions in the paper, but I got lost in Cluster soup.

Finally, maybe I missed it, but I didn't see a complete conclusion or presentation of the regressions for each cluster. I'd be very interested in seeing that.

Still, interesting research that takes a good whack at player fit. I agree with his underlying assumption that fit is a critical part of roster construction.

[EvanZ]

Thanks for posting the links and starting this thread.

Just went through the Spatial Analytics paper. Thought it contained really interesting data and took an interesting approach. I thought it was a bit preachy about the need for spatial analytics -- especially since Goldsberry didn't acknowledge shot location charts that many NBA teams use as part of their scouting process. I also thought his summary of how court locations are talked about by "analysts" was strawman stuff.

Eli Witus looked at shot locations several years ago. Is there overlap in this paper with his work? Was he cited?


http://www.countthebasket.com/blog/2008 ... ake-shots/

[kjb]

I'd forgotten about Eli's analysis. Yes, it looks like there's some overlap. No, Eli is not cited. Eli's data is for the entire league -- Goldsberry is looking at individual players. But, it's interesting to see that Eli eventually went with 3x3 squares after trying 1x1 to "reduce noise."

[Crow]

Evan, by "adjusted shot location" did you mean adjusted scoring impact based on shot location or something close to that?

Literal adjusted shot location might be interesting, though maybe different- for every shot location look at a player's eFG% then match that to a distance / location from which the league on average shoots that eFG%. So if you shot 12 footers the way the league shoots 20 footers the adjusted shot location would be 20 feet. Is that anything like what your adjusted shot location idea would do? Maybe this is too convoluted but I thought I'd at least throw it out there...

[Crow]

The panel on sports betting might be worth mentioning to those involved in the market or just tracking it.

[EvanZ]

Evan, by "adjusted shot location" did you mean adjusted scoring impact based on shot location or something close to that?

Literal adjusted shot location might be interesting, though maybe different- for every shot location look at a player's eFG% then match that to a distance / location from which the league on average shoots that eFG%. So if you shot 12 footers the way the league shoots 20 footers the adjusted shot location would be 20 feet. Is that anything like what your adjusted shot location idea would do? Maybe this is too convoluted but I thought I'd at least throw it out there...

I actually meant adjusted the same way as in APM. The dependent variable would be shot location instead of point margin.

[Crow]

I like the use of 14 clusters in theBig 2's and Big 3's paper. If I have time I will lineup this data up as well as possible with past efforts by Ed Kupfer and David Sparks for fewer clusters. Unfortunate that NBA efficiency was used. All or almost all of the major modern boxscore metrics are better. When will the NBA update it, if ever?

[kjb]

I liked the clusters too, but thought the paper would have been more easily understood if he'd used descriptive names for the categories instead of Cluster 1, Cluster 2, etc. Without referring back to the paper, what's a Cluster 1 player? Hell if I know.

[KirkG]

Hi guys,
I wrote the CourtVision paper...

I hope to meet some of you at the Sloan conference.

this is my first post - I love the board and it's been cool to see you guys discussing my project; I've definitely appreciated your feedback.

In response to kjb, you raise many good points! I'm not sure if the 1x1 squares are too small or not; they may be... My goal was to reveal which players are effective from the most court spaces, and I think my analysis did a reasonable job of that. However, I ran a quick analysis of the zones you suggested from this image: http://www.rawbw.com/~deano/images/scorloc.gif - *I've shared some interesting results below*. My issue with this kind of partitioning is that it is quite coarse. For instance, I know from my own awful shooting performances, I am much more effective from certain places within Zone A than others. There is considerable "intrazone" variability within those zones. How should we partition the court for optimal shooting analysis? To be honest, I'm unsure what the best partitioning strategy would be. Also, to be clear, I'm not familiar with the "shot location charts" used by NBA teams; is there a place where I could see some examples?

In response to Mystic, I like your range/spread idea. I have been thinking about that a lot. The top 3 guys in that metric are: Nash (.62), Calderon (.57), and Ray Allen (.56); it's really quite amazing how far Nash is ahead of everyone else, but what does this metric really tell us?

Most efficient shooters in kjb's zones: http://www.rawbw.com/~deano/images/scorloc.gif
Zone 1: Brand, Nash, Garnett
Zone 2: Lebron, Calderon, Dwight
Zone 3: Joe Johnson, Nash, Okur
Zone A: J. Terry, N. Robinson, Calderon
Zone B: Nash, Mo Williams, David West
Zone C: Nash, Daniel Gibson, Dirk
Zone X: Anthony Morrow, S. Curry, Steve Blake
Zone Y: Troy Murphy!, Martell Webster, Boris Diaw
Zone Z: S. Curry, A. Morrow, A. Parker

kg

[kjb]

Kirk -- I have a different chart, but I don't think it'll yield substantially different results than the one DeanO has up on his old site. I don't know the right size of the box, but Eli's 3x3 feels about right. Just my opinion.

I like the overall approach, though. Would really like to see the next step -- taking it to passing, off-ball positioning, screen-setting, etc.