Spacing Rating

Home for all your discussion of basketball statistical analysis.
Jacek
Posts: 2
Joined: Wed Dec 14, 2016 9:35 am

Spacing Rating

Post by Jacek » Thu Jul 06, 2017 8:38 pm

Nicholas Sciria from Nylon wrote txt about new metric - spacing rating. What do you think about it? What would you change in this metric? http://fansided.com/2017/07/05/nylon-ca ... berwolves/

jgoldstein34
Posts: 240
Joined: Fri Sep 30, 2016 6:38 pm

Re: Spacing Rating

Post by jgoldstein34 » Thu Jul 06, 2017 10:04 pm

I like the idea behind it and the methodology is sound. I think it might miss on guys who get respect spacing without actually being good shooters i.e. Wade or Westbrook or LeBron. It's a good way to guess spacing at the very least. With positional tracking data, it would all be moot because we'd know how closely a defender stays to the offensive player and could really calculate gravity of players and use that to figure out true lineup spacing.

Nate
Posts: 123
Joined: Tue Feb 24, 2015 2:35 pm

Re: Spacing Rating

Post by Nate » Fri Jul 07, 2017 1:09 pm

I'm not sure that "spacing" is a well-defined concept. That said, the article doesn't do a very good job of justifying 3 point rate as a proxy of some kind of spatial thing, and I pretty much lose the plot there.

I have to agree with jgoldstein that "spacing" (whatever that term really means) seems very much like a topic that should be attacked using position data.

I would also like to see better clarification about what "spacing" means so that metrics can be falsified.

jgoldstein34
Posts: 240
Joined: Fri Sep 30, 2016 6:38 pm

Re: Spacing Rating

Post by jgoldstein34 » Fri Jul 07, 2017 5:16 pm

Agreed, he just sort of says 3pt shooting and attempt rate matters and is good without justifying it entirely. While I think it does matter, I wish there was at least a regression showing correlation with offensive rating or something. Conceptually, 3pt shooting matters and increased spacing should result from that, just don't think he really proves that.

Rd11490
Posts: 111
Joined: Mon Sep 29, 2014 4:54 am

Re: Spacing Rating

Post by Rd11490 » Mon Jul 10, 2017 4:20 pm

I'm going to hijack this thread to discuss what people think should go into a spacing metric using the player data. I started playing around with the 2015-16 movement data and have a couple of ideas for a metric but wanted some input before I spun up a cluster to process an entire season of data.

A couple of ideas that I have:

Simple:
Average distance between player and closest defender over the course of all plays
- Closest defender or assign defenders based on current positioning of all players (i.e. one defender can not be assigned to two offensive players)
-Pretty sure this stat is already out there. Gravity?

More complex:
Allowed Space = area of smallest connected pentagon of offensive players - Defensive area of smallest connected pentagon of offensive players
- Average over the possession / at time of shot?
-Treat this similar to raw plus-minus?
-Calculate some form of AdjustedAllowedSpace?

I'm open to any suggestions, critiques, ideas, etc.

Crow
Posts: 5928
Joined: Thu Apr 14, 2011 11:10 pm

Re: Spacing Rating

Post by Crow » Mon Jul 10, 2017 4:41 pm

If you go with space to nearest defender approach, the model should value where the offensive player is in returns of expected ppp. I.e. being near rim or 3 pt line with space is probably more valuable than floating in mid-range with space.

If you go with offense / defensive polygons or both or net how would you apply plus-minus technique? Might still want to compare offensive area to sum of the expected ppp of the offensive players (on ball different from off ball). Try to get at effective space (a product) or maybe what you called adjusted allowed space. Space itself shouldn't be the goal. It might be overstressed at the expense of actual impact. If you wanted to make the model more elaborate you could try to consider spacing pattern impacts on the non shooting or scoring factors, turnovers and offensive rebounds. At some point you have to consider the global impact of spacing beyond shooting / scoring efficiency. Including impact on next defensive possession, made basket or miss.

Ultimately it matters most at time of shot / defense but you could look to see if high momentary net offense values early in play tend to increase end values. I.e. big threat leads to defensive scramble, deterioration even if not immediately capitalized on. If the offense knows what it is doing. Which offenses are missing their peak advantages most or improving their standing least (or least per second)? Which defenses lapse into high threat most, who recovers, who gets burned most? Does study of entire possessions help understand strengths / weaknesses more than just actual outcomes? Could go either way. Look beyond randomness or get confused by non-boxscore noise? Understanding processes beyond just outcomes is probably worthwhile effort, if done right and with adequate respect for outcomes.

If folks are going to study spacing, it should probably get taken to play / set level. What plays produce max space or max effective space value? What plays are really good half plays and could be used for a different end point? Which leave good improvisation options after they end, especially against well-run defenses? Plays need to more than just "look good", be clever, have one video proof of working, etc. They should be scored on results and potential, potential for even better results. How much should plays be run toward one goal outcome vs. multiple decent or better outcomes? Depends on expected PPP and % of times you can get / actually use that option. Plays that depend on a non top 2 the am decision maker to make a key decision are mor suspect than plays where they know what to do and have high likelihood of execution or plays that rely on a top 2 decisionmaker and to make the choice.

jgoldstein34
Posts: 240
Joined: Fri Sep 30, 2016 6:38 pm

Re: Spacing Rating

Post by jgoldstein34 » Mon Jul 10, 2017 5:04 pm

I really like you're idea for the allowed space and treating it as a plus-minus. It would effectively be looking at how the spacing changes with different offensive/defensive players to try and work out individual impact, yes?

Rd11490
Posts: 111
Joined: Mon Sep 29, 2014 4:54 am

Re: Spacing Rating

Post by Rd11490 » Mon Jul 10, 2017 5:25 pm

I really like you're idea for the allowed space and treating it as a plus-minus. It would effectively be looking at how the spacing changes with different offensive/defensive players to try and work out individual impact, yes?


Yeah, i'd just run ridge regression similar to APM
If you go with space to nearest defender approach, the model should value where the offensive player is in returns of expected ppp. I.e. being near rim or 3 pt line with space is probably more valuable than floating in mid-range with space.


Does this data exist in the public domain? I know I can calculate based on shot data from that year as I have access to it. I think I added the shot chart endpoint to my API wrapper last week, so I should be able to pull it all, but would prefer to just upload a csv to s3.

Ultimately it matters most at time of shot / defense but you could look to see if high momentary net offense values early in play tend to increase end values. I.e. big threat leads to defensive scramble, deterioration even if nit immediately capitalized on. If the offense knows what it is doing. Which offenses are missing their peak advantages most or improving their standing least (or least per second)? Which defenses lapse into high threat most, who recovers, who gets burned most? Does study of entire possessions help understand strengths / weaknesses more than just actual outcomes? Could go either way. Look beyond randomness or get confused by non-boxscore noise? Understanding processes beyond just outcomes is probably worthwhile effort, if done right and with adequate respect for outcomes.
Someone in Goldsberry's lab wrote a paper on this a few years ago. It would be a massive undertaking to recreate, but would be interesting to work on.
If folks are going to study spacing, it should probably get taken to play / set level. What plays produce max space of max effective space value? What plays are really good half plays and could be used for a different end point? Which leave good improvisation options after they end, especially against well-run defenses? Plays need to more than just "look good", be clever, have one video proof of working, etc. They should be scored on results and potential, potential for even better results.
I really like this idea, but I think we are pretty far away from being able to classify this type of stuff. I know there is some work from the sportvu team on classifying PnR, basic actions, but I don't think we have enough public data to start classifying specific plays with any form of accuracy (at least those of us in the public domain, I'm sure there's enough training data if you work for a team).

Crow
Posts: 5928
Joined: Thu Apr 14, 2011 11:10 pm

Re: Spacing Rating

Post by Crow » Mon Jul 10, 2017 5:46 pm

Yeah, somebody did a study of moment to moment Expected Value. Harvard guys and / or Dr. Phillip Mayim. Gravity&Levity looked at play decision trees. Usually room to pick up and take things further.

Crow
Posts: 5928
Joined: Thu Apr 14, 2011 11:10 pm

Re: Spacing Rating

Post by Crow » Mon Jul 10, 2017 5:52 pm

How do you create space?

Dribble
Pass
Speed
Agility (body and / or ball)
Stopping
Acceleration
fake drive
Ball distraction
Lull to sleep
Stealth changes in orientation / creeping
Picks
Leading passes
Body leverage (arms, shoulders, elbows, butt, whatever)
Fakes & multiple maneuvers
Length
Hops

what else?

other player gravity / motion / threat
footwork, including stepback, sidestep, runner, leaner, spin, eurostep
quick release or shot fake
pass fake
palm ball
run defender into somebody
hesitation dribble / crossover, back peddle
doing anything different than expected, by degree or radically.

Nate
Posts: 123
Joined: Tue Feb 24, 2015 2:35 pm

Re: Spacing Rating

Post by Nate » Mon Jul 10, 2017 8:14 pm

Rd11490 wrote:I'm going to hijack this thread to discuss what people think should go into a spacing metric using the player data. I started playing around with the 2015-16 movement data and have a couple of ideas for a metric but wanted some input before I spun up a cluster to process an entire season of data.

A couple of ideas that I have:

...

I'm open to any suggestions, critiques, ideas, etc.
I'd be more inclined to go with some kind of voronoi tesselation model where you assign value to each spot on the floor, and then give the player that's closest to that spot credit for controlling that spot on the floor. I'll spend some time thinking about how to sensibly incorporate ball location and player dynamics.

A somewhat simpler model might be an open man model that tracks how often there's an open man (just using proximity) and the rate that the ball finds that player.

A third approach would be to look at how the defense moves when a shot is taken.

Nate
Posts: 123
Joined: Tue Feb 24, 2015 2:35 pm

Re: Spacing Rating

Post by Nate » Tue Jul 11, 2017 9:14 pm

There was a Sloan paper a couple of years back where people were looking at offensive player - ball - hoop triangles to work out match-ups. I don't remember the year or title.

Something to look at that I haven't seen a lot of is relative player motion. I.e. if a player on the offense moves, which players on the defense move similarly, and, using that kind of information to identify pairings, how much space does the defense give.

A simple idea for identifying picks: Do a voronoi tesselation using just the players on offense, and see when one of them has no defensive players in his partition.

Crow
Posts: 5928
Joined: Thu Apr 14, 2011 11:10 pm

Re: Spacing Rating

Post by Crow » Tue Jul 11, 2017 9:36 pm

https://en.wikipedia.org/wiki/Voronoi_diagram

I didn't know.

How many points should the model include? I might start with about 11.



Application to rebounding http://www.sloansportsconference.com/wp ... unding.pdf

Poster
http://dcervone.com/slides/court_realty_poster.pdf
"This reveals that point guards tend to have the lowest
average portfolio values as ballcarriers, since they're often in control well
beyond the arc. Big men usually have higher portfolio values when possessing
the ball, since these opportunities usually occur close to the basket."
This seems like just a start and not necessarily compelling... but if true, it should impact play design, passing responsibilities and usage. Look out for Jokic!



Anybody aware of or want to look into this? https://github.com/jdaaph/NBAHackathon

Nate
Posts: 123
Joined: Tue Feb 24, 2015 2:35 pm

Re: Spacing Rating

Post by Nate » Fri Jul 14, 2017 2:09 pm

Rd11490 wrote:I'm going to hijack this thread to discuss what people think should go into a spacing metric using the player data. I started playing around with the 2015-16 movement data and have a couple of ideas for a metric but wanted some input before I spun up a cluster to process an entire season of data.

...
Can we discuss what people think should "come out" of a spacing metric as well? Is there something tangible (ideally something quantifiable) that we can talk about which corresponds closely to the notion of 'spacing'?

Rd11490
Posts: 111
Joined: Mon Sep 29, 2014 4:54 am

Re: Spacing Rating

Post by Rd11490 » Fri Jul 14, 2017 2:21 pm

Good point. When I started thinking about this problem, all I wanted was a way to measure the term "spacing" that so many talking heads like to use. i.e. Jimmy Butler and Taj Gibson will kill the Timberwolves offensive spacing.

Post Reply