Measuring situational shooting, when players call their own number?

[TeamEd]

Hi there.

Earlier this season, I noticed something watching the Raptors. This was that when the team fell behind, Kyle Lowry seemed more likely to drive or pull up. Conversely, when the team was ahead he seemed more comfortable running plays for his teammates.

From this observation, I looked at his in game splits at NBA.com http://stats.nba.com/player/#!/200768/stats/. At a glance, they confirmed that he did shoot more often when trailing. This was true even though the Raptors had a very positive MOV and at the time he had played many more minutes while ahead.

This got me thinking about a way to measure his, and other NBA players’ situational shooting and passing tendencies. This in turn took me down an Excel rabbit hole that ended with me coming up with a formula for what I’ve been calling “situational shooting mentality.” I introduced this idea over on the forums at raptorsrepublic.com.If you want, you can go look at how my thinking developed there: http://www.raptorsrepublic.com/forums/showthread.php?14417-Kyle-Lowry-and-the-Raps-come-from-behind-offence But, now that I’ve settled on an idea that seems to work I want feedback from people who actually care about this stuff.

So, hello RealGM forums! Be nice. Or don’t, whatever.

Situational Shooting Mentality measures how much more often a player takes true shooting attempts in comeback situations relative to his overall distribution of minutes. I think of it as measure of a player’s tendency to say “I got this” to spark a comeback.

The basic formula for SSM is:

((MP (behind/tied) ÷ MP) - (TSA (behind/tied) ÷ TSA)) ÷ (MP (behind/tied) ÷ MP) x -100, where MP is minutes played and TSA is True Shooting Attempts defined as FGAs + FTA x 0.44.

This basic formula produces a percentage that shows how much more likely a player is to take a shot in a comeback situation vs. an even distribution of shots. It is easy to adjust the measure for any situational split. I could use this to examine clutch shooting tendencies, for example. I settled on looking at behind/ tied tendencies for reasons of practicality.

Using this, a high SSM tells you a player increases his shooting when behind, and a negative SSM tells you he takes less attempts. The same formula works if you replace TSAs with assists, with the caveat of much smaller sample sizes. I’ve been calling that measure the Situational Passing Mentality.

Early on, I found one problem with this formula besides how labour intensive it is to pull the data. That is that there is a league-wide tendency for all players to take more TSAs while trailing. The trailing team produces around 5% more TSAs, on average (this trend exists even if I exclude TSAs taken while tied).

This is interesting by itself, but it’s not the point of my measure. Since the point of the measure is to say how much more likely is a player to take a TSA against expectations, I added a simple adjustment against league average. This depresses everyone’s SSM, but creates a more logical zero point.

You can see why I did this by comparing a chart of basic teamwide SSMs versus the same chart adjusted to league average.





The corrected formula is therefore the player or team’s basic SSM minus league average SSM.

Here’s a chart of what SSM and SPM look like in practice. These are results from last week for a run of players with more than 10 FGA a game, plus a selection of ballhandling guards who don't quite fit that criteria. The cutoff for the SPM chart is those players in that sample with, IIRC, more than ~7 assists per 100 possessions.



As you can see, I was correct in noticing that Kyle Lowry shoots more often in comeback situations, though he’s not at the top of that list. Unsurprisingly, Kobe takes many more shots and records many many fewer assists while trailing. Meanwhile while a player like John Wall takes far fewer TSAs in the same situation, while recording assists more frequently.

Here's the same two charts for teamwide numbers:



You can see here that the Cavs fail to produce both TSAs and assists when trailing, relative to league average. The tendency is especially marked in the assists category -- note the much wider leaguewide variation in the scale.

So, what do you think? Does this measure make sense? Is it at all valuable? What mistakes am I making? Is the name stupid?

I have a sense of some areas you might think this measure fails to achieve what I want it to, but I’d like to hear your honest feedback without my guidance.

UPDATE: Due to a glitch in my Excel sheets, the numbers in these charts are incorrect. Oops.

[TeamEd]

Ok, here's a look at how this stat can be useful. I've done a full new run of the numbers, fixing a dumb mistake that made the previous set almost completely useless.

---

The Wizards' John Wall and the Raptors' Kyle Lowry are both driving PGs who rack up assists and get to the line.

They're both advanced stats darlings with > 20 PERs, > 4.5 Win Shares, stellar NetRtgs, top 20 Real +/- and top 10 WAR.

Wall passes more. Lowry scores more. But otherwise, by the old and the new numbers they look like similar players.

And of course, Lowry and Wall are both Eastern Conference All-Star starters. #NBABallot

But this is all misleading. You wouldn't notice by the numbers, but Lowry and Wall are very different point guards.

How? It's about mentality: When Wall and Lowry are trying to lead their teams to a comeback, they take their offence in opposite directions.

With the Raptors trailing, Lowry shoots more often than he normally does--a healthy 7 per cent more than his overall rate. *

Wall meanwhile actually shoots less. He decreases his shooting rate by 8 per cent when behind. *

In fact, if you rank the 50 players who've shot the most by how their shooting mentality changes in comebacks, Lowry is #1 and Wall is #50--by a fair margin:



They pass differently in comebacks too.

Wall increases his assist rate 9 per cent when the Wizards are trailing and Lowry decreases his by 7 per cent when the Raps are behind. *

Of the 30 NBA players with the most total assists, Wall has the 5th most aggressive passing mentality when trailing.

Lowry meanwhile decreases his assist rate in comebacks more than anyone on the list (except Lebron):



So, even though their games are similar overall, Lowry and Wall are polar opposites when their teams' need them.

In comebacks, Wall becomes a consummate team player while Lowry says "I got this," and goes to the rim.

Funny thing: both approaches work. With Lowry on the court, the Raps are +5.9 in +/-per48 overall and +16.1 when trailing.

Meanwhile, with Wall playing the Wiz are +7.0 in +/-per48 overall and +22 when trailing. **

* As before, these rates are adjusted for the fact that the average NBA player shoots and records assists more often when trailing. The average NBA player is 5.4% more likely to take a true shooting attempt when trailing. He's also 3.9% more likely to record an assist. For unadjusted numbers, Lowry increases his True Shoot Attempt rate by 12.3% when trailing. Wall decreases his by 2.6%. Wall increases his unadjusted assist rate by 13.2 while trailing. Lowry decreases his by 3.0.

** There's also a sizeable league-wide tendency to record a better +/- while trailing.

[blabla]

Very interesting and cool stuff. Good job. Interesting that M.Gasol would shoot so much and Jennings so little

[Knosh]

So just so I understand what you are looking at: Being behind means being down 0-2 10 seconds into the game or being down 20 with 1 minute left or being down 2 points with <24 sec left. It's all the same?

I

[TeamEd]

So just so I understand what you are looking at: Being behind means being down 0-2 10 seconds into the game or being down 20 with 1 minute left or being down 2 points with <24 sec left. It's all the same?

I

Yes.

[Illmatic12]

Very cool, you should post this on reddit r/nba. Nice clean and readable charts/graphics too

[Knosh]

So just so I understand what you are looking at: Being behind means being down 0-2 10 seconds into the game or being down 20 with 1 minute left or being down 2 points with <24 sec left. It's all the same?

I

Yes.

So when I read your intro:

Earlier this season, I noticed something watching the Raptors. This was that when the team fell behind, Kyle Lowry seemed more likely to drive or pull up.

I was thinking that I noticed that too, but really I was more thinking about situations where the Raptors were down by not too much at the end of games. So clutch situations really. So did you really want to look at being behind vs. being ahead (even if it's in blowout situations or at the start of games) or were you more thinking clutch situations too? I'm not watching that many Raptors games, so it could be that it's the first, but if you wanted the latter, you probably have a lot of noise in the data.

Btw, forgot to say: Very interesting post and thanks for sharing!

[TeamEd]

So just so I understand what you are looking at: Being behind means being down 0-2 10 seconds into the game or being down 20 with 1 minute left or being down 2 points with <24 sec left. It's all the same?

I

Yes.

So when I read your intro:

Earlier this season, I noticed something watching the Raptors. This was that when the team fell behind, Kyle Lowry seemed more likely to drive or pull up.

I was thinking that I noticed that too, but really I was more thinking about situations where the Raptors were down by not too much at the end of games. So clutch situations really. So did you really want to look at being behind vs. being ahead (even if it's in blowout situations or at the start of games) or were you more thinking clutch situations too? I'm not watching that many Raptors games, so it could be that it's the first, but if you wanted the latter, you probably have a lot of noise in the data.

Btw, forgot to say: Very interesting post and thanks for sharing!

Well, clutch is worth looking at, but the sample sizes are much lower in clutch situations.

There's nothing to stop me from doing a run looking at last 5 mins +/- 10 pts, or whatever. The formula is going to be essentially the same, since all this is a simple percentage change in attempt/assist rate corrected for league average. The problem is I expect I'll end up with wildly unreliable data due to small samples if I just look at clutch situations (That's a huge problem with almost all clutch stats).

Doing it this way, even though I may be capturing situations that are uncompetitive I'm at least looking at a large sample of ~50% of attempts. And really, the only times a game gets truly uncompetitive is if a lead is >20. And even then, that's only really the case if it's also under about 8 mins or so.

[TeamEd]

Very cool, you should post this on reddit r/nba. Nice clean and readable charts/graphics too

Done. Thanks for the tip.
http://www.reddit.com/r/nba/comments/2t ... mentality/

[Ursusamericanus]

Awesome thread, OP.

[TeamEd]

So, I did a redo of the team numbers since the OP was all wonky with an couple of errors in my excel tables.

Here's the comeback shooting mentality of all 30 teams:


And here's the comeback passing mentality of all 30 teams:



Some interesting things going on here. Again, look at where the Cavs fall on both. It seems the team struggles to get both attempts and assists when trailing. Considering this is essentially a measure of team offensive effectiveness, you can see how they've struggled so fae (less so recently, I'm sure).

It's not easy to be sure, but it also definitely looks like there's a solid correlation between overall team strength and comeback shooting mentality. If you look, you see that the top of that chart is filled with top teams. There might be a similar thing going on with the passing mentality numbers, but it's less clear.

Not sure what's going on there with the Mavs.

[TeamEd]

Here's an update, with one wrinkle.

I've run a look at how player's TS%s change when behind. There isn't much of a correlation between that and shooting mentality, but I think it's worth a look:



And, here are the updated Shooting Mentality and Passing Mentality charts:





As you can see, Kyle Lowry -- who was the inspiration for this stat -- has mellowed a little. Meanwhile, Mike Conley is way up at the top of both charts now.

There's a lot of Conley/ Gasol happening whenever Memphis trails.

[TeamEd]

So, I pulled the entire NBA.com database of player in-game splits. This is a database of around 9,000 different player stat cards going back to 1996-97: MASSIVE UPDATE TIME.

My formulas are the same. I just now have a gigantic database.

Here are the links to this data. It's all in Tableau, so feel free to play around with the numbers. They go back almost 20 seasons! I'm looking for help with the analysis, so I'll drop these links here talk briefly about this season's data and mostly leave the rest up to you.

Shooting Mentality: https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/SMBehindorTied?:embed=y&:display_count=no

The chart shows the comeback Shooting Mentality of all NBA players who have taken >850 true shot attempts so far in 2014-2015. There's some interesting stuff here. For one, I love how I've got Jamal Crawford at the top at +9.5 and Lou Williams at the bottom at -7.4. It shows you can't really generalize about 6th-man gunners. I also find the contrast between Lebron's -3.8 and Kyrie's +4.3 noteworthy.

TS% change: https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/ShootingEfficiencyBehindorTied?:embed=y&:display_count=no

This chart shows a player's change in TS% in comebacks. This number is in percentage points and it is not adjusted for league average. There are some interesting connections here with the other charts. Blake's TS% goes down a tone in comebacks, but he also passes a hell of a lot more.

Passing Mentality: https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/PMBehindorTied?%3Aembed=y&%3AshowVizHome=no#1

The chart shows the passing mentality of every player with >250 assists so far in 2014-2015. The sample sizes are smaller here, so I expect there are some issues with Blake's insane +16.3. That said, Mike Conley doesn't have any sample size problems and he's got an almost equally insane +13.3. Also noteworthy, Kyle Lowry's still sitting at the bottom of this chart. He's still one of the guys with one of the biggest splits between shooting mentality and passing mentality. He shoots more and passes less. John Wall's the opposite. He shoots less and passes more. Lebron meanwhile, both shoots less and passes less.

Rebounding Mentality: https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/RMBehindorTied?%3Aembed=y&%3AshowVizHome=no#1

This one is new. It shows the rebound mentality of all players with >500 rebounds so far. I'm actually surprised to see any significant ahead/behind splits at all in rebounding, but there they are. You've got Duncan up there at +8.3 and Love trailing near the bottom at -9.1.

Finally, here's a tool you can use to find a player's career numbers in all of these measures. https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/PlayerFinderBehindorTied?%3Aembed=y&%3AshowVizHome=no#1

I love this thing because it lets you look at career tendencies. Tim Duncan and Steve Nash, for instance, absolutely shine in Shooting Mentality. They have both played every single full season of their careers with positive shooting mentalities. Nash even posted a ridiculous +16.3 in 06-07. Chris Paul and Tracy McGrady meanwhile posted negative shooting mentalities for most of their careers.
And, as a bonus. I've also used the same methods in close-score situations. So, here's a full set of the same links for player performance in scores within +/- 5 points.

Shooting Mentality (Close): https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/SMClose?:embed=y&:display_count=no
Passing Mentality (Close): https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/PMClose?:embed=y&:display_count=no
Rebound Mentality (Close): https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/RMClose?:embed=y&:display_count=no
Player finder (Close): https://public.tableausoftware.com/views/NBASituationalStatistics1996-2015/PlayerFinderClose?:embed=y&:display_count=no

Whew. That's a lot. Anyway. I'm posting all this here so you can help me with the analysis and testing of all this.
What do you think? Do you understand what these numbers show? Do you see any problems with my methodology or my presentation? Have you found anything interesting in the data from previous seasons? Any errors?

As far as I know, I'm the first person to do this on this scale. If you know of anyone else who's done the same thing before, let me know. This is all based on a previous post. You can see how my methodology developed there: