Do the NBA Hustle stats matter?

The NBA's own analytics department a few years back, under the leadership of then Director of Basketball Analytics Jason Rosenfeld, brought out a slew of new statistics they titled "Hustle Stats". While these days so much of the new data we see has come from the SportVU/Second Spectrum camera systems with primarily machine learning derived 'auto tracking', the Hustle stats are old school -- actual humans watch every game and mark down actions of interest, specifically:

• Screen Assists
• Deflections
• Loose Balls won
• Charges taken
• Contested Shots (broken out by 2's, 3's)
• ...and recently for the '17-18 playoffs they added Boxouts

There are many ways to evaluate these numbers; a simple place to start is in-game. Does the team with a better "hustle stat" during the game win more often?

...If a team has 6 loose balls won, and their opponent 4, the team would be +2 and have 'won' the loose balls for the game...

W-L Record for the team that wins the Hustle Stat category

Category	Regular Season W-L Record	Regular Season Win%	Playoffs Win%
Loose Balls Won	645-424	60%	61%
Screen Assists	668-466	59%	57%
Contested 2's	658-537	55%	64%
Contested Shots (all)	639-543	54%	54%
Deflections	612-539	53%	56%
Charges Taken	311-305	50%	64%
Contested 3's	564-608	48%	46%
Boxouts	N/A		56%

If you prefer using 'Contested Shot%' instead of number of contests (as many NBA coaches do), the Win% for having an edge is
54% (58% playoffs) for overall Contested Shot%, 56% (54%) for Contested 2%, and 49% (53%) for Contested 3%

So at a quick glance there's some merit to the interest in these numbers, with both net Loose Balls Won, and the net Screen Assists being pretty good at indicating a team with an advantage. Those are also two stats with very tangible gains -- the loose ball is winning possession (you hear these also referred to as 50/50 balls) which is worth about a point, and then the Screen Assist is only given for a screen leading to a made basket, so naturally having higher screen assists is a promising sign in a game.

To add perspective though, these win percentages aren't spectacular compared to the basic box score stats, where the team with the higher field goal percentage in a game won 81% of the time in '17-18, the team with more field goals made won 80%, the team with more Assists won 71%, more total rebounds won 67%, more blocks won 61% etc.

Some of the other Hustle categories are more nebulous in their benefits, providing reasons to be skeptical -- contesting a shot may lower the percentage of the shot attempt, gaining fractions of a point for the team over an uncontested shot, but giving up the shot can be a negative, as seen by a team with more contested 3's on defense having a losing record!

An overall edge in Hustle

Another approach is to see which team won more of the hustle categories and if there's predictive value there. Since Boxouts only began for the playoffs we will leave that out, and as there are three separate Contested Shots numbers and we don't want to over-value one area, we will go with just Contested 2's. This means there are five categories up for grabs in a game:

Loose Balls won, Screen Assists, Contested 2's, Deflections, Charges Taken.

A full point for the team with better numbers in a category, a half point if they tie (so for example, win three, tie one, lose one is 3.5 'category points').

W-L Record for the team that wins the Hustle Stat category

Category Points	Regular Season W-L Record	Regular Season Win%	Playoffs Win%
5	15-5	75%	100%
4.5	41-11	79%	100%
4	140-61	70%	57%
3.5	194-116	63%	75%
3	235-175	57%	60%

2.5 Category Points means it's the same for both teams, so a 50% win rate, and the rest are simply the inverse of the other team,
e.g. 0 Category Points teams went 5-15...

So that looks a little more interesting, with the team with an overall edge (3+ Category points) winning 63% of the games in the regular season, 68% in the playoffs.

The Mystery of Deflections

When it comes to the beloved Deflections stat, arguably the most "prized" of the NBA hustle stats, I would suggest a distinction sheds new light on their value...

The image commonly associated with deflections is hard playing, up into the opponent, scrappy, physical, alert, active, close guarding, pressure packed, unrelenting defense leading to the ball being under constant attack by the defenders' hands. Devin Harris has a good story about playing for Jerry Sloan in Utah -- Sloan would tell his team "you can foul 30 times a game and get called for 30 fouls, or you can foul 60 times a game and get called for 30 fouls!" While not specifically about Deflections this is the mentality some coaches are after in their quest for higher defensive "activity" and deflections are thought to be one of the more visible signs of effort. A classic plea to the defense is "Make it Hard!" [Side note: this year's referee Points of Emphasis includes calling more fouls on defenders for holding/grabbing/impeding movement...we will see]

My issue with deflections is simple, there are really two types: deflections leading to steals, which have obvious immense benefits, and deflections where the ball still belongs to the offense. In discussing the Deflection charting with the NBA league analytics staff for clarification, blocked shots do not count as a deflection, but otherwise a defender intentionally touching the ball while the offense has it is a deflection. This is slightly imperfect for the purpose of simply subtracting steals from deflections to get "non steal deflections" as there are a small number of situations where a steal may not also count as a deflection (pass into the back, player dribbles ball off his foot that bounces to a defender who picks it up)...but these are rare. So let's look at the won-lost records again...

Category	Regular Season W-L Record	Regular Season Win%	Playoffs Win%
Steals	663-429	61%	67%
Non-Steal Deflections	518-607	46%	50%

If we look at games where the team had a 'significant advantage' in non-steal deflections, say a net of +5 or more, the W-L is 163-214 (43%)
...or if you don't like the net concept and want straight non-steal deflections for the team only, with 10+ non-steal deflections teams went 200-255 (44%)

Suddenly the importance of Deflections is in question. While Steals bring the dual gains of a defensive stop and a chance to unleash an offensive possession with a much higher than average points per possession expectancy, Deflections that don't wind up in a steal showed negative predictive power in this past regular season as to which team would win. Why?

Several plausible explanations come to mind:

• a non-steal deflection represents a "missed opportunity" and much like with loose balls, failing to corral the ball comes at a cost of about a point when possession is up for grabs
• in the moments after a deflection, chaos can ensue which may benefit the original offensive team if they retain the ball (say if 2 or 3 defenders, thinking it's a steal, start running the other way!)
• a team getting lots of deflections may be gambling/over extending which in itself may be a negative sign
• ...or could it be in a matchup of teams, the underdog feels the need to be more aggressive/go for steals and deflections, while the superior team on paper is casual with the ball on offense/content to 'stay home' on defense to where even if the underdog loads up on deflections, it's not enough to overcome the talent discrepancy? Or a team with a big lead eases up, while the team far behind amps up the pressure?