“Hustle” Caveats

Apr 17, 2016; Cleveland, OH, USA; Detroit Pistons guard Kentavious Caldwell-Pope (5) shoots over Cleveland Cavaliers forward Kevin Love (0) during the first quarter in game one of the first round of the NBA Playoffs at Quicken Loans Arena. Mandatory Credit: Ken Blaze-USA TODAY Sports
Apr 17, 2016; Cleveland, OH, USA; Detroit Pistons guard Kentavious Caldwell-Pope (5) shoots over Cleveland Cavaliers forward Kevin Love (0) during the first quarter in game one of the first round of the NBA Playoffs at Quicken Loans Arena. Mandatory Credit: Ken Blaze-USA TODAY Sports /
facebooktwitterreddit
Apr 17, 2016; Cleveland, OH, USA; Detroit Pistons guard Kentavious Caldwell-Pope (5) shoots over Cleveland Cavaliers forward Kevin Love (0) during the first quarter in game one of the first round of the NBA Playoffs at Quicken Loans Arena. Mandatory Credit: Ken Blaze-USA TODAY Sports
Apr 17, 2016; Cleveland, OH, USA; Detroit Pistons guard Kentavious Caldwell-Pope (5) shoots over Cleveland Cavaliers forward Kevin Love (0) during the first quarter in game one of the first round of the NBA Playoffs at Quicken Loans Arena. Mandatory Credit: Ken Blaze-USA TODAY Sports /

With one game in the book for each of this year’s first round series, the new “Hustle Stats” have survived their first challenge – they were tracked for all 8 games and even updated regularly as the games were proceeding!

As of yet, it’s difficult to draw many conclusions from the new metrics. Sure, the Clippers had far and away the most “screen assists” with 25, but it’s simply too soon to tell if this was a result of unusual shotmaking, poor defense by Portland, something about the Clips’ offensive style or a reflection of good play. On the last point, it’s worth noting the Thunder had the fewest screen assists, with only two. And they did just fine scoring the ball.

So while the sample sizes are too small to use these new numbers for much in the way of team and especially player-level analysis, there are some things to look at in broader terms, most notably with the “shots contested” counts. First of all, there is the danger of terminological confusion. By my count there are four different categories of “contested shots” available on NBA.com.

  • Contested 2s and 3s under Hustle Stats  – defined as closing out with a hand raised. It appears from quick analysis that multiple players can be awarded contests on the same shot and these will be added together to form the team totals, so we can’t be quite certain how many individual shots were contested or not.
  • Field goals defended at the rim – shots where the player in question was within 5 feet of both the shooter and the rim. These form the backbone of my Rim Protection metrics.
  • “Contested” and “Uncontested” shots in player tracking box scores – Shots where a defender is within 3.5 feet of the shooter[1. As with all SportVU based stats, this measurement is torso-to-torso, so 3.5 feet basically works out to “arms-length” from the shooter or not.] are contested. I noted some issues with this one-size-fits-all definition last postseason.
  • Shots by defender distance – For example see here for Detroit’s breakdown.

And of course when the SportVU shot logs were still available[2. To paraphrase a cinematic masterpiece – Bring! Them! Back! Bring! Them! Back!], one could break down the effect of defensive proximity on shooting percentages even further.

In any event, that’s a lot of definitions, and they are all slightly different. Which can lead to some confusion, especially given the colloquial understanding of what “contesting a shot” really means – a defensive effort that makes the shooter materially less likely to score.

Take for example three pointers. Using the SportVU logs for previous seasons’ data, it becomes clear that the NBA.com definitions[3. 0-2 Feet = very tightly contested; 2-4 Feet = tightly contested; 4-6 Feet = open, 6+ Feet = wide open.] aren’t quite right on three pointers – very few threes are attempted with a defender closer than 2 feet and the point where shots go from being below average to above average looks seems to be around 6 feet, rather than 4 feet. Threes a with a defender between 4 & 6 feet away were made at around 34% last season, as opposed to 38% with the defender between 6 and 10 feet and 40% with defenders greater than 10 feet away. Thus to my way of thinking, using 6 feet as the cut off between contested and not[2. If one were being binary about it, which has some value in terms simplicity.] makes more overall sense than 3.5 or 4 feet, or whatever distance a player must reach for the hand up “hustle” contest to be counted.

Still, the NBA.com definitions do indentify some buckets in which to place and measure groups of shots, and this is where the difficulties with the “Hustle Stat” contests come into play. Take Game One of Detroit-Cleveland for example. The Cavs “contested” threes 24 times.[4. I’ll make the assumption that most of these were a single player contesting a shot, because if you’re shooting a three with two players close enough to contest, what are you doing? ] The Pistons took 29. The “eye test” can be quite flawed on these things, but it certainly seemed watching the game in real time as if Detroit got many more more open looks than that from deep.

Looking at SportVU-based shooting splits, that seems closer to correct. Ten of the 29 threes had no Cavs defender within 6 feet, and another 13 had the closest defender within 4 and 6 feet. So by that measure, 23 of the 29 threes taken by the Pistons were uncontested. Those totals seem inconsistent, though neither is necessarily wrong given the varying definitions. But this is why the next step should be to combine the “Hustle contest” with individual shots to determine if there is a specific distance at which having a hand up starts or stops having an impact. Or to put it another way, to start to empirically prove the “hand down, man down” theory. At this point, we simply don’t know if these contests are doing anything on the floor, or if they represent what commentators like to call “fake hustle” with players making a show of closing out when in actuality they are too far away and too late to really bother the shot.

If the above all reads like a bit of hand-waving, that’s what it is to a large degree. More info is better than less, but until we have a better idea what the additional data means in terms of impact on the game, my hope is people can refrain from citing the hustle stats as definitive of “good defense because they contested shots better.” At least for the time being.