We Know Better Now – A Note On Outdated Statistics

February 9, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) shoots the basketball against Houston Rockets forward Trevor Ariza (1) during the third quarter at Oracle Arena. The Warriors defeated the Rockets 123-110. Mandatory Credit: Kyle Terada-USA TODAY Sports
February 9, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) shoots the basketball against Houston Rockets forward Trevor Ariza (1) during the third quarter at Oracle Arena. The Warriors defeated the Rockets 123-110. Mandatory Credit: Kyle Terada-USA TODAY Sports /
facebooktwitterreddit
February 9, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) shoots the basketball against Houston Rockets forward Trevor Ariza (1) during the third quarter at Oracle Arena. The Warriors defeated the Rockets 123-110. Mandatory Credit: Kyle Terada-USA TODAY Sports
February 9, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) shoots the basketball against Houston Rockets forward Trevor Ariza (1) during the third quarter at Oracle Arena. The Warriors defeated the Rockets 123-110. Mandatory Credit: Kyle Terada-USA TODAY Sports /

Properly employed, statistics add important context and narrative precision to a sporting context. Consider how the following description provide important detail to Tuesday night’s marquee matchup, a rematch of the 2015 Western Conference Finals:

  • The Golden State Warriors defeated the Houston Rockets.
  • The Warriors beat the Rockets 123-110.
  • The Warriors beat the Rockets 123-110, with Stephen Curry putting up 35 points and 9 assists.

And so on.

Each bit of numerical information brings to light additional detail about the game. Deciding which details to emphasize as important is a vital part of properly describing the game. Brandon Rush playing 3 minutes is no less factually accurate than Curry’s point total. However, in retelling the story of the game, not to mention the Warriors’ season as a whole, one is far more important than the other. While that is a rather obvious example, there are many more instances where settling on the right number of metric to “tell the story” is far more difficult.

As more detailed, varied and simply better frames of analysis become available, making those choices correctly becomes even more of a challenge. Some old standbys might be supplanted as better data allows for more exactitude. I want to look at a couple of these commonly cited measures which, while not bad stats, might have been surpassed by the growing refinement in our understanding of the game.

Points Off of Turnovers

For as long as I can remember, this has been a stat commentators have pointed to as a demonstration of how sloppy offense is compromising a team’s defense. Which is an important thing to know – a weak point of possession-based analysis is ignoring the degree to which the end of one play affects how the next might play out. But all turnovers are not created equal, in large part because all ensuing possessions are not equal. In fact, the difference could not be starker between a live ball turnover[1. AKA steals.] and a travel, or pass thrown out of bounds. As one of my high school coaches liked to thunder after a lazy ball reversal was intercepted in practice: “if you’re going to throw that pass, just throw the ball into the stands. At least then we can get back on D!”

Possessions after steals are some of the highest expectation plays in possible – hence the high value assigned to steals in many player projection and valuation systems. By contrast, deadball turnovers do not lead to any greater likelihood of scoring than other stoppages of play.

This isn’t to say forcing deadball turnovers is a bad thing, ever. There are certain advantages in terms of fast break opportunities to gaining possession via defensive rebound, on the order of about about a quarter of a point per possession on aggregate. That bump pales in comparison to the edge gained by not allowing a shot at all – with offensive rebound chances factored in, even the very worst shot attempt an opponent may throw up has a much higher expectation than the extra chance of transition play from a miss, so ensuring a zero point possession is a great outcome for a defense.

But in terms of describing how defending well can lead to good offense, steals and deadball turnovers are apples and rutabagas. Points on plays following steals can at least conceivably flow from the now-defense being scrambled as a result of the sudden and disorganized end of the previous offensive possession. By contrast  points which occur on the possession following a turnover resulting in a stoppage are simply the points scored on the next possession and could easily have come following a made basket, or a trip to the free throw line or even out of a timeout.

Just as teams aren’t uniform in how often they turn the ball over, there is a sizable variation in terms of how those turnovers occur. Thru Tuesday, here is a comparison of all 30 teams in terms of total turnovers versus opponents’ steals:

Dashboard 1 (77)
Dashboard 1 (77) /

One of the few criticisms leveled at the Warriors this year is their propensity to throw the ball around at times on offense. And it’s true, they do turn the ball over, with a higher than average proportion as a result of steals. As the chart above shows, the Celtics have a similar number of total turnovers, but significantly fewer as a result of steals. Thus Golden State is compromising their own defense a little more often, possibly indicating an area where their looseness (or profligacy, depending on how critical one wishes to be of a historically stupendous offensive unit) might hurt them slightly.

All of this is basically demonstrating that ‘points off of turnovers” isn’t telling quite the story commentators are intending, and that perhaps “points allowed off of steals” would better illustrate that specific point.

Points in the Paint

If there’s one sure thing, even post-Warriors’ 2015 championship run, it will be the calls for teams to “pound the ball inside.” Whether Kevin Love in Cleveland, Chris Bosh during the Heatles Finals runs or the oldest of old saws, Dwight Howard, there is seemingly no offensive problem for which more post ups isn’t the answer. What is at work here is a confused understanding of terms. “Easy baskets” and “shots at the rim” are good, efficient things. That said, those categories only overlap , and are not coextensive with, “post ups” or “shots in the paint”.

At this point, it’s fairly well established that tossing the ball into the post and expecting efficient isolation scoring is a triumph of hope over extensive data. Sure, there is value to getting the ball inside, but that’s as much about drawing help to open up kickouts to three point shooters or allowing for better offensive rebound chances as it is about “barbecuing chicken” on the block.

By virtue of “the paint” being a readily identifiable area on the court, the very real differences between a dunk or layup right at the rim and a 7-foot floater or jump hook get collapsed. But with more detailed “X,Y” information about every shot taken[1. Even prior to SportVU this information was available, though there tend to be some discrepancies between data sets generated by human coders and motion tracking software.], it becomes quite clear that not all shots in the paint are created equal:

PITP
PITP /

As the collated data from the 2014/15 season shows[1. This data comes from PBP logs rather than SportVU, as it includes “shot zone” information used to populate NBA.com shot charts – the results would be substantially similar using only SportVU data though the distances might stretch out another foot or two in terms of high value shots ending at around 4 feet instead of 2.], the efficiency of “inside” shots drops off very quickly as the shooter is farther and farther from the rim. Some of this is the result of defensive pressure – if you are wide open at 5 feet, you are probably pretty open at 2 feet for an even better shot – but it also demonstrates that the best “last line of defense” is much closer to the rim than the borders of the key. Preventing shots at the rim is the basis of good defense, which is better accomplished by forcing that 7 foot floater than collapsing to prevent it and allowing an open three on a kickout.

So, in terms of describing whether a team is or is not protecting the valuable real estate near the basket well, “points in the paint” doesn’t always tell us much. Getting beat by the opposition raining teardrops is a much different outcome than giving up a layup line. In the latter instance, there probably needs to be a defensive adjustment. In the former, perhaps some modifications are needed to prevent the initial penetration, but simply letting the percentages sort themselves out over team is quite possibly the best “remedy.” A more accurate analysis would probably look more to attempts and baskets “at the rim[2. However one wants to define the term, whether shots taken in the restricted area, within 5 feet or the hoop, or some other empirical definition.]” as the best indicator of interior defense.

In the case of both examples, the original language was capturing something real and important. But as the data gets better and analysis more detailed, it is important to ask more specific questions, which unlock the story not just of what happened in a game or over a season, but why those things came to pass.