A Graphical History of Moreyball

One of the most prominent narratives of the 2018-19 NBA season is about the dominance of Moreyball across the league – an overriding aversion to low-efficiency mid-range shots, in favor of Morey-approved looks at the rim and from three. What does this shift look like? Is the typical three-pointer really that much more efficient than the typical 18-footer? How radically has league-wide shot selection really changed? And can this change be traced to Morey’s Rockets?

Field goal frequency by distance from rim, for all regular season attempts in stats.nba.com database (through Dec. 14, 2018)

Figure 1: Field goal frequency by distance from rim, for all regular season attempts in stats.nba.com database (through Dec. 14, 2018)

Points per field goal attempt by distance from rim, for all regular season attempts in stats.nba.com database (through Dec. 14, 2018)

Figure 2: Points per field goal attempt by distance from rim, for all regular season attempts in stats.nba.com database (through Dec. 14, 2018)

Figures 1 and 2 give an animated look at how shot distributions have shifted from 1996 to the present, along with how these shifts reflect the average value of different shots. A few observations:

  • As shown in Figure 1, league-wide adoption of Moreyball started around ~2014 and has been progressing rapidly through the present, with a precipitous decline in 15- to 20-footers and an increase in the frequency (and depth) of threes. Frequency of short-range twos has remained fairly steady (initial big spike in shots at rim seems like a data artifact; see below).

  • As shown in Figure 2, the shift away from long twos and toward threes represents a durable difference in shot value – long twos have been nearly steady at ~0.8 points per shot, while threes from within ~28 feet have been consistently more valuable on average.

  • 2018-19 has seen a major jump in the value of long threes, with attempts from out through 30 feet falling at nearly the same rate as threes taken from more prudent distances. This might represent a sorty of Steph Curry-ization of the league, with longer shots gaining legitimacy as good looks, and more players taking (and practicing) them as such.

  • The big spike in shots right at the rim that gets greatly attenuated in 2010 is probably just an artifact of the data. My best guess is that this reflects a bunch of shots that would have previously been coded as 0 feet from the rim instead being coded as 1-3 feet (possibly as result of better remote sensing?). This hypothesis is supported by a corresponding increase in pointers per 0-foot shot (these are now all true dunks/lay-ins, rather than various other shots in paint).

  • Shot frequency and value distributions change sharply from 1996-97 to 1997-98. This is a function of an experimentally short 3-point line (a uniform 22 feet) moving back to its original/current distance of 22 feet in the corners and 23.75 feet everywhere else.

Path chart of season-average field goal frequencies and efficiencies, from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

Figure 3: Path chart of season-average field goal frequencies and efficiencies, from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

While the above GIFs are instructive, it can be hard to simultaneously hold multiple moving images in the mind. That said, the path chart in Figure 3 shows how short twos, long twos, and threes have changed over time with respect to both frequency and efficiency. This chart neatly summarizes the analytics-nerd case for abandoning long twos: over the past two decades, they’ve been worth consistently less than short twos or threes, and their decline in prominence is rapidly moved to relfect that absence of value: while all shot types are holding roughly steady in terms of value, and short twos are holding steady in terms of frequency, threes have skyrocketed in frequency at the expence of long twos. Where long twos used to be hoisted at nearly three times the rate of threes, they’re currently taken ~25% less frequently than threes.

It’s clear that the idea of avoiding mid-range shots in favor for looks nearer and more remote has taken hold of league-wide offenses. But did this trend really originate with Morey’s Rockets? To look at this, I reproduce initial shot frequency GIF, but in static, small-multiples format and with Houston’s shot selection highlighted. In Figure 4, the Rockets’ shot attempt distributions, shown in red, diverge sharply from broader NBA averages starting in the 2012-13 season, with sharply fewer long mid-range shot attempts and sharply more three-point attempts. (Not coincidentally, 2012-13 is the season that James Harden joined the Rockets.) This trend has gotten more pronounced for Houston over time, though the rest of the league has also followed suit, as seen above; in the current season, Houston is taking vanishingly few long mid-range shots and many threes, and the rest of the league is not far behind.

Distribution of field goal attempts by season for each NBA franchise (faint grey lines), the Houston Rockets (red lines), NBA-wide seasonal average (blue lines), and multi-year NBA average (black line), from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

Figure 4: Distribution of field goal attempts by season for each NBA franchise (faint grey lines), the Houston Rockets (red lines), NBA-wide seasonal average (blue lines), and multi-year NBA average (black line), from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

Finally, Houston’s pre-eminence as a mid-range averse franchise can be seen most clearly by focusing in on the rate of such shots over time for all teams. Figure 5 shows that, starting in 2012, the Rockets became far and away the most mid-range averse team in the leage, and that, they remain so (with this year’s exception of the Bucks, who are almost as stingy from mid-range as the Rockets), even as the rest of the legue continues on a long-running trend of moving away from these generally inefficient shots.

Proportion of field goal attempts coming from midrange (2-point shots outside of 5 feet from rim), with each individual team in grey, league average in blue, and Rockets in red, from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

Figure 5: Proportion of field goal attempts coming from midrange (2-point shots outside of 5 feet from rim), with each individual team in grey, league average in blue, and Rockets in red, from stats.nba.com API (with 2018-19 shots through Dec. 14, 2018)

For the code I used to gather and process the shot chart data used in this post, see https://github.com/TrevorDavidThomas/NBA_analytics/blob/master/dataAssembly_shotCharts_full.R.

For the code I used to generate the graphs, see https://github.com/TrevorDavidThomas/NBA_analytics/blob/master/analysis_teamLevelShotDistribs_distanceFocus.R.