Spider Graphs: Charting Basketball Statistics
Spider Graphs, as a method of analyzing basketball statistics, were developed by Roger Pimentel in 2009.
They’re a new way to reflect a player or team’s statistics in a visual way that gives you an overall impression of their style. They were created and first used by the author in evaluating matchups in the 2009 NBA Finals.
The top half of the graph represents offensive statistics (FG%, points, and assists), and the bottom half represents defense (steals, blocks, and rebounds—which we strongly feel are a defensive statistic). A player whose graph has more area up top should be considered a bigger contributor on offense, and a larger bottom half suggests a defensive focus.
To add another layer, the three upper-right statistics (points, assists, and steals) tend to be associated with perimeter players, whereas the bottom-left three (blocks, rebounds, and high FG%) tend to be the jurisdiction of big men.
Since the graphs use an ordinal system, it’s possible for the maximum values of each axis to change without damaging the integrity of the graph (so long as they remain roughly proportional). The NBA Spider Graph Generator uses maximum values of 35 points, 12 assists, 2.5 steals, 4 blocks, 18 rebounds, and 60% shooting. A zero score represents zero in every case except FG%, where zero on the graph represents about 30% shooting (to amplify differences between players). Prior articles on HTWS using Spider Graphs may use different maximums.
The graphs use maximum values of 40 points, 20 rebounds, 20 assists, 5 blocks, 5 steals, and 75% shooting, normalized so that they’re all on a scale from 0-100. A zero score represents zero in every case except FG%, where zero on the graph represents 25% shooting (to amplify differences between players).
They are only meant to reflect past statistical performance, leaving out intangibles and change to a player’s game over time.
As an example, here’s Dwight Howard from 2008-2009.
Dwight epitomizes interior presence—and that’s reflected in the graph by the lower left statistics, high FG%, rebounds, and blocks.
He’s a strong offensive player, as shown by the decent-sized area in the top half of the graph. Where Howard is really a monster, however, is on the defensive end—and the area of the bottom half of his graph is accordingly larger.
Let’s do one more… Kobe Bryant, also from 2008-2009:
In contrast to Dwight’s big man graph, Kobe is stronger in the top-right three statistics, which is where perimeter players tend to come out to shine.
Likewise, while Kobe is a strong defender, he has always made his mark as a scorer and a playmaker. Accordingly, the top half of his graph dominates the bottom half.
Here are some articles with real-life applications of Spider Graphs:
Spider Graphs, like most graphical representations, are better for conveying an idea than giving hard numbers. This is especially true since Spider Graphs use normalized numbers, so each statistic is on a scale from 0-100.
One hazard of Spider Graphs is that using the area of a graph to determine the sheer impact of a player isn’t entirely accurate. The graphs punish a player who excels in one statistical category but not the statistic that is adjacent to it on the graph; for instance, a volume scorer who gets very few assists would have a large spike on the Points axis but then it would drop way back down on the Assists axis. In comparison, a player with mediocre Points and Assists numbers may actually have a graph with more area.
Another limitation is that each of the statistics can be used in either a per-game or per-minute way, except for field goal percentage. When graphing per-game statistics for a rarely-used player, it’s not uncommon for their graph to be right around zero on five of the axes, but then spike out on FG%—because even a player who plays two minutes a game can have an excellent FG% (though the rest of his numbers will be negligible).
Finally, there are likely better numbers that can be graphed, using advanced statistics. Spider Graphs would likely be much more accurate/useful if they were built on per-possession data. Further, FG% might be better replace by eFG% and the like. Feel free to email Roger if you have any thoughts along these lines (or if you know where to get per-possession data on a player-by-player basis). We’ll make a lot more progress by putting our heads together.
Questions/comments on Spider Graphs? Email Roger.