I clustered players using three per-36-minute stats that are decent surrogates for the three dimensions of modern basketball skill – minutes of ball possession (red), 3-point attempts (green), and field-goal attempts defended inside 6 feet (blue) – and plotted the groups using the RGB color model.
Next, I organized the players into 127 five-man units that played together for at least 100 minutes each during the 2017-18 season. I used k-means clustering to group lineups into similar-looking five-color combos. Then, I tabulated the median net rating for each grouping of clustered lineups. These were the best lineup structures:
These were the worst:
I matched up player height and reach data from the NBA combine with lineup data and then tested the relationships between average lineup height and defensive rating. I shared my findings in The Athletic – Bay Area, in a story called “The Warriors have revolutionized small ball, but are they really small?” For the 2016-17 season, I found there was no relationship between a lineup’s average height and its defensive effectiveness (quantified by the number of points the opponents scored per 100 possessions).
But, unlike average height, variance in height was the best indicator of a lineup’s defensive capabilities last season — that is,
Units that were comprised of players who were of similar size tended to allow fewer points.
The Warriors “small”-ball lineups were comprised of players with small ranges in height.
This seems to have translated to an effective switch-heavy defense in Golden State.