Guards against the Atlanta Dream
Both cuts of Atlanta Dream's defense against guards. Read the team-level one: the position split is ~85% noise (split-half r = -0.03).
What this metric is worth
read before the ranksPosition-split defense-vs-position is not measurable: roughly 85% of a DvP cell is noise, so its rank carries no information about tonight.
split-half reliability r = -0.033 across 12 cells (7 of 12 negative) against a shuffle null of 0.000; a placebo using the wrong position family reproduced the effect to within 0.12pp
Position-blind team defense is reliable and is the cut to reason from.
split-half reliability r = +0.688, against r = -0.033 for the same teams cut by position (same games, same season, same test)
Both cuts are computed the same way from the same games. Only one of them repeats when you split the season in half, which is why the answer leads with team level and why nothing on this page gates a pick on either.
Both cuts, side by side
trusted cut firstPosition-blind, Atlanta Dream allow 84.5 points a game, 4th of 15. Position-blind team defense is reliable and is the cut to reason from. (split-half reliability r = +0.688, against r = -0.033 for the same teams cut by position (same games, same season, same test))
Position-blind, and the cut the reliability test passed. Every team-game counts once.
Rank 1 allows the least. A positive z means this team allows LESS than the league average across 15 teams; turnovers forced is reported without a rank, because more is not obviously better for a defense.
This cut ranks Atlanta Dream 2nd of 15 for points allowed to guards, and that rank should not move a decision. Position-split defense-vs-position is not measurable: roughly 85% of a DvP cell is noise, so its rank carries no information about tonight. (split-half reliability r = -0.033 across 12 cells (7 of 12 negative) against a shuffle null of 0.000; a placebo using the wrong position family reproduced the effect to within 0.12pp)
Position-split defense-vs-position is not measurable: roughly 85% of a DvP cell is noise, so its rank carries no information about tonight.
Rank 1 allows the least. A positive z means this team allows LESS than the league average across 15 teams; turnovers forced is reported without a rank, because more is not obviously better for a defense.