Fantasy Football Defense Rankings, Week 5

Seventy percent of the points-allowed projection is the opponent's implied team total, which is the betting market's own answer to the only question that matters here. The remaining thirty is a backward blend of what the opponent has been scoring and what this defense has been allowing.

That backward half is adjusted for the offenses the defense has actually faced. A unit that has played three of the best offenses in football reads as one of the worst in the league on raw numbers, which is the most common way these rankings go wrong in September. New England is the example this week: 19.3 points allowed a game on the raw figure, 15.8 once the schedule is accounted for.

Both start from projected pass attempts, because a rate means nothing without the volume behind it. Attempts are half the posted market line for that quarterback and half his own rate per game started. The rate applied to them is the average of the quarterback's and the defense's, each shrunk toward the league rate so that four games of football is not mistaken for evidence.

One thing worth knowing, because it is the opposite of the usual advice: the spread barely moves pass attempts. Fitted on 5,755 team games since 2015, a favorite does throw at a lower rate, about 0.13 points of rate per point of spread, but he also runs more plays, about 0.09 a point, and the two almost exactly cancel. Attempts move 0.027 per point of spread with a correlation of -0.02. Adding the spread to a baseline of the team's own attempts moved the correlation from 0.435 to 0.436.

The touchdown figure is the implied probability from the longest posted price on that defense in the anytime touchdown market. Longest because the book paying the most is charging the least margin, which matters here more than anywhere since only a handful of books post defenses at all. A dash means nobody has priced that unit yet.

Fumble recoveries are held flat at the league rate. Recovery is close to a coin flip and the spread between teams is mostly luck, so a per-team number would be precision that does not exist.