Prop LabPlayer traits and matchup myths
Is Defense vs Position Useful in Fantasy Football? Mostly Not
Fantasy points allowed by position is the most-used matchup stat in fantasy football. We tested whether it repeats, whether it predicts, and whether any matchup angle beats a good projection.
By Edgehalla ResearchPublished 8 min read
What defense vs position measures, and why it misleads
Defense vs position (DvP) ranks each defense by the fantasy points it allows to quarterbacks, running backs, receivers and tight ends. “Start your tight end, he faces the worst TE defense in the league” is a DvP argument.
The problem is sample size. A defense faces one offense a week. By Week 6, its tight end number rests on five games, and a single touchdown can swing the ranking from 10th to 30th.
So we asked two questions. Does a defense’s DvP repeat within a season and across seasons? And does a first-half number predict the second half? We used 2023 to 2025, adjusted each game for the opponent’s usual output, and graded every position.
DvP reliability by position
Split-half reliability compares a defense’s odd weeks with its even weeks. A value near 1 means the number is stable. Near 0 means it is noise. We also checked the year-over-year correlation.
How much does defense vs position repeat?
PPR points allowed by position, 2023–25
Tight end DvP had zero within-season reliability and wide receiver DvP did not carry across seasons; none of the four positions passed our bar.
- Split-half reliability
- Year-over-year r
Quarterback and running back DvP had some consistency within a season. But a first-half number predicted the second half only 4% (QB) to 9% (WR) better than assuming every defense is average, and worse than that for running backs and tight ends.
Tight end DvP was pure noise within a season: 0.00 reliability. Wide receiver DvP did not carry over at all from one season to the next.
This is why our Defense vs Position page shrinks every defense hard toward average. After four games, a defense’s own results get only 7% to 17% of the weight, depending on position.
Deep balls, sidelines and the middle of the field
Maybe position is the wrong cut. We also split passing defense by where the ball went: deep or short, left, middle or right.
Which passing splits repeat within a season?
Split-half reliability (Spearman–Brown) of PPR points allowed, 2023–25
Only points allowed over the middle earned a passing grade; deep passes and tight end points were close to noise.
Getting burned deep does not stick. Deep-pass points allowed had a raw split-half r of 0.09 and a year-over-year r of 0.03. A defense that gave up three long touchdowns last month is not more likely to do it again.
Outside receivers are a trap. Points allowed to outside receivers were fairly consistent within a season (0.45), but they did not predict the second half at all.
The middle of the field is the one passing split that passes. Points allowed on throws over the middle had 0.44 reliability, cut second-half error by 8.5%, and carried over between seasons (r 0.27, or 0.35 with the same defensive coordinator).
Even that one did not help in practice. Tested as a matchup angle on top of our projections, it was worth −0.11 (WR) and +0.01 (TE) points per game for a one-standard-deviation matchup. A good projection already captures what it knows.
What does repeat is how a defense plays, not what it allows. Blitz rate, man coverage rate and two-high shells all graded A, with reliability of 0.81 to 0.88. They follow the coordinator and reset when he leaves.
Do any matchup angles beat a good projection?
DvP is only the simplest matchup tool. We also tested 21 player-versus-defense angles, from blitz and man coverage to screens, missed tackles and red-zone defense, by position. That was 64 tests in total.
64 matchup angle tests, 2024–25
No matchup angle earned an A or B grade, even though a positive control showed the test could detect real effects.
The largest effects were 0.2 to 0.5 PPR points per game for a one-standard-deviation matchup. None cut the 2025 holdout error by more than 0.25%.
Could the test just be blind? We checked with a positive control: we fed a known-good signal (a strong outside projection minus ours) through the same test. It lit up, with t-statistics of 7 to 16. A real matchup effect of about 0.3 points per game, for a one-standard-deviation matchup, would have been caught. So “nothing found” means nothing big was there.
The reason is a mismatch in every angle. Defense habits are stable, but a player’s split against them is not: “he feasts against man coverage” had a split-half reliability of 0.33. Player habits are stable, but what a defense allows on those plays is not. Every angle pairs a stable half with a noisy half.
A related trap: points over expected
The same noise shows up on the offensive side. Fantasy points over expected (FPOE) measures how much a player beat the points his usage should produce. It feels like skill. Week to week, it is mostly luck.
What repeats from one week to the next?
Week-to-week correlation, games with 5+ expected PPR points, 2023–25
Expected points from usage repeat far better than points over expected, which are close to noise week to week.
The hottest fifth of players after eight weeks scored 2.7 fewer points a game in the second half, even though their usage held. Track opportunity, not efficiency, on Over Expected.
Matchup myths, tested
Verdicts: myth, real but priced, small but real or unproven.
- Myth
“Defense vs position tells you who to start.”
TE DvP had a split-half reliability of 0.00 and WR DvP a year-over-year r of −0.07. First-half DvP predicted the second half worse than average for RBs and TEs.
- Myth
“A defense that gets burned deep will keep getting burned deep.”
Deep-pass fantasy points allowed barely repeated within a season (split-half r 0.09) or across seasons (r 0.03).
- Real, but priced in
“Some defenses are soft over the middle.”
True: middle-of-field points allowed graded B (reliability 0.44). But it added −0.11 (WR) and +0.01 (TE) points per SD beyond our projection.
- Myth
“A player who beat his expected points is breaking out.”
Points over expected repeat week to week at r 0.05. The hottest fifth after eight weeks scored 2.7 fewer points a game in the second half.
- Small but real
“A usage spike after a role change is just noise.”
Role-change risers kept 82% of their usage gain into the next game and beat their baseline 62% of the time (2024–25).
How to use matchups in fantasy football
- Never bench a starter over a DvP rank. It is the weakest input in a start/sit decision.
- Use DvP as a tiebreaker only, and prefer a version that shrinks small samples, like ours.
- Chase roles, not matchups. Usage gains after a role change kept 82% of their size into the next game. See Usage Risers.
- Ignore tight end matchups almost entirely. TE DvP had zero within-season reliability.
If you also bet player props, the same rule applies: defense rankings are public and already in the line.
Frequently asked questions
Is tight end defense vs position reliable?
No. In our 2023–25 tests, tight end DvP had a split-half reliability of 0.00: a defense’s odd weeks told you nothing about its even weeks. Treat tight end matchups as noise.
How many games before defense vs position means anything?
More than a season provides for most positions. Even with a full season, our shrinkage keeps 21–45% of the weight on a defense’s own results. After four games it gets only 7–17%, depending on position.
What is the most reliable defensive matchup stat for fantasy?
Fantasy points allowed on throws over the middle was the only passing split to pass our tests (reliability 0.44, year-over-year r 0.27). Defensive habits like blitz rate and man coverage repeat even better, but they do not translate into fantasy points.
Should I bench a player because of a tough matchup?
Almost never. None of our 64 matchup angle tests earned an A or B grade against a good projection, and the largest effects were 0.2 to 0.5 PPR points per game. Usage and role matter far more.
Do defenses that give up big plays keep giving them up?
Not reliably. Fantasy points allowed on deep passes had a split-half r of 0.09 within a season and a year-over-year r of 0.03. A few long touchdowns can make a defense look leaky for weeks.
More in Player traits and matchup myths
Sources and method
- nflverse data (play-by-play, schedules, weather, injuries, combine)
- FTN Data charting (via nflverse, CC BY-SA 4.0)
- Spearman–Brown prediction formula (split-half reliability)
- False discovery rate and the Benjamini–Hochberg procedure
- National Council on Problem Gambling (1-800-GAMBLER)
- Responsible Gambling Council (Canada)
Research and analysis, not betting advice. Past results do not guarantee future returns; bet only what you can afford to lose.