Why Coverage Matchups Matter in Fantasy

Ceiling outcomes are almost all that matter in weekly DFS tournaments. In 2024, 51 of the 61 WRs (~84%) appearing in DraftKings Milly Maker-winning lineups scored at least 20 DK points. Collectively, they did so at about half the average ownership of their RB counterparts. I’m glossing over a lot of the game theory and projections-related reasons for this, but in short, it suggests tournament-winning WRs are most often those having outlier ceiling performances that a median projection (which evenly considers a player’s full range of outcomes) won’t capture well.

It might be intuitive to you that we should primarily focus on ceiling in DFS, but the same is arguably true in regular seasonal fantasy leagues. At the WR position, a 20-25 fantasy point performance is worth roughly triple the Wins Above Replacement (WAR) of a 10-15 point performance.

This is all to say that we can’t evaluate a stat’s usefulness solely by examining its correlation to fantasy points scored (as I’m incredibly guilty of). Ceiling is also massively important — especially for WRs — and an R-squared measuring across the entire range of outcomes (from goose egg weeks to 40+ fantasy point performances) won’t necessarily tell us how well a stat predicts all-important spike weeks.

While a wide correlational approach is insufficient alone, we will still start there. As you read, just trust there’s a reason I’ve already spent so many words discussing its limitations as it relates to the all-important ceiling.

How Predictive Are Coverage-Filtered Stats Overall?

Tons of very smart people in this industry will tell you that coverage shell matchups — that is, how well a player performs against man or zone coverage, or Cover 3, or single-high looks — are rather noisy and not always worth considering. That’s because, on the whole, coverage-filtered stats are not as predictive across the entire range of outcomes as unfiltered stats.

But as the above numbers show, TPRR and 1D/RR against single-high looks are a notable exception to this rule, even before we attempt to filter for players facing single-high heavy defenses. This thread spells out my thoughts on why this is the case, but in short, most of the best receivers in the NFL enjoy better efficiency against single-high.

This makes some intuitive sense. Stats accrued against one-on-one coverage — of which single-high creates a ton on the outside — could be better than regular stats at measuring player skill. As a result, a player’s TPRR, YPRR, 1D/RR, etc. against single-high is about as useful (or perhaps slightly more) as those same stats against all coverages, despite being comprised of fewer routes.

Our only filters for the above analysis were that we only considered games in which the player recorded a minimum of 20 routes (regardless of coverage) from Week 5 onward. These are minor controls to ensure we’re only evaluating instances where a player is both on the field for a material number of routes and has a sample of previous work for us to evaluate.

We haven’t yet attempted to filter for the schematic matchup a player is facing in a particular week, and we’ve already found a “win” for coverage stats in the form of single-high metrics.

Coverage-Filtered Stats Against Ideal Schematic Matchups

In-season, we can easily target matchups in which a receiver can expect to face a lot of single-high looks (or two-high, zone coverage, Cover 3, etc.) with the Fantasy Points Data Suite’s Coverage Matrix. So our next move is to repeat the analysis above while filtering for players matched up against a defense that enters the week having played an exceptionally high rate of our chosen coverage.

In these ideal schematic matchups, you might expect our coverage-filtered stats to show improvement across the board compared to “unfiltered” or “overall” stats. But outside of a few exceptions I’m going to show off below, that isn’t the case. Despite testing several thresholds (e.g., over 25%, 30%, 35%, 38% man coverage), we usually get something like this:

While the predictiveness of man coverage stats generally improves as we get into more man-heavy matchups, they never catch up to the overall stats. In general, you are better off using unfiltered, schematic matchup-agnostic stats to predict median outcomes, with the following exceptions:

• TPRR against single-high looks is slightly more predictive than overall TPRR, regardless of the defense’s single-high rate. The gap becomes wider (a .336 vs. .312 correlation coefficient) in matchups where a 60% or greater single-high rate is expected. The same is true of XFP/RR (.345 vs. .317) at the same matchup threshold.

• TPRR (.339 vs. .317) and XFP/RR (.339 vs. .333) against zone coverage are slightly more predictive than unfiltered versions of those stats in matchups above a 78% expected zone coverage rate.

• In matchups with above a 30% expected rate of Cover 2, the Cover 2-filtered version of every stat below, aside from target shar,e becomes more predictive. (But only a handful of matchups project this Cover 2-heavy each season, and I wouldn’t blame you for shrugging at this due to sample size).

But everywhere else, unfiltered stats are still better than coverage-filtered ones at predicting fantasy points, even in ideal schematic matchups. That is, if we’re trying to predict across the entire range of outcomes or come to a median projection. If that’s your goal (as it might be for most prop betting), feel free to primarily rely on overall stats rather than coverage-specific ones outside of the exceptions listed.

But as you’ll recall, instead predicting ceiling games or “spike weeks” is also an important goal for both season-long fantasy leagues and DFS tournaments. Let’s shift our focus here.

A Ceiling-Focused Approach To Schematic Matchups

If numbers aren’t your thing, here’s the TL;DR, keeping in mind that across all matchups, players averaged a 12.7% ceiling rate (of scoring 20+ PPR fantasy points):

• In matchups against a defense that’s played at least a 42% man coverage rate, players with at least a 10% better YPRR vs. man coverage hit 20+ fantasy points 25.5% of the time. That’s a ~100% better ceiling rate than all players across all matchups.

• In matchups against a defense that’s played at least a 63% single-high rate, players with at least a 10% better TPRR vs. single-high hit 20+ fantasy points 17.1% of the time. That’s a ~35% better ceiling rate than all players across all matchups.

• In matchups against a defense that’s played at least a 36% Cover 1 rate, players with at least a 20% better YPRR vs. Cover 1 hit 20+ fantasy points 21.3% of the time. That’s a ~68% better ceiling rate than all players across all matchups.

• In matchups against a defense that’s played at least a 38% Cover 3 rate, players with at least a 5% better TPRR vs. Cover 3 hit 20+ fantasy points 17.2% of the time. That’s a ~35% better ceiling rate than all players across all matchups.

• In matchups against a defense that’s played at least a 22% Cover 4 rate, players with at least a 5% better TPRR vs. Cover 4 hit 20+ fantasy points 21.1% of the time. That’s a ~66% better ceiling rate than all players across all matchups.

• For zone coverage, two-high looks, and Cover 2, I could not find any statistically significant increase in ceiling rate with any threshold of a reasonable sample size.

To summarize the impact of the above, paying attention to schematic matchups can help you hit a ceiling performance between 35% (for the more common matchups) and 100% more often (for the exceedingly rare matchups). That is absolutely massive for DFS tournament players, who will rack up hundreds to thousands of entries across an NFL season. It’s also useful for ladder bettors, who aim to identify mispriced alternate props to profit from players hitting high-end ceiling outcomes.

But even the average redraft fantasy manager playing in their home league can easily factor this into their lineup decisions, so long as they have a Fantasy Points Premium subscription. That’s because I’m going to highlight these spots in my Advanced Matchups column each week.

And there, I’ll have the advantage of being able to discern additional context that this study didn’t; for example, it wouldn’t have been smart to pick Tyreek Hill in most ostensibly single-high-heavy matchups last season. Since the Dolphins had a lackluster run game, virtually every defense (no matter how single-high-heavy they were normally) played elevated rates of two-high looks against Miami in 2024 to limit explosive downfield passing. I didn’t fully catch on to this until partway through the year, but this is just one example of the advantages a human mind can have over a spreadsheet.

I explain how I arrived at my results in meticulous detail in the “Nerdy Stats Explanation” section below. But first, a few high-level musings and disclaimers:

Why do schematic matchups work for predicting ceiling outcomes, but largely not for predicting the entire range of outcomes, as the rest of this article explains?

Big fantasy performances (ceiling outcomes) are often the result of one or two big plays; a 30-yard receiving touchdown is worth 10 fantasy points in a PPR league. Even in an ideal schematic matchup, a player usually sees only a handful more routes against the coverage in question than they otherwise would (going from a 25% man coverage rate to a 42% man coverage rate in a 40-route game adds about 7 routes against man coverage).

That’s a handful more opportunities to make a big week-winning play — which indeed occurs some of the time as our results suggest. But most of the time (in general, across the full range of outcomes), a few additional routes on which a player historically has a good matchup won’t move the needle.

Therefore, schematic matchups aren’t that useful for median projections (and largely present lower correlation coefficients than unfiltered stats) but are quite helpful for predicting ceiling performances at the extremes.

Why use YPRR for man and Cover 1, but TPRR for other schematic matchups?

The short answer is “because they work better,” as you can reference from the matrix in the Nerdy Stats Explanation.

As for why that likely is, YPRR is better at capturing the impact of big plays than TPRR. That’s because a receiver can gain anywhere from 0 to 99 yards on a single route, but only 0 or 1 target. And it’s probably more important to be extra-sensitive to big plays when analyzing man coverage compared to zone coverage, since they tend to happen more against man.

Conversely, against zone (including many single-high looks, Cover 3, and Cover 4), it’s harder to create big plays, so TPRR as a volume metric performs better for our purposes.

Why are your thresholds for each coverage set so high? What does that do to your sample sizes?

I tested several possible thresholds for each coverage (e.g., a 25%+, 30%+, 35%+, 40%+, 42%+ rate of man coverage) and found that I had to stick with relatively high rates to find any signal.

Ideally, I’d have 10+ years of charted coverage data to work with, allowing me to maintain a larger sample size even after splitting it apart. That’s my long-term hope.

However, so far, we have charted only 2022 through 2024, and I excluded results from the first four weeks of each season to allow each player’s YPRR and TPRR time to stabilize (since the analysis relies on cumulative YTD stats entering each week). After enforcing our 20-route minimum, that left us with just under 4,000 “player weeks” to analyze, which is still fairly solid. But those numbers shrink quickly after we filter for coverage rates.

I performed significance testing for the rates I reported at the top of this section against the inverse of the sample, and all were significant at p < .05. But for transparency, here’s where our sample size ends up for each coverage.

• 175 player weeks entered a matchup expected to face at least a 42% rate of man coverage. 55 of those player weeks also entered with at least a 10% better YPRR against man coverage.

• 470 player weeks entered a matchup expected to face at least a 63% rate of single-high. 146 of those player weeks also entered with at least a 10% better TPRR against single-high.

• 165 player weeks entered a matchup expected to face at least a 36% rate of Cover 1. 47 of those player weeks also entered with at least a 20% better YPRR against Cover 1.

• 766 player weeks entered a matchup expected to face at least a 38% rate of Cover 3. 319 of those player weeks also entered with at least a 5% better TPRR against Cover 3.

• 555 player weeks entered a matchup expected to face at least a 22% rate of Cover 4. 237 of those player weeks also entered with at least a 5% better TPRR against Cover 4.

Nerdy Stats Explanation

I analyzed every instance of a WR or TE running 20 or more routes in a game after Week 4 over the past three seasons. In 12.7% of those instances, the player scored 20.0 or more PPR fantasy points. This is the base “ceiling game rate” we’re trying to beat (as highlighted in the bottom-right of the matrix below).

Then, I filtered for matchups where the opposing defense entered the game having played an exceptionally high rate of a specific coverage. I tested several thresholds for each coverage type to find the optimal cutoff. The ones I ultimately chose are listed in the “Coverage and Cutoff Rate” column below. For example, in the third row, I analyzed matchups that the defense entered having played above a 63% rate of single-high.

Sticking with that example, I then isolated players who entered the matchup having averaged at least a 5% higher TPRR against single-high than against all coverage looks — players who had previously been more efficient against single-high. I did the same using a “10% higher” TPRR threshold, and then again using 10% and 20% higher YPRR benchmarks.

This gave me four new groups of players who were both expected to face a lot of single-high and who had previously been more efficient (to varying degrees) against single-high. For each of these groups, I again calculated their rate of exceeding 20.0 fantasy points.

For matchups above a 63% single-high rate, I found that 17.1% of players who’d previously commanded at least a 10% TPRR boost against single-high went over 20.0 fantasy points. That was quite a step up from our original base rate (12.7%). In other words, if you’d identified one of these players in an ideal single-high-heavy matchup, you’d be about ~35% more likely to hit on a ceiling game than if you picked a player and week at random.

Finally, to confirm that the player’s efficiency indeed affected their ceiling rate, I calculated the ceiling rate for the inverse: players with less than a 10% single-high TPRR boost (so including players who were average or worse against single-high). This is displayed in the “comparison column”; 10.2% of all other players in single-high heavy matchups hit at least 20.0 fantasy points. (For the three people who care, this is where I tested statistical significance.)

I repeated this process for every coverage shell, with the ideal matchup rate and player efficiency threshold noted via bold and underlines in the above image.