NFL Fantasy Projections: How the Lab Approaches Football

NFL fantasy projections sit at the intersection of statistical modeling, football operations intelligence, and scoring-format math — and the gap between doing them adequately and doing them well is wider than most platforms let on. This page explains how Fantasy Projection Lab builds its NFL projections: the inputs, the logic chains, the classification decisions, and the places where honest analysts disagree. The goal is transparency, not mystique.

• Definition and scope
• Core mechanics or structure
• Causal relationships or drivers
• Classification boundaries
• Tradeoffs and tensions
• Common misconceptions
• Checklist or steps (non-advisory)
• Reference table or matrix

Definition and scope

An NFL fantasy projection is a pre-game (or pre-season) numerical estimate of a player's expected fantasy point output under a specific scoring system. That last clause matters more than people tend to think. A tight end projected at 11.4 points in a standard scoring league might project to 14.1 in full PPR — not because his football changed, but because the scoring math rewards reception volume differently. The projection is always format-conditional.

Scope-wise, NFL projections cover 32 teams, roughly 17 regular-season weeks (expanded from 16 beginning with the 2021 season per the NFL's official schedule structure), and a player pool that stretches from starting quarterbacks down to backup running backs who are one injury away from 20 carries. The Lab produces projections across all positional groups — QB, RB, WR, TE, K, and team defenses — with the understanding that each position requires a meaningfully different model architecture. Kicker projections don't share logic with wide receiver projections any more than a seismograph shares logic with a wind gauge.

The full methodology context lives at Projection Models Explained, which covers the framework that applies across all sports. What follows here is specific to how that framework is tuned for football.

Core mechanics or structure

The projection engine for NFL players operates in three layers.

Layer 1: Baseline volume estimates. Every projection starts with an estimate of how much work a player will see — pass attempts for quarterbacks, carries and targets for running backs, targets for receivers, snap counts for everyone. The snap count and target share data underlying these estimates comes from play-by-play tracking (Pro Football Reference, NFL Next Gen Stats). A receiver who averaged 7.4 targets per game over the prior 8 weeks gets a target volume prior before any game-specific context is applied.

Layer 2: Efficiency rate application. Volume estimates get multiplied against per-opportunity efficiency metrics — yards per carry, yards per route run, catch rate, touchdown rate per red zone target. These rates are regressed toward positional means based on sample size, a process explained in detail at regression to mean in fantasy. A running back who scored touchdowns on 14% of red zone carries over 3 games gets regressed hard toward the NFL average (roughly 8–10% depending on era and carry type); one who hit that rate over 50 red zone touches gets much lighter regression treatment.

Layer 3: Context adjustments. Game environment variables — opponent defensive rank by position, Vegas implied team totals, weather at outdoor stadiums, pace of play, and injury-related depth chart shifts — modify the baseline × efficiency product. The Vegas lines and fantasy projections page covers how market-implied game totals are incorporated as priors for pass/run volume splits.

The output of all three layers is a projected stat line (attempts, completions, yards, touchdowns, carries, receptions, etc.), which is then converted to fantasy points using the target scoring format. Scoring format impact on projections documents how significantly a format shift — standard to half-PPR to full PPR, for example — can reorder projections at the positional level.

Causal relationships or drivers

Projections are not guesses dressed up in decimals. They reflect causal chains that can be traced from input to output.

Offensive line health → carry volume and efficiency. When a starting offensive guard misses a game, the downstream effect on a running back's yards-after-contact number is measurable. Studies using NFL play-by-play data (available via nflfastR, an open-source R package maintained by Sebastian Carl and Ben Baldwin) consistently show line personnel quality as a statistically significant predictor of running back efficiency beyond just volume.

Defensive coverage scheme → receiver target distribution. A defense that plays heavy Cover 2 suppresses deep targets and shifts volume toward short-to-intermediate routes, which in PPR formats benefits slot receivers more than outside boundary wideouts. This is a causal mechanism, not a correlation artefact. Matchup-based projection adjustments walks through how coverage tendencies are incorporated.

Game script → position-group weighting. A team projected as a 10-point underdog (implied by Vegas lines) will run the ball less in the second half as they chase the score. That shifts the projection environment away from running backs and toward pass catchers. The causal arrow runs from line movement to projected game script to position volume, not from historical averages alone.

Injury depth chart position → opportunity absorption. When a WR1 misses a game, the slot receiver doesn't always absorb the targets. Target distribution is partly scheme-dependent — some offenses spread targets across the depth chart, others funnel them to a clear second option. Injury adjustments in projections covers the elevation logic in detail.

Classification boundaries

NFL projections are classified along two primary axes at the Lab: position group and projection horizon.

By position, quarterbacks use a completely separate model architecture from skill position players — pass-game volume is quarterback-first, while receiver projections are downstream of that volume estimation. Quarterback projection methodology, running back projection methodology, wide receiver projection methodology, and tight end projection methodology each document position-specific modeling decisions.

By horizon, week-level projections differ from rest-of-season and preseason projections in the weight given to recent data versus career priors. A week-level model might weight the last 4 games at 60% of the prior construction; a preseason model relies almost entirely on prior-year data and offseason intelligence. In-season vs preseason projections covers this distinction structurally.

The projection vs ranking difference page addresses a classification question that trips up new users: a projection is a point estimate with associated uncertainty, while a ranking is an ordinal sort that depends on the projection and positional scarcity and roster construction context. They are related outputs, not the same output.

Tradeoffs and tensions

Recency bias vs. mean regression. Weighting recent games heavily makes projections responsive to real changes (new offensive role, improved scheme fit) but also amplifies noise from small samples. Weighting long-run averages more heavily stabilizes projections but can miss genuine breakouts for 3–5 weeks. There is no universally correct calibration — sample size and projection reliability lays out the tradeoff quantitatively.

Precision vs. honesty about uncertainty. Publishing a projection as "14.7 points" implies a precision that no model actually has. Projection confidence intervals makes explicit that NFL weekly projections carry wide variance — a player projected at 15 points might have a 90% confidence interval spanning 4 to 38 points. The precise number is a central estimate, not a promise.

Injury information timing. Practice participation data (limited/full/DNP) released Wednesday through Friday moves projections materially, but the information is often incomplete or strategically ambiguous until the official injury report. Projection update schedule documents when updates are pushed in relation to official report windows.

Roster construction tension with lineup optimization. A projection system optimized for expected value will sometimes conflict with roster construction logic that prioritizes floor in high-stakes formats. Floor and ceiling projections addresses this by providing not just a point estimate but distributional shape — how often a player is likely to produce below 8 points, above 20.

Common misconceptions

"Higher projected points always means start." Not quite. Two players projected at the same point total can have radically different variance profiles. A volatile boom-or-bust receiver and a steady PPR slot might share a 14-point projection while having entirely different distributional shapes. Format and roster depth determine which profile is preferable.

"Projections account for all injuries." Projections update when information is public. They cannot incorporate private medical evaluations, undisclosed soft-tissue limitations, or practice observations not reflected in the official injury report. The data sources used in projections page is explicit about which inputs are available and which are not.

"A projection system that was accurate last year will be accurate this year." Backtesting projection accuracy is necessary but not sufficient. Player personnel changes 30–40% of the league's starting roster composition each offseason. A model that excelled on 2022 data didn't necessarily capture the features that drive 2024 performance, especially if scheme trends shifted. The comparing projection systems page evaluates systems on rolling out-of-sample accuracy, not in-sample fit.

"Machine learning models are always better." Machine learning in fantasy projections addresses this directly. Neural network and ensemble models can overfit to historical patterns and fail to generalize when the NFL changes rules, rosters, or schemes. Regression-based models with strong feature engineering frequently outperform black-box approaches on out-of-sample accuracy metrics.

Checklist or steps (non-advisory)

The following sequence describes how a single NFL week-level projection is constructed at the Lab from initial inputs to final published output.

1. Depth chart ingestion — Official depth charts, practice reports, and beat reporter updates are collected from NFL.com and team transaction wires.
2. Volume prior construction — Rolling weighted averages (last 4 and last 8 games) for target share, snap rate, and carry share are calculated per player.
3. Efficiency metric calculation — Per-opportunity efficiency rates (yards per target, catch rate, red zone TD rate) are computed and regressed toward positional baselines.
4. Game environment pull — Vegas closing lines and implied team totals, weather data for outdoor stadiums, and opponent defensive rankings by position are ingested.
5. Context adjustment application — Volume priors are modified by game-script probability (from Vegas spreads), matchup quality, and injury-related depth chart shifts.
6. Projected stat line generation — Adjusted volume × adjusted efficiency = projected raw stats (yards, TDs, receptions, etc.).
7. Fantasy point conversion — Raw stats are converted to fantasy points under the target scoring format (standard, half-PPR, full PPR, TE-premium, etc.).
8. Confidence interval attachment — Historical variance for each player-type and role is used to generate floor (25th percentile) and ceiling (75th percentile) estimates alongside the median.
9. Quality review — Projections with extreme outlier values (>3 standard deviations from positional average) are flagged for manual review before publication.
10. Publication and update scheduling — Initial projections publish Wednesday; updates follow Thursday injury report release and Sunday morning practice participation reports.

The full Fantasy Projection Lab home serves as the entry point for accessing these outputs across all active sports and formats.

Reference table or matrix

NFL Position Group: Key Projection Inputs and Their Weight

Projection Horizon: Input Weighting Shift

Rest of season projections explains why the late-season weighting shifts so heavily toward recent games — by Week 12, role changes and injury histories have produced enough 2024-season data to outweigh career baselines for most players.