Regression to the Mean: What It Means for Fantasy Projections
Regression to the mean is one of the most misunderstood forces in fantasy sports — and one of the most exploited by managers who understand it. This page explains what the concept actually means statistically, how it operates in practice across NFL, NBA, and MLB contexts, and where it should and shouldn't change a roster decision.
Definition and scope
Here is something that trips up a lot of experienced fantasy managers: a running back who scores 6 touchdowns in the first four weeks of the season is not "hot." He is probably lucky, and the math already knows it.
Regression to the mean is the statistical tendency for extreme observations — a performance well above or below a historical baseline — to be followed by results closer to that baseline. The concept was formalized by Francis Galton in the 19th century studying hereditary height data, but its operational logic applies anywhere random variance is mixed with underlying signal. In fantasy sports, that's everywhere.
The scope matters here. Regression to the mean is not the same as decline. A player who scored 6 touchdowns in 4 games has not suddenly become worse by Week 7 when he has scored 0. The underlying talent level hasn't changed. What changed is that variance, which gave him those extra scores, normalized. The projection models explained at FantasyProjectionLab account for this explicitly, separating sustainable rate-based production from variance-driven spikes.
The statistical machinery behind regression is straightforward: the more extreme an observed value is relative to a population average, the more of it is attributable to random noise rather than skill, and the more strongly the next observation will pull back toward the mean.
How it works
The mechanism hinges on a distinction between true talent and observed performance. Every box score result is a combination of the two. When a wide receiver posts a 35% target share over two games, that number is almost certainly not his true talent level — it is his true talent level plus a favorable variance event. The expected value for his next game is closer to his career or seasonal baseline.
A useful way to think about the split:
- Identify the baseline — career averages, positional norms, or situational context (e.g., a quarterback's red-zone touchdown rate relative to league average)
- Measure the deviation — how far above or below baseline the recent sample sits
- Estimate sample size reliability — small samples amplify noise; sample size and projection reliability covers the thresholds at which statistics become meaningful
- Apply a shrinkage factor — projection systems weight observed performance back toward the prior, with the weight determined by sample size
Touchdown rate is the canonical regression candidate in NFL fantasy. League-average touchdown rate on red-zone carries sits around 33-38%, depending on the era and source (NFL Next Gen Stats). A back who scores on 6 of his first 8 red-zone carries is at 75% — an extreme outlier. Without a dramatic change in role or opponent quality, a pull back toward league norms is nearly certain.
Common scenarios
Regression scenarios tend to cluster into recognizable patterns across sports:
Touchdown variance (NFL) — As noted, touchdowns are the highest-variance scoring event in fantasy football. A tight end who scores 4 touchdowns in 3 weeks on 12 targets hasn't become Travis Kelce. His underlying target rate may be excellent; his touchdown conversion rate is almost certainly temporary.
Batting average on balls in play, or BABIP (MLB) — This is regression to the mean's most documented application in baseball. League-average BABIP for hitters sits around .300 (FanGraphs BABIP reference). A batter carrying a .420 BABIP through April is experiencing a significant positive variance event. Pitchers below .250 BABIP over a short stretch are similarly candidates for negative regression — meaning their ERA is likely to rise, lowering their fantasy value in standard formats.
Three-point shooting variance (NBA) — A shooter making 52% of his threes over the first month of the season, against a career rate of 37%, is running hot. The NBA fantasy projections framework treats sustained shooting outliers with skepticism until the sample crosses roughly 150 attempts — a threshold supported by research from FiveThirtyEight's CARMELO and Raptor model documentation.
Decision boundaries
Not every extreme performance regresses at the same speed or to the same degree. Knowing where the decision boundary lies — act on the regression signal vs. hold — is where the concept translates into roster moves.
Buy-low candidates — A player posting well below his expected output due to high BABIP luck (pitchers), red-zone target drought, or cold shooting is a trade acquisition target. The underlying role and opportunity are intact; the results will normalize upward.
Sell-high candidates — Conversely, a running back who has scored 7 touchdowns in 5 games without a commensurate increase in target share or opportunity volume is a sell. Trading him while the box score looks gaudy captures peak market price before regression arrives.
The contrast between these two situations is the practical core of the concept: same statistical principle, opposite roster action, depending on whether the player is below or above his true talent line.
One important caveat: regression applies to rate statistics, not to workload. A receiver who genuinely earned a larger target share through a change in offensive scheme or a teammate injury is not regressing — that is a structural shift in opportunity. Usage rate adjustments in projections addresses how to distinguish noise from signal when opportunity levels change mid-season.
The FantasyProjectionLab home applies regression-adjusted baselines across all major sports, which means the projections already incorporate shrinkage toward population norms rather than extrapolating hot or cold streaks forward.