Projection Confidence Intervals: Understanding Uncertainty in Forecasts

A projection without a confidence interval is like a weather forecast that says "72 degrees" with no mention of the storm system sitting offshore. Projection confidence intervals quantify the range of outcomes a player is realistically expected to produce, turning a single-point estimate into a probability distribution. This page covers how those intervals are constructed, what drives their width, where they break down, and how to read them without falling into the traps that trip up even experienced fantasy managers.

• 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

A confidence interval, in the statistical sense formalized by Jerzy Neyman in his 1937 work published in Philosophical Transactions of the Royal Society A, describes a range around a parameter estimate that would contain the true value in a specified percentage of repeated trials under the same sampling conditions. In fantasy sports projection systems, the term gets borrowed — sometimes loosely — to describe the band within which a player's actual fantasy point total is expected to fall with a stated probability, most commonly 80% or 90%.

The distinction matters immediately. A 95% confidence interval on a running back's weekly projection does not mean "he will almost certainly score inside this range." It means that if the same projection model ran 100 equivalent weeks, approximately 95 of those intervals would capture the actual outcome — a concept the American Statistical Association has repeatedly clarified in public guidance on statistical inference. Fantasy applications rarely hit textbook conditions, but the framework still carries real signal.

Scope-wise, confidence intervals apply at multiple horizons: single-game projections (narrow scope, high variance), weekly lineup decisions, and rest-of-season projections where uncertainty compounds across future games. The interval width grows with the time horizon — sometimes dramatically — which is a feature, not a flaw.

Core mechanics or structure

Most fantasy projection systems derive confidence intervals through one of three mechanical approaches.

Bootstrap resampling draws thousands of random samples from historical performance distributions to generate an empirical output range. The model runs, say, 10,000 simulations, and the 5th and 95th percentile outcomes define an 90% interval. No assumption about the underlying distribution's shape is required — which makes bootstrapping attractive when player performance is clearly non-normal (wide receivers with occasional blowup games skew right hard).

Parametric modeling assumes a specific distribution — often a negative binomial for discrete counting statistics like receptions or rushing attempts — then calculates the interval analytically. This is faster computationally and works well when historical data is abundant, but it misfires when the assumed distribution doesn't fit the player type.

Monte Carlo simulation is structurally similar to bootstrapping but typically incorporates correlated inputs — opponent defensive ratings, weather variables, teammate usage — across the simulation draws. A model might run 50,000 game-state scenarios, each drawing from separate distributions for snap count, target share, and yards-per-target, then combine them. The output is a full probability distribution, not just a point estimate.

The interval itself is then expressed as a floor-to-ceiling range. Floor and ceiling projections are often just the 10th and 90th percentile outputs of this process, relabeled for accessibility.

Causal relationships or drivers

Four factors drive interval width with the most consistency across sports.

Sample size is the most direct driver. A veteran quarterback with 4 seasons of 16+ starts has more distributional data than a rookie with 6 career games. Projections built on thin samples carry wider intervals almost by definition — a principle covered in depth at sample size and projection reliability.

Role volatility expands uncertainty sharply. A running back in a true committee where no back has exceeded 55% of carries over the past 8 games will produce an interval 2 to 3 times wider than a feature back who handles 70%+ of attempts. Usage rate adjustments in projections feed directly into this dynamic.

Opponent unpredictability matters because a defense's performance distribution has its own variance. Projecting against a defense that allows 28 fantasy points to wide receivers one week and 8 the next introduces compounded uncertainty into the receiver's interval.

Injury history creates asymmetric intervals. A player with a recurring soft-tissue injury doesn't just have a wider interval — the interval skews toward the floor. The probability mass shifts downward even when the midpoint projection stays the same. Injury adjustments in projections is where that asymmetry gets modeled explicitly.

Classification boundaries

Not all confidence intervals in fantasy projection systems mean the same thing, and the label alone is insufficient to interpret the number.

The most important boundary is between a confidence interval and a prediction interval. A CI estimates where the mean outcome falls; a prediction interval estimates where a single game's actual score will fall. For fantasy purposes, the prediction interval is almost always the relevant construct — but projection tools frequently publish the narrower CI, which looks tighter and more reassuring than it should.

Tradeoffs and tensions

Wider intervals are more honest and less useful. Narrower intervals are more useful and often less honest. That tension doesn't resolve cleanly.

A system calibrated to produce 90% coverage — meaning 9 out of 10 actual outcomes fall inside the projected band — will generate intervals so wide for volatile players that they lose practical value. A tight-end's 90% interval in a week he's verified as questionable might span 2 to 28 fantasy points. Technically rigorous. Operationally useless.

Projection systems respond by shrinking intervals toward the mean through regularization techniques. This improves the cosmetic appearance of precision but deliberately sacrifices calibration. The tradeoff is openly discussed in the academic forecasting literature, including work published in the International Journal of Forecasting, where Gneiting and Raftery's 2007 paper on proper scoring rules established that sharpness without calibration is a measurable failure mode.

There's also a presentation tension. Fantasy players prefer a clean number. Displaying "Patrick Mahomes: 24.3 ± 9.1 points (90% PI)" is accurate but cognitively harder to act on than "24.3 projected." The reading and interpreting projection outputs page addresses how to use both pieces of information together.

Common misconceptions

"A narrow confidence interval means a safe player." Width reflects variance, not ceiling. A player with a narrow interval centered at 8.2 points is predictably mediocre. Narrow intervals appear for players with high usage, low volatility, and limited upside — not necessarily "safe" in fantasy terms.

"The point projection is the most likely outcome." The mean of a right-skewed distribution sits above the median and mode. For explosive skill position players whose distributions skew right due to touchdown probability, the single-number projection often overstates the most probable outcome. 50% of actual scores will fall below the median, which is below the mean.

"80% and 90% intervals are interchangeable." A 90% prediction interval is meaningfully wider than an 80% interval. For a receiver projecting 14 points with typical variance, the 80% interval might span 6–22 points while the 90% interval stretches to 4–26. The extra 10 percentage points of coverage require substantial additional range.

"Confidence intervals account for unknown unknowns." They don't. A model built on historical distributions cannot assign probability to a game-time scratch that wasn't in the injury report, a fumble lost, or a coaching decision to abandon the run game by halftime. The intervals capture known variance; black swan events sit outside the model entirely.

Checklist or steps (non-advisory)

The following sequence describes how a projection confidence interval is typically constructed in a quantitative fantasy system:

1. Estimate distribution parameters from historical data, applying regression-to-mean adjustments for small samples — see regression to mean in fantasy
2. Incorporate contextual modifiers: opponent defensive rank, projected snap share, weather inputs, Vegas implied totals from Vegas lines and fantasy projections

Reference table or matrix

The table below maps interval characteristics to their practical interpretation in a fantasy lineup context, drawing on the classification framework from the glossary of projection terms.

For comparison across sports, the home page at Fantasy Projection Lab aggregates model outputs across NFL, NBA, MLB, and NHL — each sport carrying structurally different variance profiles that affect interval construction at the player level.