Analytics Strategy

College Football Receiver Receiving Yards Prediction Model

College Football Receiver Receiving Yards Prediction Model

A college football receiver receiving yards prediction model estimates how much opportunity a receiver will get, then translates that opportunity into catches and yards. A practical starting point is expected targets multiplied by catch probability and expected yards per catch. The stronger version also accounts for changing roles, different types of targets, and the range of outcomes around the average.

The first question is where the targets will come from. A receiver coming off a huge game might have a stable role, or he might have benefited from one busted coverage. Those situations should produce different expectations for the next matchup.

Below is a transparent framework you can use to evaluate a receiving projection. Every numerical example is hypothetical. None represents a live player forecast, a tested performance claim, or a description of ATSwins’ proprietary model.

Table of Contents

Start With the Data You Actually Have

Decide which player-game statistic you are predicting and when the forecast will be issued. For this framework, the target is a receiver’s official receiving yards in a specific game. The forecast uses only information available at its stated publication time.

Game logs are a reasonable starting point, but they do not tell you everything about opportunity. Receptions show completed passes. Targets include unsuccessful opportunities. Routes describe involvement before the quarterback chooses where to throw. Offensive snaps are broader still, because some involve blocking or other assignments.

When route information is available and reliable, it gives you a way to separate playing time from target demand. When it is missing, use an explicitly labeled alternative, such as targets per team pass attempt. Do not rename a team-level denominator “routes” just because the resulting number looks plausible.

Check player identifiers, school changes, duplicate games, missing targets, and late corrections before fitting a model. A missing value is not automatically a zero. If a feed fails to identify the intended receiver, treating the play as evidence that nobody earned a target can bias your estimates.

For broader background, the ATSwins introduction to college football player prop models explains how different statistical targets require different inputs. This article concentrates on allocating receiving opportunities consistently within one offense.

Build a Team Passing Opportunity Budget

Project the team’s passing volume before handing targets to individual receivers. If an offense is expected to attempt 30 passes, separate player projections should not quietly assume a combined 40 targeted throws. Player estimates need a shared opportunity budget.

Keep the denominator consistent. Dropbacks include plays that do not become pass attempts, such as sacks. Pass attempts may include throwaways or other plays without an assigned receiving target. A target-share model should state whether its denominator is all attempts or only attempts with identified targets.

Suppose a hypothetical offense projects for 32 pass attempts, including two attempts without an assigned receiver target. That leaves 30 targeted attempts. Giving one receiver a 24% share produces 7.2 expected targets. Fractional targets describe an average across possible games; nobody needs to receive part of a throw.

The remaining shares must fit inside the remaining budget across wide receivers, tight ends, running backs, and other eligible options. If a teammate is unavailable, redistribute opportunities based on expected personnel and assignments. Giving every healthy player a full increase would count the same missing workload several times.

If you have dependable route data, a different path starts with expected dropbacks, route participation, and targets per route. Reconcile those player estimates with team volume too. The ATSwins discussion of snap counts and player opportunities provides related context for that approach.

Translate Targets Into Receiving Yards

A basic college football receiving yards model can multiply expected targets by catch probability and expected yards per reception. Using 7.2 targets, a 65% catch probability, and 14 yards per catch produces 65.52 expected receiving yards. That is a transparent baseline, not proof that the inputs are accurate.

There is a reason to separate the pieces. A short completion and a deep sideline attempt do not need the same catch probability or expected gain. If the receiver’s route mix changes, carrying forward both his old catch rate and his old yards per catch may tell an inconsistent story.

One improvement is to model target types separately, then add their expected contributions. Estimate the opportunities, completion chance, and conditional receiving gain for each category. Categories should be mutually exclusive and supported by your data, rather than detailed enough to sound impressive but too thin to estimate.

Another option uses expected yards per target directly. That measure already incorporates incomplete targets, so multiplying it by catch probability again would discount incompletions twice. Choose one coherent calculation and follow its definitions throughout the pipeline.

Receiving gains include what happens after the catch. If your expected yards per reception already represents the full credited gain, do not add yards after catch a second time. Accounting mistakes can make a projection look optimistic without any actual disagreement about the player.

Stress-Test a Receiver Projection

Return to the hypothetical 65.52-yard baseline. Suppose the receiver’s expected share of targeted attempts drops from 24% to 20%, with the team budget still at 30. His expected targets fall from 7.2 to six. Holding catch probability and yards per catch fixed, the estimate becomes 54.6 yards.

That is a 10.92-yard change caused entirely by opportunity. You do not need a new opinion about the receiver’s talent to explain it. A returning teammate, a different personnel package, or a revised playing-time assumption could justify investigating such a scenario, although none guarantees that exact adjustment.

Now keep the original 24% share but reduce the team budget to 26 targeted attempts. Expected targets become 6.24, and the same efficiency inputs produce 56.784 yards. Report that as roughly 57 yards rather than presenting unnecessary decimal places as confidence.

These controlled changes identify which assumptions drive the forecast. They are not complete simulations, because real inputs can move together. A game with fewer passes might also feature a different route mix, a different defensive response, or fewer late opportunities for the starter.

Create a central scenario and plausible alternatives before seeing the result. Explain what new information would move you between them. If an estimate only looks attractive under the most optimistic combination of volume and efficiency, that dependence belongs in the assessment.

Keep the Full Range of Outcomes

A receiver does not accumulate yards at a steady rate. An incomplete deep target can produce zero, while a completed version of the same opportunity can create a large gain. Low-volume outcomes and explosive plays both belong in the forecast.

A simulation might sample team passing opportunities, allocate targets, determine completions, and sample receiving gains. That is a proposed design, not evidence that a model has been built or validated. The assumptions connecting those steps matter as much as the number of simulation runs.

For example, receivers on the same offense share passing volume and compete for targets. Drawing every player’s targets independently can generate impossible team totals. Conditioning player opportunities on the same simulated game environment keeps those relationships visible.

Report the expected value, median, and a defined prediction interval when the model supports them. An interval describes projected outcome uncertainty, not a promise. If the model’s central 80% intervals repeatedly miss far more than 20% of comparable results, their historical coverage needs attention.

Avoid forcing a symmetric distribution just because it is easy to calculate. A receiving-yard forecast may need room for zero catches, a low typical outcome, and an occasional long gain. The right shape is something to test against unseen games.

Choose and Test the Modeling Approach

Start with a baseline that is easy to audit, such as a recent target average multiplied by a consistently estimated yards-per-target figure. Compare it with a component model before deciding whether a more complicated machine-learning method adds useful accuracy.

A direct regression model predicts receiving yards from its inputs. A component model predicts intermediate quantities first. Either can be useful, and either can fail. Component models offer clearer explanations, while errors in several stages can compound. A direct model can capture relationships without making its opportunity assumptions obvious.

Use chronological evaluation. Train on earlier games and test on later ones, with preprocessing and tuning confined to the training process. Keep games from the same week together where appropriate, so information from a later kickoff cannot slip into an earlier forecast.

Do not feed the model a season average calculated after the game it is supposed to predict. The same restriction applies to end-of-season defensive ratings and retrospective injury labels. Historical data must represent what could have been known at the forecast time.

Measure yardage error and probabilistic performance separately. Mean absolute error describes typical prediction error. Calibration examines whether events assigned similar probabilities occur at approximately those rates across enough observations. Prediction interval coverage checks the reliability of the projected ranges. No single metric establishes that a model is useful for every purpose.

Use Probability When Evaluating a Line

A 66-yard projection against a 59.5-yard line leaves a 6.5-yard gap. That difference does not reveal the probability of exceeding the threshold. You need the distribution, because two receivers with the same average can have very different chances of reaching 60 yards.

Suppose a hypothetical model estimates a 55% chance of exceeding the line. At decimal odds of 1.90, the break-even probability is approximately 52.63%, calculated as one divided by 1.90. Modeled expected profit per $1 staked is $0.045, calculated as 0.55 multiplied by 1.90, minus one.

Those calculations do not validate the 55% estimate. If uncertainty about the player’s role moves the estimate below break-even, the apparent advantage disappears. Record the available threshold, price, and forecast time together, and reassess when any of them changes.

The ATSwins guide to comparing college football projections with prop lines covers that broader comparison. This receiver framework supplies the opportunity and efficiency assumptions that should come before it.

Receiver Prediction Model Questions

Are targets more useful than last week’s receiving yards?

Targets describe opportunity that final yardage can hide, but they still need context. Eight targets in a temporary replacement role mean something different from eight targets in an established starting assignment. Use recent yardage to describe what happened, then examine targets, playing time, and the team’s passing plan to decide what is likely to carry forward.

What if reliable route data is unavailable?

Build a simpler forecast using consistently defined team pass attempts, player targets, and receiving efficiency. Label that limitation clearly. You can still evaluate a target-based baseline, but you should not claim to measure route participation without route observations. Missing detail is a reason to adjust confidence and model scope, rather than invent a more sophisticated input.

Does a quarterback change require a new projection?

It deserves a review of the assumptions. Reconsider expected passing volume, target allocation, completion probability, and receiving efficiency using whatever evidence is available. A change does not automatically mean every receiver gets fewer yards. When comparable evidence is limited, use alternative scenarios and a wider uncertainty range instead of applying the same unexplained adjustment to every player.

Make the Forecast Explainable Before Using It

Before trusting a college receiver projection, ask what team passing budget supports it, how the receiver earns his share, and whether the efficiency assumptions fit his expected role. If those pieces cannot be explained, a precise total is difficult to evaluate.

Keep a record of the assumptions before kickoff. Afterward, separate a missed workload expectation from poor receiving efficiency or an ordinary unusual result. That distinction gives the next revision a purpose. Changing the model after every disappointing game without identifying the error can make the process worse.

Use ATSwins’ linked college football analysis as a starting point for evaluating forecasts, then bring the same scrutiny to your own numbers. A useful receiving prediction model gives you an estimate you can inspect, a range you can question, and a clear reason to update when the evidence changes.