December 14, 2023. A baseball writer is preparing his Cy Young ballot. The candidates include a left-handed starter who has produced a career year — 2.71 ERA, 224 strikeouts, an FIP that suggests his underlying performance was even better than the headline numbers. The writer’s gut, after a season of watching the pitcher, is to vote him first. The writer’s analytical training, however, asks a different question: what was my prior expectation for this pitcher entering the season, and how strongly should I update on the evidence of a single season? The pitcher’s career numbers, before 2023, suggested a solid but not elite starter. The 2023 evidence is strong. The Bayesian update on the pitcher’s true quality is meaningful but, when properly calibrated, less dramatic than the year’s headline numbers alone would suggest. The writer votes the pitcher second. Two months later, the pitcher signs a six-year, $215 million contract with another team. By the second year of the contract, his ERA has regressed to 3.85. The writer’s Bayesian discipline, applied to the award vote, was vindicated.
Bayesian updating in sports — the structured method of revising beliefs about player and team quality in light of new evidence — is, in 2026, the analytical posture that most reliably distinguishes durable writing from the cycle’s shorter-half-life takes. The framework is not new. It dates to 18th-century probability theory. Its application to sports has been the project of analysts since at least the 1990s. What has changed in the last decade is the data infrastructure that makes Bayesian thinking actionable rather than aspirational. With clean play-by-play data, sufficient career-length samples, and robust prior expectations, the modern analytical writer can apply something close to formal Bayesian updating in real time, producing evaluations that survive the cycle of single-season hype.
I have been writing about analytical method across sports since 2018, and the framework that has shaped my own approach to evidence and revision is the one this article is about. Bayesian updating in sports — what the method actually is, how it applies to player evaluation, team analysis, and award voting, and how to incorporate the discipline into writing without sounding like a probability textbook, is the subject of this article.
The origin: where Bayesian updating came from
The Bayesian framework was articulated by Reverend Thomas Bayes in his posthumously published 1763 paper “An Essay towards solving a Problem in the Doctrine of Chances.” Bayes’ theorem provides a mathematical rule for updating the probability of a hypothesis as new evidence becomes available. The mathematical formalism remained largely academic through the 19th and 20th centuries, gradually entering applied statistics and decision theory in the post-war period.
The sports application emerged through the work of probabilistic forecasters in the late 20th century. Bill James‘ early sabermetric work used Bayesian-flavored reasoning without naming it as such — projecting future performance by combining prior expectations with new evidence, weighted by sample size and the inherent variance of the metric being measured. Nate Silver‘s PECOTA system at Baseball Prospectus in the 2000s formalized the framework. Silver’s later work at FiveThirtyEight extended Bayesian reasoning to political forecasting, sports, and other domains.
The contemporary sports application is most visible in preseason projections, award-voting writing, contract negotiations, and retrospective player evaluation. Each of these uses Bayesian-flavored reasoning to combine prior expectations with new evidence in ways that the naive “trust the headline numbers” approach systematically fails to do.
How Bayesian updating works
The basic structure has three components. The prior: your starting belief about the quantity of interest (a player’s true quality, a team’s expected wins). The evidence: the new data you’ve observed. The update: how strongly the evidence should shift your prior, calibrated to the evidence’s strength and the prior’s confidence.
The mathematical formula is Bayes’ theorem: P(H|E) = P(E|H) × P(H) / P(E). In plain language: the probability of your hypothesis given the new evidence equals the probability of the evidence assuming your hypothesis is true, multiplied by your prior probability for the hypothesis, divided by the overall probability of the evidence. For practical analytical writing, the formal math is less important than the conceptual habits.
The key habits are: start with a clearly articulated prior before looking at the new evidence; weight the evidence by sample size and variance, not just by its headline values; update incrementally rather than swinging to extremes; and maintain calibrated uncertainty about the updated belief, recognizing that the next round of evidence may shift it further.
The critical component: priors matter
The single most important habit in Bayesian sports writing is making your prior explicit. What did you expect from this player or team before the new evidence arrived? A clearly stated prior makes the writer accountable to a position that the new evidence is being measured against. It also helps the reader calibrate how strongly to update.
A player with a career-long ERA of 4.10 who posts a 3.50 ERA in one season has produced evidence. The Bayesian update depends on the prior. If the prior was 4.10, the new evidence shifts the estimate of true quality to somewhere between 3.80 and 4.00 — meaningful but not extreme. If the prior was “this pitcher might be on the verge of a breakout based on changed mechanics, scouting reports, and underlying metrics,” the same 3.50 evidence might shift the estimate to 3.65. The same evidence, different priors, different conclusions.
Bayesian vs alternative frameworks: a comparison
| Framework | How it handles evidence | Strengths | Weaknesses |
|---|---|---|---|
| Bayesian updating | Combines prior with evidence, weighted by variance | Calibrated uncertainty, durability | Requires explicit prior; less narratively dramatic |
| Frequentist null-hypothesis | Tests whether evidence is statistically significant against null | Established statistical framework | Doesn’t directly answer “what should I believe” |
| Naive trust-headlines | Treats latest evidence as definitive | Simple, immediate | Ignores prior; vulnerable to single-sample hype |
| “Sample size disclaimer” | Acknowledges variance but doesn’t formally update | Captures uncertainty | Often used as hedge rather than discipline |
| Eye-test pure | Subjective integration of evidence with prior intuition | Captures non-quantifiable context | Vulnerable to bias; not transparent |
The honest analytical writing combines Bayesian discipline with eye-test context. The framework provides the structure; the watching provides the context.
What the data needs
Bayesian writing requires career-length data for the prior, recent data for the evidence, and variance estimates for the metrics in use. Most modern sports databases (Baseball Reference, Basketball Reference, FBref) provide career-length data. Variance estimates require either domain knowledge or formal statistical computation.
Building the analysis
- State your prior explicitly before reviewing new evidence.
- Pull the new evidence with appropriate sample-size context.
- Calibrate the update based on the evidence’s strength.
- Hold the updated belief with appropriate uncertainty.
- Write the piece with explicit acknowledgment of how the update moved you.
Where this gets weird: common mistakes
The “I always thought so” hindsight inflation. After evidence confirms a position, it’s easy to retroactively claim that was always your prior. The cleaner version writes down the prior in advance.
Over-updating on small samples. The most common mistake is treating a 50-game sample as if it were a 1,000-game one. Sample-size discipline is the most underused tool in sports writing.
Under-updating on strong evidence. The opposite mistake is refusing to revise priors even when the evidence has shifted decisively. Bayesian discipline is symmetric — it should produce updates in both directions.
The “Bayesian Reasoning is Just Hedging” trap. Used carelessly, the framework can produce writing that always says “well, with uncertainty…” and never commits to claims. Good Bayesian writing makes claims with calibrated confidence, not perpetual hedging.
When Bayesian updating shines
Award voting. Single-season awards are the canonical Bayesian use case. Career priors plus single-season evidence produce more durable votes than headline-trusting approaches.
Trade and contract evaluation. A player coming off a career year may or may not be a smart contract bet. The Bayesian frame is the framework for that judgment.
Long-form retrospectives. A piece evaluating a player’s career arc should explicitly handle how priors evolved as evidence accumulated. The framework produces more honest retrospective writing.
Predictive writing. Forecasting next-season performance benefits from explicit Bayesian discipline rather than naive trust in last season’s numbers.
A working example: Aaron Nola’s contract negotiation
Aaron Nola’s 2023 free-agency winter is the case study. Nola was coming off a career year — 32 starts, 199 innings, 3.78 ERA but underlying FIP of 3.49 suggesting his actual performance was even better than the headline. The Phillies offered him a six-year, $172M extension. Other teams floated similar offers. The Bayesian frame asked: what was Nola’s career prior, how much should we update on the 2023 evidence, and what contract value does the updated estimate justify?
Nola’s career through 2022 — solid #2 starter, occasional Cy Young consideration, durable but not elite. The 2023 evidence shifted the prior modestly upward. The Bayesian-calibrated contract estimate was probably $140-$160M over five-six years, not the $172M the Phillies extended. The team paid the Bayesian-overpriced contract for narrative and clubhouse reasons that are legitimate but not strictly Bayesian.
By 2024-25, Nola’s actual performance had regressed closer to his career prior. The Phillies, in retrospect, had paid a Bayesian-inflated price for a player they valued for non-Bayesian reasons. The contract is not, by any standard, a disaster; it is just a clean illustration of why Bayesian discipline matters in long-term contract evaluation.
The limits
Bayesian writing cannot eliminate uncertainty. The framework produces calibrated uncertainty, which is better than ignoring uncertainty entirely but doesn’t produce false confidence.
Bayesian writing cannot resolve disagreements about priors. Two analysts with different career-baseline expectations will produce different updates from the same new evidence.
Bayesian writing can feel deflating in the moment. A career-year that the framework treats as partly variance and partly genuine improvement is less narratively satisfying than “this player is now elite.”
One additional limit: Bayesian discipline is hardest to apply to extraordinary cases. A player whose career arc produces multiple extreme outcomes — Joe Burrow’s rookie year, then ACL recovery, then 2023 elite production — is structurally hard to model because each new data point can dominate the prior.
FAQ
Do I have to use formal probability math?
No. The conceptual habits — stating priors, weighing evidence by sample size, updating incrementally — produce most of the practical benefit without formal calculations.
How do I know if I’m being Bayesian correctly?
The cleanest test is to write down your prior explicitly before reviewing the new evidence. If your final position can be reconstructed as “I had X prior, the evidence moved me by Y, I now hold Z” — you’re being Bayesian.
Does this work for predicting games?
Yes, in the structured forecasting sense (probability of outcome). For individual game outcomes, variance dominates. For long-run forecasts (playoff probabilities, season win totals), Bayesian reasoning produces measurably better predictions than the alternatives.
Who writes Bayesian-style sports analysis well?
Nate Silver (FiveThirtyEight era and current), Zach Lowe (NBA), Bill Connelly (CFB), the Baseball Prospectus team. The framework, even when not named, is the methodological backbone of careful sports writing.
Sources and further reading
- FiveThirtyEight — the canonical home of Bayesian-style sports forecasting in English.
- Baseball Prospectus — the sabermetric outlet that has long applied Bayesian-flavored reasoning to baseball.
- Bayes’ theorem (Wikipedia) — the underlying mathematical framework.
- Nate Silver, The Signal and the Noise — book-length treatment of probabilistic forecasting across domains.
The Cy Young vote that opened this article — second place for a pitcher who had a great year but whose career prior suggested regression — was the kind of Bayesian discipline that the cycle’s alternative would not have produced. The writer was right. The framework is, in a real sense, the disciplinary backbone of careful sports writing. For the broader frame on analytical method in sports, our primer on sports analytics is the natural starting point.
