Survivorship Bias: Why Every Champion Looks Inevitable When You Already Know Who Won

June 19, 2016. Game 7 of the NBA Finals. The Cleveland Cavaliers, down 3-1 in the series ten days earlier, complete a comeback that no NBA team had ever pulled off, and Kyrie Irving’s go-ahead three-pointer with 53 seconds left becomes one of the most rewatched shots in the league’s history. In the years since, the 2016 Cavaliers have been narrated, in retrospective coverage, as a team of destiny. The pieces talk about the team’s resilience, LeBron James’s leadership, the way Cleveland’s championship drought ended on a play that “had to” go in. The pieces are warm. The pieces are well-written. The pieces are also, by a wide margin, the wrong story about why Cleveland won. Survivorship bias is the reason.

Survivorship bias is the analytical sin at the heart of most retrospective sports coverage, and it is the most common reason that smart writers, after a championship, end up writing dumb pieces. The mechanic is simple: we know who won, and we work backwards from the result to construct an inevitability that did not exist at the time. The losing team’s similar attributes — the same coaching philosophy, the same roster construction, the same late-game habits — vanish from the analysis because they did not produce a trophy. The 2016 Warriors and the 2016 Cavaliers had more in common, structurally, than most championship-versus-runner-up duos. The narrative cleaved them. The math, in retrospect, never did.

I have been writing about how to read sports analytics carefully since 2014, mostly through the lens of fantasy football communities and the criticism of takes-driven coverage, and the cognitive error I find myself confronting in my own writing most often is the one I am about to walk you through. Survivorship bias, how it warps every sports retrospective, where it shows up in the data, and how to write around it without falling into its traps, is the subject of this article.

The origin: where survivorship bias came from

The concept has a famous origin story, and it is a real one. During World War II, the U.S. Air Force was trying to figure out where to put additional armor on its bombers to reduce the rate at which they were being shot down. The natural instinct was to study the planes that returned from missions, identify the locations where they had the most bullet holes, and reinforce those spots. The statistician Abraham Wald, working with the Statistical Research Group at Columbia, pointed out the problem. The planes that returned, by definition, had survived the damage they took. The locations with the bullet holes were the locations where damage was survivable. The locations without bullet holes — engines, fuel lines, cockpits — were where the planes that did not return had been hit. Wald’s recommendation was to reinforce the un-hit areas.

Wald’s paper, published in 1943 and re-discovered by a generation of behavioral economists in the 1980s, became the foundational text for one of the most underappreciated cognitive errors in any field. The pattern is universal: when you study only the successes, you mis-read what success required. The failures, which often had identical inputs, are invisible to the analysis. The conclusions you reach about “what works” are, more often than not, wrong.

The concept entered sports analytics writing through a few channels. Bill Barnwell wrote about it at Grantland in the early 2010s, applying the frame to NFL coaching decisions and front-office construction. Nate Silver’s work at FiveThirtyEight repeatedly returned to the theme in his coverage of championship probabilities. Michael Mauboussin’s The Success Equation (2012), which is not a sports book but should be on every sports writer’s shelf, devotes most of its argument to survivorship and selection bias in performance evaluation.

Twelve years later, the lesson has been incorporated into the academic conversation around sports analytics. It has not, in my opinion, been adequately incorporated into the public-facing writing. Most championship retrospectives are still structured as if the winners did something distinctive that the losers did not, when, in reality, the gap was often a single bad bounce.

How it works: survivorship bias in plain language

Survivorship bias in sports has two distinct forms. The first is narrative survivorship: we tell stories about the teams that won, and we structure those stories around traits that we infer caused the winning. The second is data survivorship: when we calculate the average behavior of “successful” teams or players, we exclude the failures by definition, which makes the average look more distinctive than it actually is.

The narrative version is more common in sports writing. A team wins a title. A columnist sits down to write the explanation. The columnist identifies the team’s distinctive traits — a free-spending owner, a charismatic coach, an unconventional roster construction — and frames those traits as the cause of victory. The piece is well-written. The piece is also vulnerable to a simple counterfactual: how many losing teams had the same traits? If the answer is “most of them,” the traits are not the cause. The trait is just the trait. The victory is the variance.

The data version is more insidious. Consider a public analytics piece arguing that “championship-winning teams have an average regular-season net rating of +6.5.” That number is true. It is also, in isolation, almost useless, because the average regular-season net rating of teams that lost the title in their conference finals is +5.8. The gap between champions and “almost champions” is much smaller than the gap between champions and the rest of the league. The article that quotes only the champions’ average has selected on the outcome variable, which means the conclusion it draws — that you need a +6.5 to win — is technically true but practically misleading.

The cleanest test for survivorship bias in any sports argument is: does this analysis include the teams that had the same inputs but did not win? If the answer is no, the analysis is, almost by definition, biased toward the result.

The critical component: the counterfactual

The single most important habit for writing around survivorship bias is to always, explicitly, name the counterfactual. The counterfactual is the world in which the inputs were the same but the outcome was different. A team won a Super Bowl on a 4th-and-1 quarterback sneak in the closing minute. The counterfactual is the world in which the sneak was stopped. What does the analysis look like in that world?

If the answer is “the analysis is unchanged” — that is, if the team’s underlying process was sound and the result was variance — then the article should describe the process, not the result. If the answer is “the analysis is opposite” — if the team’s underlying process was, in fact, badly flawed and the result was an outlier — then the article should describe the process honestly and call the result lucky. Most championship retrospectives do neither. They describe the result as inevitable.

Survivorship bias vs other cognitive errors: a comparison

Survivorship bias is one of a cluster of cognitive errors that pollute sports writing, and the distinctions matter:

The honest version of analytical writing names which errors it is most vulnerable to, and structurally protects against them. Survivorship bias is most acute in retrospective coverage; recency bias is most acute in weekly coverage; outcome bias is most acute in coaching evaluations. Each requires a different antidote.

What the data needs: building counterfactuals

The methodological correction to survivorship bias, in sports analytics, is to always include the comparison set of teams or players that did not win. The mechanics:

For team-level analysis: if you are writing about a championship team’s regular-season profile, include the conference finalists, the second-round losers, and the play-in seeds in the comparison set. The “what champions look like” question becomes “what playoff teams look like, and where on that distribution the champions fall.” The distribution is, in most cases, considerably less distinctive than the champions-only data suggests.

For coaching analysis: if you are writing about a coach’s decision-making, include the same decisions made by other coaches that did not produce the same result. A 4th-and-1 conversion that won a game looks different in retrospect than a 4th-and-1 conversion that lost one, but the pre-snap probabilities were nearly identical. The decision was good in both cases, even though the outcomes diverged.

For player analysis: if you are writing about a Hall of Fame profile, include the players with statistically similar peak years who did not, for various reasons, end up in the Hall. The “what makes a Hall of Famer” question becomes “what separates the borderline cases,” which is, almost always, the more interesting question.

The mechanical correction is, in practice, easy. You just have to ask “and how does this look against the comparison set” before every conclusion. The disciplinary challenge is that the answer is usually less dramatic than the champions-only narrative, and editors prefer drama.

Building the analysis: a working framework

The practical framework for writing around survivorship bias:

1. Identify the survivorship trap at the start. Before you write a retrospective, ask: am I about to attribute the result to traits that the losing teams also had? If the answer might be yes, structure the piece around process, not outcome.
2. Pull the comparison set explicitly. Champions plus conference finalists. Hall of Famers plus their borderline contemporaries. Successful coaches plus the ones who made the same decisions and lost.
3. Write the lede from the comparison data, not from the winner’s profile alone. “The 2016 Cavaliers were structurally similar to the 2016 Warriors; the difference was, in retrospect, three high-leverage possessions in Game 7” is a better lede than “The 2016 Cavaliers were a team of destiny.”
4. Name the counterfactual in the body. “In a world where Kyrie’s three doesn’t fall, this article reads differently” is a sentence you should be willing to write. If you are not willing, you are writing inevitability fiction, not analysis.
5. Edit out the inevitability language. “Had to win,” “destined to,” “only the Warriors could have.” These phrases are flags for survivorship bias. Cut them.

Where this gets weird: common mistakes

Even writers who know about survivorship bias fall into related traps.

The “survivor-of-survivors” trap. Once you start including conference finalists in your comparison set, you have not solved the problem; you have merely moved it one level deeper. The conference finalists themselves are survivors of the regular season. The truly fair comparison includes every team in the league, weighted by playoff probability. The math gets ugly. Most pieces stop one level short.

The “small-sample, big claim” trap. A piece arguing that “championship teams have an above-average emphasis on three-point shooting” is built on a sample of one team per year, over a few decades. That is, at most, a few dozen data points. Drawing strong claims about championship profile from such a small N is itself a form of bias — survivorship combined with sample size pretense.

The “we know they were great because they won” trap. This is the meta-version of survivorship bias. A team wins a title. The analyst, asked to evaluate them, defaults to “well, they won, so they must have been great.” The reasoning is circular. The team won because they were great; they were great because they won. Without an independent assessment, the analysis has no anchor outside the result.

The “comeback narrative” trap. When a team comes back from a deficit and wins, the comeback gets framed as evidence of underlying resilience. Often, the comeback is evidence of nothing more than the team having always been the better team — they fell behind because of variance, then reverted toward their true talent level. The comeback story flatters the team. The math, frequently, says they were never the underdog in the first place.

When the counterfactual approach shines: use cases

The strongest applications:

Coaching evaluation. A coach with three Super Bowl rings is, by survivorship logic, an elite coach. A coach with three Super Bowl appearances and one ring is, by the same logic, less elite. Both are oversimplifications. The decisions that distinguish elite coaching are usually distributed across regular-season games as much as championship moments, and the evaluation works better when you include process metrics — expected wins from decisions, fourth-down conversion logic, in-game adjustment quality — rather than ring counts alone.

Hall of Fame and award debates. The cleanest version of “is Player X a Hall of Famer” arguments includes the borderline cases who did and did not make it, with their statistical profiles laid out side by side. The argument shifts from “did they win enough” to “what does the distribution of Hall-eligible careers look like, and where does this player fall.” The conversation is more productive. The conclusions are less dramatic.

Dynasty assessment. Was the 2014-19 Warriors run a dynasty? The question is more interesting when you compare them to the 2010-13 Heat, the 2000-04 Lakers, the 2010-13 Spurs almost-dynasty, and the actual losses they took along the way. The “what makes a dynasty” question shifts from “five titles in eight years” to “what are the structural conditions that produce sustained excellence.” The conditions are usually less heroic than the narratives.

Roster construction post-mortems. A general manager’s tenure is usually evaluated by the championships won. The more useful evaluation includes the contemporaneous GMs who made similar moves, with similar inputs, and produced different outcomes. The “good GM” narrative often masks “lucky GM,” and vice versa.

The limits: what the counterfactual approach cannot tell you

The honest version of this writing names the limits.

The counterfactual approach cannot tell you who is going to win the next championship. It can correct retrospective inevitability bias and produce more honest evaluations of past performance. The prediction problem — which is a different problem, governed by different math — does not always benefit from the same toolkit.

The counterfactual approach can feel deflating. The narrative that champions are distinctively excellent is, in some ways, more pleasurable to read than the narrative that champions are slightly-better-than-runners-up who got the variance breaks. Editors and audiences often prefer the former. Writers who insist on the latter sometimes lose readers.

The counterfactual approach is itself vulnerable to bias. The comparison set has to be chosen, and the choice of comparison set can itself be motivated. A writer who wants to argue that Team X was a “true champion” can choose a narrower comparison set that flatters them; a writer who wants to argue that Team X was lucky can choose a broader one. The discipline is to make the comparison set explicit and defensible.

The counterfactual approach does not preclude the celebration of greatness. A team that wins because of underlying strength is a great team. A team that wins because of one bounce is, often, also a great team — they were one of the best in the league all year. The corrective is not to deny greatness; it is to be honest about the role of variance in any single result.

A working example: the 2021 Tampa Bay Buccaneers

The 2021 Tampa Bay Buccaneers are a useful case study in how survivorship bias warps championship narratives. The Bucs won the Super Bowl following the 2020 season with Tom Brady at quarterback, and the retrospective coverage was, almost uniformly, a story about Brady’s leadership, the team’s veteran experience, and the inevitability of a championship-caliber quarterback elevating a strong roster to a title. The narrative was warm. The narrative was popular. The narrative was also, by the underlying numbers, less distinctive than the coverage suggested. The Bucs’s regular-season point differential was good but not historic. Their road through the postseason involved a series of opponents who had each been favored, by Vegas, in their own playoff matchups. The Bucs were, in the aggregate, about a 30-35% favorite to win the title at the start of the playoffs, given their seeding and matchups. They cashed in on those probabilities. That is impressive. It is also not inevitability.

The 2022 Buccaneers were the same franchise with largely the same roster, and they were eliminated in the divisional round. The retrospective coverage of the 2021 win did not, in most cases, get updated to reflect that the 2022 result was, by the same math, equally probable. The 2021 piece had described inevitability; the 2022 result revealed that the inevitability was always a narrative imposed on an outcome that variance helped along. Holding both seasons together is, in my opinion, what honest sports retrospective should do. The franchise was very good in both years. The trophy was decided by margins that the writing rarely acknowledges.

Frequently asked questions

Is survivorship bias the same as outcome bias?

Closely related but distinct. Outcome bias is evaluating a specific decision by its result rather than its process — “the coach was wrong to go for it because they didn’t convert.” Survivorship bias is broader: studying only successful cases and inferring causes from them. Outcome bias is, in a sense, a single-instance form of the survivorship problem. The fixes are similar: in both cases, evaluate process and probabilities, not just outcomes.

How do I write a championship retrospective without falling into the trap?

Structure the piece around process, not outcome. Lead with what the team was doing well over the course of the season, not with the championship moment. Include the conference finalists or wild card teams in your comparison set explicitly. Name the high-leverage moments where variance could have flipped the result. Cut inevitability language. The result is a piece that respects what the team did without inflating it into prophecy.

Doesn’t this kind of writing make sports feel less special?

It makes the writing more honest, which is, in my experience, what sports analytics writing should aim for. Sports retain plenty of drama when described accurately; the drama does not require inflated narratives to function. The 2016 Cavaliers won an extraordinary championship. The math says the result could easily have gone the other way. Both things are true. Holding both in a single piece is what good analytical writing looks like.

What about the “champion’s mentality” argument?

It is largely survivorship bias. The teams that won are described as having a champion’s mentality; the teams that lost in the same circumstances are described as having “fallen short.” The mental traits — composure, resilience, late-game focus — are often inferred from the result rather than measured independently. When sports psychologists try to measure these traits prospectively, the predictive value is much smaller than the retrospective narratives suggest. The mentality narrative is, mostly, the story we tell ourselves about the team that survived.

Sources and further reading

• Abraham Wald’s wartime statistical work — the original survivorship-bias case study, still the most teachable example.
• Michael Mauboussin’s The Success Equation — a book-length treatment of survivorship, luck, and skill in performance evaluation.
• FiveThirtyEight’s sports archive — particularly Nate Silver’s repeated explorations of championship probabilities and the difference between process and outcome.
• Bill Barnwell’s archive at Grantland — the writing that consistently brought the counterfactual frame into NFL coverage.
• Farnam Street’s accessible explainer on survivorship bias — a clean, non-sports introduction to the cognitive error and its variations.

The 2016 Cavaliers, in retrospect, were not a team of destiny. They were a team that, in any given replay of those Finals, wins three or four times out of ten. The variance in their favor in the seven games we actually got was real but unspectacular: a hot Kyrie shooting night, a Draymond Green suspension, a Stephen Curry off-night for the ages. The pieces written in the months after framed it differently. The pieces written in 2026 still do. The counterfactual — the version where the Warriors close it out in five — would have changed the narrative completely without changing the underlying teams at all. For the broader frame on reading team trends without falling for the streak, our guide to reading team trends is the natural companion piece.