The most common email I get after a game-analysis piece runs comes in two flavors. Half say I missed the obvious thing the data could have told me. The other half say I leaned too hard on stats and missed what the cameras showed.
Both kinds of email are often right. The piece below is the working method I use to avoid each failure mode separately. The numbers do work the eye test cannot. The eye test does work the numbers cannot. Pretending otherwise is the cleanest way to write a piece that ages badly inside a week.
This is the workflow. Five steps, two short reference tables, and the cases where each side of the conversation has to defer to the other.
Quick read: the five-step method in 60 seconds
- Watch the game first. Form an opinion before opening any stat page. Note one or two specific moments that shaped your read.
- Pull three layers of data. Box score, season context for the same metrics, opponent profile. Three layers, no more.
- Check where the data agrees and disagrees with what you saw. The disagreement is the article.
- Identify which side wins each disagreement. Sometimes the tape. Sometimes the stats. Knowing which uses experience and humility.
- Write the version that names both. Pieces that hide one side or the other read shallow within months.
The two failure modes
Most weak game-analysis writing fails in one of two specific ways. Naming them makes both easier to avoid.
The first failure mode is data without sight. A writer who never watched the game pulls a box score, finds a striking number, and writes 1,500 words built around it. The piece reads coherent until you realize it could have been written from a different city. It misses the substitution that swung the game, the scheme adjustment in the third quarter, the visible body-language shift after the missed call. The numbers describe what happened. They do not always tell you why. A piece built on numbers alone tends to mistake the symptom for the cause.
This is the most common failure mode on sites that scale aggressively. The Ringer and the longer-form analytics outlets generally avoid it because their editorial workflow requires watching. The faster the publishing cycle, the harder this failure mode is to dodge.
The second failure mode is sight without data. A writer watched the game, took notes, and writes the piece directly off those notes without checking whether the numbers agree. The piece reads confident until you realize the writer’s read of the game came from twelve high-leverage minutes and missed the fifty minutes where the underlying performance disagreed. A team can lose a game by twelve and still have been the better team across forty-eight minutes by every possession-based metric. The numbers know that. The eye test sometimes does not.
This failure mode dominated sports writing before 2010 and still appears in plenty of mainstream coverage. It is more frequent in football than in basketball, partly because NFL samples are smaller and partly because the game’s structure makes single moments feel more decisive than the data ultimately supports. The companion read on how single moments turn into trends lives in the small samples piece.
The five-step workflow, in a single table
The workflow below is the version I run before publishing any game-analysis piece. It does not require advanced data. It requires the discipline to do all five steps in order.
| Step | What it does | Common failure if skipped |
|---|---|---|
| 1. Watch the game with an opinion-forming intent | Builds the qualitative read before the numbers anchor you | Piece becomes a stat dump with no narrative spine |
| 2. Pull three layers of data (box score, season context, opponent profile) | Gives quantitative grounding without overwhelming the piece | Either too little data (assertion) or too much (data salad) |
| 3. Identify where the data agrees with what you saw | Confirms the obvious read; you can usually cut these sections | Piece spends words restating things the reader already accepts |
| 4. Identify where the data disagrees with what you saw | Surfaces the actual story; this is the part worth writing | Piece misses the interesting tension between eye and number |
| 5. Write the piece around the disagreement, naming the limits of both sides | Produces analysis that survives the next two months | Piece reads either as hot take (sight) or as cold spreadsheet (data) |
The workflow is simple. The discipline is doing all five on the same piece. The pieces that age well are the ones where the writer ran all five even when the deadline did not respect them.
Specific stats to look up by sport
The table below maps the sport to the three layers of stats most likely to either confirm or productively contradict your watching impressions. Pulling more than these per piece almost always hurts the writing.
| Sport | Box-score layer | Season context layer | Opponent profile layer |
|---|---|---|---|
| NBA / WNBA | True shooting %, usage, assists, turnovers, plus/minus | Player season averages, lineup minutes, on/off splits | Opponent defensive rating, recent results, key player availability |
| Soccer | xG, xGA, shots on target, possession, pressing intensity | Season xG per match, points-per-xG ratio, set-piece efficiency | Opponent recent xG profile, formation, key absences |
| NFL | EPA per play, success rate, third-down conversion, turnovers | Season EPA, DVOA, situational splits (red zone, third-and-long) | Opponent defensive EPA, pressure rate, injury report |
| College football | EPA per play, success rate, explosive play rate | SP+ rating, returning production, recent results | Opponent SP+, conference strength, key injuries |
| MLB | wOBA, BABIP, hard-hit rate, exit velocity | Season splits, recent rolling averages | Opponent pitcher tendencies, bullpen workload |
The public sources for these layers are mostly the same handful — Basketball Reference and the NBA’s official stats pages for basketball, FBref and Understat for soccer, Pro Football Reference and rbsdm.com for football. The trap is loading up on more layers than the table above suggests. Three layers is usually enough. Five becomes unreadable.
Where the eye test beats the stats
The eye test wins reliably in several specific categories. Knowing where it wins lets you weight your reading correctly when the two sides disagree.
Defensive scheme and coverage. Public box-score data captures very little about whether a team played zone or man, dropped or blitzed on third down, switched picks or fought over them. A beat writer watching tape can describe these in a sentence. The data, in 2026, mostly cannot — at least not at the public level. When the question is “what did the defense actually do,” the tape is ahead.
Personnel matchups and game-script adjustments. A team that scored 28 second-half points against a defense missing its primary safety is generating different evidence than the same score against a healthy unit. Stats summaries rarely surface the injury context in real time. The eye test, paired with the injury report, catches this faster.
Body language, momentum, and emotional state. The analytical research on momentum effects in sports is genuinely mixed. Sometimes the eye test is reading variance and calling it momentum. Sometimes it is reading something the data underweights. Either way, the eye test is the source for these claims; the stats cannot generate them.
Coaching adjustments inside the game. The third-quarter scheme shift, the rotation tweak, the timeout-driven momentum break — these are visible on tape and largely invisible in box-score data. The careful version of any analytics piece names these adjustments out loud, even when the underlying numbers do not flag them.
Where the stats beat the eye test
The numbers win, with equal reliability, in their own set of categories.
Sample-size honesty. The eye test naturally overweights memorable moments — the late-game three, the goal in stoppage time, the fourth-down conversion. The numbers price the full forty-eight minutes (or ninety, or sixty) on equal terms. A team that won a close game by three lucky plays at the end may have been outplayed across the rest of the contest. The stats know this. The tape, in real time, does not always.
Opponent-quality adjustment. The eye test sees a dominant offensive performance. The stats know whether the opponent’s defense was bottom-five in the league. Opponent-adjusted metrics like DVOA, opponent-adjusted EPA, and SP+ catch this faster than any single broadcast can. The framework on how these metrics earned their place lives in our durability piece.
Variance versus signal in shooting and finishing. A 47% three-point night by a 36% shooter is variance, not improvement. A striker scoring three goals from one xG is a finisher who got lucky, not necessarily a finisher who has improved. The eye test calls these breakouts. The numbers, paired with our regression to the mean piece, contextualize them honestly.
Cumulative pattern detection across multiple games. Single-game reads sit at the mercy of the night. Cross-game patterns — a team’s defense regressing toward its season baseline, a player’s role shifting, a coordinator’s scheme evolving — are easier to detect in the data than in any single game’s tape. The full vocabulary that supports these reads sits in our analytics field guide.
Frequently asked questions
What if I do not have time to watch the full game?
Watch the most leverage-rich segments and the final ten minutes, then rely on the data for the rest. Crucial caveat: name the gap in the piece. “I watched the second half and the closing minutes” is a different kind of analysis than “I watched the full game.” Readers can calibrate to that. They cannot calibrate to the silent gap.
How do I know which stats to pull when I do not know what I’m looking for?
Start with the box score and the season context layer. The opponent profile layer is the optional one that becomes useful when your eye-test read is unusual. If you noticed something that the score did not capture — a possession that should have produced more, a defensive sequence that punched above its weight — go to the opponent profile to see whether the matchup explains it.
What if the eye test and the data agree completely?
That is usually the boring case. A short note confirming the agreement is enough. The piece worth writing is the one where they disagree. If your eye-test read is “Team A was the better team” and every possession-based metric says the same, the article tends to be shorter and confirmatory. Save the long-form for the disagreements.
How is this different from just “doing analytics” or “watching tape”?
Pure analytics workflows skip the watching. Pure tape workflows skip the cross-check. The combined version takes longer, produces fewer pieces per week, and ages much better. The premium analytical sites — The Athletic in basketball coverage, the better soccer tactics sites, the careful football outlets — almost all run some version of this workflow even when they do not say so explicitly.
The takeaway, in one paragraph
Reading a game well in 2026 is not a choice between watching and reading the data. It is the discipline of doing both, naming where they agree, and writing the piece around where they disagree. The eye test wins schemes, matchups, body language, and in-game coaching adjustments. The data wins sample-size honesty, opponent adjustment, variance separation, and cross-game patterns. The pieces worth publishing run all five steps even when the deadline does not respect the workflow. For the wider vocabulary this method sits inside, our sports analytics field guide is the natural next read.
