Trust

How we decide what to publish.

Airlines are very good at producing numbers that flatter them. The only way to be useful in that environment is to be harder on ourselves than anyone else would be. These are the rules every Row30 article is held to, and most of them are enforced by software before a human reads the draft.

Sourcing

Primary sources, always. Every statistic comes from a federal dataset we pull ourselves: the Bureau of Transportation Statistics, the DOT Air Travel Consumer Report, the Senate Lobbying Disclosure Act database, the Federal Election Commission, the eCFR, and the federal courts. We do not cite a blog citing a press release citing a study.
One number, one source, one date. A figure without its period is a defect. Every stat on this site names the year or range it covers, and links the dataset it came from.
Self-interested numbers are labelled as claims. When an airline publishes a figure about itself, or a compensation company publishes a figure about how much money goes unclaimed, we attribute it to them rather than stating it as fact.
If we computed it, we show the arithmetic. Derived figures get their full working published, including the assumptions and the range the estimate moves across.
We publish the caveats that weaken our own story. Every statistic in our database carries a caveats field, and those caveats survive into the article. Cutting one for length is not permitted.

Statistical standards

We publish rankings and comparisons, which is exactly the kind of content that misleads most easily. These rules exist to stop us doing it by accident.

Uncertainty ships with the estimate. Point estimates travel with an interval. Where two airlines or airports have overlapping intervals we say they are statistically indistinguishable rather than ranking one above the other, because the margin of error on a difference between two numbers is roughly twice the margin on either one.
We do not say "statistically significant". The American Statistical Association recommends retiring the phrase, and the thresholds behind it. We report the interval and describe what it means for the question at hand.
Percent and percentage point are not the same word. A move from 20% to 22% is two percentage points, not two percent. Our automated checks reject the confusion.
Small samples do not get ranked. Below a credible sample size an estimate is context only. It never takes a rank position or a headline, and it is listed separately as insufficient data.
We disclose what we searched. Our database holds roughly 7,000 pre-computed comparisons. Picking the most surprising one and publishing it is textbook data dredging, so every ranking states the universe it scanned and the filter it applied, both decided before the results were looked at.
Correlation is never written as cause. Associated with, coincides with, moves alongside. Never causes, drives, or leads to. Our validator blocks causal verbs near a correlation statistic.
We never multiply a count by a ceiling. The $2,150 bumping cap and the $4,700 baggage cap are federal maximums. Multiplying either by a passenger count would overstate reality by orders of magnitude. We show the count and the cap side by side, never their product.
Definitional breaks are annotated, never crossed silently. The DOT changed both its on-time reporting threshold and its reporting basis on January 1, 2018, and changed what counts as a complaint in mid-2023. Data on either side of those lines describes different populations, and we say so rather than drawing one trend line through them.

One trap worth naming outright. The federal government computes average delay minutes across delayed flights only, and cancelled flights sit outside that denominator. That means an airline can improve its average delay by cancelling its worst flights. So any delay ranking we publish shows the cancellation rate for the same airline and period in the same place, because the delay figure on its own is misleading.

What we refuse to publish

Advice. We publish what the rules provide and what courts have held. We never tell a reader what they should do, because that is legal advice and we are not a law firm.
Outcome promises. No service can guarantee that an airline pays. We do not imply otherwise, and we never attach a dollar figure to an individual reader’s situation.
Testimonials, reviews, or invented experiences. We do not publish success stories or reviews of any kind. Federal rules prohibit fake testimonials whether written by a person or generated by software, and any figure about typical outcomes has to be a measured number rather than a flattering anecdote.
"Results may vary" as a fix. The Federal Trade Commission has tested that disclaimer and found it does not undo a misleading impression. If we ever publish an outcome figure it will sit next to a substantiated typical result, not a hedge.
A page whose only content is a reworded regulation. The government publishes the regulation better than we can. A rights page on Row30 earns its place through what surrounds the rule: its amendment history, the cases interpreting it, and our data on how often the underlying problem actually happens.

Our methodology page lists the specific widely-circulated airline statistics we decline to use, and why.

Review and accountability

Automated validation runs before a human reads the draft. Every number in an article is matched against the database row it claims to come from. If a digit does not reconcile, the article cannot be published. The same check confirms the period is stated, the citation is carried, the licence permits the use, and no comparative claim outruns its interval.
A human reviews it, and someone is accountable for it. Analysis produced in-house is published under an organizational byline, the Row30 Data Desk, which is a common convention for statistical reporting. Editorial responsibility for it rests with Row30’s founder and editor, reachable through our contact page. Guest contributors who write for us carry their own personal byline and their own profile page.
We never put a name on work that name did not do. We do not buy bylines, we do not credit a reviewer who did not review, and we never invent an author, a headshot, or a credential. Software is never listed as an author. If no individual did the work, the byline is the desk, which is always the honest option.
Currency is stated, not implied. Each article shows the period its data covers separately from when the page was last edited. We only advance the modified date when the text or a number actually changed, never on a rebuild.

Corrections

If we get a number wrong we correct it, say so on the page, and date it. An error involving a number almost always merits a correction rather than a quiet edit, and we use the word “corrected” rather than a softer one.

A correction that changes a figure propagates to everywhere that figure appeared: the article, any ranking table built on it, the newsletter, and anything embedded elsewhere. Federal statistical practice requires corrections in all product formats, and we hold to the same standard.

Found an error? Tell us. Every correction we have made is logged at /corrections.

Independence and funding

Row30 takes no money from airlines. There is no airline advertising on this site, no sponsored content, and no affiliate relationship with any carrier. We are funded by the flat fee passengers pay to have a complaint filed, and by nothing else.

That fee creates its own conflict of interest, so it is worth stating plainly: Row30 makes money when passengers file complaints, which gives us an incentive to make airline performance look worse than it is. The statistical rules above, and the automated checks behind them, exist to remove our discretion in exactly the places where that incentive would bite.

What Row30 is not

Row30 is not a law firm, does not provide legal advice, and does not represent passengers in legal proceedings. We publish data journalism and education. For advice about your own situation, consult an attorney. More on that here.

The sources we pull

The standards we hold ourselves to

None of the rules above are original to us. They are the published standards of federal statistical agencies, the American Statistical Association, and the fact-checking community, applied to airline data.

How our research is produced, including our use of AI

Row30’s analysis is produced with AI assistance and reviewed by a named human before publishing. The pipeline that pulls the federal data, computes the statistics and calculates the confidence intervals is conventional software, not a language model. Drafting and structuring the write-up is AI-assisted.

Row30 Data Desk. Before a page can publish, an automated check matches every number in the draft against the stored source row. If a digit does not reconcile, the article is blocked. The same check confirms the period is stated, the citation is carried verbatim, the licence permits the use, and no comparative claim outruns its stated uncertainty.

Software is never credited as an author. A named person is accountable for every published figure, and every article records who reviewed it and when. We do not generate author profiles, headshots, testimonials, reviews, or quotes attributed to people who do not exist.

Last reviewed July 2026. Maintained by the Row30 Data Desk. Editorial responsibility rests with Row30’s founder and editor, reachable via our contact page.