Classify the dataset before interpreting the model
Plane-model data can answer different questions depending on what each row and denominator represent. Start by classifying the table before interpreting the model name: is it a record of events, a rate using an exposure measure, or a comparison whose inputs do not align? This classification keeps a descriptive record from being read as a conclusion it was not designed to support.
Use the event unit first. A source may count crashes, accidents, incidents, or fatalities, and those categories are not interchangeable. Crash definitions vary substantially across methodologies, so numerical crash-rate comparisons require caution when the analyses do not use a common definition.
Next write the table’s scope beside its unit: the aircraft population, geography, observation period, reporting coverage, and denominator. A comparative aircraft-safety study used exposure-adjusted results rather than relying solely on raw accident counts. That detail identifies the kind of comparison being attempted; it does not show that every model table measures risk in the same way.
The audit’s output is a bounded description of what was counted, what exposure was used, and which conclusion the available inputs can support. When those notes are missing, keep the result at the level of the dataset rather than treating a model label as a safety conclusion. Recorded frequency also does not establish what caused an individual crash or who bears legal responsibility.
Write a one-line description of the table: state its event unit, observation period, geography, aircraft population, reporting coverage, and denominator. Then classify it as a recorded-event count, an exposure-based comparison, or a comparison whose inputs do not align. This small record helps keep the table’s descriptive content separate from conclusions about model safety, the cause of an individual crash, or legal liability.
Separate crashes, accidents, incidents, and fatalities
Aircraft event vocabulary comes before aircraft-model comparisons. “Crash,” “accident,” “incident,” and “fatality” do not describe the same measurement, so combining them can change both the numerator and the apparent ranking.
A crash may be used informally for a severe event, but the cited methodologies use different criteria for what counts as an accidental crash. The assessment reviewed multiple approaches and found that definitions of events considered aircraft crashes vary substantially. It did not convert those rates to a common definition. That means two datasets can use the same word while counting different events.
In the United States, NTSB guidance describes an aircraft accident as an occurrence involving a fatality, serious injury, or substantial aircraft damage. These reporting requirements help organize official records, but they do not make every dataset interchangeable.
A fatality is an outcome, not a substitute for the event definition. A dataset counting all accidents will answer a different question from one counting fatal accidents, fatal injuries, hull losses, or total deaths. Before comparing models, write down exactly what the numerator includes.
Use this event-definition check:
- Event label: Does the source count crashes, accidents, incidents, or another category?
- Severity threshold: Does it require death, serious injury, substantial damage, or no particular outcome?
- Counting unit: Is the unit an aircraft, an occurrence, a flight, a person, or a fatality?
- Outcome filter: Are fatal and nonfatal events combined or separated?
- Reporting rule: Are events included because they meet a regulatory reporting threshold, appear in a voluntary report, or satisfy a study-specific rule?
- Treatment of multiple aircraft: Does one collision count as one event or more than one aircraft-level observation?
If you cannot answer these questions for both models, describe the totals as recorded counts within their respective datasets. Do not present them as comparable model-level crash rates.
Define the aircraft population, geography, and period
The aircraft population determines what a model total means. A comparison of large commercial jets cannot answer the same question as a comparison that combines turboprops, private aircraft, helicopters, military aircraft, and general-aviation airplanes. Even the phrase “commercial aircraft” needs a stated boundary.
The FAA’s 2022 commercial-jet summary covered certified jet airplanes greater than 60,000 pounds maximum gross weight. It included aircraft in temporary nonflying status and aircraft used by non-airline operators, while excluding commercial airplanes operated in military service and aircraft manufactured in the CIS, USSR, or PRC. That scope is narrower than “all planes,” and the reported results should not be extended beyond it.
A model comparison should disclose at least these population details:
- Aircraft type, such as jet or turboprop.
- Weight, certification, or operational category.
- Commercial, private, military, or general-aviation use.
- Whether temporarily inactive aircraft are included.
- Whether the unit is a model family, variant, generation, or individual type designation.
The model name also needs precision. Combining them may hide differences; separating them may leave too few observations for a stable comparison. The correct choice depends on the question and the data structure, so the published method should state it rather than leave you to infer it.
Geography and time period are equally material. A worldwide dataset cannot automatically establish a United States pattern, and a 2025 global figure cannot be compared directly with a U.S. series from an earlier period without explaining the mismatch. Record the country or region, the operating environment, the start and end dates, and whether the period refers to the accident date, reporting date, or publication date.
The FAA summary illustrates why dates matter: its fleet, flight-hour, departure, and aircraft counts describe the worldwide commercial-jet fleet in 2022. They are a 2022 reference point, not a timeless description of the fleet. Use the same observation period for each model whenever possible. If periods differ, label the difference and avoid a direct rank.
Use exposure denominators instead of raw event totals
A count tells you how many events were recorded. An exposure denominator helps express how much opportunity existed for those events to occur. Without that denominator, a model with more aircraft or more departures may look worse simply because it operated more often.
Boeing describes accident rates as accidents per million departures and says departures, or flight cycles, have a stronger statistical correlation with accidents than flight-hours, the number of airplanes in service, passenger miles, or freight miles. That does not make every other measure irrelevant. It means the denominator must match the event and operation being studied, and the source should explain why it selected that measure.
The FAA’s 2022 summary reported 58.7 million flight hours and 26.3 million departures for its covered worldwide commercial-jet fleet. It also reported 28,385 total airplanes in that fleet, including 13,492 Boeing airplanes. Those figures show why fleet size, flight hours, and departures are different quantities. They should not be substituted for one another without an explanation.
When you review a model comparison, ask what each measure contributes:
- Departures or flight cycles: Useful when the analysis measures event frequency per takeoff-and-flight cycle and the source supports that choice.
- Flight hours: Relevant when time in operation is the exposure being studied, but not interchangeable with departures.
- Fleet size: Shows how many aircraft were available or included; it does not show how heavily each aircraft was used.
- Aircraft age: Helps you identify whether the comparison mixes different service histories or changing fleet composition. The source should state how age is measured and used.
- Operating context: Route length, airport environment, weather, operator practices, and aircraft use can affect what the data represent. A defensible comparison should say whether those conditions are combined or separated.
A useful rate is not simply a smaller number. It is a number whose numerator and denominator refer to the same population, period, geography, and operating question. If a source provides only raw totals, call them raw totals and keep the conclusion narrow.
The cited comparative study provides the right general discipline: it used exposure-adjusted results so its statistical interpretation would not rely solely on raw accident counts. You should look for the same separation in any model comparison. If the source shows only counts, it has not yet shown a comparable accident rate.
What broad aviation statistics can and cannot show
Broad aviation statistics can establish the scale and direction of a defined aviation population, but they do not automatically produce an aircraft-model ranking. In 2025, ICAO reported 38.08 million scheduled-commercial flight departures worldwide, compared with 37.09 million in 2024 and 38.79 million in 2019. Those figures describe scheduled-commercial departures worldwide, not the exposure of every aircraft model or every U.S. operation.
IATA reported an all-accident rate of 1.32 accidents per one million flights in 2025, with 51 total accidents. Its five-year average for 2021–2025 was 1.27 accidents per one million flights. IATA also reported 1.03 jet accidents per one million flights and 4.08 turboprop accidents per one million flights in 2025.
Those figures can help you distinguish a rate from a count and a jet category from a turboprop category. They cannot tell you that one named jet model is safer than another, because the cited figures are not presented as a compatible model-by-model comparison with matching model exposure, definitions, and geography.
Use broad statistics as context, not as a shortcut. Keep the following questions visible:
- Does the statistic cover worldwide scheduled commercial operations, a reporting organization’s members, or a U.S. regulatory dataset?
- Does “flight” mean a departure, a completed flight, or another unit?
- Does “accident” use the same definition as the model-level dataset?
- Are jets and turboprops separated, and are individual models separated within those categories?
- Does the observation period match the model comparison?
A category-level rate may show that aircraft groups have different reported rates in a specified dataset. It does not identify the cause of an individual event, and it does not support a definitive ranking among models within either group.
What the FAA's 2022 Commercial-Jet Snapshot Can and Cannot Show
An informative way to read the FAA’s 2022 commercial-jet summary is to start with the population it measures. The summary covers certified jet airplanes greater than 60,000 pounds maximum gross weight. It includes aircraft in temporary nonflying status and aircraft used by non-airline operators, while excluding commercial airplanes operated in military service and CIS/USSR/PRC-manufactured airplanes. A figure from this summary therefore describes that defined worldwide fleet, not an undifferentiated set of all aircraft operations.
Within that scope, the summary reports 28,385 airplanes, 58.7 million flight hours, and 26.3 million departures in 2022, including 13,492 Boeing airplanes. Those figures provide exposure context for interpreting event totals. They do not, on their own, show how events were distributed among individual aircraft models.
Boeing describes accident rates as accidents per million departures and says departures, or flight cycles, have a stronger statistical correlation with accidents than flight hours, the number of airplanes in service, passenger miles, or freight miles. The practical implication is narrow: a model comparison needs a denominator that matches the population and period of its event count. The fleet totals are useful for seeing the scale of exposure, but they are not a model-level result.
6. Check reporting coverage. Ask which authority, operator group, database, or study supplied the events. Determine whether the source includes every event within its stated scope or only events that met a reporting threshold. Preserve the source’s own scope instead of treating it as a complete census of all aviation activity.
7. Check the method and uncertainty. Look for confidence intervals, significance testing, sensitivity checks, or an explanation of how rare events were handled. A small difference between model rates may be less informative than the difference in the data’s uncertainty. If the method does not show how the result was calculated, describe the comparison rather than ranking it.
8. Check revisions. Boeing says departure data are continually updated and revised as new information and estimating processes become available, so accident rates may vary between editions of its publication. Record the edition or retrieval date and note whether later revisions could change the denominator or rate.
9. Check the conclusion against the inputs. A source may support “Model A had more recorded events in this dataset” without supporting “Model A is less safe.” It may support a category rate without supporting a model ranking. Keep the conclusion no broader than the aligned inputs.
The audit produces a clean decision rule: proceed to comparison only when the event definition, population, geography, period, reporting coverage, and denominator are compatible. Otherwise, report what the source records and identify the comparison as incomplete.
Why a published comparison may still be incomplete
A published table can look precise while still leaving material questions unanswered. Aviation data change as records are corrected, new information is added, and exposure estimates are revised. Boeing says departure data are continually updated and revised, so accident rates may vary between editions of its publication. A rate should therefore be read with its edition, date, and methodology rather than treated as a permanent number.
Reporting coverage also matters. The FAA’s Aviation Safety Reporting Program is voluntary and collects reports about actual or potential aviation-safety deficiencies and discrepancies. The program invites pilots, controllers, flight attendants, maintenance personnel, other airspace users, and any other person to submit written aviation-safety reports. A voluntary safety report can be valuable information, but its presence or absence does not by itself establish the frequency of accidents for a model.
Treat incomplete coverage as a question to test, not a detail to hide. Ask whether the source includes only events reported to a particular authority, only a defined operator group, or only records that meet a threshold. Ask whether the exposure denominator covers the same aircraft population as the event records. Ask whether aircraft that left service are represented consistently in both the event history and the exposure calculation.
Fleet composition can also change over time. A model may have a different number of aircraft in service, a different distribution of ages, or a different operating mix at the beginning and end of a study period. If the method treats those conditions as constant, you need to know that before interpreting a small difference.
Survivorship deserves a direct check. If the denominator reflects aircraft that remain in service but the numerator covers a longer historical period, the comparison may not describe the same population. You should ask how withdrawn, destroyed, retired, or otherwise absent aircraft were handled. Do not assume that a surviving fleet is a neutral sample of the original fleet.
Rare events also create statistical noise. A few events can move a rate substantially when exposure is limited, while a large exposure base can make a difference appear stable even when operational conditions changed. An exposure-adjusted result is a stronger starting point than a raw total, but it still needs compatible definitions, scope, and uncertainty reporting.
When a published comparison leaves these questions unanswered, use it as a description of the published dataset, not as a final safety judgment. Record the limitation next to the number. That keeps a data gap from becoming an unsupported ranking.
Read U.S. aviation records without turning them into a national ranking
U.S. aviation records can help you identify and investigate individual events, but an event record is not automatically a model-level risk dataset. The FAA accident-and-incident webpage contains individual records identified by aircraft or flight, location, and date. That structure is useful for locating a particular occurrence and separating one event from another.
Those requirements explain how certain records enter the official process; they do not turn every report into a common statistical denominator.
The FAA’s voluntary Aviation Safety Reporting Program adds another kind of information. It collects reports about actual or potential aviation-safety deficiencies and discrepancies and invites submissions from aviation personnel, other airspace users, and any other person. A safety report may identify a hazard or concern without being an aircraft accident. Keep those records separate from accident counts unless the methodology explicitly combines them.
For a U.S. comparison, use a record-by-record workflow:
- Identify the aircraft or flight, location, and event date.
- Confirm whether the record is labeled an accident, incident, or safety report.
- Note the aircraft category and model designation exactly as recorded.
- Check whether the record covers one aircraft or multiple aircraft in the same occurrence.
- Keep the U.S. observation period separate from worldwide statistics.
- Find the matching exposure measure before calculating a model rate.
A list of FAA records can show that events were recorded within the page’s scope. It cannot, by itself, show that every event involving every model was reported in the same way, that the models had equal exposure, or that one model has a higher underlying risk. Keep the U.S. record useful for identification without turning it into a national ranking it was not designed to establish.
A crash pattern does not establish cause or legal liability
A model-level pattern and the cause of a particular crash answer different questions. A count or rate describes how events are distributed within a defined dataset. It does not explain why one event occurred, identify the conduct or condition that caused it, or establish who is legally responsible.
Several events involving the same model could have different causes. They might involve different operators, maintenance histories, flight phases, weather conditions, airport environments, crew decisions, component failures, or other circumstances. Conversely, a single event can involve more than one contributing factor. A model frequency cannot replace the investigation of the specific occurrence.
The same boundary applies to legal responsibility. A higher recorded frequency does not prove that an aircraft manufacturer, airline, owner, maintenance provider, crew member, or another party caused a particular injury or crash. A lower frequency does not rule out responsibility in an individual event. Liability depends on the facts, applicable law, evidence, and the parties involved.
Use model data for context only. To understand a specific crash, separate the aircraft’s identity from the event’s facts, preserve the relevant records, and distinguish established findings from early reports or assumptions. Do not use a raw model total as a substitute for a cause analysis or an individualized legal assessment.
The safe analytical boundary is this: frequency can describe a pattern, but it cannot by itself establish cause or legal liability. Keep those conclusions open until the particular event and governing legal questions have been evaluated.
What a defensible aircraft-model comparison must disclose
A defensible aircraft-model comparison should let you answer six questions before you trust its conclusion:
- What event is counted? The source defines crashes, accidents, incidents, fatalities, or another outcome consistently across the models.
- Which aircraft are included? The model, variant, aircraft category, operator type, and fleet boundary are stated.
- Where and when were events observed? Geography and dates match, or the difference is plainly disclosed.
- What exposure denominator is used? Departures, flight cycles, flight hours, fleet size, or another measure is tied to the same population and period.
- How complete and changeable is the record? Reporting coverage, revisions, fleet changes, survivorship questions, and statistical uncertainty are addressed.
- What does the conclusion actually establish? The wording stays at the level supported by the data and does not turn frequency into causation or liability.
Crash definitions vary across methodologies, the comparative study used exposure-adjusted results, and Boeing says departure data can be revised between publication editions. Those conditions make raw totals insufficient for a reliable model comparison.
When you encounter a claimed ranking, apply the audit before relying on it: align the event definition, aircraft population, geography, dates, reporting coverage, and denominator. If one of those pieces is missing, keep the conclusion descriptive and provisional. If you are dealing with an individual aviation injury or crash, a qualified aviation lawyer must assess whether the facts support a particular legal path; a model-level statistic cannot make that determination for you.
Questions readers may still have about plane-crash comparisons
Can FAA or NTSB records show which aircraft model has the most crashes?
They can help identify recorded events by aircraft or flight, location, and date, but they do not automatically provide a complete, comparable model ranking. You would still need a consistent event definition, aircraft population, observation period, geographic scope, reporting coverage, and exposure denominator. Without those elements, the records support a description of reported events—not a reliable conclusion about model-specific crash risk.
What should an aircraft-model crash comparison table include?
It should state the event category being counted, the exact model or model family, aircraft population, geography, observation dates, reporting source, and denominator. The denominator might be departures, flight cycles, flight hours, or another exposure measure, but it must match the numerator’s population and period. The table should also identify revisions, uncertainty, and any excluded aircraft or events.
Can worldwide aircraft accident rates be used to rank U.S. aircraft models?
Not by themselves. Worldwide statistics and U.S. records may cover different operations, definitions, periods, and reporting systems. For a U.S. model comparison, use data with a defined U.S. scope and matching model-level exposure whenever possible. A worldwide category rate can provide context, but it does not establish a U.S. ranking among named aircraft models.
How does aircraft age affect comparisons of crash rates by model?
Aircraft age helps reveal whether a comparison combines fleets with different service histories or changing fleet composition. It should be examined alongside the model definition, exposure measure, and operating context rather than treated as a standalone explanation for an observed difference. A sound source should state how age is measured and incorporated; otherwise, keep the conclusion descriptive and qualified.

Emery Brett Ledger brings 28+ years of experience to personal injury law. He founded & led The Ledger Law Firm in securing over $100 million in compensation for clients with life-altering injuries & complex claims. Licensed in California, Texas, & Washington, Emery earned his law degree from Pepperdine University School of Law. His practice areas include car & truck accidents, wrongful death, catastrophic injuries, maritime claims, & mass tort litigation. He has been recognized by The National Trial Lawyers’ Top 100, Mass Tort Trial Lawyers Top 25, and America’s Top 100 Personal Injury Attorneys. Emery also received the 2025 Elite Lawyer Award & holds a perfect 10.0 Avvo rating with Platinum Client Champion status.
