The Changing Shape of Hot-Air Balloon Accidents
How aircraft size, commercial operations, and the limits of accident data shape what we can learn about balloon safety.
The story in four findings
01 · The mix changed
Recorded events fell from per year in the 1980s to in 2020–25, and their size mix shifted: giant balloons were involved in of recorded events in the 1980s and of in 2020–25.
02 · The commercial share rose with size
Revenue sightseeing was identified in of recorded events involving balloons of AX7 or less, and in of those involving giants (1982–2025).
03 · Larger balloons carried more people
In recorded revenue-sightseeing events, the median number of people aboard rose from for AX7 or less to for giants.
04 · The limit is essential
A smaller share of people aboard giants was injured or killed () than aboard AX10 balloons (). The difference is exploratory: points, 95% interval . Without flight exposure by balloon size, these records cannot establish relative accident risk.
All findings describe recorded U.S. NTSB events, not the full population of balloon flights.
Chapter 1 · The changing mix
The mix of balloon sizes in recorded events changed while the yearly count fell
Excludes events involving more than one balloon. Events per year is a descriptive normalization by calendar years, not an exposure-adjusted rate. Source: NTSB CAROL; balloon sizes from The Hot Air Balloon Database catalog. Giant = over 211,888 cu ft (FAI AX‑11+).
Chapter 2 · The commercial connection
The larger the balloon, the larger the revenue-sightseeing share
Share of U.S. single-balloon NTSB records identified as revenue sightseeing, by rated envelope size, 1982–2025.
“Not identified” does not mean private or noncommercial flying. 19 events flagged as revenue sightseeing also record the operation as “Personal”; they are counted by the sightseeing flag. The flag is recorded for every event in 1982–2025. These counts describe recorded events, not flights or the fleet.
Excludes events involving more than one balloon. Source: NTSB CAROL; balloon sizes from The Hot Air Balloon Database catalog. Giant = over 211,888 cu ft (FAI AX‑11+).
Chapter 3 · More people aboard, different outcomes?
In recorded revenue-sightseeing events, larger balloons carried more people
Each mark is one U.S. single-balloon NTSB event flagged as revenue sightseeing, 1982–2025: people aboard and people injured or killed. Larger marks stand for several events with the same counts.
Outcomes are NTSB counts for people aboard: killed, seriously injured, minor injury or uninjured. Each person is counted once, at their most severe injury.
Of everyone aboard these events, the share injured or killed
Among people aboard recorded events, not a measure of accident risk.
Shares pool all people aboard within a size class. The giant vs AX10 difference is exploratory; its uncertainty is in Chapter 4.
of recorded events, 1982–2025: excludes involving more than one balloon, not flagged as revenue sightseeing ( of them with operation type unknown; not counted as revenue sightseeing) and whose balloon size is unknown. Describes recorded events only; not a rate or a measure of risk. Source: NTSB CAROL; sizes from The Hot Air Balloon Database catalog.
The lower observed share for giant balloons raises a question: does that difference persist when we examine different periods and exclude influential events?
Chapter 4 · What the data cannot tell us
The giant share hurt was lower in every prespecified cut, with wide intervals
Giant minus AX10: difference in the share of people aboard who were injured or killed, in U.S. single-balloon NTSB events flagged as revenue sightseeing, 1982–2025. Exploratory.
Lines are 95% bootstrap intervals: uncertainty about the difference among these recorded events, not the range of outcomes in individual events.
The exploratory analysis examined 25 intervals (six cuts by four outcome measures, plus one); the outcome switch shows all 30 exported. Some excluding zero would be expected even with no true difference. The bootstrap resamples events as if independent; some operators appear in several events and of events have no operator recorded, so the intervals may be too narrow. No decade-adjusted model was fitted. 10,000 replicates, seed 20261110. Source: NTSB CAROL; sizes from The Hot Air Balloon Database catalog.
What the records show, and what they cannot
Recorded balloon events became fewer per year while their size composition changed: larger balloons accounted for a greater share of recorded events, and the larger the balloon, the larger the share of its recorded events identified as revenue sightseeing.
Within events flagged as revenue sightseeing, giant balloons carried more people, but a smaller share of those aboard was recorded as injured or killed than in AX10 events. For that measure, the difference was negative in all six prespecified comparisons, but the 95% intervals for the full period and both within-decade comparisons include zero. The comparison is exploratory.
The essential distinction: these are patterns among recorded events, not measures of accident risk. Without reliable information on how often balloons of each size flew, the records cannot establish whether one size was safer than another.
Sources and references
Data
- National Transportation Safety Board. CAROL — Aviation Investigation Search. Category: Balloon, all dates; JSON export downloaded 10 November 2026 (895 events). The record of every event in this story. Onboard injury counts are present from 1982; the revenue-sightseeing flag is recorded for every event from 1982.
- The Hot Air Balloon Database. thehabdatabase.com, a community archive of recorded balloons, via Kaggle: Matt Huff (matthuff4), “Hot Air Balloon” dataset (licence listed by the uploader: MIT). Used only to look up each balloon model’s rated envelope volume (a catalog rating, not a measurement); 790 of 870 single-balloon events could be sized.
- Makeover Monday 2026, Week 45: Hot Air Balloons. The workbook used for the size lookup. This story began as my makeover for that week: “The typical balloon stayed the same size. Giants became more common.” (9 November 2026; source).
Definitions
- Fédération Aéronautique Internationale (2026). Sporting Code Section 1 – Aerostats, version 3.0, effective 1 August 2026, para 2.1.2. Size classes AX7 or less to AX11+. “Giant” = AX-11 or larger: rated volume over 6,000 m³ (211,888 cu ft).
- 49 CFR § 830.2, Definitions (NTSB notification and reporting rules). An aircraft accident is an occurrence in which a person suffers death or serious injury, or the aircraft receives substantial damage. The record also holds a small number of incidents (3 of the 227 events in Chapter 3).
Context and checks
- Ballard SB, Beaty LP, Baker SP (2013). Hot-air balloon tours: crash epidemiology in the United States, 2000–2011. Aviation, Space, and Environmental Medicine 84(11):1172–1177. doi:10.3357/ASEM.3719.2013; PMC4888601. Cross-check of the record filter: they report 78 balloon-tour crashes with 518 occupants in 2000–2011; the CAROL export holds 77 revenue-sightseeing accidents involving 78 tour balloons and 524 occupants. Supportive, not an exact match.
- Federal Aviation Administration. General Aviation and Part 135 Activity Survey, CY2024. Reports active aircraft and flight hours for “lighter-than-air” aircraft (balloons and airships together), not by balloon size. That is why these records cannot estimate accident risk by size.
Methods and tools
- Analysis: reproducible R pipeline with an analysis record (findings, dated corrections, tests). Every number on this page is exported by R and checked; the browser only draws. Code, derived data and analysis record: github.com/poncest/balloon-accidents-story. The balloon-size lookup derived from The Hot Air Balloon Database is not redistributed.
- Graphics: D3 v7.9.2 (ISC licence). Type: Big Shoulders, DM Sans and JetBrains Mono (SIL Open Font License 1.1).
These are descriptions of recorded NTSB events. They are not accident rates, and nothing here estimates the risk of flying in a balloon of any size.