The Changing Shape of Hot-Air Balloon Accidents
An interactive data story about how balloon size, commercial operations, and the limits of accident data shape what we can learn about balloon safety
→ The Changing Shape of Hot-Air Balloon Accidents
This page explains the question, the findings, their limits, and how the project was built.
What can accident records tell us about hot-air balloon safety, and what can’t they tell us?
I analyzed U.S. NTSB hot-air balloon records and built a four-chapter interactive story using R and D3. Recorded events became fewer per year, and larger balloons accounted for a growing share of them. Among events identified as revenue sightseeing, larger balloons carried more people.
The central finding is also the central limitation: these records describe accidents and incidents, not how often balloons of each size flew. They cannot establish whether one balloon size is safer than another.
Project at a glance
| Duration | November 2026 |
| Role | Analysis, editorial design, visualization, implementation |
| Deliverable | Four-chapter interactive data story |
| Tools | R, D3 v7, HTML/CSS, Playwright (automated checks) |
| Data | NTSB CAROL balloon records; rated sizes from The Hot Air Balloon Database |
| Live story | balloon-accidents.netlify.app |
1. The question and why it matters
The project began as my Makeover Monday 2026 Week 45 chart. Among hot-air balloons recorded in US states in The Hot Air Balloon Database, giants made up a growing share of recorded builds (2.2% of those built in 2000–2009, 9.9% in 2020–2026), while the median rated size stayed at 90,000 cu ft. Those are catalog records, not a census of the active fleet.
The natural next question was what the accident record shows about balloon size, and especially whether bigger balloons are safer or more dangerous. That question is easy to ask and easy to answer badly: counting accidents by size looks like evidence about safety, but it is not. The goal was a story that shows everything the records can say, and makes what they cannot say just as visible.
2. What the records show
The mix changed. Recorded events fell from 26.7 per year in the 1980s to 9.5 in 2020–25. Giant balloons were involved in 0 of 267 recorded events in the 1980s and 15 of 57 in 2020–25.
The revenue-sightseeing share rose with size. Revenue sightseeing was identified in 22 of 260 (8%) recorded events involving balloons of AX7 or less, and in 63 of 66 (95%) involving giants (1982–2025).
Larger balloons carried more people. In recorded revenue-sightseeing events, the median number aboard rose from 3 for AX7 or less to 11 for giants.
Injury shares differed, with wide uncertainty. Among people aboard recorded revenue-sightseeing events, 242 of 752 (32%) aboard giants were injured or killed, compared with 145 of 334 (43%) aboard AX10 balloons. This exploratory difference was −11.2 percentage points (95% interval: −26.0 to +3.6).
3. What the data cannot tell us
The analysis describes recorded NTSB accidents and incidents, not all balloon flights. Three limitations govern its interpretation:
- No size-specific exposure. Without knowing how many flights or flight hours each balloon size accumulated, event counts cannot be converted into accident rates or relative risk. The FAA reports flight hours only for “lighter-than-air” aircraft as a whole (balloons and airships together), never by balloon size.
- Selection into the record. An event is recorded as an accident only when someone is killed or seriously injured, or the balloon is substantially damaged. The records are not a representative sample of all flights, so the injury shares apply only to the recorded events analyzed.
- No causal inference. Differences across balloon sizes coincide with differences in operation, time period, and other factors. The analysis cannot establish that size caused any observed outcome.
The interval for the giant-versus-AX10 difference in injury share includes zero, and even a clearer difference would not resolve the missing exposure.
4. How I approached the problem
Data and populations
Every event comes from the NTSB CAROL export (895 balloon records). Each balloon’s model was matched to its rated envelope volume and grouped into FAI size classes, from AX7 or less to giant (AX11+, over 211,888 cu ft); 790 of 870 single-balloon events could be sized. Each chapter uses a defined subset. Chapter 1 covers 1980–2025 and shows events of unknown size as their own category. Chapters 2–4 begin in 1982, the first year with both the revenue-sightseeing flag and onboard counts: Chapter 2 includes 704 single-balloon events from 1982–2025, including 47 whose balloon size is unknown, and Chapters 3 and 4 use the 227 sized events flagged as revenue sightseeing with onboard counts present.
Four chapters
- The changing mix: recorded events per year by size and decade.
- Revenue-sightseeing classification by size: the share of recorded events identified as revenue sightseeing, by balloon size.
- People aboard: occupants and injury outcomes in revenue-sightseeing events.
- The limit: giants versus AX10 balloons across six prespecified comparisons, with bootstrap 95% intervals.
How I kept the analysis defensible
- Visible denominators. Every share is shown with its count (“63 of 66 (95%)”, not just “95%”), so small groups cannot pass for large ones.
- Limits beside the charts. Each chapter states its limitation next to the graphic, not in a footnote, and the uncertainty interval always appears with the word “exploratory”.
- Results computed in R and checked. I used R to calculate and validate every published number, then D3 only to draw it. Automated checks verify the displayed numbers, layouts at desktop and mobile widths, keyboard accessibility, and the fixed comparison scale. Separating the analysis from the browser made the results reproducible and the presentation testable.
Data attribution
Balloon sizes come from The Hot Air Balloon Database, a community archive, credited on the page with a note on exactly what they were used for. Because its redistribution terms are not confirmed, the size lookup stays private; the public repository holds the code, the exported story data, and the analysis record.
5. Outcome and reflection
Outcome
The result is a four-chapter interactive data story that connects historical accident records to balloon-size classifications, lets readers examine the evidence, and makes uncertainty and data limitations visible throughout. Rather than presenting accident counts as evidence of safety, it shows both what the records reveal and where interpretation must stop.
Reflection
The hardest part was editorial, not technical: deciding what not to claim, then making that restraint visible rather than burying it in footnotes.
6. Explore the project
Citation
@online{ponce2026,
author = {Ponce, Steven},
title = {The {Changing} {Shape} of {Hot-Air} {Balloon} {Accidents}},
date = {2026-11-11},
url = {https://stevenponce.netlify.app/projects/standalone_visualizations/sa_2026-11-11.html},
langid = {en}
}