Flyvbjerg Project Database: The Iron Law Source
Bent Flyvbjerg's Oxford project database is the most-cited empirical source on big-project performance. We give the sample, methodology and citation URL.
What is the Bent Flyvbjerg database?
The Bent Flyvbjerg project database is the largest dataset of its kind in the world: more than 16,000 large projects, assembled over decades by Flyvbjerg and his team at Oxford's Saïd Business School and still expanding. For each project it records the budget at the decision to build against the final outcome, across rail, road, tunnels, dams, energy, IT, defense, and the Olympic Games. It is the empirical basis for Flyvbjerg's Iron Law of Megaprojects and for reference class forecasting. Only 0.5% of the projects in it come in on budget, on time, and with the promised benefits.
Average overrun by category (Flyvbjerg-Gardner 2023)
| Project category | Mean cost overrun | % of projects in the fat tail (50%+ overrun) | Mean overrun of that tail |
|---|---|---|---|
| Solar power | +1% | 2% | +50% |
| Energy transmission | +8% | 4% | +166% |
| Wind power | +13% | 7% | +97% |
| Tunnels | +37% | 28% | +103% |
| Rail | +39% | 28% | +116% |
| Airports | +39% | 43% | +88% |
| Buildings | +62% | 39% | +206% |
| IT | +73% | 18% | +447% |
| Nuclear power | +120% | 55% | +204% |
| Olympic Games | +157% | 76% | +200% |
| Nuclear waste storage | +238% | 48% | +427% |
From the base-rate appendix of Flyvbjerg and Gardner (2023), How Big Things Get Done (book; figure as quoted, not checked against the page). This is a book, and we have not checked every row against the printed page; the rows are as quoted in published summaries of the appendix, and a 2026 Public Management Review paper cites the same appendix page (192) for the nuclear-waste ranking. Overruns are real (inflation-adjusted) against the budget at the decision to build. The IT fat tail is the source of the widely-quoted 447% figure: the 18% of IT projects that go more than 50% over budget overrun by 447% on average.
What counts as a megaproject
Flyvbjerg's rule of thumb is that megaprojects are measured in billions of dollars, major projects in hundreds of millions, and projects in millions and tens of millions (Flyvbjerg 2014, Project Management Journal, read 8 October 2026). The 16,000-plus project database is wider than megaprojects: it spans 23 project types, from solar farms and IT systems to the Olympic Games.
The unit of analysis is the project, not the contract. Cost overrun is measured in real terms (inflation-adjusted) against the originally approved budget at the decision-to-build stage, not against later re-baselined figures. This is critical: many megaproject business cases re-baseline the budget partway through, which would otherwise mask the overrun.
Can you access or download the database?
Short answer: not the raw data. The full database is a proprietary research asset, assembled and curated over decades by Flyvbjerg and his team (originally at Oxford's Saïd Business School), and it is not published as a downloadable file. There is no public spreadsheet, API, or open data portal for the 16,000-plus project records themselves.
What is public is the analysis built on it, and that is what you cite:
- The category base rates (the table above) are published in the appendix of How Big Things Get Done (Flyvbjerg & Gardner, 2023). This is the most accessible form of the data.
- The methodology and underlying figures appear across Flyvbjerg's peer-reviewed papers, most of which are free to download from his arXiv author listing. The canonical overview is “What You Should Know About Megaprojects and Why”, read 8 October 2026 (Project Management Journal, 2014).
- The book itself: How Big Things Get Done (publisher page) (book; figure as quoted, not checked against the page).
If a source claims to link “the Flyvbjerg database” as a downloadable dataset, treat it with caution: it is almost certainly a derived table of the published base rates, not the original record-level data.
How to cite
Documents: How Big Things Get Done (book; figure as quoted, not checked against the page) and Flyvbjerg 2014 (arXiv author copy), read 8 October 2026.
Related references on this site
Frequently Asked Questions
What is the Flyvbjerg megaproject database?
Bent Flyvbjerg's project database is the largest known dataset of large project outcomes, assembled at Oxford's Said Business School. In its most-cited 2023 form (Flyvbjerg and Gardner, How Big Things Get Done), it covers over 16,000 projects across rail, road, dams, oil, gas, IT, defense, Olympics, nuclear, and other categories. It is the empirical basis for reference class forecasting.
What share of projects in the database actually deliver?
In the 16,000-plus project database (Flyvbjerg and Gardner 2023): 47.9% of projects come in on budget, only 8.5% come in on budget and on time, and only 0.5% come in on budget, on time, and with the promised benefits. This is Flyvbjerg's Iron Law: over budget, over time, under benefits, over and over again. Note that the separate, often-conflated '98% of megaprojects overrun' (by more than 30%) and '80% average cost increase' are from McKinsey's 2015 article The Construction Productivity Imperative, not from the Flyvbjerg database.
Which project type performs best in the database?
Solar power projects show the smallest average overrun (1% over budget), followed by energy transmission (8%) and wind power (13%). Nuclear waste storage facilities show the largest, with a mean overrun of 238%. The Olympic Games row reads 157%. A 2026 peer-reviewed update by Ansar, Flyvbjerg and Budzier puts the mean at 159% across the 23 Games from 1960 to 2024 with data, every one of which overran.
Can you access or download the Flyvbjerg database?
The full raw database is not publicly downloadable. It is a proprietary research asset assembled and curated by Flyvbjerg and his team, originally at Oxford's Said Business School. What is public is the analysis: the category base rates appear in the appendix of How Big Things Get Done (2023), and the methodology and many of the underlying figures are set out in Flyvbjerg's peer-reviewed papers, many of which have free author copies on arXiv. For citation you reference the book or the papers, not a dataset file.