Private projects overrun too. The difference is who counts.

The story is familiar to anybody who has worked on both sides. A government project is tendered and goes to the lowest bidder, who may have won by underbidding, and then reality sets in. Scope is added all the way through. Every payment sits behind the supervising consultant, the cost adviser and the finance function. Ownership is spread across agencies, utilities and municipalities, none of which has the project as its priority. A private project has one owner who holds the land, the decisions, the design and the money, pays quickly, and finishes.

Most of that is true as a description of the rules, and two later pieces take the rules one at a time: the payment chain, and the technical score. This one is about the conclusion, because the conclusion is where the story goes wrong, and the evidence for that is not an opinion. It is the first dataset in this field to test the question.

The technical score, taken on its own terms

, read: The score that is not a numberThe score that is not a number

What the study of ownership found

The largest sample of transport project cost data of its kind at the time, 258 rail, road, bridge and tunnel projects in 20 countries, was used in 2004 to test three explanations for cost escalation: length of implementation, size of project, and type of ownership. Ownership was known for 183 projects, and the authors split it three ways rather than two: private, state-owned enterprise, and ordinary public ownership by a ministry.

Every limit of that finding travels with it. Fifteen projects with ownership data, four of them private, is a small table, and the authors say so. There were no private rail or road projects in the sample at all. The state-owned-enterprise result may be confounded by geography, since three Japanese high-speed lines drive it. So the defensible reading is narrow: the data do not support the private-good story, and the one ownership type that stands out is neither private nor ordinarily public. It is the corporation owned by government, which the authors describe as falling between two stools, lacking the public control of a ministry and the market pressure of a private firm. For any owner whose capital work is delivered through state-owned companies, that is the finding to sit with rather than the public-private one.

Two years earlier the same authors had written the sentence that should have settled the debate before it started: the data are insufficient to decide whether private projects perform better or worse than public ones, and nobody knows. They also explained why nobody knows. Public projects have to reconstruct actual costs from fiscal-year accounts, which is slow but possible. Private data are "often classified to keep them from the hands of competitors", and both kinds of owner hold back actual costs because escalation is an embarrassment.

That is the first half of the answer. Government projects are counted. Their overruns are in a monthly report, an audit finding, a parliamentary answer. A private project's overrun is in a board paper nobody outside will read, and if the company is healthy the project completes and operates, which is what the story at the top calls finishing without apparent disruption. Both things are true of the same project. It overran and it finished. The public one did both in public.

Where private and public projects have been counted together, the gap closes. A study of 1,471 IT projects found an average cost overrun of 27% and one project in six over 200%; the sample was 92% public agencies, and the authors report little difference between those and the private companies in the rest. The same paper carries a regional case it presents as a success: a bank's core-system replacement, doubled in scope by a merger mid-way, delivered 18% over cost and 7% late. In this field that counts as spectacular, and it is still an overrun.

One body does hold a large record of private owners' projects, and it sells access to it. Independent Project Analysis benchmarks industrial capital projects, oil and gas, chemicals, minerals and power, for the companies that own them. Its own page describing its founder's book on megaprojects puts the failure rate against the owners' own business objectives at 65%, drawn from that proprietary database. The figure is therefore the consultancy's published account of its own data, and is carried here as that. It is nonetheless the closest thing in public to an answer to the question "how do private projects do", and the answer it gives is not "they finish on budget".

What the counting shows when it is done

The story's remaining points, utilities and municipalities and clearances, are the multi-principal problem, and the best dataset for it is India's statistics ministry's monthly count of central government projects above Rs 150 crore, now in its 486th issue, which until recently carried the reasons implementing agencies give for delay. The list of causes that circulates in the region, land, clearances, financing, detailed engineering, equipment supply, is that report's list.

Two things in that table are worth more than the headline. The gap between 831 delayed against the original date and 554 against the latest approved date is the rebaselining this site has written about before: 277 projects that are on schedule only because the schedule was moved. And the reasons are self-reported by the agencies being measured, which is the same instrument the dashboard article showed returns the reporter's own beliefs. Land acquisition and clearances are real and they are the multi-principal problem in its purest form, but a list compiled by the delayed party is evidence of what the delayed party says.

Every programme performs well against a baseline it can move

, read: Every programme performs well against a baseline it can moveEvery programme performs well against a baseline it can move

Then the series changed. The April 2026 report, generated from the ministry's new portal, gives 1,981 ongoing projects, an original cost of Rs 37,12,662 crore and a revised cost of Rs 42,78,402 crore. It carries no count of delayed projects and no reasons for delay; the document was searched in full for both. A cost overrun of 15% is visible by subtraction. Time overrun, which two years earlier was 831 projects and 35 months, is no longer in the public report at all. Whatever the reason for the change, the effect is the one this piece is about: a public report that made public projects look bad no longer carries the number, and the public projects now look like private ones, which is to say uncounted.

The region's own auditors have counted too, and what they count is where the delay sits. Jordan's Audit Bureau, in its 73rd annual report, lists the findings it meets repeatedly on government works tenders, and Saudi Arabia's General Court of Audit publishes thematic summaries of its audit results for 2023 and 2024.

Two auditors, two countries, and the list is the readiness list: studies, land, permits, coordination, supervision. Neither auditor's first finding is the winning bid. And the two reports differ in the one way this piece keeps returning to. The Jordanian report names the tender, the value in dinars and the date of the Bureau's letter; the Saudi report gives the theme, the sector and the count of findings closed. Both are public, both are the auditor's own selection of what to publish, and a reader can check one of them clause by clause.

What the private owner actually has

The story's second half deserves the same attention as the first, because it is right about something. A private project that overruns still completes and operates, and the reason is not virtue. It is structure, and the structure can be named.

One principal. The owner of a private project holds the land, the design, the money and the decision, and the same person answers for all four. That is the accountability the 2004 authors said mattered more than ownership, and its absence is what they found in the state-owned enterprises that escalated by 110%, owned by government but answerable neither to a ministry's budget nor to a market.

Money committed rather than appropriated. A private sponsor who approves a project has, in principle, approved its cash; a public owner has approved this year's budget line, and public procurement law commonly provides for next year's falling short. A project whose money is committed does not stretch by statute.

A payment chain the owner sets. Fourteen days or 80 is a choice for a private owner and a regulation for a public one, and the contractor's price follows the choice. The second piece in this set measures that chain against the international contract form and three public rule-books.

The freedom to design the incentive. The 2002 paper's remedy for strategic misrepresentation was not privatisation but accountability, and one of its four instruments was private risk capital in the project, so that somebody who loses money when the estimate is wrong is present when the estimate is made. A private owner can put their own money at risk in the contract and shape the contractor's reward around the outcome; a public owner is bound to approved forms.

None of these is an outcome. The IT study's regional success, 18% over and 7% late, is what these four things produce when they work. The private owner's advantage is not that the project comes in on budget. It is that fewer parties stand between the decision and the money, so that when it does not, somebody decides quickly what to do, pays for it, and opens the doors. That is copyable.

The principal-agent point, which the story has right

Beneath the rules is the argument most of the literature converges on. The 2002 paper set out two families of explanation for the 86% of projects whose costs were underestimated: optimism, which is self-deception and produces error; and strategic misrepresentation, which is deliberate and produces what the authors, defining the term carefully, call lying. Their finding was that the second fits the data better than the first, because error would be random and the bias is not. The 2009 paper gave the two their names, delusion and deception, and its abstract says the private sector, the public sector and the partnerships between them all have a dismal record on large-project promises.

The principal-agent framing is right, then, but it does not divide public from private. It divides every party that estimates from every party that pays. A contractor's estimator, a consultant's cost adviser, a ministry's planning department and a company's project sponsor all have reasons to present a number the approver will accept, and the only ones who face a published correction afterwards are the public ones. The remedy the 2002 paper proposed is not privatisation. It is accountability: transparency, performance specifications, explicit regulatory regimes, and private risk capital even in public projects, so that somebody who loses money is in the room when the estimate is made.

Which reading this takes

The claim that public projects overrun and private ones do not is widely believed and unsupported by the published data. That could be read as showing the belief simply false, or as showing that it describes something real in the wrong words. This piece takes the second. What people who hold the belief have seen is true: public projects are counted and private ones are not, public money is annual, public owners are many where private owners are one, and the rules that govern the award and the payment are different. Each of those is a design with a stated reason, and each can be argued with on its own terms. None of them is ownership.

The limits. The ownership test rests on 15 projects. The IT sample was 92% public. The IPA figure is a consultancy's account of its own proprietary data. The Indian data are self-reported by the agencies measured, and the two audit reports are the auditors' own selection of findings. No private-sector dataset of comparable size exists in public, which is the finding rather than a gap in it.

What to examine

• Ask who, by name, answers for the land, the design, the money and the decision. If the answer is four organisations, the project has the structure the auditors' findings describe, whoever owns it.
• Ask whether the execution period was set to the appropriation or to the work. If to the appropriation, the schedule has an extension built in, and the recovery plan should say so.
• Ask what the studies, land, permits and approvals looked like on the day the contract was signed. The Jordanian Bureau's recurring findings are the checklist, and five of the six are things done or not done before anybody tendered: were the designs checked against the site, was the duration calculated rather than assumed, was there prior coordination with the other bodies concerned, were the approvals and licences obtained, and did the employer resource its own supervision.
• Ask, of anything already finished, whether it is occupied. The Bureau's sixth finding is completed projects standing empty for want of the employer's own approvals, which is a cost with no contractor in it at all, and the one an owner is least likely to have counted.
• Ask what will be counted at the end, against which baseline, and who will see it. If the answer is the latest approved baseline, the project will finish on time by definition.
• Ask, of any private project held up as the model, what it cost against its first approved budget. If nobody can say, that is the difference.

---

Sources. Flyvbjerg, Holm and Buhl, What Causes Cost Overrun in Transport Infrastructure Projects?, Transport Reviews 24(1), 2004, pp. 3-18, read in full from the authors' copy on arXiv (1304.4476); Table 1 for the ownership figures, the text for the tests and the authors' caveats. Flyvbjerg, Holm and Buhl, Underestimating Costs in Public Works Projects: Error or Lie?, Journal of the American Planning Association 68(3), 2002, read in full (arXiv 1303.6604); the 86% and 28%, the statement that nobody knows whether private projects perform better, the passage on private data being classified, and the two explanations. Flyvbjerg and Budzier, Why Your IT Project May Be Riskier than You Think, Harvard Business Review, September 2011, read in full (arXiv 1304.0265). Flyvbjerg, Garbuio and Lovallo, Delusion and Deception in Large Infrastructure Projects, California Management Review 51(2), 2009, abstract read (arXiv 1303.7403). Independent Project Analysis, its own page on its founder's book, 11 January 2012. MoSPI, Flash Report on Central Sector Projects, May 2024, and Flash Report on Central Sector Infrastructure Projects, April 2026, both read in full. Audit Bureau of Jordan, 73rd Annual Report, for 2024, chapter 8 (report page 60) and pages 51 to 52, read at the document. General Court of Audit, Saudi Arabia, Reports on the results of audit work for fiscal years 2023 and 2024, read in full. All read 6 and 7 September 2026.