← All posts

The Sydney Opera House and the planning fallacy: the cure is your own past

Everyone tells it as a parable about optimism, and the first budget was also set low on purpose.

Construction on the Sydney Opera House started in 1959, before the drawings were resolved and before the money was fully in place. Four years was the plan. It took fourteen.

The rest of the numbers are nearly furniture by now. 233 designs were entered in the 1956 international competition and Jørn Utzon of Denmark won it. The first estimate was £3.5 million, which reads as A$7 million after Australia's 1966 move to decimal currency at two dollars to the pound, with completion set for early 1963. Queen Elizabeth II opened the building on 20 October 1973, roughly ten years later than the original plan, at a final cost of A$102 million, about fourteen times the estimate, much of it paid for by a State Lottery (496 of them were drawn between 1958 and 1986).

That gap is why the Opera House opens almost everything written about the planning fallacy, including the 1994 paper by Buehler, Griffin and Ross that made it the standard illustration. Told that way it's a story about optimism: good people, a beautiful idea, an honest guess that went wide.

The budget that was argued for, not believed

Flyvbjerg, Garbuio and Lovallo went back through the project in 2009 and found something less flattering sitting beside the optimism. The A$7 million was a political, low-balled figure aimed at winning approval before an election, and the 1959 start came before drawings or funds were ready. Utzon's son Kim has argued the published budget was about an eighth of the real one, which is a family member's view quoted in research rather than an audited number, though it points the same way.

Two more things the retelling gets wrong. The overrun, which Flyvbjerg rounds to about 1,400%, bought a scaled-down version of the winning design. And Utzon resigned, on 28 February 1966, after a dispute with the New South Wales government, so every version of the story that has him sacked is wrong. A$102 million wasn't the end of the bill either, since another A$45 million was allocated in 2002, partly to bring the place closer to what he'd drawn.

Neglecting distributional data

Kahneman and Tversky named the planning fallacy in a technical report dated June 1977, "Intuitive Prediction: Biases and Corrective Procedures", published in 1979 in TIMS Studies in Management Science and reprinted in 1982 in Judgment under Uncertainty. (Not in Thinking, Fast and Slow, which is where most people file it.) Their sentence is exact:

The planning fallacy is a consequence of the tendency to neglect distributional data.

Distributional data is how cases like this one actually turned out. The same passage allows that the fallacy is sometimes down to motivational factors such as wishful thinking, which describes an election-year budget well enough.

Kahneman and Lovallo put names to the two stances in 1993. The inside view works from the plan in front of you, its parts and its scenarios, and it feels like diligence. The outside view works from how comparable efforts went. Kahneman's own example, told from memory with figures that shift a little between tellings, is a curriculum team that expected about two years and took eight, in a field where comparable teams had needed seven to ten years or had never finished at all.

Remembering isn't enough

Buehler, Griffin and Ross asked 37 students writing honours theses when they would hand in. Average prediction: 33.9 days. Best case 27.4, worst case 48.6. They took 55.5 days, and only about 30% were done by the date they had named for themselves. (Those figures come from summaries of the paper rather than its full text, so hold them loosely.) The disaster scenario they could bring themselves to imagine still came in a week short.

The same paper watched how people predict. They narrate the future step by step instead of consulting the past, and when their own overruns do come up they get explained away as external, passing and specific. Bad flu that term. A printer that died in week three. Asked how long someone else would take, the same people guess too long, so the bias points one direction and only at you.

Roy, Christenfeld and McKenzie reviewed the evidence in 2005 and found the hole underneath all of it. People do predict durations from memories of past durations, and those memories are themselves systematic underestimates. The record in your head has been quietly edited in your favour.

In Study 4 of the Buehler paper, the optimistic bias was eliminated for participants told to connect relevant past experiences to their prediction. Make the past specific at the moment of guessing and the bias stops working.

The cure, at two sizes

At national scale the method is reference class forecasting: predict from the real outcomes of a class of similar completed projects. Flyvbjerg calls it the cure to the planning fallacy. In summer 2004 the UK Department for Transport and HM Treasury adopted it for large transport projects, and the American Planning Association endorsed it in April 2005. In practice it's an uplift taken from what actually happened to comparable schemes: for roads, +15% if the funder will accept a 50% chance of overrun, +32% for 80% certainty, and for rail the 50% figure is +40%.

Nobody is going to apply a percentage uplift to "write the weekly update". The person-sized version is the record: what your own similar work actually took, kept somewhere that doesn't shorten it on the way back.

That's what an estimate in I Heart Checklists is. The time on a task comes from measured history, yours first and then from people who have timed the same work, and the source shows on the task so you can see where the number came from. Then Do places today's work against the hours you really have, less your meetings, and tells you the minute you'll be free (we've written about why that's a time and not a percentage). Most task managers hold what you have to do and when it's due. This one also works out what you can do. There's a free trial if you'd like to watch your own numbers come in.

One caveat, since the piece rests on research. Reference class forecasting was built and tested on bridges and railways, mostly using other projects' outcomes. Treating your own task history as the reference class is a reasonable extension of it, supported by that Study 4 result, rather than a claim anyone has tested directly on task apps.

The Opera House got a postcard out of being ten years late. Your Tuesday doesn't have that kind of upside.

Sources: Kahneman & Tversky, Intuitive Prediction: Biases and Corrective Procedures (1977, published 1979); Buehler, Griffin & Ross (1994); Flyvbjerg, Garbuio & Lovallo (2009).

Your day has a size. Your list should know it.

I Heart Checklists plans around the hours you actually have, then tells you when you're finished. $2.99/mo, 14 days free.

Start free trial