Tim Harford's Blog, page 14

Cautionary Tales – The Panama Canal Series

Panama Disaster 1: Ferdinand De Lesseps, “the Great Frenchman”, was convinced that he was the man to build the Panama Canal. No, he wasn’t an engineer and no, he’d never actually been to Panama before. But he’d managed to dig the Suez Canal, and everyone had said that would be impossible too. How hard could it be?

Panama Disaster 2: Sixteen years have passed since Ferdinand De Lesseps’ catastrophic failure in Panama, and the dramatic collapse of the French Panama Canal company. Now, President Theodore Roosevelt has picked up the task. “No single great material work,” Roosevelt tells Congress, “is of such consequence to the American people.”

The Americans have their work cut out. Enter chief engineer John Stevens. How does he spot a problem no-one else has noticed? And what does he do to solve it?

The Panama Canal series is available exclusively to subscribers of Pushkin+.

[Apple] [Spotify] [Stitcher]

Further reading

Pushkin+ listeners seeking further reading for our episodes on the Panama Canal should consult The Path Between the Seas by David McCullough, Hell’s Gorge by Matthew Parker, John Frank Stevens, Civil Engineer by Clifford Foust, and Framers by Kenneth Cukier, Viktor Mayer-Schonberger and Francis de Vericourt.

Academic studies:

Michael Hogan “Theodore Roosevelt and the Heroes of Panama” Presidential Studies Quarterly, Vol. 19, No. 1, Part I: American Foreign Policy for the 1990s and Part II: T. R., Wilson and the Progressive Era, 1901-1919 (WINTER 1989), pp.79-94

Jones, E. E.; Harris, V. A. (1967). “The attribution of attitudes”. Journal of Experimental Social Psychology. 3 (1): 1–24. doi:10.1016/0022-1031(67)90034-0. (See also Patrick Healy “The Fundamental Attribution Error“.)

AI has all the answers – even the wrong ones

Can large language models solve logic puzzles? There’s one way to find out, which is to ask. That’s what Fernando Perez-Cruz and Hyun Song Shin recently did. (Perez-Cruz is an engineer; Shin is the head of research at the Bank for International Settlements as well as the man who, in the early 1990s, taught me some of the more mathematical pieces of economic theory.)

The puzzle in question is commonly known as the “Cheryl’s birthday puzzle”. Cheryl challenges her friends Albert and Bernard to guess her birthday, and for puzzle-reasons they know it’s one of 10 dates: May 15, 16 or 19; June 17 or 18; July 14 or 16; or August 14, 15 or 17.

To speed up the guessing, Cheryl tells Albert her birth month, and tells Bernard the day of the month, but not the month itself. Albert and Bernard think for a while. Then Albert announces, “I don’t know your birthday, and I know that Bernard doesn’t either.” Bernard replies, “In that case, I now know your birthday.” Albert responds, “Now I know your birthday too.” What is Cheryl’s birthday?* More to the point, what do we learn by asking GPT-4?

The puzzle is a challenging one. Solving it requires eliminating possibilities step by step while pondering questions such as “what is it that Albert must know, given what he knows that Bernard does not know?” It is, therefore, hugely impressive that when Perez-Cruz and Shin repeatedly asked GPT-4 to solve the puzzle, the large language model got the answer right every time, fluently elaborating varied and accurate explanations of the logic of the problem.

Yet this bravura performance of logical mastery was nothing more than a clever illusion. The illusion fell apart when Perez-Cruz and Shin asked the computer a trivially modified version of the puzzle, changing the names of the characters and of the months. GPT-4 continued to produce fluent, plausible explanations of the logic, so fluent, in fact, it takes real concentration to spot the moments when those explanations dissolve into nonsense.

Both the original problem and its answer are available online, so presumably the computer had learnt to rephrase this text in a sophisticated way, giving the appearance of a brilliant logician. When I tried the same thing, preserving the formal structure of the puzzle but changing the names to Juliet, Bill and Ted, and the months to January, February, March and April, I got the same disastrous result. GPT-4 and the new GPT-4o both authoritatively worked through the structure of the argument but reached false conclusions at several steps, including the final one. (I also realised that in my first attempt I introduced a fatal typo into the puzzle, making it unsolvable. GPT-4 didn’t bat an eyelid and “solved” it anyway.)

Curious, I tried another famous puzzle. A game show contestant is trying to find a prize behind one of three doors. The quizmaster, Monty Hall, allows a provisional pick, opens another door to reveal no grand prize, and then offers the contestant the chance to switch doors. Should they switch?

The Monty Hall problem is actually much simpler than Cheryl’s Birthday, but bewilderingly counterintuitive. I made things harder for GPT4o by adding some complications. I introduced a fourth door and asked not whether the contestant should switch (they should), but whether it was worth paying $3,500 to switch if two doors were open and the grand prize were $10,000.**

GPT-4’s response was remarkable. It avoided the cognitive trap in this puzzle, clearly articulating the logic of every step. Then it fumbled at the finishing line, adding a nonsensical assumption and deriving the wrong answer as a result.

What should we make of all this? In some ways, Perez-Cruz and Shin have merely found a twist on the familiar problem that large language models sometimes insert believable fiction into their answers. Instead of plausible errors of fact, here the computer served up plausible errors of logic.

Defenders of large language models might respond that with a cleverly designed prompt, the computer may do better (which is true, although the word “may” is doing a lot of work). It is also almost certain that future models will do better.

But as Perez-Cruz and Shin argue, that may be besides the point. A computer that is capable of seeming so right yet being so wrong is a risky tool to use. It’s as though we were relying on a spreadsheet for our analysis (hazardous enough already) and the spreadsheet would occasionally and sporadically forget how multiplication worked.

Not for the first time, we learn that large language models can be phenomenal bullshit engines. The difficulty here is that the bullshit is so terribly plausible. We have seen falsehoods before, and errors, and goodness knows we have seen fluent bluffers. But this? This is something new.

*If Bernard was told 18th (or 19th) he would know the birthday was June 18 (or that it was May 19). So when Albert says that he knows that Bernard doesn’t know the answer, that rules out these possibilities: Albert must have been told July or August instead of May or June. Bernard’s response that he now knows the answer for certain reveals that it can’t be the 14th (which would have left him guessing between July or August). The remaining dates are August 15 or 17, or July 16. Albert knows which month, and the statement that he now knows the answer reveals the month must be July and that Cheryl’s birthday is July 16.

**The chance of initially picking the correct door is 25 per cent, and that is not changed when Monty Hall opens two empty doors. Therefore the chance of winning $10,000 is 75 per cent if you switch to the remaining door, and 25 per cent if you stick with your initial choice. For a sufficiently steely risk-taker, it is worth paying up to $5,000 to switch.

Written for and first published in the Financial Times on 5 July 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.

Counting the cost of my ‘major keying error’

Receiving a demand for a parking fine is always annoying. Even more so when you know you paid for a ticket. But there it was, the letter from Euro Car Parks demanded payment, and they had photographs to prove it. It took 10 minutes for me to send evidence showing that at the time and place in question, the ticket machine had charged the price of an evening’s parking to my debit card.

After pondering this for a while, Euro Car Parks took a different tack: it withdrew the demand for payment of a fine, and, instead, demanded a £20 administrative fee for all the trouble I had caused it. My crime, it turns out, was that I had only entered the first four characters of my vehicle registration. This “major keying error” violated the car park’s terms and conditions.

But such mishaps merely spark curiosity. Why do “major keying errors” occur and is there anything we can do to prevent them?

In May, a pair of UK regulators fined Citigroup more than £60mn for several failures of risk control, most spectacularly when a trader planned to sell $58mn of shares but, in a major keying error, issued an order to sell $444,000mn of shares instead. Some of this order was blocked, but the remainder was more than enough to fleetingly crash stock markets across Europe.

The system made such an error unnervingly easy: the trader typed a number into the wrong box, asking the system to sell 58 million units instead of $58mn worth of units. Each unit was worth thousands of dollars, and there’s the problem. It is a bad idea to have a share trading system that lets you accidentally sell nearly half a trillion dollars worth of shares — which goes some way to explaining the £62mn fine. (How I wish regulators could be persuaded to levy such a magnificent fine on Euro Car Parks.)

What can be done to prevent such horrors? One possibility is to tell a system’s users not to make any mistakes. This seems to be the position of Euro Car Parks, and it is not wholly satisfactory. Nobody plans to enter the wrong registration number when paying for parking and, no doubt, Citigroup traders endeavour not to accidentally sell half a trillion dollars worth of shares. But mistakes will be made.

An alternative is to program the software to notice the mistake. Euro Car Parks could have flashed up a message saying “you have only entered four digits, are you sure that’s right?” or even “LOL sucker you’ll hear from our lawyers” would serve as a warning.

Citi’s system did flash up 711 warnings, of which only the first 18 lines were visible. That is only slightly better than no warnings at all, because trigger-happy warnings tend to be ignored as a matter of habit. And the Citi warnings must have been somewhat obscured by the fact that the system sometimes defaulted to assuming that shares had a unit price of -1, which means that if you mistakenly type 58 million units instead of $58mn, the system might tell you you’re selling -$58mn rather than the more obviously unnerving figure of, ahem, $444,000mn.

We can take comfort that this is not the most costly keying error in history. In fact it is not even Citigroup’s most costly keying error this decade. In 2020, the bank accidentally transferred $900mn of its own money to some creditors of Revlon, the cosmetics firm, again because of a software system that made such a slip all too easy. Some of those creditors decided to keep the money, on the grounds that Revlon did indeed owe it to them. US regulators fined Citi $400mn for having deficient systems.

We may laugh, but when a system requires perfection from operators, the consequences can be tragic. Nancy Leveson, an MIT professor who specialises in software safety, has documented an infamous case: the Therac-25. The Therac-25 was a radiation-therapy device in the 1980s that could fire high-energy beams either of electrons or X-rays into patients.

The type of beam matters. The X-ray beam was fired through a “flattener” to spread the treatment to the right area, but which also absorbed much of the energy. If the X-ray beam was somehow fired with the flattener out of position, disaster would result.

Disaster resulted. In one case, in a Texas hospital in 1986, the operator entered an “e” for the electron beam, then realised she had meant to type “x” for the X-ray, and swiftly moved the cursor back to correct the entry. The hidden flaw in the system was that rapid edits could bewilder it. If such an edit was made during the eight seconds it took to set everything up, the flattener would not be rotated into place and the software would be confused about the machine’s configuration.

The upshot? The X-ray beam was fired without the flattener, delivering an extreme dose of radiation. The computer then told the operator that only a low dose had been administered, and invited her to press “P” to proceed with a second attempt. The patient, suffering burning pains, was already trying to get off the treatment table when he was hit by the second beam. It was later estimated that he had received around 100 times the intended dose. He lost the use of his arm, was paralysed by radiation burns to his spine, and died five months later from numerous complications. It was not the only fatal accident involving the Therac-25. A major keying error, indeed.

There is no such thing as a foolproof computer system, but software can be designed to fail gracefully or disgracefully. On reflection, perhaps £20 wasn’t such an extortionate fee for a lesson in life.

Written for and first published in the Financial Times on 12 July 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.

Cautionary Tales – Embracing the Escape Fire (with Adam Grant)

Cautionary Conversation: Steve Jobs hated his phone so much that he smashed it against a wall. He also referred to mobile carriers as “orifices”. Yet he went on to invent the world’s most popular smartphone. Why did he change his mind?

Tim Harford and organisational psychologist Adam Grant (Think Again, Hidden Potential) discuss the consequences of letting our ideas become part of our identity; when it’s essential to adapt; and whether frogs really do stay sitting in slowly boiling water.

[Apple] [Spotify] [Stitcher]

Further reading

Adam Grant Think Again

Adam Grant Hidden Potential

How to fix the UK? Let me count the ways

Here’s the bad news: it is going to take more than a change of government to cure what ails Britain. The symptoms are wearyingly familiar, but worth summarising. Waiting lists for NHS treatment have soared above 7.5mn, from 4.2mn in 2019 and 2.5mn in 2010. Prisons are at capacity, and the court system long since exceeded it. Local government funding has been squeezed for years, with obvious effects on local services such as social care, libraries and leisure centres.

The simple explanation for all this is that 14 years of Conservative-led governments have cut taxes, preferring to trust citizens with their own money even if it leaves the public realm looking threadbare. But that’s not what has happened. While headline taxes on average earners are indeed low, as a proportion of national income the total tax burden is — infamously — near the highest level since the 1940s, while the UK continues to borrow and add to the largest pile of debt in living memory. In short, we are spending more than ever and somehow getting less than ever for it.

Those are the symptoms. The cause is familiar, too: productivity has stagnated since 2008. Productivity is the value that an average worker produces in an hour of work. Productivity means getting more for less, and for decades, we were able to expect that living standards — both in terms of the public services we enjoy and in terms of the money we ourselves are able to spend — would gradually improve. The past 16 years have been different. “Real wages are roughly at the same level as they were in 2007,” say Anna Valero and John Van Reenen of the LSE’s Centre for Economic Performance.

Why has productivity been so dismal? There are several culprits, some easier to fix than others. The UK’s finance-heavy economy suffered from the 2008 financial crisis. Regional in­equality is high, and this isn’t just a story about “left-behind” towns, but cities such as Greater Manchester and Birmingham not reaching the potential suggested by Europe’s great secondary cities such as Barcelona, Frankfurt and Toulouse.

The education system offers little to young people who aren’t on track to get a high-quality degree. The decades-long reluctance to build enough houses means that our homes are ageing, cramped, poorly insulated and excruciatingly expensive, but it also damages productivity. That is because it is prohibitively costly to take one of the simplest and most basic steps in search of a better life, which is to move to a place with a more dynamic economy than the place you grew up.

Finally, there are the decades of under-investment, which is even more striking in the private sector than the public sector, but which bites in both. “Virtually all of the productivity gap with France is explained by French workers having more capital to work with,” the Resolution Foundation declared in last year’s epic “Ending Stagnation” report.

Investment means taking a long-term view, sacrificing consumption now in search of higher living standards tomorrow. The UK has preferred “jam today” for 40 years, and we are now suffering the consequences.

So what to do? Two policy gambles have failed. The first, Brexit, knocked business investment flat on its back and complicated all sorts of trade and travel that was once simple. The independent Office for Budget Responsibility estimates that Brexit is on course to knock 4 per cent off the UK’s economic potential over the next few years.

The second, Liz Truss’s attempt to kick-start growth with an unfunded tax cut, fell flat within days. Having correctly diagnosed the disease, she almost certainly prescribed the wrong cure. We’ll never know for sure, because her bedside manner was so clumsy that the patient rebelled. The best thing one can say about Truss’s shortlived premiership is that her policies were much easier to reverse than Brexit.

To list the problems with the British economy is to see the difficulties of fixing them, but there is some hope. Long-promised planning reform could unlock an enormous increase in investment, productivity and most importantly the affordability of everyday life. One simple but radical idea, proposed by the Centre for Cities, is a presumption in favour of development on greenbelt within half a mile of existing commuter stations, while protecting sensitive land. Less than 2 per cent of the greenbelt would be affected but more than 2mn sustainable homes built.

Tiptoeing closer to the EU customs union and the single market would be a useful first step towards eventually undoing the damage of Brexit. The tax system remains full of loopholes and absurdities. More logical, efficient taxes would raise more money while causing less economic damage.

And with luck, one serious drag on the UK economy will be lifted with a new government: we might reasonably expect less uncertainty. No doubt there will be shocks ahead, but with luck they will not be worse than Brexit, Trump, Covid-19, war in Ukraine and a revolving door of prime ministers, each with their own idiosyncratic vision of what must be done. With less uncertainty we can certainly hope for more business investment. The next government faces a huge challenge. Simply steadying the ship would be a start.

Written for and first published in the Financial Times on 28 June 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.

Cautionary Tales – Flying Too High: AI and AirFrance Flight 447

Panic has erupted in the cockpit of AirFrance Flight 447. The pilots are convinced they’ve lost control of the plane. It’s lurching violently. Then, it begins plummeting from the sky at breakneck speed, careening towards catastrophe. The pilots are sure they’re done-for.

Only, they haven’t lost control of the aircraft at all: one simple manoeuvre could avoid disaster…

In the age of Artificial Intelligence, we often compare humans and computers, asking ourselves which is “better”. But is this even the right question? The case of AirFrance Flight 447 suggests it isn’t – and that the consequences of asking the wrong question are disastrous.

[Apple] [Spotify] [Stitcher]

Further reading

Jeff Wise, “What Really Happened Aboard Air France 447,” Popular Mechanics, December 6, 2011

William Langewiesche, “The Human Factor” Vanity Fair, October 2014

“Children of the Magenta,” 99% Invisible podcast, June 23, 2015

Nick Oliver, Thomas Calvard, Kristina Potočnik (2017) Cognition, Technology, and Organizational Limits: Lessons from the Air France 447 Disaster. Organization Science 28(4):729-743

Cockpit Transcripts in French and English

Fabrizio Dell’Aqua Falling Asleep At the Wheel – Working Paper

Luchins, A. S. (1942). Mechanization in problem solving: The effect of Einstellung. Psychological Monographs, 54(6),

You Are Not So Smart Episode 281 – on AI and Brainstorming

James Reason Human Error

In Broken Britain, even the statistics don’t work

From the bone-jarring potholes to the human excrement regularly released into British rivers, the country’s creaking infrastructure is one of the most visceral manifestations of the past 15 years of stagnation. To these examples of the shabby neglect of the essential underpinnings of modern life, let me add another: our statistical infrastructure.

In her new book, How Infrastructure Works, engineering professor Deb Chachra argues that infrastructure is an extraordinary collective achievement and a triumph of long-term thinking. She adds that a helpful starting point for defining infrastructure is “all of the stuff that you don’t think about”.

Statistical infrastructure certainly matches those descriptions. The idea that someone needs to decide what information to gather, and how to gather it, rarely crosses our mind — any more than we give much thought to what we flush down the toilet, or the fact that clean water comes from taps and electricity from the flick of a switch. As a result the UK’s statistical system, administrative databases, and evidence base for policy are suffering the same depredations as the nation’s roads, prisons and sewers.

Easiest to measure are the inputs: the Office for National Statistics faces a 5 per cent cut in real terms to its budget this year, has been losing large numbers of experienced staff, and is hiring dramatically fewer than five years ago. But it is more instructive to consider some of the problems. The ONS has struggled to produce accurate estimates of something as fundamental as the unemployment rate, as it tries to divide resources between the traditional-but-foundering Labour Force Survey, and a streamlined-but-delayed new version which has been in the pipeline since 2017.

That is an embarrassment, but the ONS can’t be held responsible for other gaps in our statistical system. A favourite example of Will Moy, chief executive of the Campbell Collaboration, a non-profit producer of systematic reviews of evidence in social science, is that we know more about the nation’s golfing habits than about trends in robbery or rape. This is because the UK’s survey of sporting participation is larger than the troubled Crime Survey of England and Wales, recently stripped of its status as an official National Statistic because of concerns over data quality. Surely nobody made a deliberate decision to establish those curious statistical priorities, but they are the priorities nonetheless. They exemplify the British state’s haphazard approach in deciding what to measure and what to neglect.

This is foolishness. The government spends more than £1,200bn a year — nearly £18,000 for each person in the country — and without solid statistics, that money is being spent with eyes shut.

For an example of the highs and lows of statistical infrastructure, consider the National Tutoring Programme, which was launched in 2020 in an effort to offset the obvious harms caused by the pandemic’s disruption to the school system. When the Department for Education designed the programme, it was able to turn to the Education Endowment Foundation for a solid, practical evidence base for what type of intervention was likely to work well. The answer: high-quality tutoring in small groups.

This was the statistical system, in its broadest sense, working as it should: the EEF is a charity backed by the Department for Education, and when the crisis hit it had already gathered the evidence base to provide solutions.

Yet — as the Centre for Public Data recently lamented — the DfE lacked the most basic data needed to evaluate its own programme: how many disadvantaged pupils were receiving tutoring, the quality of the tutoring, and what difference it made. The National Tutoring Programme could have gathered this information from the start, collecting evidence by design. But it did not. And as a result, we are left guessing about whether or not this was money well spent.

Good data is not cheap to collect — but it is good value, especially when thoughtfully commissioned or built into policymaking by default. One promising avenue is support for systematic research summaries such as those produced by the Cochrane Collaboration for medicine and the Campbell Collaboration for social science and policy. If you want to understand how to promote literacy in primary schools, or whether neighbourhood policing is effective, a good research synthesis will tell you what the evidence says. Just as important, by revealing the gaps in our knowledge it provides a basis for funding new research.

Another exciting opportunity is for the government to gather and link the administrative data we all produce as a byproduct of our interactions with officialdom. A well-designed system can safeguard personal privacy while unlocking all manner of insights.

But fundamentally, policymakers need to take statistics seriously. These numbers are the eyes and ears of the state. If we neglect them, waste and mismanagement are all but inevitable.

Chachra writes, “We should be seeing [infrastructure systems], celebrating them, and protecting them. Instead, these systems have been invisible and taken for granted.” We have taken a lot of invisible systems for granted over the past 20 years.

The Resolution Foundation has estimated that in this period, UK public investment has lagged the OECD average by a cumulative half a trillion pounds. That is a lot of catching up to do. The next government will need some quick wins. Investing in better statistical infrastructure might be one of them.

Written for and first published in the Financial Times on 21 June 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.

Who’s responsible for our accountability problem?

I was recently scheduled to present my Cautionary Tales podcast live on stage as a curtainraiser for a podcast conference. Talented voice actors, live sound effects and even a trombone would weave a dramatic soundscape for a full house. There was only one problem. Nobody seemed to have the authority to let me in.

The people issuing conference passes wouldn’t give me one — not unreasonably, since I wasn’t attending the conference proper and hadn’t registered to do so. They advised that I speak to “that lady there”. That lady there seemed puzzled: the conference wouldn’t officially open until tomorrow, so I didn’t need a pass. Just walk in, she said.

The security guard had a different view. He had been given strict instructions that nobody gets in without a pass. Talk to the conference team, he told me. Nothing to do with us, they said — talk to security.

I realised I was facing what the writer Dan Davies has named an “accountability sink”, in which it was somehow nobody’s fault. In his recent book The Unaccountability Machine, Davies explains the basic logic of an accountability sink: decision-making power is removed from individuals you might want to shout at, and made instead by an algorithm or some distant committee both ignorant of and immune to your objections. Everyone I spoke to insisted that they were powerless to act. And if you cannot change a decision, you cannot be held accountable for it. That’s the accountability sink at work.

“Computer says no” first became a comedy catchphrase 20 years ago, so although the accountability sink is a useful new term it describes an older problem. Just ask 440 luckless squirrels who in 1999 arrived at Schiphol airport in the Netherlands, en route from Beijing to Athens. Unfortunately, particularly for the squirrels, they had been shipped without the right paperwork. What to do? It wasn’t legal to send them on to Athens. It wasn’t possible, apparently, to send them back to Beijing or to fix the paperwork. So the airport ground staff started up an industrial shredder normally used in poultry farms for killing male chicks and threw all 440 of the squirrels into it.

When this debacle became known to the public, it was universally agreed that shredding 440 squirrels had been The Wrong Thing To Do, but it was less clear who should be accountable. The basic problem, Davies explains, was that the decision to destroy paperwork-less animals had been turned into a matter of policy emerging from the Dutch department of agriculture. Policy is policy, not something that gets overruled by manual workers in airport sheds wearing protective gloves and standing next to an industrial shredder and a few crates of squirrels.

The policy itself was probably sensible. The problem was that there was no way to identify and deal with exceptional cases. The question, “What if the policy demands that we hurl hundreds of squirrels into an industrial shredder?” arose too late.

Accountability sinks can have a legitimate purpose. When the pressure is on, it’s tempting to play favourites or take short-cuts, prioritising the noisiest complaints or putting off painful consequences. “Computer says no” may be a punchline, but the computer is a useful way to deflect awkward customers. It also often makes the right choice.

Consider the idea of an independent monetary policy, when an elected government sets an inflation target and then an un­elected central bank has the task of trying to hit that target. It’s a sensible response to the fact that democratically elected governments always have a short-term incentive to stoke inflation, reaping the briefest of benefits followed by some serious headaches.

When the Bank of England hurts millions by raising interest rates, it can point to its inflation mandate. When people complain to the government, the government can say that interest rates are a matter for the Bank of England. It’s an accountability sink again, but monetary policy is almost certainly better as a result of the fact that somehow nobody is to blame.

Yet all too often accountability sinks are created for the simple, ignoble reason that nobody wants to be held accountable if they can avoid it. Even worse, the sink may be fictional, a mere scapegoat. A generation of British politicians learnt that whenever anyone complained about a policy, it was always easy to blame Brussels, or unelected judges, or health and safety gone mad.

Accountability sinks range from the algorithm that raises your insurance premium to the institutional denial that has delayed many reckonings for the British state: infected blood, persecution of sub-postmasters, Hillsborough, the Windrush scandal and so many others.

Faced with such an enormous range of misfiring accountability sinks, is there anything useful we can say about what we could change to make things better? Perhaps there is. The common thread here is an imperviousness to feedback. No doubt the people involved in the squirrel incident wondered if there might be a better way to deal with paperwork problems, but there was no line of communication open to the relevant ministry, unaware of the gruesome consequences. Without feedback, nothing ever gets better.

The world is a complicated place and humans are imperfect. Mistakes will be made. The question is, what happens after the mistake? Does news ever reach anyone with the ability to change things? If so, do they? Some accountability sinks exist for a good reason, some do not. Either way, if there is no way to learn from mistakes, the mistakes will eventually bury us all.

Written for and first published in the Financial Times on 14 June 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.

Cautionary Tales – Run, Switzer, Run: The Women Who Broke the Marathon Taboo

Until the 1960s, women couldn’t compete in Olympic events any longer than a sprint – and commentators declared that a marathon would kill them, or leave them unable to have children. Rubbish, of course. But when Kathrine Switzer signed up for the 1967 Boston Marathon, it wasn’t the distance that bothered her – it was the enraged race officials trying to assault her.   

Thanks to pioneers like Kathrine, women have made huge strides in long distance running – and are now challenging the times of men in the very races they were banned from for so long.  

[Apple] [Spotify] [Stitcher]

Further reading

On Bobbi Gibb

Ailsa Ross, “The Woman Who Crashed the Boston Marathon” JSTOR Daily 18 March 2013

Olivier Guiberteau, “Bobbi Gibb: The Boston Marathon pioneer who raced a lie” BBC Sport 29 August 2023

Brigit Katz “The incredible story of Bobbi Gibb, the first woman to run the Boston Marathon” Women in the World, New York Times, 20 April 2015

On Kathrine Switzer

Kathrine Switzer “The Girl Who Started It All” Runners World – excerpt from Kathrine Switzer Marathon Woman;

The Spirit of the Marathon by Gail Waesche Kislevitz Breakaway Books, 2002

“I ran with the men and changed history” BBC Outlook

On The Spine Race

Nick Van Mead “Montane Spine Race: 268 miles of pain”

Dave Lee “Spine Race 2013”

“Spine” (Amazon Prime documentary)

Jasmin Paris “Spine Race”

Episode 6 – Spine Race 2019

BBC Sport

BBC Scotland

Other sources

Roger Robinson “Eleven Wretched Women” – What really happened in the first Olympic women’s 800m. Runner’s World 14 May 2012 

Colleen English “Not a Very Edifying Spectacle”: The Controversial Women’s 800-Meter Race in the 1928 Olympics 08 October 2015

A Marathon Legend Revisited

Natalie Angier “2 Experts Say Women Who Run May Overtake Men” The New York Times 7 January 1992

Run Repeat State of Ultra-running

BBC More or Less

What zebras can teach us about international trade

It’s not often that you can squeeze zebras into a column about trade tariffs, but against the backdrop of a trade war over electric vehicles, with the US election, the Chinese economy and the global climate at stake, let’s try. The Biden administration is imposing heavy tariffs on Chinese goods, especially electric vehicles.

Medium term, the effect will be to block cheap EVs entering the US market, which is bad for the planet, bad for American consumers and great for anyone else who wants to make EVs in, or sell EVs to, the US.

But long term? The long game is to try to shift the structure of the US economy towards the manufacturing of green technologies such as solar panels, batteries and electric cars. Might that work? That’s where the zebras come in.

Consider a simplified model of a savannah. Grass grows in the sun. Zebras eat the grass. Lions eat the zebras. And because it’s not much of a model without a technical term, let’s introduce one: the trophic level. The trophic level of the sun is zero. The grass has a trophic level of one, the zebras two and the lions have a trophic level of three.

Of course it all gets more complicated. Warthogs eat plants, but they might eat a dead zebra or even a dead lion. So a warthog might have a trophic level of, say, 2.1. All this is useful stuff to think about if you’re modelling the ecology of the savannah. Useful, too, if you’re thinking about the structure of an economy.

Two complexity scientists, James McNerney and Doyne Farmer, have suggested looking for analogies to trophic levels in economies. It’s not that an economy has a food chain or an apex predator, as such. But economies do have lots of interdependent industries, and the mathematics of trophic levels offers a promising way to analyse them.

In an economic setting, let’s define the trophic level of zero as being individuals. A widget industry that uses only human labour has a trophic level of one. A sprocket industry that uses a 50:50 mix of workers and widgets has a trophic level of 1.5, and so on. The more links there are in an industry’s supply chain, the higher its trophic level. Does that mean that industries with a high trophic level are more sophisticated? No more than lions are more sophisticated than zebras. But the trophic level does matter.

McNerney, Farmer and colleagues used data from the World Input-Output Database to calculate the trophic levels of different industries in the US, China and other countries. They found that the Chinese economy is full of industries with a trophic level higher than four, whereas the highest trophic level of a major US industry is food manufacturing, at just over 3.5. Many large US industries, including health, retail and defence, have a low trophic level of about 2. Trophic levels aren’t fixed. US agriculture is highly mechanised and has a trophic level above 3, while Chinese agriculture is a labour-intensive activity with a trophic level below 2.5.

Policymakers in the US say they want to defend US manufacturing jobs from Chinese competition. There are some plausible security reasons, and some implausible ones, but this is also an attempt to raise the trophic level of the US economy. Is that desirable? Low trophic levels notwithstanding, the typical US citizen enjoys a far higher standard of living than those in China. But, as Farmer explains in his recent book Making Sense of Chaos, there is an advantage to high-trophic-level industries. They tend to get more efficient more quickly.

The reason is simple, almost mechanical: an industry with no suppliers has only one possible source of technological improvement, itself. An industry with a deep supply chain profits when any company in that chain improves. McNerney has found that, for the typical industry, about two-thirds of technological improvements come from suppliers and only one-third are made internally.

This simple theory makes some assumptions that may be wrong, but when McNerney, Farmer and colleagues looked at the data, they found the evidence accorded with the theory. Economies with higher trophic levels are more innovative and tend to grow more quickly. The theory also explains the vague, yet widely held, belief that there is something special about manufacturing. What’s special is that manufacturing often has a high trophic level.

Many voters will applaud the new US tariffs on China. Should they? Farmer tells me that “an industrial policy that supports industries with deep supply chains, raising the trophic level of the economy, should result in faster GDP growth and stronger increases in productivity”.

That leaves open the question of whether tariffs are the right way to nurture such industries. Decades of rhetoric about protecting “infant industries” have tried to obscure the fact that tariffs usually protect old, fading industries rather than young, growing ones. These new tariffs, by contrast, are protecting young, fast-growing market sectors. So perhaps this time things will be different.

I would dearly like to believe that the tariffs will be a springboard to healthy global competition to make zero-emission technologies. But even economists are sometimes prone to wishful thinking. Maybe I have been swept away by the romance of the savannah.

Written for and first published in the Financial Times on 7 June 2024.

Loyal readers might enjoy the book that started it all, The Undercover Economist.

I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.