Wednesday, 21 October 2020

Moving my writing to substack

 I'm moving a lot of my non-academic writing to substack. Expect more about society, politics and social science. You can read it online, or subscribe to get it in your inbox at irregular intervals.

Here's a sample of new stuff, some more serious than others:

Democracy: Fire or Slow Decay – populism is only one of two trends eating away at democracy

Li Bai and the Unicorns Led By Donkeys

Li Bai and the Abominable Prague Machine in which two Tang-era Chinese poets discuss the end of civilization and Tesla's stock price


Saturday, 18 July 2020

Is the future socialist?

This is a hard blogpost to write. My usual view of socialism is absolutely this tweet by Elon Musk:

But it is a good mental exercise to consider options that you've long ruled out.

The original case for socialism in the 19th century was that industry was growing more and more organized, concentrated and monopolistic. This wasn't just a pathology – there were really good reasons why you might want, say, a single company running railways all across America. People could see this, and thought: "why not continue this process?" Industry was going to become a single vertically-integrated monopoly, and it could then be taken over on behalf of the people, instead of run to profit a single capital owner. I think that was the nub of Marx's argument about the economic historical inevitability of socialism.

From 2020, the case looks the same. I stopped buying stuff online in 2008 because my credit card got stolen from a little specialist internet shop, who just couldn't have adequate security. When I started again, it was with Amazon, because I thought I could trust them. And I still buy 90% of my stuff from Amazon.

In short, the internet has created huge forces making monopoly efficient. (Remember back when everyone had email provided by their own ISP? It sucked.) More recently the need for big data to feed to AI does the same.

Yet, this pressure to monopoly inevitably makes the incumbents lazy, and stifles innovation, because any firm with a new idea can see it copied by one of the Big Four. Firms like Evernote, Dropbox and Slack are struggling against incumbent copies (Apple Notes / iCloud and Google Drive / Microsoft Teams).

At this point, with huge profits being made by a few big and not especially innovative firms, why not take the next step and socialize them? Monopoly has outcompeted competition. We should acknowledge that.

I see the appeal. Perhaps unsurprisingly, in the end I'm not convinced.
  • The advantage of Big Tech may be a passing phase, driven by the relative rarity of high-level tech skills. During lockdown I started buying things from local firms online. My favourite coffee shop put up a website with a nice clean interface, probably provided by Shopify. They delivered within a day. I no longer felt scared to order. Similarly, much research in AI is about working with smaller datasets which don't require a planet-scale infoharvesting operation.
  • Breaking up monopolies is just as plausible a remedy as taking them over.
  • Administering huge companies "for the benefit of the people" still suffers from the usual principal-agent problems. It might look less like putting Google under state control, and more like lending Google the power of the state.
It seems to be a universal rule of human cognition that we confuse local improvement with global optimality. "A step in this direction will make things better, so let's continue in this direction forever and reach Utopia!" So far, socialism has always proved to be a mirage like that. But who knows how many steps we will take towards it this time?

Thursday, 16 July 2020

A data point

On the plane to India in 1994, I flicked through the in-flight magazine. Between adverts for Scotch whisky and American suits, it had a feature article: "Is India Racist?" 

I read on, wondering if it would cover discrimination against Muslims (about 15% of the population) or Dalits (about the same). 

In fact, the writers had sent a black person into various Bombay shops, to see how he was treated. 

I have not been able to find estimates of the number of people of African or African-American descent in India. I think it is safe to say it is not large, and was probably smaller still in the 1990s, although there is apparently one tribe.

Tuesday, 14 July 2020

Can we trust the historians?

As statues fall, and debate roils about the West’s colonial past, we need expert witnesses. Does Edward Colston belong in the street, a museum, or the river? What about Cecil Rhodes or Oliver Cromwell… or Winston Churchill? We want historians’ evidence and informed judgment to deal with complex questions of guilt and responsibility.

But is that what we’ll get? Some examples suggest not.

David Starkey denied in an interview that the slave trade was genocidal. It couldn’t have been, or “there wouldn’t be so many damn blacks”. He resigned from his position at Cambridge and lost his job at another university. Professor Starkey has form for careless language. He claimed “whites had become black” during the 2011 riots. Meanwhile, Priyamvada Gopal from Cambridge tweeted “white lives don’t matter” – “qua white lives”, she added. She kept her job, but received death threats and switched her Twitter account to private.

Naomi Wolf, author of The Beauty Myth, was caught out live when she claimed that gay men were executed for sodomy in the 19th century. She had misinterpreted a key phrase in the court records. An honest mistake, maybe, but the research came out of her 2015 PhD in history from Oxford. You might have expected better fact-checking.

Okay, Dr Wolf is a full-bore conspiracy theorist, and Professors Starkey and Gopal are trolls – and perhaps their remarks don’t reflect the quality of their research. But consider Nancy Maclean’s Democracy in Chains. It told the story of James Buchanan, the economist who helped to invent “public choice” – the economic study of government decisions. The book described Buchanan as a conspirator with the billionaire Ed Koch to tie down democracy. It was a finalist for the US National Book Award. But libertarians on the internet found that Maclean had truncated quotes misleadingly. For example, she quoted Tyler Cowen:

 The weakening of checks and balances would increase the chance of a very good outcome.

In fact he wrote:

While the weakening of checks and balances would increase the chance of a very good outcome, it would also increase the chances of a very bad outcome.
She described Cowen as “creating a handbook for how to conduct a fifth-column assault on democracy”. Cowen’s essay said: “I explicitly favor more democratic systems”.

An academic review in the Journal of Economic Literature described the book as “at best sketchy… replete with significantly flawed arguments, misplaced citations, and dubious conjectures.” 

Closer to today’s issues, historians of the “New History of Capitalism” (NHC) view slavery as central to early capitalist development. Their books have been controversial. The Economist magazine published a critical review of Edward Baptist’s The Half Has Never Been Told, then retracted it in the face of online criticism. But economic historians of slavery were scathing. Reviews in two journals described the school as “substitut[ing] more and better anecdotes for tables of regression results”, and accused Baptist of selectively quoting and merging evidence from slave records so as to mislead the reader. One review ended:

As the NHC matures, it might embrace the enduring strengths of traditional historical scholarship, including citing sources correctly, conducting close (and accurate) readings, drawing inferences that are actually supported by the evidence, and integrating its findings into the broader historiography. It should also stop making stuff up.
Zing. Yet the New History of Capitalism played a major role in the New York Times’ 1619 project, which put slavery and racism at the heart of American history, and Baptist’s thoroughly debunked statistics were quoted by Ta-Nehisi quotes at a Congressional hearing.

There are often fierce debates between different intellectual schools. That’s not the issue here. These books were credibly accused of misleading their readers. That is, they were dishonest.

Do not think this is just a left wing thing. Orlando Figes was caught pumping his own books with anonymous reviews on Amazon. Niall Ferguson, the biggest media don of all, misleadingly edited quotes  when he joined the debate about the effects of Obamacare. (In 2018, he resigned from Stanford after asking for “opposition research” on a progressive campus activist.)

Dishonesty exists in all of academia. But mostly, when it is found out, it is punished. Diederik Stapel – the “Lying Dutchman” – who invented all of the data for his experiments in social psychology, is no longer an academic. Nor is Michael Lacour, who ran a fraudulent study on persuading people to support gay marriage. Discovered scientific fraud ends careers. Quotations are a key tool of historical research, so you’d expect similar standards to apply: misleading quotation is data falsification. In fact, most of the above historians are still in academia, and didn’t even lose their jobs or their book awards.

What’s up with that?

One underlying reason might be intellectual trends. There was a period when quantification was fashionable in history. Times changed, and historians drew back from this – perhaps because they reasonably believed that historical narrative cannot be supplanted by data analysis. (The same arguments are now being refought in a new debate on “cliometrics”.) History became influenced by the post-structuralist and post-modern turn, including skepticism towards truth in narrative. Speaking to historians at my university recently, I found that these ideas remained influential: “of course, all remembering is misremembering,” said one, as if this were a truism. This attitude may have made it easier to get away with phoney, but rhetorically persuasive, arguments.

Another possibility is loose standards. Historians surely have shared ideas of what counts as good work. But they may not be as precise or technical as in other disciplines. Econometricians obsess over identification of causal effects. Statisticians have rigorous mathematical conditions for estimating parameters of a model. History might by nature be more diverse than that, and this could let shady work fall through the gap. Historians often rely on the telling anecdote. This is not a flaw: just as in a law court, a single fact really can demolish a historical case. But when the supply of possible examples gets large, the ability to make your case by careful selection increases along with it.

Lastly, incentives matter. Engineers with a new idea can build and sell a better mousetrap. Medical advances turn into treatments that face randomized controlled trials. Economists can get paid making policy recommendations. None of these leads to perfect incentives for scientific honesty. But one thing that makes you a big name in history is writing a successful book. This gives a certain incentive to please the public, perhaps by telling it what it wants to hear. Indeed, the Amazon reviews for Maclean’s work are glowing. There’s an audience for the story that economist intellectuals conspired to overthrow democracy... even if it is just a story.

Whatever the reason, we need historians to get their house in order. The “hard” data-crunching social sciences, on their own, cannot teach us about our past. Narratives matter, but they can’t just be any old story. Psychology is processing its reproducibility crisis. Economics has had its credibility revolution. Where is history’s civil war of the footnotes?

Sunday, 21 June 2020

On two pieces of graffiti in the town centre of Norwich

Two kinds of graffiti have popped up in town.

One is in the underpass near my house: a massive BLACK LIVES MATTER, written over about 20 yards. In between there are some extra slogans: "Don't trust the media! Think for yourself!"

It is signed by Knapple, Norwich's best known graffiti artist. Knapple's signature is a stylized, coloured pineapple. Sometimes, she gives lectures. Her website says "Graffiti is about everyone being entitled to have a voice, regardless of whether you like it or not."

The work is big and stylish. It took time. Knapple works in daylight. The council (or maybe the Business Improvement District) hires her to paint the underpass. She has an on-going battle with the local kids. They graffiti her graffiti.

The second kind of work is all around the town centre. The council has marked social distancing signs on the pavement: "this way" and "keep 2m apart". The unknown artist, or artists, have commented on this, in brightly coloured chalk: "No evidence!", "Don't think for yourself", "Do as you're told" and in one place, scornfully, "Baa! Baa!". Several buildings have been marked with a stencil: COVID IS A HOAX.

Knapple has several advantages over the anonymous artist, like "not being insane" and "being unlikely to kill anyone". But the other wins in one way: they are more sincere. When Knapple says don't trust the media and think for yourself, she probably means "don't trust the Daily Mail". The Guardian might get a pass. If someone were to actually bypass the mainstream media and start reading 5G conspiracy websites, then she would be horrified. (I am making assumptions, but they are generous ones.) If she were discovered writing COVID IS A HOAX, her paid work would dry up.

The anonymous artists really mean "don't trust the media" – any of it. They really want you to think for yourself, and to find your own truth, on Facebook or Youtube, in the internet's dark corners.

A lot of the left wing is a bit Knapplish today. They want to be radicals. They have radical styling. Black Power clenched fists get posted to corporate Twitter feeds. Young couples, feet on the property ladder, declare allegiance to socialism.

It is fun to feel like a rebel. It is hard to give up this feeling. But if you don't, you might end up succouring people who really do want to overturn everything, and who are too stupid, mad or heartless to care about the side effects.

The property owners have quickly painted over COVID IS A HOAX. People in Norwich aren't daft.

Update: the underpass graffiti has been joined by antisemitic theories ("Rotschilds made millions"),  an anti-police acronym ("ACAB") and a White Power symbol and slogan.

Sunday, 5 January 2020

Political arguments, custard elephants and the law of iterated expectations

The law of iterated expectations is a statistical fact which looks like this:

E[X] = E[ E[X | Y] ]

(Don't run away. The point is coming.)

In English, this says that the expected value of some variable X is equal to the expectation of its conditional expectations, given different values of another variable Y. (Don't run!)

Let's make it a bit more concrete. Suppose Y is a statement which may be true or false. Then:

E[X] = Probability(Y is true) * E[X, given Y is true]   +
            Probability(Y is false) * E[X, given Y is false]

The point to focus on is that if E[X, given Y is true] is higher than E[X], then E[X, given Y is false] cannot also be higher than X. You can show this with a bit of algebra, using that the probability of Y is strictly between 0 and 1. (Don't run! I swear that is the last of the algebra.)

For example, the expected value of a fair die throw is 3 and a half, and this is equal to

1/2 * 3 + 1/2 * 4

which is the expected value if the die lands on an odd number, times the chance of that; plus the expected value if the die comes up even, times the chance of that. You can cut the die roll up many different ways; the formula will always give you the right answer. Notice that 3.5 is strictly between 3 and 4.

Now, let's begin.

Here is an interesting sentence from Martin Jacques' excellent book When China Rules The World.
India's [post-colonial economic] performance was transformed, as the figures cited earlier for its economic growth illustrate, but Africa was left debilitated by the experience of the slave trade and then colonialism.
Jacques is referring to the fact that after the end of colonialism, GDP growth in Africa collapses – and even GDP levels, in some cases. As the context makes clear, the first part of this sentence claims:

1. India's GDP growth rose after independence, and this shows how bad colonialism was.

while the second part claims:

2. Africa's GDP growth fell after independence, and this shows how bad colonialism was.

Claims 1 and 2 violate the law of iterated expectations. Put Y as "a given country's GDP grows faster after independence from colonialism" and X as "how harmful colonialism was". If Y being true is evidence for the bad effects of colonialism, then Y being false cannot also be evidence for it. In fact, it must be evidence against it.

Someone might say: "You are oversimplifying absurdly. Of course there are complex links from colonialism to Africa's bad economic performance after independence. And of course India has its own, very different, history, which also features colonialism." This is true, but it doesn't affect the point. If your theory of history uses both Y and not-Y as evidence for the same overall claim, then your theory has a problem – not in the historical facts, but in how they are being interpreted.

I suspect that a lot of political arguments secretly violate the law of iterated expectations. The above was a particularly clear example in one sentence. More often, one person sees some evidence and goes "that proves my point"; but if they had seen the opposite, they would have said the same thing.

Or sometimes different people, on the same side of the argument, use contradictory pieces of evidence in different contexts. So, typically, scholars of African history will make the African argument above, and those of India will make the Indian argument. Either one may be right, but they cannot both be.

Here are some other examples. They are hypothetical, but they may sound familiar.
  • An ethnic minority underperforms for a certain educational qualification. This shows that the relevant education institution is structurally racist. Students from the minority outperform for a different educational qualification. This alo shows that the education institution is structurally racist: it is setting the admission bar too high for those students.
  • The government borrows in response to an economic shock. You think this is risky, and you predict rising inflation. But when inflation doesn't rise, that doesn't change your view. You just keep predicting it will rise in future.
Why political arguments, especially? For the usual reason: in politics and other morally-charged domains, we decide on our views and then find the evidence to support them.

Oh, and the custard elephants.

Why do elephants paint their feet yellow?
So they can hide upside down in custard.
Have you ever seen an elephant upside down in custard?
No? Shows what a good disguise it is, then.

Saturday, 26 October 2019

Testing the robustness of the NHB migration result

Never say "beyond reasonable doubt". No sooner had I made this boast than Panu Pelkonen called my bluff on twitter:

This could indeed be a concern. For example, we show that polygenic scores for educational attainment  are higher for those leaving coalmining areas than those staying in them. But what if that is only true in those clustered areas (which represent UKBiobank Assessment centres, by the way - not all of them are necessarily big cities)?

One way to address this critique is to check whether the result varies much within the sample. This isn't perfect, because there still could be variation beyond the sample. (To understand this, consider another variable – age. Almost all UK Biobank respondents are over 40. Our results hold across all the age groups within the sample: for example, among 40-50 year olds, 50-60 year olds, and so on. But that doesn't prove the result would hold for 20-30 year olds. Maybe mining areas have started to attract talent back among the younger generation – I doubt it, but it is not a question our data can answer.)

Nevertheless, if we see the result is pretty robust wherever we look within our sample, that should give us some comfort that it is unlikely to be different outside the sample.

So, this afternoon I took a look. I reran the basic analysis for each separate UKBiobank assessment centre. I discarded any assessment centres which had no people living in coalmining areas (or everybody living in a coalmining area). Then, I looked among people born in a coalmining area, and living near that centre. Within this group, I looked at the average polygenic score for educational attainment (PSEA) among people currently living in a coalmining area, and currently living outside a coalmining area. And I took the difference of the two. Our overall result was that PSEA was higher among those not living in a coalmining area ("movers away"). Would this hold across all the different assessment centres?

Here's the answer:

Difference in PSEA between "stayers" (still living in a coalmining area) and "leavers" (born in a coalmining area but not living in one now). Lines show 95% confidence intervals (standard errors clustered by census geography)

Basically, yes. Stayers have lower PSEA than leavers for almost all the 17 centres I looked at, and significantly so for most of them. There is a single assessment centre where stayers have higher PSEA than leavers. This comprises just 361 people; I would guess that near this centre, the ex-coalmining area is richer than non-coalmining areas. Overall, then, the result seems quite geographically robust.