We haven’t just pulled the figures out of the sky. Well, not often.

From At least five interesting things for your weekend (#15) by Noah Smith.  The subheading is Working-class wealth, the education gap, zero-sum thinking, data uncertainty, and indigenous knowledge.  

The two topics of zero-sum thinking and data uncertainty are near to my heart.

3. Strong arguments about zero-sum thinking 

 

Sahil Chinoy, Nathan Nunn, Sandra Sequeira and Stefanie Stantcheva have written a very interesting and potentially important paper about the rise of zero-sum thinking in the United States.

A lot of pro-growth people — myself included — believe that one reason economic growth is good is that it prompts people to think about growing the pie rather than fighting each other over how the pie is divided. We notice that before the Industrial Revolution made growth the norm, there was a lot more conquest and endemic war. This sort of made sense, because if you lived in the year 1200, the way to get rich was to take your neighbor’s land. In the year 1900, the way to get rich was to start a business. Obviously we still have some wars, and there are many non-economic reasons for conflict, but if a rising tide lifts all boats, it seems reasonable to think that people will be more content not to beggar their neighbors.

Anyway, in order to investigate this story, Chinoy et al. do a big survey of Americans, to try to figure out how zero-sum their attitude is. They ask people four questions, each of which can have five responses (from “strongly agree” to “strongly disagree”):

1. Ethnic: “In the United States, there are many different ethnic groups (Blacks, whites, Asians, Hispanics, etc). If one ethnic group becomes richer, this generally comes at the expense of other groups in the country.”

2. Citizenship: “In the United States, there are those with American citizenship and those without. If those without American citizenship do better economically, this will generally come at the expense of American citizens.”

3. Trade: “In international trade, if one country makes more money, then it is generally the case that the other country makes less money.”

4. Income: “In the United States, there are many different income classes. If one group becomes wealthier, it is usually the case that this comes at the expense of other groups.”

They combine the responses to these questions into an index of zero-sum thinking, and they find that the index is higher for Americans who were born more recently.

Note that this doesn’t necessarily mean zero-sum thinking is increasing over time! Since they’ve only done the survey once, all they know is that older Americans show less zero-sum thinking than younger ones. Sure, if people’s attitudes are fixed, then yes, this does show more zero-sum thinking over time. But we don’t know if attitudes are fixed! Maybe age and experience teach people to be less zero-sum in their thinking? We won’t know until this survey is repeated over decades.

I share with Smith a conviction that zero-sum thinking is an anchor holding back beneficial progress.  I do not have a strong view whether zero-sum thinking is becoming more pervasive in recent decades.  One the one hand recent decades have repeatedly demonstrated the capability to create new prosperity to everyone's benefit; classic non-zero-sum thinking.  On the other hand postmodernism, social justice, critical race theory, wokeness all seem to have become more prevalent (though perhaps only more loudly touted by a small minority).  Social justice is classic zero-sum thinking.  

Smith's point in the whole essay is that for methodological reasons: 

So basically, we have no idea if this is a big important change or a tiny insignificant change.

Keep in mind that I do basically believe in the story the paper (and the FT) are telling here. But that’s just my own prior. Chinoy et al. have written a good and interesting paper, but it’s hard to say how much this one paper should strengthen my belief.

I encounter this a lot.  Our universities produce a tidal wave of research every year, much of it of dubious quality and integrity.  Some of it supporting any of our personal priors, or not.  We can construct whole worldviews simply by selectively cherry-picking among weak research, even if done by otherwise reliable researchers.  

There is good research out there.  With a lot of skepticism and attention to detail as well focus on logical integrity, we can build a better epistemic understanding of the world.  But it is a heavy lift and I know I am prone to simply trust some work by some researchers more than others simply because that work agrees with what I already believe for other, ostensibly good reasons.  Sustained due diligence is hard.

The fourth point he makes in the column is closely related.  I am interested in history, the history of technology, economic development, and the history of economic development.  I am also a keen devotee of logic, reason, and empirical rationalism.  I am accustomed to wanting to assess things through objective measurement.

4. Getting good data is really hard

I toss out a lot of economic data on this blog, and it’s very natural to just accept this data as true and correct. But it’s important to remember that a lot of this data is really hard to gather, and there are usually a lot of assumptions that go into constructing it.

For example, check out this data of historical GDP per capita in the UK over the past thousand years, courtesy of the Maddison Project:

But the UK didn’t keep really good economic statistics until very recently. So how do we know these numbers for the 1700s, much less for the 1100s? The answer: assumptions about the technology available to the people at the time. These assumptions go into the numerator (total economic output) and the denominator (population), because we assume that a certain mix of agriculture, shipping, handicrafts, etc. would be able to produce a certain amount of goods and would be able to sustain a certain number of people. Change those technological assumptions, and the historical data can change a lot. I recommend the recent article by Timothy Guinnane about how sketchy our guesses about past population levels can be.

The graph is the classic hockey stick of economic development.  The world was almost universally poor and short-lived up until approximately 1500 when a magical combination of technological development, Age of Renaissance thinking, Age of Enlightenment thinking, Classical Liberalism, etc.  all came together to break the Malthusian trap.  

We know this transitioned occurred.  Magnitudes and timings differed based on what we are measuring and how we are measuring it.  Magnitudes and timings also differed by location.  As Smith points out, the very phenomenon we are interested in (economic, technological, and epistemic development) allows us to measure things better over time.  We cannot retroactively create the data which did not exist.  We might estimate what it might have been, but that is always a poor second- or third-best.

The example of population estimation is illuminating.  From We Do Not Know the Population of Every Country in the World for the Past Two Thousand Years by Timothy W. Guinnane.  From the Abstract.

Economists have reported results based on populations for every country in the world for the past two thousand years. The source, McEvedy and Jones’ Atlas of World Population History, includes many estimates that are little more than guesses and that do not reflect research since 1978. McEvedy and Jones often infer population sizes from their view of a particular economy, making their estimates poor proxies for economic growth. Their rounding means their measurement error is not “classical.” Some economists augment that error by disaggregating regions in unfounded ways. Econometric results that rest on McEvedy and Jones are unreliable.

“… we haven’t just pulled the figures out of the sky. Well, not often.”

—McEvedy and Jones (1978, p. 11)

I know that our historical estimations of population are crude estimates with gross margins of error.  When I read about something referring to Britain having a population of 8 million at the time of the American Revolution in 1775, I know with certainty that I ought to read that at as a population between 5 and 12 million.  And British records are among the best in the world for that particular time period.  

I know I should attach large margins of error.  But all my thinking is absolute.  As I consider historical cause and effect, comparative contexts, etc. in all that process, Britain has 8 million people.  At least for me, thinking in margin of error ranges for cause and effect or for comparative context is simply to contingent and exhausting.  

Continuing the example.  Common estimates of population for Britain in 1775 are about 8 million and for the American colonies about 2.5 million.  Britain is about 3 times more populous than the American colonies.  Taking those as absolutes you can quickly fill in a range of conclusions.  8 million for the size of Britain makes Britain demographically dense compared to the 2.5 million Americans in the 13 colonies.  It has implications for size of armies which can be raised, the degree to which force can be brought to bear, etc.  What percentage of the 2.5 million actively supported independence.  The questions and leapt-to conclusions cascade.

If we accept the military rule of thumb that attackers need 3 times as many soldiers as defenders, then Britains 3.2 times population advantage suggests that the revolution could have swung either way.

But everything is upended if we attach ranges.  What if Britain hat between 5-12 million and America had 2-5 million?  These are not outlandish estimates.  Now we have to consider simultaneously scenarios in which 5 million British conducted war against 5 million Americans or that 12 million British fought 2 million Americans.  The possible scenarios spiral out of control.

This issue of weak historical estimations extends far beyond economics and demographics.

It is one of the Achilles heels of Anthropogenic Global Warming.  Our array of temperature measurement devices (thermometers and satellite measurement) is recent and of varying precision, accuracy, and comprehensiveness.  At most, fifty years.  Then we stitch together to that dendrological and pollen proxies which hugely variant degrees of precision and accuracy to create a historical record of highly dubious consistency or fit for purpose.  It might possibly be directionally correct but we cannot lose sight of just how weak and proximate that manufactured record really is.