Theories of multiple equilibria, explicitly or implicitly, are the foundation for a great deal of regional policy. Such policies promise to attain large and permanent impacts on regional industrial structure through small and temporary interventions at critical moments. Our findings that even wholesale destruction of cities and industries via incendiary, high explosive and nuclear bombs have little impact on the long-term level and structure of manufacturing should give pause to those for whom a small subsidy is the only arrow in their quiver.
Davis and Weinstein, “A search for multiple equilibria in urban industrial structure“
1. Introduction
This is Part 2 of my series “The scope of longtermism,” discussing a paper of the same name.
Part 1 introduced the scope question for longtermism: how wide is the class of decision situations for which longtermism holds? I clarified my target, swamping axiological strong longtermism (swamping ASL) and my view: the scope of swamping ASL, while probably nonempty, is smaller than many longtermists suppose. To witness nonemptiness, I suggested that Congress’ decision to fund the Space Guard Survey may well have fallen within the scope of swamping ASL.
The next order of business is to limit the scope of swamping ASL. To do that, the next three parts of this series will introduce three scope-limiting factors: probabilistic and decision-theoretic factors that are present in many contemporary decision problems and which, when present, substantially reduce the prospects for swamping ASL to hold of that problem.
Today’s post introduces the first scope-limiting factor: rapid diminution.
2. Rapid diminution
Consider the increase (ΔV) in far-future value that an action may bring. Plausibly, as ΔV increases, the probability P of realizing this increase decreases. It is, after all, harder to make a larger difference to far-future value than it is to make a smaller difference.
Longtermists correctly emphasize that ΔV can grow very large. However, it is important to emphasize that P can also grow very small. This leaves us with a race dynamic: when ΔV grows more quickly than P diminishes, far-future value may dominate the expected value of our actions. But when P diminishes more quickly than ΔV grows, far-future value may make at most a modest contribution to the expected value of our actions.
Say that acts exhibit rapid diminution when P diminishes more quickly than ΔV increases.
For example, if the long-term impact of an act follows a normal distribution, centered around the origin, with a standard deviation equivalent to the value of ten lives saved, then the probability of long-term impacts exceeding the value of fifty lives saved is less than one in a million. Very large values of ΔV, while physically possible, are unlikely enough to make little difference to the expected value of our action.
To say that rapid diminution is a scope-limiting factor is to say two things: (1) that rapid diminution often occurs, and that (2) rapid diminution, when present, significantly reduces the prospects for swamping longtermism.
(2) is plausible enough. The case for longtermism often relies on high-value, low-probability impacts. If an act exhibits rapid diminution, then high-value, low-probability impacts make little difference to the expected value of the act. This tends to scuttle the case for longtermism to hold within that decision problem.
Why should we believe (1), that rapid diminution often occurs? Here we confront two dueling narratives about how the macro structure of the world reacts to change.
On the first story, the world is a chaotic system that is easily perturbed and set on an entirely different course. If you prevent a child from going blind, they may found a business, combining with others to make their village prosper, then their region and their nation, shifting the global balance of power and eventually leading to the formation of a world government that will pay renewed attention to existential risks affecting us all. To quote Neil Gaiman:
It used to be thought that the events that changed the world were things like big bombs, maniac politicians, huge earthquakes, or vast population movements, but it has now been realized that this is a very old-fashioned view held by people totally out of touch with modern thought. The things that really change the world, according to Chaos theory, are the tiny things. A butterfly flaps its wings in the Amazonian jungle, and subsequently a storm ravages half of Europe.
The second story is rather different. On this story, the world is a stable system that takes a great deal to perturb. If you prevent a child from going blind, they may found a business, but even if they do, the fortunes of their village, region and nation may be shaped much more strongly by broader macroeconomic and geopolitical trends than by the addition of a single business. At this level, the long-term trajectory of the village, region and nation is likely to remain largely unaltered. And as for the flapping of a butterfly, there is, on this story, a good reason why meteorologists don’t study butterflies halfway across the world before predicting the weather: their wing flappings are unlikely to matter much on this scale.
Those who think that rapid diminution is common lean towards the second story: the world is a relatively stable system, and it takes a lot to shift the long-term trajectory of the world. Those who think that rapid diminution is less common lean towards the first story. But which story is correct?
3. Persistence studies: Negative findings
A good way to get a handle on the matter is to look at persistence studies. Persistence studies is an academic field which studies how changes tend to persist across time. If persistence studies identifies many persistent and highly impactful changes, then that is some evidence for the first story. But if persistence studies struggles to identify persistent and highly impactful changes, then that is some evidence for the second story.
There has been an explosion of work in persistence studies in recent decades. Persistence studies is still a young field, and the jury remains deeply divided, but many observers have been struck by a spate of surprising negative findings. Even many sizable short-term changes that we might have expected to have persistent long-term impacts turn out to have few persistent effects.
For example, suppose we bomb a city. That is, presumably, more likely to impact the long-term future of the city than a simple act such as curing a child’s blindness. Will the long-term fortunes of the city be changed forever? Or will the city quickly spring back to the course that it was on, such that a few decades later, it will be difficult for any but historians to know where the bombs fell?
Often, the results of bombing are rather less persistent than we might expect. Consider the worst case: in 1945, the United States dropped atomic bombs on the cities of Hiroshima and Nagasaki. The results were as devastating as you might expect: at least 100-200,000 people died out of about half a million inhabitants. Infrastructure was devastated, and land and air irradiated. Large-scale depopulation followed. But did it last? Surprisingly, it did not: it is widely agreed that the populations of Hiroshima and Nagasaki returned to their prewar levels by the mid-1950s.
Perhaps other bombed cities in Japan fared differently? Columbia economists Donald Davis and David Weinstein (2008) find that Japanese manufacturing took a beating across the board due largely to allied bombing:
Surely it would be reasonable to suspect that the cities hit this hard, if not worse, would be assured a worse economic future than those left untouched? Surprisingly, Davis and Weinstein find strong evidence that this was not the case: cities that were hardest hit during the war experienced stronger growth in the decades following the war, until they had regained their initial trajectories. Perhaps the most striking figure from Davis and Weinstein is this one, showing a strong negative correlation between pre- and post-war growth – that is, a tendency for harder-hit cities to grow more quickly towards their pre-war production levels.
Not only the cities of Hiroshima and Nagasaki, but in fact the bombed cities of Japan in aggregate showed a remarkably quick recovery towards their pre-war manufacturing trajectories. Industry-segregated results are similar.
Theories of multiple equilibria stress, like Gaiman’s witch, that small changes to an economic system may have large and persistent consequences. By contrast, if even the dramatic bombing of Japan during the Second World War was not enough to permanently alter the course of manufacturing, then we may want to rethink the sensitivity of economic and other systems to small shocks. As Davis and Weinstein put it:
Theories of multiple equilibria, explicitly or implicitly, are the foundation for a great deal of regional policy. Such policies promise to attain large and permanent impacts on regional industrial structure through small and temporary interventions at critical moments. Our findings that even wholesale destruction of cities and industries via incendiary, high explosive and nuclear bombs have little impact on the long-term level and structure of manufacturing should give pause to those for whom a small subsidy is the only arrow in their quiver.
However, perhaps these findings from Japan are a fluke? We know, after all, that Japan experienced a remarkable post-war economic miracle, and certainly post-war Japan was an unusually stark example of societal transition. Fair enough. Let us consider Vietnam.
American bombing in Vietnam during the Vietnam War was highly devastating. Berkeley economists Edward Miguel and Gerard Roland (2011) study the regional impact of Vietnam War-era bombing. They find no evidence of negative impacts from US bombing on population density, poverty rates, consumption levels, infrastructure, or literacy rates by the year 2002.
For example, here are their estimates across six models of the impact of various factors on 1999 poverty rates.
In many models there is, if anything, a slight negative impact of bombing on poverty rates, though even when this impact is statistically significant it is miniscule. More importantly, bombing rates were significantly less predictive not only than the expected factors, such as facts about land area or membership in North versus South Vietnam, but even than relatively less influential factors such as average temperature.
If these studies are any indication, even very large changes to something the size of a city may not have long-lasting effects. This should give pause to those who think that even small interventions on something as large as the world will often have significant and long-lasting effects.
4. Persistence studies: Positive findings
To be fair, the field of persistence studies has presented some examples of potential large and persistent effects. For example:
Alberto Alesina and colleagues (2011, 2013) have suggested that the development of the plough may have permanently affected fertility norms and gender roles, by giving men a lasting advantage in agricultural labor due to the muscular demands of working the plough. This may have cemented the gendered division of labor and led to downstream normative changes.
Nathan Nunn and colleagues (2008, 2011) have suggested that the African slave trade stably reduced social trust in the most affected regions. Those who could be sold into slavery by their neighbors, family or rulers and carted off by traveling merchants came, understandably, to be distrustful of others.
Jonathan Schulz and colleagues (2019) have suggested that the Catholic Church may have been responsible for the spread of `WEIRD’ (Western, Educated, Industrialized, Rich and Democratic) personality traits. They suggest that the spread of the Catholic Church throughout many of the WEIRDest societies today allowed the Church to stably shape norms, values and other elements of WEIRD societies.
That is fair enough. One should not deny that any effects at all can persist for centuries. However, three points are worth bearing in mind here.
First, there aren’t very many compelling positive findings. To motivate swamping longtermism, we do not need to find a few examples of long-term persistent effects. We need to find many examples of long-term persistent effects. Otherwise, there will be little evidence for the Gaiman-style worldview on which many acts have significant long-term persistent effects, and comparatively more evidence for the Davis-and-Weinstein-style worldview on which most acts lack long-term persistent effects.
Second, many of these findings are controversial. In particular, it is very hard to get reliable data on these timescales, and there are some compelling competing explanations of many findings that will be difficult to dispel without better data.
Finally and most importantly, the effect sizes aren’t large enough. We were after examples of swamping longtermist options: options whose long-term effects greatly exceed in magnitude the short-term effects of any option. Let us make two highly concessive moves. First, let us compare the long-term effects of the studied changes only to the short-term effects of these changes themselves, rather than to all possible short-term effects of alternative options. Second, let us take up an ex post perspective which fixes long-term and short-term effects at their actual values, so that long-term effects do not have to be discounted for uncertainty. Even with these concessive moves in place, none of the phenomena described above seem like plausible examples where the magnitude of long-term effects greatly exceeded that of short-term effects.
Begin with the plough. It may be true that the introduction of the plough stably affected fertility- and gender-norms, though this is among the most controversial of the cited findings. However, the introduction of the plough also significantly increased agricultural outputs, leading immediately to societal growth, urbanization, increased specialization of labor, and any number of other revolutionary short-term societal changes.
Turn next to the African slave trade. It may be true that the slave trade stably reduced social trust in affected regions. It is also true that the slave trade sold millions of people into slavery. This is, one presumes, at least as large an axiological effect as the decline of social trust.
Turn finally to the spread of WEIRD personality traits. It may be true that the Catholic Church made us WEIRD. It also exerted a dominant influence over almost every aspect of personal and civic life for over a millennium, and continues to be one of the very most influential institutions in society today.
None of this is meant to suggest that the long-term effects of the plough, the African slave trade, and the Catholic Church were not important. But it is not clear that these long-term effects exceeded in magnitude the short-term effects of the same developments.
5. Conclusion
In this post, we discussed the first scope-limiting challenge: rapid diminution. Rapid diminution occurs when the probabilities of large long-term evaluative impacts decrease more quickly than the sizes of those impacts increase. In the presence of rapid diminution, large long-term impacts may genuinely contribute to the expected value of acts, but their contribution tends to be dominated by the contribution of smaller, more probable contributions.
To assess the prevalence of rapid diminution, we looked to persistence studies. Persistence studies is the primary academic field concerned with finding large, persistent long-term effects over a period of centuries. If rapid diminution is prevalent, then we should expect persistence studies to find few sizable results, where by contrast if rapid diminution is not prevalent, then we should expect persistence studies to do very well for itself.
We saw that many findings in persistence studies are negative: even something as dramatic as US bombing in Japan or Vietnam, up to and including the atomic bombing of Hiroshima and Nagasaki does not seem to have had a persistent impact on economic indicators even in the affected cities themselves. We saw that while there are a few potential positive findings in persistence studies, these findings are controversial, few and far between, and plausibly outshone by the short-term impacts of the studied phenomena.
However, it might be objected that rapid diminution is not always present. Perhaps the Romans, upon sowing the soil of Carthage with salt and thus cementing the rise of Rome, may have been justifiably confident that their acts would have a large impact on the future value of the world, even if they were not sure whether that impact would be for good or for ill.
However, this last sentence points to a second problem: even when we have good reasons to suspect that an action will have a large impact on the long-term value of the world, we may not be in a good position to assess the direction of this impact. Possible positive and negative impacts then cancel to a large degree, greatly diminishing their importance from an ex ante perspective. I discuss this second problem in the next post, under the heading of washing out.
Leave a Reply