The Economist reports on Greg Clark’s work using surnames to track social mobility. I have posted about this work before, but The Economist piece makes an important point. When tracking mobility across generations, you cannot simply extrapolate the results of a single generation into the future.

Corak’s work draws on recent studies that compare income levels between just two generations: fathers and sons. That is out of necessity; good data covering three or more generations are scarce. But reliance on limited data could lead to overestimates of social mobility.

Gregory Clark, an economist at the University of California, Davis, notes that across a single generation some children of rich parents are bound to suffer random episodes of bad luck. Others will choose low-pay jobs for idiosyncratic reasons, like a wish to do charitable work. Such statistical noise makes society look more changeable than it is. Extrapolating the resulting mobility rates across many generations gives a misleadingly sunny view of long-term equality of opportunity. Mr Clark suggests that family history has large effects that persist for much greater spans of time. Fathers matter, but so do grandfathers and great-grandfathers. Indeed, it may take as long as 300-500 years for high- and low-status families to produce descendants with equal chances of being in various parts of the income spectrum.

In line with his earlier work, Clark fingers a genetic basis to the finding.

Mr Clark’s conclusion is that the underlying rate of social mobility is both low and surprisingly constant across countries and eras: the introduction of universal secondary education scarcely affects intergenerational mobility rates in Britain, for example. This consistency, he suggests, shows that low mobility may be down to differences in underlying “social competence”. Such competence is potentially heritable and is reinforced by the human tendency to mate with partners of similar traits and ability.

This work reminds me of an article in which Sam Bowles and Herb Gintis explore the inheritance of inequality. They estimated that intelligence is responsible for a correlation between parent and child income of 0.01, with total correlation due to genetic factors of around 0.12. I played with the results to show that a larger proportion of the observed variation could be explained by genetics if you tweak some of the assumptions. However, Clark’s work shows that a larger limiting factor on their result was a single generation comparison in their analysis.

Postscript: A debate followed the Economist article, with contributions by Miles Corak, Fransicso Ferreira, Greg Clark (and again), and Jason Long. They are all worth reading.

Last week I presented a summary of the method and findings of Ashraf and Galor’s American Economic Review paper The ‘Out of Africa’ Hypothesis, Human Genetic Diversity, and Comparative Economic Development (for the latest ungated version, go here). As discussed in that post, one limb of Ashraf and Galor’s argument is that genetic diversity provides a greater range of traits for the development and implementation of new technologies (which I’ll call innovation). In this post, I look at that claim in more detail.

The measure of genetic diversity used by Ashraf and Galor is based on non-protein coding regions of the genome. This is common in population studies to prevent selection from distorting attempts to track evolutionary history. However, Ashraf and Galor note that the genetic diversity shaped by the founder effect as populations moved out of Africa likely included other phenotypically expressed genetic diversity. For example, they note one study in which craniometric (head shape) diversity declines with distance from Africa.

At a population level, increased genetic diversity can have evolutionary benefits. Increased diversity provides a broader suite of traits on which natural selection can act. Where the environment changes, it is more likely that traits favourable in that environment will be present, and if the environment is particularly unstable, genetic variation will provide a basis for some of the population to be viable through the range of conditions.

Ashraf and Galor refer to a couple of studies in support of their hypothesis, but their examples do not build a strong case for their specific argument linking genetic diversity and innovation. In one case, they reference a study in which it was found that less diverse, inbred populations of fruit fly became extinct at lower concentrations of salt. However, the authors of that study were unable to differentiate whether inbreeding or lack of diversity drove the outcome. Other studies that isolated the effects of low diversity did not always show the same result (for example, in this study of flies).

Another more relevant study found that more diverse honeybee colonies have higher workforce efficiency (I can’t give a a link to this study, but a summary of work in the area is here). This finding provides evidence that diversity has an effect on production outcomes. Ashraf and Galor expand their discussion of the effect of diversity on bee colonies in the Web Appendix, where they discuss an experiment published in Science. In that study, colonies with various levels of diversity were released, with more diverse colonies founding their colonies faster and accumulating food stores more quickly. The authors of the study proposed that greater response variation to changing conditions may be behind the higher fitness of diverse colonies. If there is a trade-off between response thresholds and activity efficiency, or if response thresholds result in some behaviours missing from a worker’s repertoire, higher genetic diversity will provide a basis for the full suite of required traits.

One question overhanging these bee studies is the direction of the effect. As the Ashraf and Galor hypothesis has a countervailing force whereby diversity decreases cooperation, diversity could also be expected to harm colony production. Whether diversity is beneficial or not would depend on the relative effect of the two forces. Some of the bee studies note the potential for intra-colonial conflict and ask whether factors such as recombination rates may reduce conflict. But when comparing to the hump shaped relationship between human genetic diversity and economic development, it is not easy to place the bee colonies on the curve. Any effect of diversity on bee colony success could be justified after the fact as being at a certain point on the curve.

Towards the end of the paper Ashraf and Galor examine the hypothesis for human populations by regressing the number of scientific papers published per year per person against genetic diversity and a range of controls including social infrastructure, years of schooling, risk of malaria and distance to waterways. Continent fixed effects are used, which eliminates comparisons across continents, as well as other controls for sub-Saharan Africa and OPEC. The result of the regression is that genetic diversity is a significant factor in scientific output, with a one per cent increase in diversity linked to an increase in scientific articles per person per year of 0.02.

Despite this statistical evidence, I am not convinced. Partly, I suspect missing variables – that is, the qualitative traits that have been under selection since the Out of Africa event. The inclusion of controls such as social infrastructure also makes the analysis difficult, as it is closely related to potentially relevant factors such as IQ and levels of trust (the data is available on the AER website, so I should test my musings).

Still, if I were to favour a hypothesis of diversity affecting innovation, I can see two hypothetical pathways. First diversity may increase innovative activity as it allows for gains to trade. If there are different kinds of intelligence or other innovative traits, a broader basket of traits may provide more opportunity for technological innovation. The bee studies seem to fall into this camp.

Second, diversity may provide a higher probability of a favourable innovative trait being present in a new environment. Diversity would lead to a useful trait being present, but that trait would then spread and genetic diversity related to that trait would be eliminated by selection.

I lean toward the second pathway. Different forms of intelligence are highly correlated (hence the search for the g factor), and we would expect that a group with higher absolute intelligence would outperform a group with more diverse intelligence levels (unless, of course, those at the top end are exceptional).

However, having favoured this second pathway, I am not convinced that either is the case. I prefer a hypothesis that selection pressures pushed traits in a certain direction, with diversity a secondary factor (if a factor at all). Partly this comes from the nature of innovative activity, which is affected by diversity but relies heavily on the innovative capacity of those involved. Innovative capacity is also only a subset of the traits that affect evolutionary success. If the hypothesis was one of evolutionary success, I would give more weight to a diversity hypothesis. This is also one of the many reasons that extrapolating bee or other studies to humans can be problematic. For the bees, we are measuring proxies for evolutionary success, while innovative activity is only one of many factors that may affect evolutionary success for humans.

Given the above, further research is required to give the diversity-innovation hypothesis standing. But what approach should be used to tease out this question? Cross-species comparison provides one potential avenue, but how would levels of diversity across species be compared or levels of innovation measured? The low genetic diversity in humans relative to other species may complicate the analysis. I also suspect that a cross-species approach might be more useful in assessing the effect of diversity on levels of cooperation (the subject of a forthcoming post) as it may be too difficult to develop useful indexes of innovative behaviour in other species. Alternatively, further work on isolated human populations could be useful.

My preferred direction of research would be to directly analyse the selection that has occurred on the various populations, genetic evidence permitting. I would also examine what other controls are useful in the analysis. IQ is one option, although a Flynn-like hypothesis of economic development and IQ ratcheting up together would make that difficult.

As an end note, some of the debate about the paper raises the question of whether Ashraf and Galor were directly relating genetic diversity to economic development, or whether genetic diversity is a proxy for phenotypic diversity unrelated to that genetic diversity (such as language). I have deliberately skirted that issue for this post, but as you can see below, my thoughts are forthcoming.

My posts on Ashraf and Galor’s paper on genetic diversity and economic growth are as follows:

1. A summary of the paper methodology and findings
2. Does genetic diversity increase innovation? (this post)
3. Does genetic diversity increase conflict?
4. Is genetic diversity a proxy for phenotypic diversity?
5. Is population density a good measure of technological progress?
6. What are the policy implications of the effects of genetic diversity on economic development?
7. Should this paper have been published?

Earlier debate on this paper can also be found here, here, here and here.

Four links this week:

1. Paleofantasy: What Evolution Really Tells Us about Sex, Diet, and How We Live by Marlene Zuk. (HT: Evolvify)
2. Ignorance, the Ultimate Asset has some good bits. Some interesting lines: Taleb, however, is so consumed with the downsides of a complex world and advises strategies so “hyperconservative,” that he often ends up preaching futility and paralysis. The Santa Fe Institute, likewise, has for years looked to complexity as the next economic paradigm but has mostly emphasized the chaotic nature of the economy and has thus run into explanatory dead ends.
3. On a similar theme, Ronald Coase laments the disconnect between economics and entrepreneurship.
4. The latest fertility crisis piece that manages to ignore that fertility has increased in most developing countries over the last decade.

Although the debate it triggered has been going for a few months (see here, here, here and here.), Quamrul Ashraf and Oded Galor’s paper The ‘Out of Africa’ Hypothesis, Human Genetic Diversity, and Comparative Economic Development has been published in the February edition of the American Economic Review (for the latest ungated version, go here – although you can download the data and supplementary materials from the AER site without a subscription).

Over the next few weeks I will dissect parts of Ashraf and Galor’s argument, and look at some of the criticisms that people have made. As a start, however, I’ll present a basic description of the method and findings.

Ashraf and Galor’s hypothesis is that genetic diversity affects economic development through two pathways. First, genetic diversity has a positive role in development as it expands a population’s production possibility frontier. That is, the wider mix of  traits available in the population means that there are more likely to be traits present that can advance and implement new technologies.

The second is a negative effect of genetic diversity, whereby heterogeneity increases distrust, thereby reducing cooperation. This increases the chance of conflict and generally reduces the level of social order in the population.

Ashraf and Galor use expected heterozygosity as their measure of genetic diversity, which is a measure of the probability that two randomly selected people from the population differ with respect to a given gene, averaged over the measured genes. Genetic diversity is affected by what is known as the founder effect. When a new population emerges from a larger population, such as when a group of humans migrate, they take only a subset of the genetic diversity available in the initial population. As humans migrated out of Africa and spread across the world, each new migration took a smaller set of the available diversity. Diversity tends to decline as we move from Africa to Europe to the Americas.

Depending on the relative strengths of the negative and positive effects of genetic diversity on economic development, this pattern may result in a hump-shaped relationship between the two. Populations with more extreme levels of diversity may suffer from insufficient diversity for pushing out the production frontier, or high levels of conflict due to dissimilar individuals.

Ashraf and Galor tested this hypothesis using genetic data from the Human Genome Diversity Cell Line Panel, which comprises 53 ethnic groups, each of which are believed to be native to the area in which they are found and relatively isolated from gene flow from other groups. As a result, they represent a reasonable measure of genetic diversity in those areas before modern-day mobility.

Ashraf and Galor recognised that this dataset is not very big. For example, it comprises only 2 populations from Oceania and 4 populations from the Americas. As a result, they also developed an index of predicted genetic diversity based on migratory distance for a larger group of 145 countries.

As their initial analysis is for 1500 CE, Ashraf and Galor use population density as the measure of development. In a Malthusian world, any improvements in technology that might improve living standards are quickly swallowed by population increases. Population grows to its carrying capacity. The Industrial Revolution was the first time in human history where this pattern was broken. Thus, per person income is a poor measure of technology in a Malthusian world as everyone is at subsistence. Technology only changes how many people can live on subsistence in a given area – hence the use of population density. Population density is also used for their analysis of 1 CE and 1000 CE (which is contained in the Web Appendix). For their analysis of 2000 CE they use income per person.

Ashraf and Galor ran regressions of genetic diversity against economic development using a range of control variables, including latitude, the percentage of arable land and the suitability of land for agriculture. They also use continent fixed effects as part of the controls, which should account for any unobserved continent specific factors. The interpretation of the use of continent fixed effects is that the findings hold within the continents.

Their first set of results using the smaller 53 ethnic groups found a hump-shaped relationship between genetic diversity and development, as would be predicted by the opposing costs and benefits to diversity. The size of the effect is such that a 1 percentage point increase in diversity for the least diverse society would increase population density by 58 per cent. A 1 percentage point decrease in diversity for the most diverse society would increase population density by 23 per cent. Ashraf and Galor also ran some tests to show that it is diversity and not migratory distance that is affecting development.

The headline results from their larger predicted diversity set is again a hump-shaped relationship between diversity and development for 1500 CE. A one percentage point increase in diversity for the least diverse society would increase population density by 36 per cent, while a one percentage point decrease for the most diverse society would increase population density by 29 per cent.

The analysis is then done for 2000 CE, with country diversity calculated by examining the mix of ethnicities that make up the country. Again, the hump shaped pattern holds. Given countries can be identified, these results have attracted some of the most attention. Ashraf and Galor summarise them as follows:

The direct effect of genetic diversity on contemporary income per capita, once institutional, cultural, and geographical factors are accounted for, indicates that (i) increasing the diversity of the most homogenous country in the sample (Bolivia) by 1 percentage point would raise its income per capita in the year 2000 CE by 41 percent; (ii) decreasing the diversity of the most diverse country in the sample (Ethiopia) by 1 percentage point would raise its income per capita by 21 percent; (iii) a 1 percentage point change in genetic diversity (in either direction) at the optimum level of 0.721 (that most closely resembles the diversity level of the United States) would lower income per capita by 1.9 percent; (iv) increasing Bolivia’s diversity to the optimum level prevalent in the United States would increase Bolivia’s per capita income by a factor of 5.4, closing the income gap between the United States and Bolivia from a ratio of 12:1 to 2.2:1; and (v) decreasing Ethiopia’s diversity to the optimum level of the United States would increase Ethiopia’s per capita income by a factor of 1.7 and thus close the income gap between the United States and Ethiopia from a ratio of 47:1 to 27:1. Moreover, the partial R² associated with diversity suggests that residual genetic diversity explains about 16 percent of the cross-country variation in residual log income per capita in 2000 CE, conditional on the institutional, cultural, and geographical covariates in the baseline regression model.

The paper closes with a brief look at the evidence for the costs and benefits of genetic diversity, such as various measures of trust and innovation within countries. Ashraf and Galor show that there is some evidence that these relationships are in the right direction. I’ll delve into that evidence in more detail in later posts.

My posts on Ashraf and Galor’s paper on genetic diversity and economic growth are as follows:

1. A summary of the paper methodology and findings (this post)
2. Does genetic diversity increase innovation?
3. Does genetic diversity increase conflict?
4. Is genetic diversity a proxy for phenotypic diversity?
5. Is population density a good measure of technological progress?
6. What are the policy implications of the effects of genetic diversity on economic development?
7. Should this paper have been published?

Other debate on this paper can also be found here, here, here and here.

Following my last post suggesting that there was no quality-quantity trade-off in modern societies (at least to an extent that mattered), I wanted to point to a nice model that makes this argument. The interesting thing about the model is that the purpose of the authors in developing it was not primarily to make that point.

The model is by Oded Galor and Omer Moav model in their article Natural Selection and the Origin of Economic Growth. I’ve posted on the paper before and my first working paper simulated this model (a post about that paper is here).

The basic idea of the model is as follows. In the population there are two genotypes – those who prefer higher quality children, and those who prefer a greater quantity of children. Those with a preference for higher quality invest more in educating their children.

In the Malthusian state before the Industrial Revolution, those who invest in their children’s quality have a fitness advantage. Their children are better able to take advantage of the scarce resources. As a result, the quality-preferring types increase in prevalence, increasing the average level of education in the population. Technological progress depends on the level of education in the population, so this also increases.

This process creates a feedback loop. Faster technological progress increases the return to human capital, so the quality-preferring types further increase their children’s education. Eventually the returns to human capital become so large that even the quantity-preferring types invest in some education. As this point technological progress greatly accelerates and the economy enters into a new modern growth state.

However, this changes the dynamics of investment in education. With technological progress so high, the quality-preferring types over-invest in educating their children and their prevalence starts to decline. Those with a lower investment in child quality now have the evolutionary advantage.

The optimal strategy in this new plentiful environment is to put all the resources into having more children, not increasing their quality. Resources are now so plentiful that all children can survive and reproduce in the next generation regardless of the level of investment. In my working paper, we showed that the less an agent invested in quality in the modern growth state, the greater fitness advantage they had over the other types in the population.

This particular model implies that, depending on the triggers for investment in children in modern economies, many people may be over-investing, at least if you consider evolutionary success the benchmark. Those types who cut the investment in quality will grow in number.

Together with evidence that the costs to child quality of having another sibling are small, this model suggests that, from an evolutionary perspective, the trade-off is not one of quantity versus quality. Rather, the trade-off is between quantity and investment in quality. That investment in quality simply does not generate the evolutionary returns that it once did.

Following his disappearance from Psychology Today, Satoshi Kanazawa has reappeared in big think (with not all happy with this move [Update: he’s now gone again, along with the post]). In a recent post, Kanazawa asks an interesting question – Why do people with many siblings have many children? Kanazawa writes:

Studies show that fertility is substantially heritable; genes partly influence how many children one has. Children of parents who have many children also have many children; children of parents who have few children also have few children. As a result, there is a strong positive correlation between the number of siblings one has (which equals the number of children that one’s parents had minus one) and the number of children one has.

… [T]he positive correlation between the number of siblings and the number of children makes no evolutionary sense. Evolutionary theory actually predicts a negative correlation. …

This is because people with many siblings have the option of investing in their younger siblings and increasing their reproductive success by doing so. Humans are just as genetically related to their full siblings as they are to their own biological children; both share half their genes. … So investing in and “raising” younger siblings is just as good genetically and evolutionarily as investing in and raising one’s own genetic children.

A second reason for expecting a negative correlation is that the parents face a quality-quantity trade-off. As a parent increases the number of their children, they have fewer resources to invest in each child. If that harms their children’s reproductive success, this will reduce the number of children in the following generation.

I would argue that the answer to this puzzle (as Kanazawa terms it) lies in what happened after the demographic transition – that is, when fertility rates declined with industrialisation. Before the transition, fertility was generally not heritable and any correlation in family size between generations was due to environmental factors and chance. But after the transition, heritability increased. This suggests that different people responded to the transition in different ways.

One of those differences is how they respond to the quantity-quality trade-off in the new, plentiful environment that accompanies industrialisation. Human history is one of Malthusian conditions, where available resources constrain the population. In that environment, any additional children will affect the resources available for the other children.

In the modern environment, there is essentially no such constraint. Increased quantity does not harm quality in a way that reduces the quantity of children in the next generation. Adoption studies generally show a low cost to more children, and those small costs are to income or education, not reproductive success. As a result, the optimal strategy in the modern environment is to have as many children as you can. If you have a genetic predisposition to do that (or even a culturally transmitted predisposition), your children are likely to too, leading to the positive correlation that Kanazawa observes.

This argument also applies to Kanazawa’s argument that we should expect a negative correlation through care for siblings. In the modern environment, that care for siblings does not assist the sibling in having more children. They are not resource constrained. The optimal strategy is for each sibling to have as many children as they can. Those that follow this strategy have more children in successive generations with little cost to quality.

This argument relates to my latest working paper. With the heritability of fertility increasing after the demographic transition, “high-fertility genotypes” can be expected to increase in number. One day a cost for those excessive children may emerge as the population starts to hit Malthusian constraints, but until then, fertility is not constrained by resources and the evolutionary dynamics point to fertility going up.

Four links this week:

1. Quamrul Ashraf and Oded Galor’s paper on genetic diversity and economic growth has been formally published in February 2013 edition of the American Economic Review (ungated working paper version here). When its release was foreshadowed a few months ago, it generated some interesting debate (such as here and here, particularly in the comments). Starting next week I’m going to write series of posts examining the threads of Ashraf and Galor’s argument.
2. The gorilla is invisible, even to radiologists.
3. Was Herbert Spencer a Social Darwinist? (From late 2011, but a good read)
4. Stephen Corry of Survival International lays into Diamond and Pinker. (HT: Evolvify)

In my latest working paper, co-authored with Oliver Richards, we argue that recent fertility increases in developed countries may only be the beginning. From the abstract:

We propose that the recent rise in the fertility rate in developed countries is the beginning of a broad-based increase in fertility towards above-replacement levels. Environmental shocks that reduced fertility over the past 200 years changed the composition of fertility-related traits in the population and temporarily raised fertility heritability. As those with higher fertility are selected for, the “high-fertility” genotypes are expected to come to dominate the population, causing the fertility rate to return to its pre-shock level. We show that even with relatively low levels of genetically based variation in fertility, there can be a rapid return to a high-fertility state, with recovery to above-replacement levels usually occurring within a few generations. In the longer term, this implies that the proportion of elderly in the population will be lower than projected, reducing the fiscal burden of ageing on developed world governments. However, the rise in the fertility rate increases the population size and proportion of dependent young, presenting other fiscal and policy challenges.

We’re certainly not the first to hint at the idea that selection of high fertility individuals will increase fertility. Fisher noted the power of higher fertility groups in The Genetical Theory of Natural Selection. I’ve seen Razib Kahn, Robin Hanson and John Hawks mention the idea in blog posts. There is one great paper by Murphy and Wang (which I will blog about soon) that has part of this argument buried in the micro-simulation. Many papers on the heritability of fertility hint at it. Rowthorn’s paper on fertility and religiosity also points in this direction. But what we couldn’t find was someone who sought to tie down the idea – particularly in the way we have.

I actually thought a paper of this nature would already be written. We were interested in the economic implication of the argument, but because there was no clear statement of the evolutionary foundations that we could use in the way we wanted, we decided to build our argument from the ground up. We’re hoping that this working paper receives some solid critique that will allow us to decide whether our angle of attack is useful or can be improved. We have constructed three basic genetic models, but are they useful? Are there better alternatives? Once we address those questions, we have some ideas for empirical tests and we hope to use the concept in some more detailed economic and cultural analysis. Ultimately, this paper will need to be tied in with a large and growing literature on the biosocial basis of fertility.

I’m the first to admit we could be wrong in the prediction of a fertility increase. What other shocks are still to come? Will the continually changing environment drown out the underlying evolutionary dynamics? Our instinct is that most of the shocks that can affect fertility have played out in the developed world – increased incomes, effective contraception, female choice and so on. But what further shocks could reduce fertility?

In presenting this paper, we tend to receive  two major classes of response. The first and obvious question is whether these dynamics will play out in time frames that matter. As its been 200 years since some populations underwent the demographic transition, there has  been enough time for selection to have occurred on a trait as important to fitness as fertility (in some populations we have evidence of this). The more interesting question is what will be the magnitude of the effect over the next 50 or 100 years? I’m not sure of the answer to this, but even a small total fertility rate increase of 0.1 children per female can have material effects on population size and structure.

The other response we tend to receive is that fertility is affected by policy, incentives, female opportunities and so on. Any trend we see today is a response to those factors. And that may be true. But to the extent there is variation in the response to the policy, incentives or opportunities and there is a genetic basis to that variation, we can see selection of those with higher fertility.

Having said that, throughout the paper we are deliberately agnostic about the merits of the various theories of what has caused the fertility decline in the developed world to date. For the purposes of our hypothesis, it is sufficient to know that there was a decline in fertility and that variation in fertility is heritable after the decline. As the first law of behaviour genetics is that all human behaviour is heritable, its not a very high bar to clear. It would be difficult if it was otherwise, particularly when you consider the raft of current theories. And even if you believe a certain factor is behind the decline, what is the causative pathway? As an example, consider the spread of the pill and the factors which are relevant to it reducing fertility. First, there is desire of someone with access to the pill to have children. Then there is their desire to take the pill to control pregnancy. Do they take the pill as instructed (possibly related to conscientiousness)? Is the pill physiologically effective? Do they experience side-effects that deter continued use? Variation along any of those dimensions would affect fertility.

The biggest simplification in the way we present our models is that, unlike our models, developed countries did not receive a single fertility shock across the population. Rather, multiple shocks hit different parts of the population at different times. This is why fertility has generally declined for much of the last 200 years, rather than suddenly suffering a single large drop. Of note, fertility tended to decrease among the wealthy first. As our framework would suggest that fertility rates will increase first among groups that experienced the shock earlier, we would predict that groups with a history of higher socioeconomic status will tend to increase their fertility rates earlier.

Immigration also presents some interesting issues. Immigrants tend to have higher fertility when arriving in a country that has undergone the demographic transition. But our framework would suggest that following generations will experience a decline as they undergo the fertility shock. To the extent that the immigrant population has not been exposed to the shock before, their fertility may decline and recover later than native populations.

These issues offer some basis for testing the hypothesis. But first, we’re keen to nail down some good ways of thinking about the problem. The working paper and the models within are part of that process. So, if you have any thoughts or criticisms, we’d be grateful to hear them.

Another interesting old paper off my reading pile has been Robert Sugden’s Spontaneous Order.

Sugden asks us to picture a scenario where you are travelling down a road towards another car. You have two alternatives – move to the left or right. The other car has the same choice. If you pick the same, you pass safely. A game theoretic analysis tells us that there are three Nash equilibria – that is, three sets of strategies that neither driver would have incentive to deviate from (or put yet another way, each driver’s strategy is a best response to the other driver’s strategy). The three are: both cars move left; both cars move right; or each car moves left or right with 50 per cent probability.

But which strategy do they choose? Sugden writes:

[T]he ultimate objective of game theory is to show that rational analysis uniquely prescribes a particular strategy for each player in a game. It is as if each player sits in a room by himself, knowing nothing about the other player except his utility function and that he is rational, and knowing nothing about how the game may have been played by other people. Each player must decide what to do, applying unlimited powers of rationality to this severely restricted information and to nothing else. …

One of the major achievements of Thomas Schelling … has been to show that games like Chicken cannot be “solved” in this way. The ideally rational but completely inexperienced players of classical game theory would find they had insufficient data to determine what they should do. In contrast, ordinary people with limited rationality but some degree of experience and imagination might have no difficulty in coordinating their behavior. On this view, the program of classical game theory is a blind alley: it requires us to throw away the information that players need if they are to work out what it is rational for them to do. …

If we are to coordinate our behavior, as we both wish to do, we must rely on some shared notion of prominence. Our common experience of English driving provides the clue we need. Steering left is prominent because it is common knowledge that this is what people generally do: we have each observed this, we can each assume the other has observed it, and so on.

But this raises a new question. Where does this experience come from? Why steering to the left?

If we are to explain why one convention is found rather than another, it is not very useful to start from a comparison between a world in which everyone follows one convention and a world in which everyone follows the other: either of these worlds, once achieved, would be self-perpetuating. Instead we must consider the process by which conventions evolve. More particularly, we must look at how they start to evolve. Once a convention has started to evolve – once significantly more people are following it than are following any other convention – a self-reinforcing process is in motion. The conventions that establish themselves will be the ones that can take root (biological metaphors are almost unavoidable) most quickly in a convention-free world.

I have always been a fan of arguments that recognise the path dependent nature of evolution. There are many possible outcomes, but various historical contingencies shape the result.

A convention can start to evolve as soon as some people believe that other people are following it. But what gives rise to this initial belief? One possibility is that the same forces are at work as enable people to coordinate their actions without communication in unrepeated games. Some forms of coordination are more prominent than others, and people have a prior expectation of finding the most prominent ones. But, I have argued, prominence is largely a matter of common experience. The implication is that conventions may spread by analogy from one context to another. If it is a matter of common knowledge that a particular convention is followed in one situation, then that convention acquires prominence for other, analogous situations. For example: on my journey to work there is a narrow bridge, not wide enough for two vehicles to pass. If two drivers approach from opposite directions, which of them should give way? Coming on this problem for the first time, my prior expectation was when the drivers came into view of one another, whoever was closer to the bridge would be given the right of way. This expectation-which proved correct-was based on an analogy with the “first come, first served” principle.

This line of reasoning has some relevance to arguments about the evolution of cooperation.  In A Cooperative Species, Bowles and Gintis note that where there are repeated interactions between people, there can be infinitely many equilibria. This concept is known as the folk theorem. Due to this possibility, Bowles and Gintis suggest that it is unlikely that cooperation could evolve in such a situation purely by self-interested rationality. How would the parties reach equilibrium, and why would it be the cooperative one? But as Sugden’s argument suggests, if certain equilibria have prominence, possibly by historical accident and the shared experience of the actors, there may be more scope for cooperation. Humans are not perfectly rational beings solving a problem from scratch. We are evolved creatures with a history and experience.

This week we have paleo for dogs, irrational consumers and the question of what is the right size:

1. John Hawks posts on a new Nature paper that provides a reason for “why we feed dogs kibble instead of raw beef” – well, at least those who don’t have their dogs on the paleo diet. Like humans with a long agricultural history, dogs have extensive duplication of the amylase gene, which assists in the digestion of starch. Razib asks “Is Paleolithic man to Holocene man what the wolf is to the dog?”
2. The Atlantic covers a new paper by Daniel McFadden that summarises the case against the neoclassical economic view of the consumer. From a quick skim, there’s not too many surprises but it’s a nice summary.
3. Nick Gruen tweeted a link to this 1928 classic by J.B.S. Haldane, On Being the Right Size. Andy Haldane of the Bank of England borrowed it for a speech on bank size. I suggest going back to the original.