Genetic diversity and economic development

Recently two economists, Quamrul Ashraf and Oded Galor, published an article in a prominent economics journal comparing genetic diversity in various countries with economic development (ungated version here).  They found the following relationship: high and low genetic diversity is associated with low economic performance.  High economic performance is associated with moderate levels of genetic diversity.  Below is the take-home “hump-shaped” graph comparing genetic diversity to per capital income for a slew of countries (lower numbers on the x-axis represent higher genetic diversity).





Ashraf and Galor conclude that this relationship supports the hypothesis that “genetic diversity within a population confers both social costs, in the form of miscoordination and distrust arising from genetic differences across members of society, and social benefits in the form of diversity-driven knowledge accumulation.

Needless to say, this interpretation caused some controversy.  A group of anthropologists (mostly from Harvard) penned a response in Current Anthropology before the original article was even published (ungated version).  Jason Collins, at Evolving Economics, has been covering the fall-out in detail.


Much of the back-and-forth has been about the specific methods employed by Ashraf and Galor.  I am going to leave that to others and instead focus on Ashraf and Galors proposed mechanisms.  For what it is worth, I mostly agree with Andrew Gelman’s take on the other issues.
 

Critique 1: If any finding is consistent with an hypothesis, finding something is not very good support for the hypothesis.

 

Ashraf and Galors’s hypothesis, as quoted above, predicts that genetic diversity should have both a positive and a negative effect on economic productivity. The great thing about this sort of hypothesis is that it can explain any observed pattern in the data.

For example, in the stylized charts below, the top chart reflects the findings of Ashraf and Galor, highlighting the positive and negative effects of genetic diversity creating a “hump-shaped” curve.  The bottom chart reflects the exact opposite findings: a “trough-shaped curve.”  Notice that this curve is also consistent with Ashraf and Galor’s hypothesis – showing regions of “negative effects” and regions of “positive” effects.  Even a flat line would be consistent with their hypothesis (the effects cancel out!).  If any empirical pattern is consistent with an hypothesis, finding a specific empirical pattern that is consistent with the hypothesis is not too surprising.




Critique 2: Genetic Difference Only Matters on (Very) Small Scales

There is a large body of work in evolutionary biology on the scale at which genetic differences should matter in cooperation. The consistent finding is that genetic differences only matter for very close relatives, for animals like humans who have fairly limited number of offspring (unlike social insects), the scale at which genetic differences might matter is something under a dozen individuals.  Any more than that and genetic relatedness is just too diluted to make a difference. 

Countries are much bigger than a dozen individuals. Within a country, people might be more cooperative with their immediate relatives, but any genetic diversity beyond that shouldn’t matter.
  
When they hypothesize that genetic differences cause “miscoordination and distrust arising from genetic differences across members of society” this sounds a lot like kin recognition.  Basically, individuals act differently towards others based on observed genetic similarities and differences. In the classic paper by François Rousset and Denis Roze, they find that even under the most ideal conditions, kin recognition only works in extremely small groups [summary here].

 

Critique 3: Genetic Diversity Cannot Explain (much) Cognitive Diversity

Political scientist Scott Page has two books summarizing research into diversity from a variety of academic disciplines. One of the books’ key points is that are that the important type of diversity in group decision-making and innovation is “cognitive diversity,”  defined as “differences in how people see, categorize, understand, and go about improving the world.”

For example, economists see and understand the world differently than population geneticists. This implies that a study about population genetics and economics would be better if conducted by a mixed group of economists and population geneticists, than by a group of only economists or a group of only population geneticists (see what I did there).

What are the sources of cognitive diversity? Are they likely to be genetic? The answer is no.  The basic argument is that in any given group there is much more cognitive diversity than genetic diversity. Therefore, genetic diversity cannot explain very much cognitive diversity. Most cognitive diversity seems to result from differences in training and experience.

(Update: After posting this, I saw that Jason Collins posted today on the claimed relationship between genetic diversity and innovation.)

 

Conclusion: Reasons to Be Skeptical

 
I am skeptical of the conclusions of this study for three basic reasons. (1) The hypothesis is consistent with any observed pattern in the data, (2) the hypothesized negative effects of genetic diversity are unlikely to matter on the scale of countries, (3) the hypothesized positive effects of diversity are unlikely to be a result of genetic diversity.

14 thoughts on “Genetic diversity and economic development

  1. Incidentally, thanks to both you and Jason Collins for posting at length on this topic. I’ve found these posts very informative in clarifying some of the questions and concerns I had about the Ashraf/Galor argument.

  2. Unfortunately, Zimmerman’s analysis appears to be based on an incomplete reading of Ashraf and Galor’s research and some confusion about scientific methodologies.

    1. Ashraf and Galor have a formal theory that generates unambiguously a hump shaped relationship between diversity and economic development. The sufficient conditions for this are diminishing marginal productivity of diversity in fostering innovations and diminishing marginal productivity of homogeneity in fostering cooperation. These are highly plausible assumptions that are at the foundations of economic theory.

    2. Ashraf and Galor establish empirically that each of these effects is supported empirically. Diversity fosters innovation and diversity reduces trust and increases the incidence of civil conflicts.

    3. A recent paper of Ashraf and Galor “Genetic Diversity and the Origins of Ethnic Fragmentation” AER P&P (May 2013) provides additional evidence about the adverse effect of genetic diversity on cooperation. They show that genetic diversity is the cause of ethnic linguistic fractionalization — a well-established force that reduces economic development.
    http://ideas.repec.org/p/bro/econwp/2013-2.html

    4. Ashraf and Galor findings confirm, unambiguously, a hump shape effect of diversity on economic development. Hence, even if theoretically the predictions would have been ambiguous that data support the Ashraf and Galor hypothesis. I thought the purpose of empirical analysis is to confirm or refute potentially conflicting theoretical predictions.

  3. Dear e10c8308-76c2-11e2-9ab9-000bcdcb471e,

    Welcome to the conversation!

    I think a key difference in our interpretation is that you write about “diversity” generally, and Ashraf and Galor write very specifically about *genetic* diversity. My skepticism, which stems from population genetics (not from economics), is that countries are at least a few orders of magnitude too large for genetic diversity to matter. Maybe there is a way to get it to work, but to convince me that the assumption is valid for genetic transmission at the scales they report, you would, at a minimum, have to justify it with a model. One that moves genes around in a realistic way with selection acting in a realistic way. Waving one’s hands around about “kin selection” isn’t going to cut it when the kin selection literature is fairly clear that it only works at very small scales in organisms that reproduce like mammals. This is because genetic similarity decreases exponentially with relatedness.

    An empirical relationship with lots of stars by it is not going to cut it. The statistical relationship may be real without the post-hoc justification for the relationship being valid. The problems of massive collinearity in cross-national data are well known.

    However, there seems to be reasonable evidence that *preference* diversity and *identity* diversity might act to decrease cooperation (though there is healthy debate about how much of the later is caused by the former). And there seems to be reasonable evidence that *cognitive* diversity might act to increase decision-making and problem-solving. But I would need to see a model, or even a clear word explanation, about the causal connection between genetic diversity and these other types of diversity to be convinced that it is the underlying cause. Scott Page’s “The Difference” book has a thoughtful discussion of how these different types of diversity might interact.

    I’ve read some economics literature related to my work and I have not come across anything suggesting that assumptions concerning genetic diversity are “at the foundations of economic theory.” But I would appreciate a reference establishing the plausibility of the assumptions Ashraf and Galor make about genetic diversity over the scale of interaction they describe in their paper. I would be especially interested an article that establishes the plausibility of “diminishing marginal productivity of homogeneity in fostering cooperation” as that is most related to my research.

    Empirical analysis has lots of purposes. Clarke and Primo’s “A Model Discipline” addresses some of these issues for social science. Burnham and Anderson’s “Model Selection and Multi-Model Inference” addresses them for a more biological science audience (but data is data). I like to think of empirical analysis as a way of comparing models. A more stereotypically Popperian view is that it can be used to refute (but not confirm) specific models. I also think to think of science as a dialog between theoretical and empirical models where. I am most convinced when they begin to support one another.

  4. Dear Matt,

    You write: “My skepticism.. is that countries are at least a few orders of magnitude too large for genetic diversity to matter.” I failed to understand your skepticism.

    Here is a simple example in which genetic diversity would have a major effect. Suppose that technological advancement can be made only by people with IQ of 120 and above (not an implausible assumption, I suppose). Consider two societies with the same mean IQ (i.e., 100), but one of the two has larger diversity and therefore fatter upper tail above 120. Isn’t it obvious that the more diverse society will be more innovative?

  5. When I wrote about kin selection, I was referring to the suggested association between genetic diversity and cooperation/trust. I don’t see this working at large scales. McElreath and Boyd’s “Mathematical Models of Social Evolution” is a good reference if you are not familiar with kin selection models. Rice’s “Evolutionary Theory” is also good, but a little more technical, introduction to population genetics modeling.

    Your story is about variance in IQ, not genetic diversity. Ashraf and Galor don’t measure or use variance in IQ in their empirical analysis. This would seem a simpler test of the idea.

    The way I understand your story is that genetic diversity causes variation in IQ which causes variation in individuals’ ability to innovate which causes variation in the number of innovations within a state which causes variation in per capita GDP. Is this essentially correct?

    This long string of causation is not at all obvious to me.

    Ignoring genetics, even if there is variance in ability to innovate (which may-or-may-not have anything to do with IQ) your story is about the number of innovators in a society. This number is heavily influenced by population size and probably much more heavily influenced by the population size than, say, genetic diversity, since states are much more varied in the population sizes than in their genetic diversity.

    Finally, the relationship between diversity and ability is complicated. See, for example, Hong and Page’s “Groups of diverse problem solvers can outperform groups of high-ability problem solvers.” http://www.pnas.org/content/101/46/16385.full

    Also, I was serious about my request for relevant references in my above comment. I really would like to look at them.

  6. You write: “My skepticism…is that countries are at least a few orders of magnitude too large for genetic diversity to matter.” …”I was referring to the suggested association between genetic diversity and cooperation/trust.”

    I am glad that you find the argument about the effect of diversity on innovations plausible. I think that your narrow interpretation of the approach of Ashraf and Galor (AG) is misleading. They view genetic diversity (GD) as a proxy for the overall level of genetic diversity and their reduced form finding that GD contributes to innovations could operate via different channels, one of which is diversity in cognitive traits.

    As to the adverse effect on cooperation, the forthcoming paper by AG “Genetic Diversity and the Origins of Ethnic Fragmantation” “advances the hypothesis that genetic diversity, determined predominantly during the migration of humans out of Africa tens of thousands of years ago, is a fundamental determinant of observed ethnic and cultural heterogeneity, as reflected by the number of ethnic groups and the levels of ethnolinguistic fractionalization and polarization within modern national boundaries. Following the out of Africa migration, the initial level of genetic diversity in indigenous settlements presumably facilitated the formation of distinct ethnic groups through a process of endogenous group selection, based on the tradeoff between the costs and benefits associated with heterogeneity and scale. While heterogeneity raised the likelihood of disarray and mistrust, reducing cooperation and thus adversely affecting group-specific productivity, complementarities across diverse productive traits and preferences stimulated productivity. Since in a given environment, diminishing marginal returns to diversity and homogeneity entail an optimal size for each group, higher initial genetic diversity would have positively contributed to the number of groups, and thus to the degree of fractionalization.”

    Genetic diversity is shown by Ashraf and Galor to lead to a higher degree of ethnic fractionalization and this in turn reduces trust and cooperation, as is well documented in the economics literature (and as 1 million people that lost their life during the ethnic conflict between the Tutsi and the Hutus could have testified).

  7. The explanation for xenophobia is partly due genetic kin selection, partly due cultural kin selection, and partly due to lack of reciprocity. I.e. you are nice to in-group members due to a history of reciprocal interactions with them and a justified expectation of more in the future. With a member of another tribe there is no such history and no such expectation – so they are more likely to bash your head in with a rock. You should not write off genetic kin selection. The phenotype of a member of a perceived different race screams that they are not a relative – and not a member of your group. It is an unrelatedness superstimulus.

  8. I believe misreading of Ashraf and Galor’s piece (AG) largely stems from a knee jerk reaction to a word “genetic” in the title. Excitement over developments in genetic decoding, tragic memory of the 20th century devastating theories of biological determinism and fascination with scientific novelty leads to a very narrow, often sensationalist and therefore imprecise reading of their paper.

    I think Mr “e10c8308….” summarized AG argument fairly well: “They view genetic diversity (GD) as a proxy for the overall level of diversity and their reduced form finding that GD contributes to innovations could operate via different channels, one of which is diversity in cognitive traits”.
    In other words, it is far less of a story “genes made me do it” rather than a story “diversity matters (a lot)”.

    AG use genetic diversity as a proxy, i.e., a tool of estimating diversity defined in far broader terms. For a historical research that spans centuries this is not a trivial matter. How else would you approach estimating diversity? How can you compare diversity in France in 2000 and 1000? A qualitative approach is one way to get the story right. And you can write a multi-volume story of a transformation of diverse regional kingdoms and vassalages into what is currently known as France and French nationality. However, it would take volumes for each particular case. And we would not get to answer the question of measuring diversity over time through it, would we?

    Another strategy is to look at ethnicity, religion, race, languages… Hold on a sec, but cultural anthropologists and sociologists have spent decades trying to educate the public that categories that exist to define differences and are denoted by terms ethnicity, religion, race, and even language (to some extent) are highly variable, socially constructed and change from t0 to t1 tremendously thereby precluding any meaningful long term historical computational studies (well, it would not stop journals from publishing those long term computational studies, but that’s a different story).
    So, ironically, AG opt to take an anthropologically informed approach. They do not look at ethnicity or religious differences as those categories are time and location specific. Instead, they identify a measure of diversity that is far less affected by cultural constructions of various times and epochs.
    Ironically, their approach subscribes to the major tenet of anthropology – a holistic perspective on human life in which culture plays a major role, but so does biology. I call this ironic because anthropologists cannot seem to see this point (but, perhaps, it is far less surpising given uncritical knee-jerk reaction to a word “genetic” that anthropologists share with the rest of the population). Thus, AG approach on diversity can be summarised as such:
    a) There is a biological component to diversity
    b) There is socio-cultural component to diversity
    c) a) and b) are not mutually exclusive
    c1) more needs to be learnt about relationship between a) and b)

    Respectfully,
    Henry James Maine

    1. Thanks Henry, for the comments.

      Jason Collins has posted on the proxy argument. The original paper was not at all clear about it. Many of the statements in the paper, not just the one that Jason Collins, quotes clearly imply a causal mechanism of genetic diversity, not just a proxy measure. http://www.jasoncollins.org/2012/10/genetic-diversity-and-economic-development-ashraf-and-galor-respond/

      Either way, I agree with Collins that they are generally muddled on mechanisms.

      You are right that finding a consistent measure of ethnicity is hard. Ethnic groups are based on identity and are often nested, overlapping, defined differently by different governments, defined differently by governments than by individuals, defined differently between individuals, defined differently over time, and individuals have different identities are more salient in different situations. (See Henderson “The Democratic Peace Through the Lens of Culture” and Fearon and Laitin “Ethnic and Cultural Diversity by Country” for a sense of the problems). This is a hard problem. I think one direction is to measure cultural variation directly. Bell et al take a stab at this using WVS data (“Culture rather than genes provides greater scope for the evolution of large-scale human pro-sociality.”)

    2. Causation and genetic diversity – I believe the authors were fairly clear in their response to the Harvard letter on this matter. I am not sure I have sufficient reasons to question their honesty here. After all, their research is not a story of biological determinism. They factor in a wide range of social, geographical, and historical factors that contributed to the process of human development. Why would they have to do it, if it were simply a story of genes and their direct impact?

      Muddled Mechanism of influence of diversity on societal processes – sure, it is muddled. AG are hypothesizing and calling for more research on this, aren’t they?

      Measuring cultural variation “directly” – well, I wish the “direct” measures were this simple and elegant as you seem to suggest. Currently, ethnicity remains the category of choice to denote diversity. While most agree that ethnicity is hard to categorize and too vague to standardize, social scientists proceed to employ it as a meaningful unit of analysis (see Alesina et al, 2003 for a list of ethnic groups without a proper theoretical foundation). It is probably not surprising that the current understanding of the role of ethnicity in social conflicts and other processes is abysmal! For instance, investigations of the effect of ethnic diversity on the onset of civil war find that different measures of ethnic diversity are positively related to conflict (Collier & Hoeffler, 2001), positively related to low-intensity armed conflict but not civil war (Hegre & Sambanis, 2001), or not at all significantly related to the probability of civil war onset (Fearon & Laitin, 2003) (for more on this see Chandra and Wilkinson 2008). Constructivist approaches are being developed, but are yet too limited in scope (for more on this see Abdelal et al, 2009)

      WVS – well, delightful as that data source is, it is highly problematic as it treats a concept of values (and culture) as a bounded whole. Critique of WVS can follow along the lines of Inglehart’s index of materialism critique (see Klein, 1995; Davis and Davenport, 1999; Clarke et al., 1999). But even leaving conceptual and methodological issues aside, WVS has one major flaw – it was designed in the 80s. Big hair, Eddie Murphie’s video and WVS…Ok, casting nostalgic digressions aside, the problem with the “direct” measure of cultural diversity is that we cannot apply it to historical research. Can one use WVS data to estimate diversity in the early 20th century? 10th century? etc.

      Finally, I see the following significance of AG’s project:
      1. They devised a clever approach to estimating diversity. Much work remains to be done on the mechanism and overall clarification of the measure. But it is an important step in operationalizing a concept of diversity and solving a problem of cultural and historical specificity of this concept.
      2. In the postmodern world of social science that largely focuses on critical critiques of critiques or voyeuristic clarifications of clarifications of earlier research, AG had courage to use their proposed measure to draw an extremely broad, large scale picture of global processes. At the very least, their take on this demands respect.

      Best Regards,

      Henry James Maine

    3. Thank you for the references on ethnicity. I haven’t seen a few of those previously.

      Their response read to me like they overreached on genetic diversity as a cause and then backpedaled. But perhaps the original paper was just confusingly written. I am prepared to chalk this up to “an empirical finding in search of a mechanism.”

      One reason to doubt their claim that they think genetic diversity is merely a proxy is the paper referenced by the anonymous commenter above. They *seem* to argue that they “empirically demonstrate that genetic diversity… is an underlying cause of various existing manifestations of ethnoliguistic heterogenetity.” This could be another example of bad/confusing writing, but it seems to me like they are fingering genetic diversity as a causal factor behind ethnic diversity. (http://www.nber.org/papers/w18738.pdf)

      In this paper, they propose a word-model of how this might work. I have a hard time following the logic of word models and would much rather have an analytic or computational model. I can not evaluate this mechanism on its face.

      However, the bulk of existing models of these mechanisms cause me to doubt them. The do not cite any existing models, so a reader might suspect that it has never been done before – thus the “need for more research.”

      I would start with these: R. McElreath, R. Boyd and P. J. Richerson. Shared Norms Can Lead to the Evolution of Ethnic Markers. Current Anthropology, 44: 122–130, 2003; R. Boyd and P. J. Richerson, The Evolution of Ethnic Markers, Cultural Anthropology, 2: 65–79, 1987; J. Henrich and R. Boyd, The Evolution of Conformist Transmission and the Emergence of Between-Group Differences.Evolution and Human Behavior, 19: 215–242. 1998;
      Henrich, J. (2004) Cultural Group Selection, Coevolutionary Processes and Large-scale Cooperation. Journal of Economic Behavior and Organization, 53: 3-35 and 127-143; Ross A. Hammond and Robert Axelrod, “The Evolution of Ethnocentrism, ” 2006. This is just a sample from social science; there are scores of models on just the genetic aspects of the argument.

      The take-home message of the body of work is that culture transmission can support cooperation in much larger groups than genetic transmission. If anything, this implies that, if there is a causal relationship between ethnic and genetic diversity between large groups, the causation arrow likely goes in the opposite direction from what they state. Actually, with a three minute literature search, I found a model of that too: Premo and Hubin. 2008. Culture, population structure, and low genetic diversity in Pleistocene hominins.

      In short, if “more research is needed” into the causal mechanisms, one quick way to do more research is to read the research that has already been done. Finding correlations in observational data, to me, is rarely sufficient to “empirically establish” causation.

  9. Matt,
    I’m not sure that Ashrav and Galore had the hypothesis before the data? Often, the scientific publication is a travesty in which the data come first and the hypothesis is rather a conclusion, but the whole writing up is done as a kind of reverse engineering so that the data seem to corroborate the hypothesis. I do not agree that any data would corroborate their hypothesis. A U shaped instead of a hump shaped curve would do the contrary. But I’m sure a reverse engineered write-up would have secured them an equally controversial publication. Anyway, this is an extraordinarily sweeping article reminding me of one with chocolate and intelligence. Anyway, even if their relation held, they might have cause and effect upside down.

  10. Hi Matt,
    I recently discovered AG’s paper on gender diversity and it got my attention which brought me here.
    From an ethical point of view, I understand how their hypothesis can be misconstrued and given the recent debate on diversity, the conclusion drawn can be very volatile in nature.
    However, since I am from Pakistan I have to reapply this theory to my country because if I go blatantly by AG’s theory of diversity, our lack of diversity works to our disadvantage when it shouldn’t. A small nation like ours faces lack of diversity if compared to the likes of India, but within the nation, it is quite diverse. So I am unable to decipher the results.

Leave a Reply

Your email address will not be published. Required fields are marked *